Downloaded from orbit.dtu.dk on: abr.

29, 2022

Remote Sensing for Wind Energy

Peña, Alfredo; Hasager, Charlotte Bay; Lange, Julia; Anger, Jan; Badger, Merete; Bingöl, Ferhat;
Bischoff, Oliver; Cariou, Jean-Pierre ; Dunne, Fiona ; Emeis, Stefan
Total number of authors:
27

Publication date:
2013

Document Version
Publisher's PDF, also known as Version of record

Link back to DTU Orbit

Citation (APA):
Peña, A., Hasager, C. B., Lange, J., Anger, J., Badger, M., Bingöl, F., Bischoff, O., Cariou, J-P., Dunne, F.,
Emeis, S., Harris, M., Hofsäss, M., Karagali, I., Laks, J., Larsen, S. E., Mann, J., Mikkelsen, T., Pao, L. Y., Pitter,
M., ... Würth, I. (2013). Remote Sensing for Wind Energy. DTU Wind Energy. DTU Wind Energy E No. 0029(EN)

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright
owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

 Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
 You may not further distribute the material or use it for any profit-making activity or commercial gain
 You may freely distribute the URL identifying the publication in the public portal

If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.
Remote Sensing for Wind Energy
DTU Wind Energy
E-Report

Alfredo Peña, Charlotte B. Hasager, Julia Lange,

Jan Anger, Merete Badger, Ferhat Bingöl,
Oliver Bischoff, Jean-Pierre Cariou, Fiona Dunne,
Stefan Emeis, Michael Harris, Martin Hofsäss,
Ioanna Karagali, Jason Laks, Søren Larsen, Jakob Mann,
Torben Mikkelsen, Lucy Y. Pao, Mark Pitter,
Andreas Rettenmeier, Ameya Sathe, Fabio Scanzani,
David Schlipf, Eric Simley, Chris Slinger,
Rozenn Wagner and Ines Würth

DTU Wind Energy-E-Report-0029(EN)

June 2013
 DTU Wind Energy-E-Report-0029(EN)

Remote Sensing for Wind Energy

Alfredo Peña, Charlotte B. Hasager, Julia Lange, Jan Anger,
Merete Badger, Ferhat Bingöl, Oliver Bischoff, Jean-Pierre
Cariou, Fiona Dunne, Stefan Emeis, Michael Harris, Martin
Hofsäss, Ioanna Karagali, Jason Laks, Søren Larsen, Jakob
Mann, Torben Mikkelsen, Lucy Y. Pao, Mark Pitter, An-
dreas Rettenmeier, Ameya Sathe, Fabio Scanzani, David
Schlipf, Eric Simley, Chris Slinger, Rozenn Wagner and Ines
Würth

DTU Wind Energy, Risø Campus,

Technical University of Denmark, Roskilde, Denmark
June 2013
Author: Alfredo Peña, Charlotte B. Hasager, Julia Lange, Jan Anger, DTU Wind Energy-
Merete Badger, Ferhat Bingöl, Oliver Bischoff, Jean-Pierre Cariou, E-Report-0029(EN)
Fiona Dunne, Stefan Emeis, Michael Harris, Martin Hofsäss, Ioanna June 6, 2013
Karagali, Jason Laks, Søren Larsen, Jakob Mann, Torben Mikkelsen,
Lucy Y. Pao, Mark Pitter, Andreas Rettenmeier, Ameya Sathe, Fabio
Scanzani, David Schlipf, Eric Simley, Chris Slinger, Rozenn Wagner
and Ines Würth
Title: Remote Sensing for Wind Energy
ISSN:
Department: DTU Wind Energy
ISBN:
978-87-92896-41-4

Abstract (max. 2000 char)

Contract no:

The Remote Sensing in Wind Energy report provides a descrip-

tion of several topics and it is our hope that students and others
interested will learn from it. The idea behind it began in year 2008
at DTU Wind Energy (formerly Risø) during the first PhD Summer Project no:
School: Remote Sensing in Wind Energy. Thus it is closely linked to
the PhD Summer Schools where state-of-the-art is presented during
the lecture sessions. The advantage of the report is to supplement
with in-depth, article style information. Thus we strive to provide Sponsorship:
link from the lectures, field demonstrations, and hands-on exercises
to theory. The report will allow alumni to trace back details after
the course and benefit from the collection of information. This is
the third edition of the report (first externally available), after very Cover: WAsP Google
successful and demanded first two, and we warmly acknowledge all Earth visualization
the contributing authors for their work in the writing of the chapters,
and we also acknowledge all our colleagues in the Meteorology and
Test and Measurements Sections from DTU Wind Energy in the PhD
Summer Schools. We hope to continue adding more topics in future
editions and to update and improve as necessary, to provide a truly
state-of-the-art ‘guideline’ available for people involved in Remote
Sensing in Wind Energy.
Pages: 308
Tables: 24
Figures: 204
References: 507

Technical University
of Denmark
Frederiksborgvej 399
4000 Roskilde
Denmark
Tel. +4546775024
[email protected]
www.vindenergi.dk
Contents
Page

1 Remote sensing of wind 11

1.1 Ground-based remote sensing for today’s wind energy research . . . . . . . . 11
1.1.1 Wind remote sensing (RS) methodologies . . . . . . . . . . . . . . . 11
1.2 Part I: Remote sensing of wind by sound (sodars) . . . . . . . . . . . . . . . 11
1.2.1 RS applications within Wind Energy . . . . . . . . . . . . . . . . . . 13
1.2.2 Recent developments . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.2.3 Summary of sodars . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3 Part II: RS of wind by light (lidars) . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.1 Introduction to lidars . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.2 Wind RS methodologies . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.3 Wind lidars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.3.4 Wind profiling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.3.5 Lidar accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.3.6 Wind lidar applications for wind energy . . . . . . . . . . . . . . . . 22
1.3.7 Summary of lidar . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2 The atmospheric boundary layer 25

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2 ABL Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 The ideal ABL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4 Surface Characteristics of real ABLs . . . . . . . . . . . . . . . . . . . . . . 37
2.5 Homogeneous Land ABL . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.6 Homogeneous Marine ABL . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.7 Inhomogenous and instationary ABL . . . . . . . . . . . . . . . . . . . . . . 42
2.8 Complex terrain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.9 Boundary Layer Climatology for Wind energy . . . . . . . . . . . . . . . . . 47
2.10 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

3 Atmospheric turbulence 52
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.2 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3 Rapid distortion theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3.1 RDT and surface layer turbulence . . . . . . . . . . . . . . . . . . . 56
3.3.2 Eddy lifetimes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.3.3 The uniform shear model . . . . . . . . . . . . . . . . . . . . . . . . 58
3.4 Fitting spectra to observations . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.4.1 Uncertainties on spectra . . . . . . . . . . . . . . . . . . . . . . . . 58
3.4.2 Spectral fitting and prediction of coherences . . . . . . . . . . . . . . 59
3.5 Model spectra over the ocean and flat land . . . . . . . . . . . . . . . . . . 61
3.5.1 Code and textbook spectra . . . . . . . . . . . . . . . . . . . . . . . 62
3.6 Comparison with the spectral tensor model . . . . . . . . . . . . . . . . . . . 65
3.7 Wind field simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

4 DTU Wind Energy-E-Report-0029(EN)

4 Introduction to continuous-wave
Doppler lidar 72
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.2 Basic principles of lidar operation and system description . . . . . . . . . . . 73
4.2.1 Brief survey of lidar types . . . . . . . . . . . . . . . . . . . . . . . . 73
4.2.2 Principles underlying anemometry by coherent laser radar (CLR) . . . 73
4.2.3 Role of local oscillator and range selection by focus . . . . . . . . . . 73
4.2.4 Doppler frequency analysis and signal processing . . . . . . . . . . . 74
4.2.5 Wind profiling in conical scan mode . . . . . . . . . . . . . . . . . . 75
4.2.6 Pioneering a revolution: ZephIR lidar . . . . . . . . . . . . . . . . . . 75
4.3 Lidar measurement process: Assumptions . . . . . . . . . . . . . . . . . . . . 76
4.3.1 Behaviour of scattering particles . . . . . . . . . . . . . . . . . . . . 77
4.3.2 Uniformity of flow and backscatter . . . . . . . . . . . . . . . . . . . 77
4.3.3 Beam positional accuracy . . . . . . . . . . . . . . . . . . . . . . . . 77
4.3.4 Optical and electrical interference sources . . . . . . . . . . . . . . . 77
4.3.5 Time-of-flight considerations . . . . . . . . . . . . . . . . . . . . . . 78
4.4 End-to-end measurement process for CW Doppler lidar . . . . . . . . . . . . 78
4.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.4.2 Transmitter optics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.4.3 Light scattering by aerosols . . . . . . . . . . . . . . . . . . . . . . . 79
4.4.4 Receiver optics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.4.5 Light beating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.4.6 Photodetection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.4.7 Fourier analysis and lidar sensitivity . . . . . . . . . . . . . . . . . . 81
4.4.8 Velocity estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.5 Ground based, vertical scan configuration wind field parameter determination 84
4.5.1 Least-squares fitting routine . . . . . . . . . . . . . . . . . . . . . . 84
4.5.2 Parameter extraction . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.5.3 Data averaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.6 Uncertainty analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.6.1 Rain/snow/cloud, solid objects . . . . . . . . . . . . . . . . . . . . . 85
4.6.2 Cloud effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.6.3 System positioning accuracy . . . . . . . . . . . . . . . . . . . . . . 88
4.6.4 Probe volume effects and operation at greater heights . . . . . . . . 89
4.6.5 Flow uniformity and complex terrain . . . . . . . . . . . . . . . . . . 89
4.6.6 Dependence on backscatter level . . . . . . . . . . . . . . . . . . . . 90
4.6.7 Beam obscuration and attenuation . . . . . . . . . . . . . . . . . . . 90
4.6.8 Wind direction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.7 Calibration, validation and traceability . . . . . . . . . . . . . . . . . . . . . 91
4.8 Turbine mounted continuous wave lidar . . . . . . . . . . . . . . . . . . . . 92
4.8.1 Least-squares fitting routine for horizontal scanning (turbine mounted)
operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.8.2 Turbine mounted CW lidar for wind turbine power curve measurement 94
4.9 Summary, state of the art, and future developments . . . . . . . . . . . . . . 96
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

5 Pulsed lidars 104

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.2 End-to-end description of pulsed lidar measurement process . . . . . . . . . . 105
5.2.1 Architecture of pulsed lidars . . . . . . . . . . . . . . . . . . . . . . 105
5.2.2 Differences between pulsed vs. continuous wave lidars . . . . . . . . . 107
5.2.3 Signal processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

DTU Wind Energy-E-Report-0029(EN) 5

 5.2.4 Coherent detection . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.2.5 Lidar equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.2.6 Spectral processing MLE . . . . . . . . . . . . . . . . . . . . . . . . 110
5.2.7 Wind vector reconstruction . . . . . . . . . . . . . . . . . . . . . . . 111
5.2.8 Fiber lidars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.3 Lidar performances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.3.1 Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.3.2 Best Focus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.3.3 Distance range and resolution . . . . . . . . . . . . . . . . . . . . . . 114
5.3.4 Velocity range and resolution . . . . . . . . . . . . . . . . . . . . . . 115
5.3.5 Time-bandwidth tradeoffs . . . . . . . . . . . . . . . . . . . . . . . . 116
5.3.6 Existing systems and actual performances . . . . . . . . . . . . . . . 117
5.3.7 Validation of measurements . . . . . . . . . . . . . . . . . . . . . . . 117
5.4 Conclusions and perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

6 Remote sensing for the derivation of the mixing-layer height and detection of
low-level jets 122
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
6.2 Mixing-layer height . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
6.2.1 Acoustic detection methods (Sodar) . . . . . . . . . . . . . . . . . . 123
6.2.2 Optical detection methods . . . . . . . . . . . . . . . . . . . . . . . 127
6.2.3 RASS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
6.2.4 Further techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
6.2.5 Comparison of acoustic and optical MLH detection algorithms . . . . 134
6.3 Boundary-layer height . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.4 Low-level jets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
6.4.1 Formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
6.4.2 Frequency and properties of low-level jets . . . . . . . . . . . . . . . 136
6.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

7 What can remote sensing contribute to power curve measurements? 143

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
7.2 Power performance and wind shear . . . . . . . . . . . . . . . . . . . . . . . 143
7.2.1 Shear and aerodynamics . . . . . . . . . . . . . . . . . . . . . . . . 143
7.2.2 Consequences on the power production . . . . . . . . . . . . . . . . 146
7.3 Wind speed profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
7.4 Equivalent wind speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
7.4.1 Standard power curve for the two groups of wind profiles . . . . . . . 150
7.4.2 A better approximation of the kinetic energy flux . . . . . . . . . . . 151
7.4.3 Equivalent wind speed definition . . . . . . . . . . . . . . . . . . . . 152
7.4.4 Choice of instrument . . . . . . . . . . . . . . . . . . . . . . . . . . 152
7.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

6 DTU Wind Energy-E-Report-0029(EN)

8 Nacelle-based lidar systems 157
8.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
8.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
8.3 The units of the lidar scanner . . . . . . . . . . . . . . . . . . . . . . . . . . 157
8.3.1 Windcube . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
8.3.2 Scanner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
8.4 Scan pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
8.5 CNR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
8.6 Wind field reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
8.7 “Visited” test sites of the SWE Scanner . . . . . . . . . . . . . . . . . . . . 163
8.7.1 Onshore test site Bremerhaven, Germany . . . . . . . . . . . . . . . 163
8.7.2 Onshore test site Risø Campus - DTU Wind Energy, Denmark . . . . 164
8.7.3 Onshore test site National Wind Technology Center (NWTC) - NREL,
USA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
8.7.4 Offshore test site “alpha ventus” . . . . . . . . . . . . . . . . . . . . 165
8.8 Measurement campaigns and some results . . . . . . . . . . . . . . . . . . . 166
8.8.1 Equivalent wind speed . . . . . . . . . . . . . . . . . . . . . . . . . . 166
8.8.2 Rotor effective wind speed . . . . . . . . . . . . . . . . . . . . . . . 167
8.9 Outlook & Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

9 Lidars and wind turbine control –

Part 1 171
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
9.2 Model Based Wind Field Reconstruction . . . . . . . . . . . . . . . . . . . . 171
9.2.1 Ambiguity in Wind Field Reconstruction . . . . . . . . . . . . . . . . 172
9.2.2 Lidar Model for Reconstruction . . . . . . . . . . . . . . . . . . . . . 172
9.2.3 Wind Model for Collective Pitch Control . . . . . . . . . . . . . . . . 173
9.2.4 Wind Model for Cyclic Pitch Control . . . . . . . . . . . . . . . . . . 173
9.2.5 Wind Model for Yaw Control . . . . . . . . . . . . . . . . . . . . . . 173
9.2.6 Wind Model for Complex Terrain . . . . . . . . . . . . . . . . . . . . 173
9.3 Modeling of the Wind Turbine . . . . . . . . . . . . . . . . . . . . . . . . . 174
9.3.1 Reduced Nonlinear Model . . . . . . . . . . . . . . . . . . . . . . . . 174
9.3.2 Estimation of the Rotor Effective Wind Speed from Turbine Data . . 175
9.3.3 Linear Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
9.4 Correlation of a Lidar System and a Wind Turbine . . . . . . . . . . . . . . . 176
9.4.1 Simulated Lidar Measurements . . . . . . . . . . . . . . . . . . . . . 176
9.4.2 Reconstruction of Rotor Effective Wind Speed . . . . . . . . . . . . . 176
9.4.3 Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.5 Lidar Assisted Collective Pitch Control . . . . . . . . . . . . . . . . . . . . . 178
9.5.1 Controller and Adaptive Filter Design . . . . . . . . . . . . . . . . . 178
9.5.2 Field Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
9.6 Lidar Assisted Speed Control . . . . . . . . . . . . . . . . . . . . . . . . . . 180
9.6.1 Controller Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
9.6.2 Simulation Using Real Data . . . . . . . . . . . . . . . . . . . . . . . 181
9.6.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
9.7 Nonlinear Model Predictive Control . . . . . . . . . . . . . . . . . . . . . . . 182
9.7.1 Controller Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
9.7.2 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
9.8 Lidar Assisted Cyclic Pitch Control . . . . . . . . . . . . . . . . . . . . . . . 184
9.9 Lidar Assisted Yaw Control . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
9.9.1 Simulation Using Generic Wind . . . . . . . . . . . . . . . . . . . . . 186
9.9.2 Simulation Using Real Data . . . . . . . . . . . . . . . . . . . . . . . 187

DTU Wind Energy-E-Report-0029(EN) 7

 9.9.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
9.10 Conclusion and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

10 Lidars and wind turbine control – Part 2 192

10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
10.1.1 Lidar Measurement Strategies . . . . . . . . . . . . . . . . . . . . . 194
10.2 Wind Turbine Feedforward Control . . . . . . . . . . . . . . . . . . . . . . . 195
10.2.1 Feedforward Control Assuming Perfect Measurements . . . . . . . . . 195
10.2.2 Feedforward Control with Imperfect Measurements . . . . . . . . . . 197
10.2.3 The Relationship between Variance and Damage Equivalent Loads . . 198
10.3 Preview Time in Feedforward Control . . . . . . . . . . . . . . . . . . . . . . 198
10.3.1 Ideal Feedforward Controller Preview Time . . . . . . . . . . . . . . . 198
10.3.2 Prefilter Preview Time . . . . . . . . . . . . . . . . . . . . . . . . . 199
10.3.3 Pitch Actuation Preview Time . . . . . . . . . . . . . . . . . . . . . 199
10.4 Blade Effective Wind Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
10.4.1 The Relative Importance of the u, v, and w Components . . . . . . . 201
10.5 Lidar Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
10.5.1 Range Weighting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
10.5.2 Geometry Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
10.5.3 Wind Evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
10.5.4 Lidar Measurements of Blade Effective Wind Speed . . . . . . . . . . 206
10.6 Lidar Measurement Example: Hub Height and Shear Components . . . . . . 207
10.6.1 Optimizing the Measurement Scenario . . . . . . . . . . . . . . . . . 209
10.7 Control Example 1: Wind Turbine Preview Control In The Presence of Evolving
Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
10.7.1 H2 Optimal Preview Control . . . . . . . . . . . . . . . . . . . . . . 212
10.7.2 Controller Performance Simulations . . . . . . . . . . . . . . . . . . 213
10.7.3 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
10.8 Control Example 2: H2 Optimal Control with Model of Measurement Coherence215
10.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217

11 Lidars and wind profiles 221

11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
11.2 Wind profile theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
11.2.1 Surface layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
11.2.2 Marine surface layer . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
11.2.3 Boundary layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
11.3 Comparison with observations at great heights . . . . . . . . . . . . . . . . . 224
11.3.1 Marine observations . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
11.3.2 Neutral observations over flat land . . . . . . . . . . . . . . . . . . . 225
11.3.3 Diabatic observations over flat land . . . . . . . . . . . . . . . . . . 226
11.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229

8 DTU Wind Energy-E-Report-0029(EN)

12 Complex terrain and lidars 231
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
12.2 Lidars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
12.2.1 ZephIR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
12.2.2 Windcube . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
12.3 Challenges and Known Issues . . . . . . . . . . . . . . . . . . . . . . . . . . 234
12.3.1 The conical scanning error in complex terrain . . . . . . . . . . . . . 234
12.3.2 Predicting the error by means of a flow model . . . . . . . . . . . . . 235
12.4 Experimental studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
12.4.1 Hilly site; Lavrio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
12.4.2 Mountainous site; Panahaiko . . . . . . . . . . . . . . . . . . . . . . 238
12.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241

13 Turbulence measurements by wind lidars 242

13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
13.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
13.2.1 Systematic turbulence errors for the ZephIR lidar . . . . . . . . . . . 245
13.2.2 Systematic turbulence errors for the WindCube lidar . . . . . . . . . 247
13.2.3 Definition of the systematic error . . . . . . . . . . . . . . . . . . . . 248
13.3 Comparison of models with the measurements . . . . . . . . . . . . . . . . . 249
13.3.1 Comparison with the ZephIR measurements . . . . . . . . . . . . . . 249
13.3.2 Comparison with the WindCube measurements . . . . . . . . . . . . 252
13.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
13.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
13.6 Future Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258

14 Ground based passive microwave radiometry and temperature profiles 260

14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
14.2 Microwave radiometry fundamentals . . . . . . . . . . . . . . . . . . . . . . 260
14.3 Upward-looking radiometric temperature profile measurements . . . . . . . . 261
14.4 Upward-looking angular scanning microwave radiometry . . . . . . . . . . . . 265
14.5 An angular scanning temperature profile radiometer . . . . . . . . . . . . . . 269
14.6 Antarctica Dome C experimental site Radiometric measurements . . . . . . . 270
14.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275

15 SAR for wind energy 276

15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
15.2 SAR technical description . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
15.3 Wind retrieval from SAR . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
15.4 Beyond C-band VV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
15.5 Current practices in SAR wind retrieval . . . . . . . . . . . . . . . . . . . . . 280
15.6 SAR wind retrieval at DTU Wind Energy . . . . . . . . . . . . . . . . . . . . 282
15.7 Mesoscale wind phenomena from SAR . . . . . . . . . . . . . . . . . . . . . 282
15.8 SAR wind fields near offshore wind farms . . . . . . . . . . . . . . . . . . . . 284
15.9 Wind resources from SAR . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
15.10S-WAsP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
15.11The wind class method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
15.12SAR wind resource maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289

DTU Wind Energy-E-Report-0029(EN) 9

 15.13Lifting satellite winds to hub-height . . . . . . . . . . . . . . . . . . . . . . 291
15.14Future advances in ocean wind mapping from SAR . . . . . . . . . . . . . . 291
15.15Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292

16 Scatterometry for wind energy 296

16.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
16.2 Principle of Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
16.3 Equivalent Neutral Wind . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
16.4 Sources of error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
16.5 QuikSCAT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
16.6 Applications of QuikSCAT Surface Winds . . . . . . . . . . . . . . . . . . . 303
16.7 Spatial Resolution of Scatterometer Winds . . . . . . . . . . . . . . . . . . . 304
16.8 Contemporary Scatterometers . . . . . . . . . . . . . . . . . . . . . . . . . . 304
16.9 Acknowledgements and Suggested Reading . . . . . . . . . . . . . . . . . . 306
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307

10 DTU Wind Energy-E-Report-0029(EN)

1 Remote sensing of wind
Torben Mikkelsen
DTU Wind Energy, Risø Campus, Roskilde, Denmark

1.1 Ground-based remote sensing for today’s wind energy

research
Wind turbines are being installed at an ever increasing rate today, on and offshore, in hilly and
forested areas and in complex mountainous terrain. At the same time, as the wind turbines
become bigger and bigger, they reach higher and higher into the atmosphere but also into
hitherto unknown wind and turbulence regimes.
The traditional method for accredited measurements for wind energy is to mount calibrated
cup anemometers on tall met masts. But as turbines grow in height, high meteorology masts
and instrumentation becomes more and more cumbersome and expensive correspondingly.
Costs for installation of tall instrumented met towers increase approximately with mast height
to the third power and licensing permits can be time consuming to obtain.
With hub heights above 100 m and rotor planes nowadays reaching diameters of 120 m
or more on today’s 5 MW turbines, the wind speed distribution over the rotor planes will no
longer be representatively measured from a single hub height measurement point, but will also
require a multi-height measurement strategy with measurements ranging in heights between
50–200 m, for the purpose of capturing the simultaneous wind distribution over the entire
wind turbine rotor.

1.1.1 Wind remote sensing (RS) methodologies

A simple way to remotely determine the wind speed is by observing marked cloud drift aloft
from the ground on a sunny day. More quantitative and accurate remote sensing measurement
techniques for wind energy applications include nowadays sound and light wave propagation
and backscatter detection based instruments such as sodar, lidar and satellite-based sea sur-
face wave scatterometry.
Today’s quest within RS research for wind energy is to find useful replacement alternatives
for expensive and cumbersome meteorology mast erection and installations. However, accu-
racy is of particular importance for site and resource assessments irrespectively of terrain,
on or offshore, and measurement errors much in excess of 1% cannot be tolerated neither
by banks nor by project developers, as 1% uncertainty in mean wind speed results in 3%
uncertainty in mean wind power.

1.2 Part I: Remote sensing of wind by sound (sodars)

Sodar (sound detection and ranging) is based on probing the atmosphere by sound propaga-
tion, lidar (light detection and ranging) is based on probing the atmosphere by electromagnetic
radiation (microwaves or laser light) and satellite RS is based on microwave scatterometry
on the sea surface and synthetic aperture radar (SAR) methods. The first two (sodar and
lidar) are direct measurements of wind speed based on Doppler shift, whereas the satellite
scatterometry are based on proxy-empirical calibration methods. First, a description of the
background and the state-of-the-art sodar is addressed. Second, the corresponding develop-
ment and application lidar RS technology is addressed.
Wind turbines operate within the so-called atmospheric boundary layer, which is charac-
terized by relatively high turbulence levels. Turbulence is here created from the strong wind

DTU Wind Energy-E-Report-0029(EN) 11

Figure 1: Commercial available sodars being inter-compared during the WISE 2004 experi-
ment: An array of sodars (and one lidar) during inter-comparison and testing against the tall
met towers (up to 168 m above ground) equipped with calibrated cup anemometers at several
heights. Venue: The Test station for large wind turbines, Høvsøre, Denmark

shear due to the proximity of the Earth’s surface. The wind speed at the ground is always
zero, both on and offshore.
Sodar is a RS methodology for measurements of the wind speed and direction aloft at
various heights in the atmosphere. Sodars are ground-based instruments that transmit a
sequence of short bursts of sound waves at audible frequencies (2000–4000 Hz) upward in
three different inclined directions into the atmosphere.
The sodar measurement technology was well established and in operational use for decades
by now, starting in the 1980’s where they served environmental protection issues and has
been extensively applied to atmospheric research for environmental protection air pollution
prediction measures well before the present burst in wind energy research and application.
In Germany for example, sodars have been commissioned on several nuclear installations to
replaced tall meteorological towers and serve now as operational monitoring devises of the
local wind speed, direction and atmospheric stability.
As the sound waves from a sodar propagate forward a small fraction of the transmitted
sound energy is scattered and reflected in all directions from temperature differences and
turbulence in the atmosphere. A very small fraction of this scattered energy reaches back into
the sodar’s detector, which in principle is a directional-sensitive microphone.
The height at which the wind speed is measured is usually determined by the time delay
in the backscatter from the transmitted pulse. Under standard atmospheric conditions with
sound propagation speed of about 340 m s−1 backscatter from a sodar measurement at 170
m height above the ground will reach back into the detector after 1 s delay time.
The wind speed component in the transmitted beam direction is subsequently determined
from the Doppler shift observed as frequency difference between the transmitted frequency
and the frequency of the received backscattered sound wave. By combining the measured
wind speed components obtained in this way from three differently inclined sound path direc-
tions, e.g. from one vertical and two inclined sound paths, the three-dimensional wind vector
including wind speed and direction and tilt can be measured by sodar from preset heights from
the ground and up to the limit determined by the sodar’s lowest acceptable Carrier-to-Noise
(C/N) ratios.
The above description is for a mono-static system, where transmitter and receivers are
co-located on the ground. But alternative configurations, e.g. in the form of so-called bi-

12 DTU Wind Energy-E-Report-0029(EN)

Figure 2: Calibration, laboratory work, and real-time Doppler spectrum obtained at Risø DTU
with the experimental bi-static CW sodar “Heimdall” (Mikkelsen et al., 2007). Upper panel:
Combined acoustic horn and parabola antenna for high-yield (+30 dB gain) backscatter re-
ceiving of sound waves. Middle panel: Two researchers at the Risø DTU Laboratory while
testing of the bi-static sodar. Lower panel: A real-time obtainable continuous Doppler spec-
trum Heimdall bi-static sodar from wind measurements at 60 m above Risø DTU

static sodar configurations exist as well, where the transmitter and receivers are separated
e.g. 100–200 meters on the ground.
Bi-static configurations have significant C/N-ratio advantages over mono-static configu-
rations for wind energy applications. Received backscatter in a bi-static configuration is not
limited to direct (180◦ ) backscatter from temperature (density) fluctuations only, but enables
also backscatter contributions from the atmospheric turbulence. And the higher the wind
speed the more turbulence.
As a consequence significant improvements of the C/N- ratios can be obtained from a so-
called “bi-static configuration”, in which the transmitter and the receiver are separated from
one another on the ground. This becomes in particular relevant during strong wind situations,
where the background noise level increases with the wind speed.
A particular configuration considered for wind energy applications is therefore the bi-static
continuous wave (CW) sodar configuration. Alternatively to the range gating in a pulsed
system, the range to the wind speed measurement in a CW system can be determined by
well-defined overlapping transmission and receiving antenna functions. At Risø DTU we have
build and investigated such a sodar system for wind energy applications.

1.2.1 RS applications within Wind Energy

Remote sensed wind speed measurements are needed to supplement and extend tall met mast
measurements, on and offshore, and within research to evaluate various wind flow models and
wind atlases for a number of purposes, including:
1. Wind resource assessments
2. Wind park development projects
3. Power curve measurements
4. Bankability
5. Wind model and wind resource (wind atlas) uncertainty evaluation
The common denominator in most of these issues is high accuracy, and with a demand
for reproducible certainty to more than 99% of what can be achieved with a corresponding

DTU Wind Energy-E-Report-0029(EN) 13

calibrated cup anemometer. A significant source for uncertainty with RS instrumentation
relative to a cup anemometer, and for sodars in particular, is the remote instrument’s relative
big measurement volumes. A sodar measuring the wind speed from say 100 m height probes a
total sampling volume of more than 1000 m3 , whereas a cup anemometer is essentially a point
measurement device in this connection. In addition the sodar’s measured wind components are
displaced in space and time, which makes the interpretation of measured turbulence by a sodar
impaired. In addition the huge sampling volumes will be putting restrictions on measurements
in non-uniform flow regimes such as found near forest edges, on offshore platforms, and over
hilly or complex terrain.
Sodar’s RS is also in demand for direct turbine control integration, wind power optimization
and turbine mounted gust warning systems, but here the demand on accuracy and reliability
is correspondingly high. Today, sodars are typically used to measure 10-min averaged vertical
profiles in the height interval between, say 20 and 200 m above the ground, of the following
quantities:
• Mean wind speed
• Mean wind directions (including azimuth and tilt)
• Turbulence (all three wind components: longitudinal, transverse and vertical)
Albeit significant inherent scatter persists in sodar measured mean wind speed and direction
data average mean wind speed compare relatively well (in most cases to within ±3%) to that
of a corresponding cup anemometer measured wind speed, cf. the slopes of the scatter plots
in Figure 3.
However, the correlation coefficients between sodar and cup anemometer data is, depending
on measurement height and atmospheric stability, relative poor as compared to a cup-to cup
anemometer correlation, where the two cups are separated by ∼ 100 m (typically less than
0.95) and reflects, among other issues, that a mono-static sodar measures the wind speed over
a huge volume whereas the cup anemometer represents a point measurement. In addition,
increased scatter will occur as a result of beam-bending due to the relative big wind speed to
propagation speed of the sound pulses. Also notable is that sodars are able to make only a
single 3D vector speed measurement about once per 6–10 s. A slow sampling rate also makes
the mean prediction of a 10-min averaged quantity uncertain, due to limited independent
sampling counts. In his note “Statistical analysis of poor sample statistics”, Kristiansen (2010)
has shown that “counting” uncertainty in terms of relative “standard deviation of the sample
variance” in a small sample can give rise to a ∼ 10% relative uncertainty when averaged
quantities are drawn from a set of only 100 independent samples.
It is also seen from the sodar vs. cup anemometer data in Figure 3 that difficulties with the
C/N ratio can occur when wind speeds exceed approximately 15 m s−1 , which by the way is
a nominal wind speed for a wind turbine. This is due to high background noise and the loss
of backscatter in neutrally stratified high wind speed regions.
Recently relative good agreements over forested areas have nevertheless been seen (< 1%
discrepancy) between sodar and cup anemometer mean turbulence intensity has been reported
by Gustafsson (2008). However, turbulence intensity, which is the stream wise turbulence
component relative to the mean wind speed, is in a 10-min averaged quantity dominated,
particularly in forested areas, by the most energy containing eddies, which in this case will be
larger than the sodar’s sampling volumes and therefore be well represented in the statistics.
However, the smaller scales including turbulent eddies with wind gusts must be anticipated
to be present also on the scales smaller than a mono-static ground based sodar will be able
to capture.
While sodars appears to be able to measure accurately both the mean winds speed and
the turbulence intensities at a turbines hub height it was found more difficult to use a sodar
for accurate measurements over the entire rotor plane due to low C/N ratio (Wagner et al.,
2008). There are several sodar manufactures on the wind RS market today including for exam-
ple Remtech, Atmospheric Systems Corporation (formerly Aerovironment), Metek, Scintec,

14 DTU Wind Energy-E-Report-0029(EN)

Figure 3: Example of scatter plots from sodar vs. cup anemometer data. The upper graph
presents unscreened sodar wind speed data plotted against corresponding high-quality cup
anemometer data measured at the Risø DTU met tower at 125 m. A data availability cor-
responding to 76% (9549 10-min averaged wind speeds) was obtained during this particular
sodar vs cup anemometer inter-comparison test of almost three month duration (12532 10-
min periods). The middle data graph shows the same data set after screening of the sodar
data for high C/N-ratios. The scatter is significantly reduced, but so is also the data avail-
ability which with only 4210 data points has been reduced to almost 34%. The bottom panel
shows (left) simultaneous measured sodar vs cup scatter plo tat 75 m height (0.989) and
(right) lidar vs the same cup for the same data period. The lidar measurements at 80 m are
seen to exhibit less scatter and high correlation coefficient (0.996)

DTU Wind Energy-E-Report-0029(EN) 15

Second Wind Inc. and Swedish AQ System to mention the most dominant. All but one base
their sodar technology on mono-static phased array antenna configurations except AQ System
sodars which are build on three solid dish parabolas offering a somewhat bigger antenna di-
rectivity (12◦ opening angle). However, only a couple of today’s sodar manufacturers address
directly the high accuracy demanding wind energy market.
The EU WISE project addressed and evaluated commercial sodars for wind energy (deNoord
et al., 2005) and concluded then that neither of the commercial sodars were particularly close
to be able to substituting standard measuring masts. In conclusion the WISE project stated
that general purpose commercial sodars were unreliable, especially in case of bad weather or
high background noise

1.2.2 Recent developments

A few improvements seem to have emerged since 2005. Particularly for the few sodars that
addresses the wind energy marked. Replacement of the phased arrays by parabola dish seems
to have contributed to the sodars overall C/N performance. Also better and improved signal
processing is apparently applied today. However, it is my personal belief that we won’t see any
significant quantum leap in sodar performance until sodars for wind energy applications are
build on bi-static configurations. Research and development along these lines are in progress,
and researchers and test engineers at Risø DTU are looking forward to see and to test possible
future bi-static configured sodars especially designed to meet the high accuracy demands set
within wind energy RS.

Table 1: Pros & Cons of sodars

Pros Cons
Portable Low duty cycle (1 pulse transmitted every 3 s,
and up to 6–10 s lapse times before all
three wind components have been sampled)
Build on well developed and well-proven Limited by low S/N- ratio at:
audio-frequency “low tech” technology 1) high wind speed conditions
2) during neutrally stratified conditions
Sodars are relatively cheap (priced down Prone to solid reflections from the
to some 25% of a corresponding wind lidar) surroundings (including wind turbines)
Low power consumption (one solar powered Prone to high background environmental noise
version uses less than 10 W)
Sound backscatter: Relatively high yield Low wavelength/aperture ratio (1:10)
(backscattered power at the detector of the results in undefined broad antenna beams
order of 10−10 W) Prone to beam bending with wind speed of
the order of 5% or higher of the speed of sound
Huge measurement and sample volumes
Signal processing limited by pulsed sodars
relative long data acquisition times
(sampling time per pulse of the order of 1 s)

Table 2: Accuracy with sodars during neutral conditions

Slope mean wind speed vs. calibrated cup anemometers ±3%

Correlation coefficients [at 125 m, neutral stratification] 0.9–0.95
Mean turbulence intensity[at 80 m] < 1% error

16 DTU Wind Energy-E-Report-0029(EN)

1.2.3 Summary of sodars
Most of today commercially available sodars are still build on “pre wind energy era” antenna
design and processing technology, which do not in particular address nor support the high
accuracy demands required within wind energy and resource assessment studies of today. The
consequence is that most – if not all – of the available sodars today still exhibit insufficient
accuracy to be accepted by the wind energy industry and society as an accurate RS tool for
precise and “bankable” wind energy investigations.
Although some improvements seem to have occurred in accuracy since our first 2005 WISE
sodar investigation, it is still not this author’s belief that sodars as they come will be able
to meet the high accuracy demands of the wind energy society in the future unless a major
quantum jump can be demonstrated in their overall performance at high wind speed, neutral
atmospheric stratification, and at present wind turbine hub heights (> 100 m).
At Risø DTU we see two venues for further research along which improved accuracy of
sodars may happen: One is to switch to fully bi-static pulsed or CW based sodar configurations,
however cumbersome, and the other is to take advantage of the immense, fast and cheap
embeddable processing power set to our disposal from the information technology industry
today, and apply these for enhanced on-line real time signal processing.

1.3 Part II: RS of wind by light (lidars)

1.3.1 Introduction to lidars
The motivation and demand in the wind energy market for wind lidars are similar to those
of wind sodars. At a continuously increasing rate today wind turbines are being installed on,
offshore, in hilly and forested areas, and even in complex or mountainous terrain. At the same
time, as the turbines gets bigger and more powerful, they also reach higher and higher into
the atmospheric flow, and thereby also into hitherto unknown wind and turbulence regimes
– on as well as offshore.
The industry’s traditional method for performing accredited and traceable measurements of
power performance is to mount a single accurately calibrated cup anemometer at hub height
two to four rotor diameters upwind in front of the turbines on a tall meteorological mast. IEC
61400-12-1 describes the accepted standard for power performance verification (power curve
measurement) and prescribes measurements of power production correlated with wind speed
measurements from a cup anemometer located at hub height in front of the wind turbine 2–4
rotor diameters upstream.
With turbines becoming bigger correspondingly high meteorology masts equipped with
wind speed instrumentation becomes progressing more cumbersome and expensive to install,
especially in mountainous and complex terrain. As wind turbines rotor planes reaches 120 m
in diameter or more it is evident that the incoming wind field over the entire rotor planes is
not measured representatively from a single cup anemometer mounted at hub height.
Accurate measurements of the inflow of today’s huge wind turbines will require multi-point
multi-height wind measurements within the entire rotor plane, to characterize the wind speed
and wind shear over the entire rotor plane. Research activities addressing detailed rotor plane
inflow and wakes is ongoing at Risø DTU in connection with the establishment of new research
infrastructure based on wind lidars, see Windscanner.dk and Mikkelsen (2008).

1.3.2 Wind RS methodologies

RS measurement methodologies for wind energy applications are today commercially available
and encompass various measurement techniques that include sound based sodars, laser based
lidars and satellite borne scatterometry. The application range for wind measurements are
also plentiful, and encompass for example:
1. Wind turbine power performance verification – Establishment of new RS based measure-

DTU Wind Energy-E-Report-0029(EN) 17

Figure 4: Windscanners in operation – CW and pulsed wind lidars engaged in measurements
of the wind and turbulence fields around a spinning wind turbine (See Windscanner.dk for
more details)

ment standards for the replacement of in-situ reference met masts. Work within the IEC
is at the moment aiming at establishment of a new international IEC-standard for remote
sensed wind measurements, as e.g. obtained by lidars, for power curve measurements.
2. Wind energy resource measurements – The global wind resources are now being mapped
globally on shore, off shore, over hilly and in mountainous terrain, etc. Here also, high
accuracy is of uttermost importance for accurate site and resource assessments. Mea-
surement errors in excess of 1% are unacceptable by project developers and investment
banks.
3. Wind turbine control – RS lidar instruments that are directly integrated into the wind
turbines hub or spinner or even into the blades are also seen as a forthcoming RS mea-
surement technology that can help improve the wind turbines power performance and
possibly also diminish fatigue wear from extreme gusts and wind shear via active steering
the wind turbines individual blade pitch or, to come one day maybe, its trailing edge
flaps.
Researchers at Risø DTU have during decades now followed and contributed to the devel-
opment of improved instrumentation for RS of wind. Starting out already in the 60’s with
more general boundary-layer meteorological investigations of flow and diffusion our present
research and experimental developments within the meteorology and test and measurement
programs at Risø DTU has recently become more and more directed towards applications
within wind energy. Wind lidars and lidar-based wind profilers, their measurement princi-
ples, their measurement performances, and also their possible future integration within wind
turbines themselves are here addressed.

18 DTU Wind Energy-E-Report-0029(EN)

1.3.3 Wind lidars
Measuring wind with a wind lidar means to probe the atmospheric flow from the ground
by use of light beams. A wind lidar is wind measurement devise able to detect the Doppler
shifts in backscattered light. The Doppler shift is proportional to the wind speed in the beam
direction in the wind lidar’s adjustable measurement volumes.
Lidars, like sodars, provide a ground-based RS measurement methodology for measuring
the winds at various ranges, angles and heights aloft. Wind lidars work by transmitting elec-
tromagnetic radiation (light) from a laser with a well-defined wavelength in the near infrared
band around 1.5 µm. They detect a small frequency shift in the very weak backscattered light,
a Doppler shift that results from the backscattering of light from the many small aerosols
suspended and moving with the air aloft.
From a meteorological point of view wind turbine are “obstacles” within the lowest part of
the atmospheric boundary layer, that is, the part of the atmosphere best characterize by high
wind shear, strong wind veers, and with the highest levels of turbulence.
A wind profiler is a ground-based wind lidar transmitting a continuous beam or a sequence
of pulsed radiation in three or more different inclined directions. A wind profiler determines
the radial wind speeds in multiple directions above its position on the ground. It does so also
by determining the Doppler shifts in the detected backscattered radiation along each beam
direction. Wind lidars, like sodars, therefore have both transmitting and receiving antennas,
which most wind profilers today combine into a single optical telescope. The three-dimensional
wind vector as function of height by measuring the radial wind speeds in three or more beam
directions above the lidar. In practice, the transmitting and receiving radiation are combined
in a single telescope and the beam is then steered in different directions via a rotating wedges
or turning mirrors.
Wind lidars in the market for vertical mean and turbulence profile measurements are avail-
able based on two different measurement principles:
1. Continuous wave (CW) lidars
2. Pulsed lidars
Several wind lidars addressing the wind energy market are commercial available today. CW-
based wind lidars are manufactured by Natural Powers (ZephIR) and OPDI Technologies &
DTU Fotonik (WINDAR) while Coherent Technologies Inc. (Wind tracer), Leosphere (Wind-
Cube), CatchtheWindInc (Vindicator) and Sgurr Energy (Galion) manufacture pulsed lidars
for the time being.
The technology imbedded in today’s CW and pulsed wind lidar systems have been spurred
from the telecommunication 1.5 µm fiber and laser technology revolution in the 90’s. There
are however, some principally differences between CW and pulsed lidar’s temporal and spatial
resolution, properties that have influence on the different lidar types ability to measure and
resolve the mean wind and turbulence characteristics of the atmospheric boundary layer wind
field.
The CW lidar focuses a continuous transmitted laser beam at a preset measurement height
and there determines, also continuously, the Doppler shift in the detected backscatter also
from that particular height. When wind measurements from more than a single height are
required, the CW lidar adjusts its telescope to focus on the next measurement height. The
measurement ranges (measurement heights) as well as the spatial resolution of a CW lidar
measurement is controlled by the focal properties of the telescope. The shorter the measure-
ment distance, and the bigger the aperture (lens), the better defined is a CW lidar’s range
definition and its radial measurement confinement. A CW lidar resolves the wind profile along
its beam in a similar manner as a photographer controls the focal depth in a big sport or bird
telescope.
The focal depth of any telescope, however, increases proportional to the square of the
distance to the focus or measurement point. This optical property limits a CW lidar build
with e.g. standard 3”optics to measurement heights below, say 150 m.

DTU Wind Energy-E-Report-0029(EN) 19

Figure 5: Two CW wind lidars belonging to the Windscanner.dk research facility being inter-
compared and tested up against the tall meteorological masts at Høvsøre, Risø DTU.

A pulsed lidar on the other hand transmits a sequence of many short pulses, typical 30 m
in effective length, and then it detects the Doppler shift in the backscattered light from each
pulse as they propagate with the speed of light. While a CW lidar measures from one height at
a time a pulsed lidar measures wind speeds from several range-gated distances simultaneously,
typically at up to 10 range gates at a time.
The pulsed lidar’s spatial resolution, in contrast to the CW lidar, is independent of the
measurement range. The pulse width and the distance the pulse travels while the lidar samples
the detected backscatter control its resolution. The spatial resolution in the beam direction
obtainable with the 1.5 µm wavelength pulsed lidar in the market today are of the order of
30–40 m.
In addition, while a CW lidar’s upper measurement distance is limited progressing uncon-
fined measurement volume at long distances, a pulsed lidar’s maximum measurement range
is limited by deteriorating C/N-ratios in measurements from far distances (height).
Moreover, while a CW lidar equipped with a 1 W 1.5 µm eye-safe laser has been tested
able to sample and process up to 500 wind speed measurements per second, a corresponding
powered pulsed lidar can handle only 2–4 wind speed samples per second. Each of these
samples, however, then on the contrary contain wind speeds from up to 10 range gates
(ranges) measured simultaneously.

CW vs pulsed lidars Overall, CW lidar features high spatial resolution in the near range and
very fast data acquisition rates, features that are well suitable for turbulence measurements.
Today’s commercial available CW lidar profilers measure radial wind speeds at ranges up to
∼ 200 m and wind vectors at heights up to 150 m.
The pulsed lidar configuration on the other hand features lower but always constant spatial
resolution properties (30–40 m) at all ranges. They are also inherently slower in their data
acquisition rate, but then they measure wind speeds at multiple heights simultaneously, and
they hold also potential for reaching longer ranges (heights) than corresponding powered CW
lidars. At the test site in Høvsøre Risø DTU, commercial available pulsed wind lidar profiles
have regularly measured the wind vector profiles up to 300 m height.

20 DTU Wind Energy-E-Report-0029(EN)

 Figure 6: CW wind lidars (ZephIRs) under testing at Høvsøre, Risø DTU

Figure 7: Pulsed wind lidars (six WLS7 WindCubes) and one Galion (far back) during testing
at Høvsøre, Risø DTU

1.3.4 Wind profiling

A wind “profiler” measures 10-min averaged quantities of the vertical wind speed profile, the
vertical direction profile, and the vertical turbulence profiles, by combining a series of radial
measured wind speed components from several, and at least three, different beam directions,
into a three-dimensional wind vector. CW-based wind lidars, e.g. the ZephIR, measure the
vertical wind profile at five consecutive heights, selectable in the range from, say 10 to 150 m
height. Pulsed lidars, e.g. the WindCube or the Galion, measure correspondingly the vertical
wind profile simultaneously at several (of the order of 10) heights, in the height interval from
40–∼ 300 m, the upper bound depending on the amount of aerosols in the air.
True for all wind profilers in the wind energy market, however, CW and pulsed lidars
irrespectively, is that they rely during combining measured radial wind speeds into a single
wind vector on the assumption that the flow over the wind lidar is strictly homogeneous.
Homogeneous wind flow means that the air stream is unaffected and not influenced by hills,

DTU Wind Energy-E-Report-0029(EN) 21

valleys, other wind turbine wakes, or near-by buildings within their volume of air scanned
above the lidar.
For this reason, neither lidar nor sodar based wind profilers will be able measure correctly
over sites located in hilly or complex terrain where the wind field is affected by the near-
presence of hills or upwind turbines. Easily, up to ∼ 10% measurement errors can be observed
between wind speeds measured by a lidar and a mast-mounted cup anemometer co-located
to take wind profile measurements from the on top of a hill. Research is therefore ongoing
in order to correct wind lidar based profile measurements for flow distortion e.g. induced by
terrain effects (Bingöl et al., 2008).

1.3.5 Lidar accuracy

Inherently, lidars can remotely measure the wind speeds aloft with much higher accuracy than
a sodar. This is due to the nature of light, which propagates ∼ 1 million times faster than
a sound pulse, and because a lidar’s antenna aperture size compared to the wavelength, i.e.
“lens diameter-to-wavelength ratio” in a lidar is about 1000 times bigger than practically
obtainable with a sodar. This result in superior beam control and also in much higher data
sampling rates.
At Risø DTU’s test site at Høvsøre, testing and calibration of wind lidar is now daily routine
and is performed by inter-comparing and correlating lidar-measured wind speeds with wind
speeds from calibrated cup anemometers in our 119-m freely exposed tall reference met mast.
During “fair weather conditions”, 10-min averaged wind speeds from lidars and the cups are
in-situ intercompared and correlated. Linear regression coefficients with both CW and pulsed
lidars could be obtained in the range of ∼ 0.99 − 1.00, and correlation coefficients as high as
∼ 0.99 (Wagner et al., 2009).
“Fair weather” means here that lidar data are screened for periods with rain, fog, mist and
low-hanging clouds and mist layers. Usually this only removes a few per cent of the data. All
lidars, CW and pulsed included, rely during determination of the wind speed from Doppler
shift measurements on the assumption that the aerosols in the measurement volumes are
homogeneously distributed and follow the mean wind flow.
Sodars for that matter, can under ideal conditions perform almost similarly well with respect
to mean wind speed (linear regression coefficients as high as ∼ 0.99 has been reported above).
The observed scatter, however, as compared to a lidar, is bigger. Correlation coefficients
observed while testing of sodars at Risø DTU’s 125 m tall met mast at wind energy relevant
neutrally stratified strong wind conditions (> 10 m s−1 ) has so far not been observed to
exceed the 0.90 level.

1.3.6 Wind lidar applications for wind energy

Wind lidar manufactures today address the market for replacement of tall reference meteo-
rology mast installations at the moment required for accredited and bankable wind resource
measurements and for ground-based wind turbine performance measurements. Lidar manu-
factures also offer their wind lidars as instruments for evaluation of model-based wind resource
estimation, on and offshore (numerical wind atlases).
Wind lidars in the market today offer the wind energy industry with RS instruments, for:
• Wind speed, wind direction and turbulence profiling.
• Wind resource assessments, on and offshore.
• Wind turbine performance testing (power curves).
• Wind resource assessment via horizontal scanning over complex terrain.

22 DTU Wind Energy-E-Report-0029(EN)

Further developments Furthermore, new and improved wind lidar data and measurement
technologies are under development for RS-based power performance measurements from the
ground but also directly from the wind turbines. A conically scanning wind lidar (Control-
ZephIR) has during the summer 2009 been tested in a operating NM80 2.3 MW wind turbine
located at Tjæreborg Enge, Denmark, with the purpose to investigate the use of wind lidars
integrated directly into the wind turbine hubs, blades or spinners. The intention is to improve
the wind turbine’s performance by use of upstream approaching wind speed measurements
from inside the turbines rotor plane as an active input to the wind turbines active control
systems. Wind lidars for turbine yaw control are already nowadays on the market (Vindicator)
and new and smaller wind lidars are in the near-future envisioned to become integrated as
“standard” on wind turbines to provide upstream lead time wind data to the turbines control
system, e.g. for:
• Enhanced wind turbine yaw control.
• Lead-time control for individual pitch control.
• Protection against fatigue from extreme wind shear and wind gusts.
• Prolonging the wind turbines longevity.
• Improving the wind turbine productivity.

1.3.7 Summary of lidar

Since the wind lidar era started at Risø DTU in 2004 new wind lidars have emerged on the
wind energy market, spurred by the telecom technology revolution of the 90’s. Today, wind
lidars, continuous and pulsed, and properly calibrated, aligned, installed and maintained, and
their volume-averaged wind measurements properly interpreted, are indeed very precise wind
measuring devises, capable of matching the wind industry’s needs today and in the future for
precise and reproducible wind profile measurements and resource assessments.
Before, however, lidar measured wind measurements can become fully certified and accred-
ited to industry standards, new and revised IEC lidar standards have first to be set and come
into effect. It is important, however, here also to apprehend the very different nature of the
previous standards point measurements as obtained from a mast-mounted cup anemometer
and a volume-averaged wind vectors as obtainable from a profiling wind lidar.
Although the first generations of wind lidars, CW and pulsed, indeed had many difficulties
with reliability, this era now seems to have been improved beyond their first children growth
pains. Today’s wind lidars offer realistic and mobile alternatives to the installation of tall
meteorological masts for many wind resource estimation assessment studies, on and offshore.
The near future will inevitably also show turbine mounted wind lidars fully integrated with
the wind turbines control systems for improving the wind turbines productivity and longevity.

Notation
CW continuous wave
C/N carrier-to-noise
lidar light detection and ranging
RS remote sensing
SAR synthetic aperture radar
sodar sound detection and ranging

References
Bingöl F., Mann J., and Foussekis D. (2008) Lidar error estimation with WAsP engineering. IOP Conf. Series:
Earth and Environ. Sci. 1:012058

de Noord M. et al. (2005) Sodar power performance measurements, WISE WP5.

DTU Wind Energy-E-Report-0029(EN) 23

Gustafsson D. (2008) Remote wind speed sensing for site assessment and normal year correction – The use of
sodar technology, with special focus on forest conditions. Master of Science Thesis in Energy Technology,
KTH, Stockholm.

Mikkelsen T., Jørgensen H. E., and Kristiansen L. (2007) The Bi-static sodar “Heimdall” – you blow, I listen.
Risø-R-1424(EN), Roskilde

Mikkelsen T., Courtney M., Antoniou I., and Mann J. (2008) Wind scanner: A full-scale laser facility for
wind and turbulence measurements around large wind turbines. Proc. of the European Wind Energy Conf.,
Brussels

Wagner R. and Courtney M. (2009) Multi-MW wind turbine power curve measurements using remote sensing
instruments – the first Høvsøre campaign. Risø-R-1679(EN), Roskilde

Wagner R., Jørgensen H. E., Poulsen U., Madsen H. A., Larsen T., Antoniou I., and Thesberg L. (2008) Wind
characteristic measurements for large wind turbines power curve. Proc. of the European Wind Energy Conf.,
Brussels

Kristiansen L. (2010) My own perception of basic statistics. Available on request from Torben Mikkelsen,
Roskilde

24 DTU Wind Energy-E-Report-0029(EN)

2 The atmospheric boundary layer
Søren E. Larsen
DTU Wind Energy, Risø Campus, Roskilde, Denmark

2.1 Introduction
The atmospheric boundary layer, ABL, is the lower part of the atmosphere, where the atmo-
spheric variables change from their free atmosphere characteristics to the surface values. This
means that wind speed goes from the free wind aloft to zero at the ground, while scalars,
like temperature and humidity approach their surface value. An illustration of the profiles is
shown on fig. 8.
Characteristics of the atmospheric boundary layer, ABL, are of direct importance for much
human activity and well being, because humans basically live within the ABL, and most of
our activities take place here. The importance stems as well from the atmospheric energy
and water cycles. Because the fluxes of momentum, heat, and water vapour between the
atmosphere and the surfaces of the earth all pass through the ABL, being carried and modified
by mixing processes here. Since these mixing processes mostly owe their efficiency to the
mechanisms of boundary layer turbulence, a proper quantitative description of the turbulence
processes becomes essential for a satisfying description of the fluxes and the associated profiles
between the surfaces and the atmosphere.
Description of the structure of the flow, relevant scalar fields, turbulence and flux through
the atmospheric boundary layers necessitates that almost all types of the flows, that occur
there, must be considered. For these objectives, there are very few combinations of character-
istic boundary layer conditions that are not of significant importance, at least for some parts
of the globe.

2.2 ABL Flows

The flow and other variables in the ABL all vary with space and time, and, neglecting the
kinetic gas theory, its variability is characterized by a huge variation of the space and time
scales that is involved. The larger spatial scales are related to the size of the globe, the
weather systems and the depth of the atmosphere, the smaller scales are in the millimeter
range. The time scales range between climate variation and milliseconds. The small scale
limits are determined by the fluid properties of the atmosphere.

Important processes for ABL produced turbulence is the production of variability from the
average velocity shear that has to exist in the ABL, as illustrated in fig. 8, because a fluid like
the atmosphere gases cannot be dynamical stable with a mean shear as shown in fig. 8, and
will start producing swirling motion, called eddies, see fig. 9. The characteristic spatial scale
is the height where it happens, and the boundary layer height h. The thus created variability
is called boundary layer turbulence, and is essential for the ABL mixing processes mentioned
in the introduction.
For the purpose of mathematical treatment, one separates the variable in mean values and
fluctuations like:

• For velocity components: ui =< ui > +u′i , < ui > in a mean value, u′i is fluctuating
turbulence., i = 1, 2, 3

DTU Wind Energy-E-Report-0029(EN) 25

Figure 8: Profiles of mean speed, temperature and humidity (u, T and q) for clear a) day,
b) night and c) cloudy conditions. Above the ABL height, h, we have the free atmosphere,
while the ABL is below h down to the surface. Humidity is specified by its mixing ration q,
being the ratio between the water and air density.
.

• For scalars, T , and q, temperature and humidity: T =< T > +T ′ . q =< q > +q ′

• Variances: < u′2

i >, < T
′2
> also denoted by σ 2

26 DTU Wind Energy-E-Report-0029(EN)

Figure 9: Large spatial scale variability for atmospheric flow at upper left, multiple temporal
scale variability at upper right, and vertical multiple spatial scale Tethersonde data at lower
left. A common name for the much of flow fluctuations in meteorology and fluid dynamics
is turbulence, a term that further implies that at least part of its analysis and description is
statistical. Production of turbulence from mean shear is illustrated at lower right (Tennekes
et al., 1972)
.

• co-variances and turbulent transport: < u′i u′j >: Transport of ui in the j direction (and
vice versa).

• < ui T ′ >, and < ui q ′ > :Transport of temperature and water vapor in the ui direction.

Multiplying the velocity co-variances by the air density, ρ ,we can say that the velocity
transport can be considered a momentum transport, similarly multiplying the temperature
transport by ρCp is converted to a heat transport, and multiplying the water vapor transport
by ρL is converted to transport of latent heat. Here Cp is the heat capacity of the air at
constant pressure, L is the heat of evaporation for water. Indeed these terms are often used
in the description, since they reflect the physical importance better than the statistical term
correlation.
Alternatively <> can be denoted with capital letters or over-bars. The coordinate system
can be described at xi , i = 1, 2, 3 or with x, y, z , with the corresponding ui or u, v, w where
u now is along the mean wind direction, w is vertical and v lateral (the second horizontal
component).

A typical behavior of 600 seconds of ABL velocity components and temperature are shown
on fig. 10.

DTU Wind Energy-E-Report-0029(EN) 27

Figure 10: A 600 second record ABL mean values and turbulence: U + u′ , V + v ′ , W + w′ , and
of T + T ′ . Notice, the coordinate system has a vertical z-axis and that the x-axis is horizontal
along the direction of the mean flow, meaning that V = 0. The mean velocity is horizontal,
because the mean vertical velocity, W ∼ 0, since w is constrained by the nearby surface.
Large spatial scale variability for atmospheric flow to the left, and temporal variability to the
right.

2.3 The ideal ABL

As the simplest ABL, we assume the ABL to be limited between a homogeneous flat surface
and a homogeneous boundary layer height, h, see fig. 11.

Figure 11: The ideal simplest atmospheric boundary layer, ABL, which still provides realistic
results. The conditions are statistically stationary and horizontally homogeneous. The wind
speed is forced to zero at the surface, and attain the free wind value Ua in the free atmosphere
above the ABL height, h. The values of wind, temperature and humidity are constant at the
surface and above h, giving rise to a vertical flux of momentum, heat and water vapor between
the surface and the h-level

The flow is assumed statistically stationary, meaning there will be variations. But these
will be statistically horizontally homogeneous and stationary. However, as seen from equation
1 we must allow a pressure gradient that, again somewhat unrealistic, is taken as constant.
Without a pressure gradient, there will be no wind.

28 DTU Wind Energy-E-Report-0029(EN)

 The three moment equations:
Du1 1 ∂ρ ∂
=0=− + f u2 − (u′ u′ )
Dt ρ ∂x1 ∂x3 1 3
Du2 1 ∂ρ ∂
=0=− + f u1 − (u′ u′
Dt ρ ∂x2 ∂x3 2 3
Du3 1 ∂ρ ∂
=0=− − 2Ω(η1 u2 − η2 u1 ) (u′ u′ )
Dt ρ ∂x3 ∂x3 3 3
The scalar equations: (1)
DT ∂ ∂T ∂T ∂T
=0= (−u′3 θ′ )(= + u1 + u2 )
Dt ∂x3 ∂t ∂t ∂x2
Dq ∂
=0= (−u′3 q ′ )
Dt ∂x3
Equation 1 summarizes the equation form the mean flow for our pseudo homogeneous ABL.
As mentioned the constant pressure gradient is necessary, but limits the horizontal scale for the
model. The temperature equation further illustrates the meaning of the substantial derivative,
for all the variables. The equation for u3 is not important in this approximation and will be
neglected in the following.
Additionally, in eq. 1 our simplified ABL is situated on the rotating planet Earth, reflected
by appearance of the Coriolis parameter, and the Earth’s rotation rate Ω, with f = 2Ωsinϕ
and with ϕ being the latitude on the globe. Further it is seen that we have introduced the
symbol θ, the so called potential temperature. This is a modified temperature including the
fact that the average pressure and density in the atmosphere decreases with height, due to
gravity. This means that an adiabatically moving air packet cools moving up and heats moving
down, but will remain in equilibrium with surroundings and at the same potential temperature.
If θ increases with height the air is denser than equilibrium at the bottom and therefore stable
against vertical perturbations. Conversely for θ decreasing with height, the air is lighter than
equilibrium at the bottom and hence unstable, if perturbed vertically. Within the boundary
layer, θ is often approximated by:

g
θ = T + Γ · z, with Γ = (2)
Cp
With Γ being about 0.01 K/m. It is noted that the only difference between T and θ is the
linear height variation.

Equation 1 shows that the vertical fluxes of scalars are constant with height, while the
momentum fluxes are slightly more complicated. Focusing on the two first equations for
the horizontal velocity components, we define the geostrophic wind, G, from the pressure
gradient, and perpendicular to the direction of this gradient:

1 ∂p 1 ∂p 1 ∂p 1 ∂p
G = (U1G , U2G ) = (− , ) = (− , ) (3)
f ρ ∂x2 f ρ ∂x1 f ρ ∂y f ρ ∂x
The two first equations in eq. 1 now take the form:

∂ ′ ′
0 = f (u2 − U2G ) − ∂x3 (u1 u3 )
∂ ′ ′
0 = −f (u1 − U1G ) − ∂x3 (u2 u3 ) (4)
This equation shows that the wind velocity approaches the geostrophic wind at the top of
the boundary layer, where the turbulence disappears. Down through the boundary layer the

DTU Wind Energy-E-Report-0029(EN) 29

Figure 12: The variation of wind speed and wind direction from the top, through the ABL
towards the ground. The wind profile can be seen as being developed through a balance
between the three forces, the pressure gradient, P , the Coriolis force, C, and the Frictional
force, F. (Larsen et al., 1983)

wind solution depends somewhat on the additional assumptions, but generally it undergoes a
spiraling motion as it reduces to zero at the ground, see fig. 12.
The detailed behavior of the wind speed through the ABL, is simplest for atmospheric
neutral stability, meaning that there is no heat flux and water vapor flux between the surface
and the free atmosphere (fig. 11). This can be assured by keeping the potential temperature,
∆Θh , difference at zero, meaning that T a−T s = Γh, compare eq. 2, and as well qs = qa . For
such situations one can derive the equations in eq. 7, using so called scaling laws, where the
momentum transport, is a new scale introduced from the co-variance, the so called friction
velocity u∗ as :

u2∗ = −u′ w′ (5)

  21
κG u∗ 2 2
ABL : = (ln( − A) + B
u∗ f z0
−B
ABL : α = tan−1 ( ) (6)
(ln( fuz∗0 − A))
u∗ z
SBL : U (z) = ln
κ z0
The first equations relate the conditions at the top of the ABL to the wind at lower heights,
the so called Surface Boundary Layer, SBL. The first of these called, the resistance laws of the
ABL, relating the friction velocity, u∗ , to the geostrophic wind, G, and a so called roughness
length, z0 , plus two three additional “universal” constants, the von Kármán, κ, and the two
constants A and B. The term fuz∗0 is denoted the surface Rossby number. Since f is of the
order of 10−4 A is about 2 and B about 5 , z0 is small. The Rossby number term dominates
the first of the resistance laws. The second of the resistance laws estimates the angle between
the geostrophic wind and the wind in the lower part of the ABL, the ASL. In this part the
wind profile is described by the last of the equations, the so called logarithmic profile.
The set of equations allow us to estimate the wind profile in the ASL for a given geostrophic
wind speed for varying roughness length. Therefore the roughness length is a very important
parameter, see fig. 14. In the principle it is a characteristic length, where the velocity extrap-
olates to its surface value, which is zero, but since it is a measure of the “roughness” of a
given landscape, much work has been done to establish consensus about the roughness of
characteristic real landscapes.
As seen from fig. 13, the roughness generally follows the intuitive images of what that
roughness is associated with. The larger and the sharper the protruding elements, the larger
the roughness. Although this image is simplistic, it still summarizes the main aspects of the
roughness characteristics of landscapes, including season variation of some landscapes. To

30 DTU Wind Energy-E-Report-0029(EN)

Figure 13: Consensus relations between the roughness lengths for different landscapes (Stull,
1988).

emphasize the importance of the roughness length for wind energy, we compare in fig. 14 the
winds in the ASL for for different roughness values and a given geostrophic wind.

Until now we have considered only situations, where the potential temperature differences
between the surface and the free atmosphere is unimportant for the dynamics, this state is
called thermally neutral. Changing T a and T s to make ∆θh different from zero, and sim-
ilarly for qs and qa , a heat flux and water vapor must flow between the top of the ABL
and the surface, and there will be a density gradient between top and bottom. For such a
situation the variation of wind speed, temperature and humidity can be described by an ex-
tension of the simple scaling used for neutral conditions in eq. 7. These scaling formulations
are normally denoted the Monin-Obukhov formulation. The set of scales is summarized below.

DTU Wind Energy-E-Report-0029(EN) 31

Figure 14: The change in surface wind for different roughness values, but the same Geostrophic
wind, G. The roughness length z0 is seen to be related to the indicated land scapes in fig. 13

Friction velocity:u∗ = −−u′ w′

′
Temperature scale:T∗ = − −w
u∗
θ

′ ′
Water vapor scale:q∗ = − −w
u∗
q

T u3∗
Monin-Obukhov stability length:L = κg(T∗ +0.61q∗ ) (7)
The water vapor concentration, q, enters into the stability measure, because both q and
Θ influence the density and thereby the stability based on density fluctuations. This will
be repeated throughout this text, but not always, because temperature is typically more
important than humidity for the stability.
The Monin-Obukhov scales are widely used, not only in the atmospheric surface layer (ASL)
that was introduced in fig. 12, but also for extension into the full ABL. In eq. 7, g is the
acceleration due to gravity, and the stability length, L, is a scale that is derived from the
energy budget as a measure of the importance of the heat and water vapor fluxes relative to
the momentum flux. As in eq. 7, κ is the von Kármán constant (∼ 0.4). The scales defined in
eq. 7, change relatively little within the ASL, above this one typically uses the ASL measured
parameters.
Corresponding to the neutral wind profile in eq. 7, we can now formulate the set of stability
influenced profiles in eq. 9, as:

u∗ z z
u(z) = (ln − ψ( ))
k z0 L
θ0 z z
θ(z) − θ0 = (ln( ) − ψT ( )) (8)
k z0T L
q∗ z z
q(z) − q0 == (ln( ) − ψq ( ))
k z0q L
Stability influenced profiles of < u(z) >, < Θ(z) > and < q(z) > . Notice, that separate
“roughness length” parameters are introduced for the scalars. These are of the order of z0
for smooth surfaces, but typically of the order of 10 times less for rough surfaces.
From the definition of L, it is seen that Lz ∼ 0 for neutral conditions, with no scalar fluxes.
ψ(0) = 0 for all three ψ-functions, and hence the logarithmic profiles for all variable are
recovered for neutral. Close to the ground Lz ∼ 0 simply because z is close to zero. Hence
all profiles start as being logarithmic close to the ground. For stable conditions, we have:

32 DTU Wind Energy-E-Report-0029(EN)

ψ Lz ∼ − L5 . For unstable conditions (meaning Lz < 0) the functions are more complicated but
can be approximated by is ψ Lz ∼ 1 − (1 − a · Lz )−n , with a between 10 and 15, and n = 0.25
for the wind and 0.5 for the scalars.
When the heat flux is positive (the ground is warmer than the air) the atmosphere is
characterized by rising air. The shear produced eddies are enhanced by the thermal structure,
and the situation is denoted thermally unstable. If the heat flux is downward, the shear
produced eddies and the general fluctuation level is diminished, for which reason the situation
is denoted stable. The enhance mixing for unstable condition reduces the wind shear, while
the reduced fluctuation level for stable condition allows a larger wind shear. The situations
are depicted in fig. 8, together with a neutral situation, (indicated by sun, moon and a cloud).
Notice that above the ABL the temperature and the potential temperature will in general
increase with height.
In reality thermal properties of the lowest atmosphere is forced either by the radiation
balance at the ground, insolation at day time, and additional cooling at night time, or/and
by advection of air masses with a temperature that differs from the surface temperature.
The balance depends on time scale considered, on the thermal properties of the ground,
cloud cover and the characteristics of the air motion around the site. We return to this when
discussing real boundary layers.
As opposed to the profiles in eq. 9, in engineering literature one often find the so called
power law profiles:

u(z)
= (z/10)α (9)
u(z = 10)
where the relevant α will be function of height, stability and roughness, as seen by compar-
ing with eq. 9. The power law profile has the advance of being simple and that α is a direct
measure of the relative shear, as can be seen by differentiation of eq. 9.

For neutral conditions we have in eq. 7 not only a wind profile in the ASL, but also for the
whole ABL a resistance law relating the G to u∗ and the geostrophic wind angle. This can in
the principle be extended also to different stabilities and to the scalar variable, letting the A
and B constants, and corresponding parameters for scalars, be function of stability, typically
in terms of h/L (Zilitinkevich, 1972),(Zilitinkevich, 1975). However, the quality of the data
fit for these extensions are worse than for the simple neutral expression in eq. 7, and therefor
they are not much used.

For our ideal boundary layer discussion, we have simply fixed the boundary layer height,
just as we could fix the temperature difference between the surface and the air on the top of
the ABL. In the real world the ABL height is determined as the outer range of the boundary
layer turbulence. For neutral conditions one find that it must be proportional to u∗ /f , since
these two parameters are the only parameters available to characterize the ABL turbulence.
Indeed one finds that this fits the data moderately well with a coefficient equal to about 0.3.
For moderate stabilities, one can use an expression like h ∼ 0.7(u∗ L/f )0.5 , and for strongly
stable conditions h ∼ L, the Monin-Obukhov stability length.
For unstable and many stable situations, it is common to use an independent rate equation,
to determine h, at least over land, that h = h(t) is now derived from an equation like
dh/dt ∼ F (··). Especially for unstable situations, a very simple and successful equation
has been developed. Assuming an unstable ABL growing, in response to sunrise, against a
background stable potential temperature gradient, γ, using the in-stationary temperature
equation in eq. 1. The result is:

 0.5 Zt
Q(t)
h(t) = 2 with Q(t) = w′ θ′ |0 (1 + 2A)dt′ (10)
γ
0

DTU Wind Energy-E-Report-0029(EN) 33

 Where A here is an entrainment coefficient of the order of 0.2, accounting for the flux is
not only from the ground, but also from the top of the growing ABL. The boundary layer
height and the roughness length in combination have importance for the existence and extent
of the surface boundary layer. Formally ASL can exist for z0 << z << h. For the large z0 in
fig. 13 and low values of h, the simple ASL expression of eq. 9 becomes invalid.
Note, that with equations like eq. 10, we have strictly speaking left the stationary boundary
layer, and are approaching the real world, where a diurnal cycle is a basic fact of life. Experience
shows, however that much of the simplified ABL described here survives, when allowing for
slower changes like many, but not all, of the diurnal changes.
Not only mean values are important to characterize the ABL, also fluctuation of the indi-
vidual signal are important. To a first approximation these can be characterized by the tur-
bulence standard deviation, where especially those associated with the velocity components
are important for many purposes. For velocity components one has developed expressions,
corresponding to the Monin-Obukhov profile.

σx z h
≈ F ( , ) with x = ui , θ, q (11)
x∗ L L
Where it has been found necessary to include h/L even within the ASL. In general the
function F increase with increasing instability, and become constant for neutral-stable condi-
tion. For velocity one often considers eq. 11 in term of the Turbulence Intensity (T I) derived
from eq. 11 by division with the local wind speed. Thereby the T I becomes an expression
for the likely relative deviation from the mean speed one will encounter for given situations,
Also distributions of the vertical instantaneous shear has become an important parameter,
e.g. distributions of α in eq. 9. Inserting the respective expressions we get, for the two:

σu κF (z/L,n/L) 1 z
TI = u = ln(z/z0 )−ψ(z/L) ≈ ln(z/z0 ) for L →0
z du φ(z/L) 1 z
α = u dz = ln(z/z0 )−ψ(z/L) ≈ ln(z/z0 ) for L →0 (12)
where ϕ(z/L) is the derivative of ψ(z/L). The behavior of data on the two functions are
shown in fig. 14.
Additional information about the fluctuations can be seen in the correlations or the spectral
structure of the signals. Atmospheric signals vary as function of both time and space. For
boundary layer turbulence, one can mostly assume that the fluctuations vary with space only,
and that measured time variation at a stationary measuring station is due to advection of a
spatial variation of the signal.
This, surprisingly simple assumption, is denoted the Taylor hypothesis. Take a measured
u(t) as an example:

u(∆t) = u(∆x/ < u >) (13)

where ∆ signifies that the Taylors Hypothesis works on differences in space and time, not
on the absolute coordinates. Equation 13 can even be used in connection with a moving
sensor, like an air plane, where the speed then must be the sum of the air velocity and the
sensor velocity.
To resolve frequency or wave number distribution of the turbulent variables, one uses Fourier
spectra computed either as frequency or wave number spectra, with connection derived from
the Taylors hypothesis, ω = k1 < u >, where k1 is the wave number along the mean wind
direction. The Fourier analysis is based on the existence of Fourier pairs. The simplest principal
way is to illustrate the spectra-correlation duality by:

34 DTU Wind Energy-E-Report-0029(EN)

Figure 15: Behavior of the turbulent intensity the dimensionless shear from eq. 12, measured
at about 60 meter over water (Wang et al., 2012; Peña et al. , 2012). The deviations between
the predicted values and the data are due to stability, processes, and also to some extent to
model failings, which we will comment later, but at the very least we see that the neutral
limit with a constant z0 is not adequate to explain the data. Specifically for the T I it is fairly
well established that the high T I for low wind speeds is due to stability effects. For higher
winds T I is expected to follow the observed mean T I is expected to be in the neutral limit
of eq. 12. However, the prediction is based on land conditions with a constant z0 , while the
over water measurements, reflects the growth of z0 with wind speed, following Charnock’s
relation as discussed later in this chapter

Z
SXY (ω) = X ′ (t)Y ′ (t + τ ) exp(iωτ )dτ
Z
1
X ′ (t)Y ′ (t + τ ) = SXY (ω) exp(−iωτ )dω
2π
(14)
Where X and Y are two turbulent variables as function of time, with a correlation as
function of a lag time, τ , with a corresponding cross spectrum SXY (ω) of frequency ω.
Similar expression could be formulated for the spatial correlation of < X ′ Y ′ (xi + δi ) > and
its corresponding wave number cross spectrum, SXY (ki ). Here the wave number analysis is
different from the frequency analysis, in that the spatial lags, δl , and the wave numbers, ki ,
are vectors. Taylors hypothesis is relating frequency to k1 only, the wave number one along
the mean velocity direction.
The spectra are in fig. 16 are scaled with the same scales as the profiles < u′ w′ >,
< w′ T ′ >, etc. Thet are plotted versus a normalized frequency n = f z/ < u >= k1 z, with
f in Hz, and, where we have again used Taylors hypothesis for the frequency-wave number
relation, the z appears from the Monin-Obukhov similarity (Kaimal et al., 1972).
The spectra in fig. 16 are empirical and other spectral expressions exist, but there is a
general consensus on their form and intensity, such that the different forms agree broadly
on the behavior of the spectra, although there are low frequency differences, that can be
important in connection with some load modeling on structures (Andersen et al., 2010). The
spectra vary systematically with height through the ABL and with stability in widely accepted
ways. Olesen et al. (1984), Hojstrup (1982), Mann (1998), and the Mann-lecture at this
course.
At present, there is a strong activity of extending profiles and turbulence expressions all the
way through the ABL, and even further up. This is both because the growing wind turbines
makes the information important and also because the breakthrough in the remote sensing

DTU Wind Energy-E-Report-0029(EN) 35

Figure 16: Neutral scaled ASL power and co-spectra from (Kaimal et al., 1972) plotted versus
the normalized frequency n = f z/U . The spectra are scaled by the relevant Monin-Obukhov
parameters defined in (7).

technology, has made systematic data gathering in these heights possible. Some progress
has been made, as indicated by the illustration in fig. 17. However, it is still a speciality in
development, because the profiles aloft become sensitive to many new important scales, like
the ABL height, but also on the detailed characteristics of the entrainment zone above the
ABL.
An often used scale for strongly unstable conditions, is a velocity scale constructed from the
turbulent heat flux, to be used when the heat flux is more important than the momentum flux,
and often in the upper part of the unstable ABL: w∗ = (h < w′ θ′ > g/T ) ∼ u∗ (−h/κL)1/3 ,
u∗ will typically at most be a few tens of cm per second, while w∗ can reach several meter per
seconds. Additionally, to the new scales entering the problem,the demands to homogeneity
becomes more severe, and several aspects of remote instationarity and inhomogeneity, not
really influencing the profile in the ASL, will influence wind speed and wind turning aloft,
like e.g baroclinity and remote changes in surface characteristics. We shall consider these
aspects later. Here we present experimental and theoretical effort to describe the wind profile
to greater heights.
We finally return to the concept of thermal stability, which as we have seen is a measure of
the importance of the atmosphere’s thermal stratification relative to the wind shear dynamics
for the flow structure. In the SBL it can be described by the stability length L (Eqs. 7 and
9). Commonly used are the so called Richardson numbers, that measures the ratio between
the potential and the kinetic energy across a layer. In the ABL it is mostly defined in terms
of gradients of the mean temperature and the mean wind as shown in the following equation
eq. 17.

g∆z ∆ρ g∆z ∆θ gz ∆θ g ∂θ/∂z

Ri = − ≈ ≈ ≈ (15)
ρ (∆U )2 θ (∆U )2 θ U2 θ (∂U/∂z)2
Where the first term is the basic definition of stability in terms of density gradients for a
layer of depth ∆, while the three last second forms are those most used in the ABL, and are
based on a relation between the potential temperature and the potential energy. The third
term characterizing a layer from the ground to height z, is denoted the Bulk Richardson
Number, while the 4th form is the differential form. Inserting the profile expressions Ri can
be described in terms of the SBL formalism. But Ri is more general validity than for the the
SBL atmosphere. To be completely consistent with eq. 7 we should have introduced water

36 DTU Wind Energy-E-Report-0029(EN)

Figure 17: Newly developed models tested against data from the Høvsøre site in Denmark.
The broken line are the extrapolations of the ASL models presented in fig. 14, the solid curves
reflect models developed by (Gryning et al., 2007), involving the boundary layer height and
one more height scale.

vapor into eq. 12 as well.

2.4 Surface Characteristics of real ABLs

We shall consider characteristic real ABLs in the light of the ideal one considered in the
former section. Characteristics of a flat homogeneous land surface, a marine ABL, the effect
of moderate inhomogeneity, and finally steep surfaces and complex terrain.
To understand some of the differences, we must consider layers below the turbulent ASL,
because this is where the frictional processes resulting in a roughness takes place, and where
the surface temperature is a result the heat flux properties. The roughness length z0 is the
height, where the logarithmic profile extrapolates to zero due to surface friction. However,
below a height of about 10 z0 the real profile is not logarithmic anymore, because of the
molecular friction and because of the flow impacting on irregularities of the surface roughness
elements.
Tracking the momentum transport through the viscous layer reveals that the momentum
and heat transport is taken over by the surface and can be written as:

   
p ∂η ∂u
−w′ u′ z = u2∗z = ρ ∂x +ν ∂z
η η
 
∂θ
−w′ θ′ z = νθ ∂z (16)
η

The real irregular surface as indicated on the figure and denoted, η, while ν and ν0 are the
molecular diffusivity of momentum and heat respectively. On the left hand side of eq. 16 the
transport taking place in the turbulent ASL have been indexed with a z. In the viscous layer
ASL turbulence cannot exist due to the nearness of the surface, and the transport is taken

DTU Wind Energy-E-Report-0029(EN) 37

Figure 18: Extrapolation of the logarithmic profile down towards the surface below height of
the viscous surface layer, δ. The logarithmic line is shown, while real velocities are indicated
by x. Small scale irregularities, which we will denote roughness elements, are indicated.

over by molecular gradient diffusivity and for momentum also by the pressure perturbation
around the irregularities. The correlation obviously becomes larger with steeper and sharper
irregularities, roughness elements. If there is no roughness element the transport is limited by
the molecular diffusion, and a scale analysis indicates the z0 ∼ δ ∼ ν/u∗ . With roughness
elements present the roughness becomes larger. On the other hand the heat transport in eq. 16
includes no pressure term, since pressure appears in the momentum equation only. Therefore
the transport of heat in the viscous sublayer is basically due to the molecular diffusion only,
and z0T , and correspondingly z0q , must be expected to be smaller than z0 , and more so, the
more rougher the surface. As mentioned earlier typically z0T /z0 ∼ 0.1 for rough surfaces.
Next we turn to the surface temperature θ0 appearing in eq. 9 and implicit in all our
arguments about the effects of stability. The surface temperature is a boundary condition in
eq. 9 but it is as well a result of the energy balance at the surface and controlled by processes
above, on and below the surface. At the surface the energy balance per unit area can be
written as (Bodyko, 1974):

dθ0
C = NR + H + E + G (17)
dt
Where, C · dθ0 /dt the heat accumulation in a thin layer at the surface, with C being the
heat capacity of this surface layer.

N R is the net radiation at the surface, consisting mainly of the incoming short wave radia-
tion from the sun and outgoing long wave radiation from the surface itself. N R is positive day
time, most positive around noon, and it is negative during night, when the surface undergoes
additional cooling. Cloud cover, season and latitude is obviously important here.

H is the turbulent heat flux = ρCp < w′ θ′ > , cooling the surface when it is warmer
than the air, and heating it when it is colder, typically due to radiation at night and day.
Obviously it can reflect also changes in the air temperature due to advection, from eq. 9 we
have H == ρCp < w′ θ′ >∼ −ρCp (θ(z) − θ0 )κu∗ / ln(z/z0T ).

E is the latent heat flux=ρL < wq ′ >, reflecting that the surface can regulate heat by
evaporating water or condensing water vapor. Just as for temperature these processes can be
driven by radiation, but also by advection.

G is the ground layer heat flux, which as all the other fluxes in eq. 17 can be both negative
and positive. But it will typically be directed downwards during day time, and upwards during
night.

38 DTU Wind Energy-E-Report-0029(EN)

 The value of q0 at the surface is driven by surface mass balances similar to the surface
energy balance for temperature. For humidity there are of course no radiation terms, on the
other hand there are two interacting balances, one for liquid water and on for water vapor,
with a strong dependency on the soil type and its water content, and- if vegetated- on the
vegetation types, all with different root systems and different strategies for exchanges with
the atmosphere. We shall not dwell further on these complexities in this context. We just
notice that if the surface is wet, it all simplifies to that q0 can be derived from the saturated
water pressure at the surface, es , where es (θ0 ), since at the surface T0 = θ0 .

2.5 Homogeneous Land ABL

For a flat fairly homogeneous land surface the ABL has a number of characteristics of relevance
for the detailed use of the equations and models, described in section 3.
The energy balance at the ground eq. 17 generally results into a significant diurnal and
annual cycle of the surface temperature and the thermal stability of the atmosphere with
instability during day, and stability during night, and the role of advection is less apparent.
With this follows also a tendency to have deep ABLs at day time, following eq. 10 and shallow
stable night time ABLs. The diurnal variation of the surface temperature, penetrate down
to about half a meter in to the soil, where the amplitude vanishes. Hence only a shallow
soil layer, with relatively little heat capacity, is involved in the diurnal heat exchange with
the surface- somewhat deeper for the annual scale. For higher wind speeds the stability is
often forced towards neutral by the roughness generated turbulence, even with fairly large
heat fluxes. From eq. 9 is seen that a given geostrophic wind will result in a larger u∗ the
larger is the z0 . From eq. 7 is seen that we can write z/L ∼ zκgT∗ /u2∗ , which diminish with
increasing u∗ . The numerics is such that we will have z/L ∼0 for moderately to high wind
for characteristic land surface roughness.
The roughness elements of the surface consist mainly of either stone like fragment, ranging
from pebbles over boulder and houses, or vegetation ranging from the tiniest to major forests.
For a surface of simple roughness elements, one can often use Lettau’s formula:

z0 ≈ 0.5 · h · S/A, A >> S, (18)

where h is the height of the roughness element, S its crosswind area, and 1/A is the surface
density of roughness elements (Lettau, 1962). For densely placed vegetation, one can often use
a simple z0 proportional to the height, and additionally introduce a so called displacement
length, d, also proportional to the height of the vegetation, with different coefficients of
proportionality for different types of vegetation fig. 19.
Vegetation based z0 , show some fairly weak variation with wind speed, reflecting that the
wind is moving straws, branches and leaves etc. Also some stability variation of z0 has been
proposed, reflecting the structure of turbulent eddies penetrating into the canopy (Zilitinkevich
et al., 2009). Further, a clear dependency on the seasonal variation of foliage is found, as in
figure 12. For this kind of land roughness, the z0T and z0q , must be expected to be about
10% of the z0 .
Figure 13 points to a seasonal variation associated with snow cover, where such happens.
The figure suggests a z0 value of about 2 mm for a natural snow surface, which could then
be the prevailing roughness for such surfaces during wintertime.

2.6 Homogeneous Marine ABL

The marine ABL has distinct features compared to the land ABL. The water is semi transpar-
ent meaning that the additional heating and cooling is distributed downwards. Additionally
the water has extensive mixing properties. The surface waves and circulation systems, like the
Langmuir cells, combined with turbulence give rise to extensive mixing. Additionally, when

DTU Wind Energy-E-Report-0029(EN) 39

Figure 19: A forest canopy showing displacement height, d, and roughness length z0 as being
proportional to the canopy height. Additionally the wind profile is indicated. Clearly, only the
wind above the canopy can be expected to be represented by an ABL profile. The displacement
height indicates the ground level for the ABL profile, while the wind inside has to be described
by other methods.

Figure 20: Measurement and modeling of seasonal variation of roughness for different types
of vegetated surfaces (Hasager et al., 2003).

heated the surface water evaporates, it will start sinking; now being heavier, because it retains
the salt from the evaporated water. If the surface water cools, it also becomes heavier due to
the cooling and sinks. All this give rise to an intense mixing in typically the upper 10 meter
of the ocean. In the heat exchange with the atmosphere the water therefore constitute a very
large heat reservoir that only can change its temperature slowly, and additionally has its own
heating and cooling from the ocean currents. Indeed when an air mass moves over an ocean it
always ends up at the temperature of the ocean. For these reasons the homogeneous marine
ABL is always close to neutral. The diurnal radiation cycle shows very little influence on the
water surface temperature, although it can be measured, but typical amplitudes are less than
a few tenths of a degree (Peña et al., 2008). The annual radiation cycle on the other hand
has significant influence on the sea temperature, because they involve enough heat to change
both the temperature and the depth of the mixed layer. However, stable and unstable condi-
tions happens over the ocean as well on shorter timescales, but they are mostly transitional,
associated with air masses moving across water surfaces with a different temperature, either
coming from a nearby land or associated with moving weather systems. We shall return to
these phenomena when coming to the inhomogenous and instationary ABLs.
The sea is also an obvious source of water vapor evaporation, indeed over the ocean since
the q0 is derived from the saturated pressure at the surface temperature. The ocean is also
a source of liquid water in the form of sea spray converting to marine aerosols. In wintertime

40 DTU Wind Energy-E-Report-0029(EN)

the spray is the source of icing on ships and offshore structures.
The roughness elements over water mostly take the form of small steep waves of a wave
length of around 5 cm, although momentum can be transferred also by larger scale breaking
waves. The ocean surface is depicted in fig. 18. Since the roughness is associated with the
waves and the waves are generated by the wind and modified by gravity. Charnock (1955)
proposed that the roughness should depend on u∗ and g. A slightly updated version of the
roughness for water looks as the following:

ν u2
z0 = 0.11 + β(c/u∗ , −) ∗ (19)
u∗ g
Where the first term reflects the molecular diffusion limit, discussed above, when only few
roughness elements are present. The coefficient, β, denotes the Charnock’s constant. It is a
function of the phase speed c of the dominant waves and u∗ . Since the roughness elements
will be mowing with the phase speed of the waves in the direction of the wind. The term
c/u∗ is denoted wave age, because c increases with the duration of the acting wind. β is
varying between 0.01 and 0.07, being smallest for mid-ocean mature waves with large phase
speed. A “typical” value is 0.015. β can be a function of other parameters as well: e.g bottom
topography, swells, and very high winds (e.g. hurricanes) resulting in foam covered waters
that reduces β further (Makin, 2005).

Figure 21: The wind profile close to the water surface, with the wave induced vorticity and
the small scale roughness element riding on the larger scale waves, with a phase speed c.

In spite of the functions shown in eq. 19 the roughness of the sea surface still remains one
of the smallest, one can encounter in nature. This means that high wind speeds will be less
efficient in forcing the stability towards neutral over water than over land, although over water
frequency of neutral stability increases with wind speed. Still high winds can be encountered,
associated with strongly stable flows over water, again reflecting an inhomogeneous situation
where warm air is advected over cold water. Here the friction almost disappears. Again we
shall return to this issue. Just as winter snow can modify the roughness of a land surface
strongly, the winter will some part of the world cover the water with ice, see again fig. 13,
and the roughness now will depend on the characteristics of the ice surface, ranging from
extremely low for smooth solid ice, to quite rough for pack ice. The small z0 also means that
the turbulence typically is lower over the water than over land, reflected also in a lower ABL
height over water than over land. The small z0 also means that the z0T and z0q are close
to z0 for low wind speeds, and start deviating only for rough pack ice or larger wind speeds,
with rough sea.

DTU Wind Energy-E-Report-0029(EN) 41

Figure 22: The Charnock’s function, β is shown versus reciprocal wave age, u∗ /c. Typical
wave ages in nature lie between 5 and 30, (Jones et al., 2001)

2.7 Inhomogenous and instationary ABL

No ABL is strictly homogenous and simple models have been developed to handle the inho-
mogneity to organise the ideas about the complex subject.

Starting with an abrupt roughness change as depicted in fig. 23. Here the wind blows from
a surface with one roughness over a surface with another roughness. The stability is neutral.
The turbulence associated with the new roughness growths into the upstream boundary layer
by diffusion. This new boundary layer is called and Internal Boundary Layer, IBL, because it
growths within the already existing boundary layer.

Figure 23: Description of the structure of the internal boundary, of height h(x), due to a step
change in terrain roughness. Rule of thumb h ∼ 0.1x

dh ∂h ∂h κA
= u(h) = Au∗0 → = → Cx h ln(h/z0 ) (20)
dt ∂x ∂x ln(h/z0 )
Where we have used that u(h) follows the logarithmic profile. C is found to be about 1.
Notice, we measure u∗ as u∗ at the surface, because u∗2 must be expect to vary with height
in the IBL. The system works for both a smooth- to -rough transition, and the opposite, as
long as one uses the largest roughness of the two in the IBL growth equation eq. 18. With
h(x) determined in eq. 18, we can find the new surface, the new u∗2 , matching the upstream
and downstream profile at h to yield:

42 DTU Wind Energy-E-Report-0029(EN)

    
u∗1 h u∗2 h
u1 (h) = ln = ln →
k z01 k z02
     
u∗2 ln zh01 ln zh01 zz02
02
ln zz02
01 M
=  =   =1+   ≡1+   (21)
u∗1 h
ln z02 h
ln z02 h
ln z02 ln zh02

Equation 21 provides a very successful estimate of the ratio between the upstream and
downstream u∗ values with fetch, x, since h = h(x). Additionally it provides us with a way
of characterizing magnitude of roughness changes, through the factor M (in the last term).
From eq. 13 we get that the roughness change between a water surface and a smooth land
surface with a roughness of 1 cm has the same roughness change magnitude as the roughness
change between the same land surface and a city or a major forest. Both roughness changes
are associated with a roughness ratio of about 1000. We may also use the results to estimate,
how wide a homogenous area that is needed for the assumption about homogeneity, need to
be. This is height dependent. The growth rate of the IBL given by eq. 20 is slightly slower than
h = 0.1x, meaning the measuring height should be less than 10 times the fetch, x, to feel
the new IBL. It can show that the height below, which the flow is in approximate equilibrium
with the new surface, is about 0.01x, meaning that we will need a homogeneous new fetch
of 100 times the measuring height to consider the upstream conditions homogeneous. Under
all circumstances the demands to a homogenous fetch, depends on the measuring height.
This is one explanation of why it has been found to be more difficult to obtain consistent
estimates of the profiles aloft. For example the profiles obtained up to 200 m in fig. 14,
demands homogeneous fetches of 20 km to be in equilibrium with the underlaying surface.
Uncritically, we could extend the theory to a full ABL, find as a rough estimate we should
have a fetch of 100h to have a new equilibrium situation.
From eq. 21 one can estimate also the ratio between the upstream and down stream wind
speeds. Unfortunately, the formulation is wrong for very large fetches. Where it predicts the
u∗ -ratio, it becomes unity, which is not in accordance with the resistance law for a new ABL
with the new z0 . Also stability effects are not included. Both deficiencies can be repaired, but
the results miss the appealing simplicity of eq. 21, and will not be treated here.
Instead we turn towards the flow over low hills: As the flow approach the hill a pressure
perturbation develops, reaching both up-stream and down-stream and breaking the air on the
front slope and back slope of the hill and and developing an acceleration on the top of the
hill.

Figure 24: The principles for flow over a low hill according to (Jackson et al., 1975) from
(Hunt et al., 1982), with the width, L, the height, H, defined.

The flow over the hill is described for three regions in the vertical. At a height L the pressure
perturbation has disappeared, and the flow is undistorted. In the inner region the maximum
flow perturbation takes place, while the wind is still forced to zero at the ground. Therefore,
this is the region with the largest wind shear. The maximum wind perturbation takes place at
the top of the inner layer, and gradually decreases to zero at the height, L. The hight of the

DTU Wind Energy-E-Report-0029(EN) 43

inner layer is of the order of 0.1 L, and the perturbation of the top of the hill, the so-called
relative speed up, is found as:

∆u H
≈2 (22)
u0 L
Troen & Petersen (1989) have developed the theory into a Fourier form, allowing one to
build an arbitrary landscape as a superposition of low hills, which when combined with the
roughness changes, becomes the model in the WAsP system Troen & Petersen (1989). If the
slope becomes too large, of the order of 20%, flow separation develops on both the upstream
and the downstream slope.
We next turn towards the situations which are driven by a changing heat flux. Here we
start with the diurnal cycle for a horizontally homogeneous area at midlatitude, where the
boundary layer show obvious diurnal changes. As we have discussed the changes, they will
be smaller if the wind is high, and the atmosphere is cloudy, than when the wind is low and
the sky is clear. But it will always be present, and limiting the degree of stationarity one can
expect for such a surface. A similar changing rate can be found in frontal passages, which
take a time of the order of one day or longer.

Figure 25: A heat flux driven diurnal change of the ABL for a homogeneous midlatitude land
surface (Stull, 1988).

We next turn towards an IBL controlled by the surface heat flux, where the IBL, here
denoted TIBL, growths against a stable upstream ABL. Here, we can use the growth of the
unstable boundary layer equation in eq. 10, also depicted in fig. 25. Just assuming that the
upward heat flux now happens as function of fetch, x, instead of time, inserting t = u/x
in eq. 10. A more comprehensive description of these types of models are found in Gryning
et al. (1990). The situation can happen for both land and water surfaces, being dependent
on differences in surface heat flux and/or surface temperature of the two surfaces involved.
However, it is quite easy to imagine, that it is developing at a coast line. Here the surface
temperature of the land and the sea is quite often different. Due to the two different surfaces
response in the surface energy budget (eq. 17) on a diurnal scale. As the time and space scales
for such coastal system increase, the differential heating between two surfaces, will influence
the atmospheric dynamics, because air will tend to rise over the relatively warm surface and
sink over the colder area. The resulting circulation is called a land-sea breeze system. At
night the land cools relatively to the sea, and the flow reverses. The land-sea breeze is the
best known one of the breeze systems. However, breeze systems occur also as breeze systems
around major cities, because of the “urban heat island” effect. When the spatial scale of a
land-breeze system increase to continental scale, one talk about Monsoons, which will typical
be seasonal rather than diurnal. However breeze systems of many scales can exist at the same
location, and will often interact.
A more complex situation appears for a transition from an unstable ABL passing over a
colder surface, where the more intensive unstable ABL has to decay before the stable IBL can
establish itself. Also, these situations occur quite often in some coastal areas.

44 DTU Wind Energy-E-Report-0029(EN)

Figure 26: Land-sea breeze system, where a hot land results in rising air, while over the relative
cold water air sinks.

Especially the transition from relatively warm unstable/neutral land ABL into a stable
IBL over the water (Melas, 1998). Several things happens. The air gradually looses heat to
the water and reach the water surface temperature at the interface, where a small neutral
layer establishes itself. This neutral layer slowly growths up through the stable IBL, until the
original temperature gradient between the two air masses is concentrated at the top of a new
boundary layer that however can take many hundred kilometers to establish itself.

Figure 27: Warm neutral air flowing over cold water, and develops a stable IBL, SIBL, gradually
transitioning to a neutral ABL with an inversion on the top. Over the water for off-winds.
(Lange et al., 2004). The broken triangular line indicates the possibility for a low level jet at
the transition.

A low level jet may happen in the transition between land and water, because the sur-
face friction in the boundary layer suddenly disappears, for the air crossing the coastal line.
Considering eq. 1, we see that the wind above the boundary layer is unchanged,friction is
unimportant here. The velocity within the ABL must accelerate because the friction is re-
duced, finally the wind has to go to zero at the surface. It is a transitional phenomenon that
gradually disappears as the full profile and stress profile reasserts themselves downstream from
the coast line. These low level jets can appears, wherever the surface stress reduces because of
increasing stability or decreasing roughness or both. It is quite well known over the nighttime
Plains of US, where they are a result of the growing night time stability.
The heat/temperature controlled IBLs may happen also especially over water, when syn-
optic air masses are advected over relatively cold or warm water. Here especially, the second
one, will give rise to a thermally induced convective IBL, which over the water typically will

DTU Wind Energy-E-Report-0029(EN) 45

be associated with mesoscale structures like rolls and cells.
The large scale temperature differences may also influence the basic equation system in
eq. 1. In this system only pressure was allowed to change to ensure a pressure gradient and
thereby a geostrophic wind. If the large scale temperature is allowed to change horizontally,
meaning that Ts and Ta changes together in fig. 11, the simple relation for the geostrophic
wind in eq. 3 changes to:

1 ∂p gz ∂Tl 1 ∂p gz ∂Tl
G = (U1G , U2G ) = (− − , − ) (23)
f ρ ∂y f ρ ∂y f ρ ∂x f ρ ∂x
In eq. 23 the original G is present at the surface only, and the deviations are seen to increase
with z. Also they will influence both the magnitude and the direction of the geostrophic wind
for baroclinic situations. Indeed a low level jet can result from this type of baroclinicity as well
(Garreaud et al., 2005). The subscript l refers to that the temperature must be considered a
layer average. The thermal contribution is often called “The Thermal Wind”.
We finally mention coast line changes that firstly happens in the water, but may still
influence the air above the water: For deep water coasts, currents and wind may carry the
surface water outwards, and upwelling of deeper colder water replaces it closer to the shore.
This means that the flow approaching the land is in the process of cooling and developing a
stable internal boundary layer over the coastal water, as in fig. 27, before making landfall. A
well known example is the Californian west coast upwelling sea- breeze system. For shallow
coasts, on the other hand the coastal water reflects the land surface temperature more,
because the radiation balance reaches all the way to the bottom. Additional wave breaking
may also enhance the mixing through the water of both momentum and heat (Johnson et
al., 1999).

2.8 Complex terrain

With complex terrain, we will understand a terrain composed of randomly steep slopes. As
we discussed in the former section, we can in the principle model flow over terrain composed
of random slopes and roughness changes at least as long as the thermal conditions are
less important, (Troen et al., 1989). But if there are too many steep slopes in an area,
there are no simple models available for the flow computation. For such work a system RIX,
Ruggedness Index (Mortensen et al., 1997) has been developed that compare steepness around
a meteorological station, from which the data are extrapolated to potential wind turbines sites,
with the steepness of the potential sites. Hereby it has been possible to estimate and reduce
the errors of the simple flow models, associated with larger steepness somewhat.
Often the varying terrain will give rise to different temperature fields, and if slopes are
additionally of a height that is comparable or larger than a typical boundary layer height, the
boundary layer can break down into several different boundary layers, at different levels above
ground or in different parts of the terrain. Therefore the diurnal radiation changes will modify
different terrain parts differently. fig. 25 show and example of the diurnal change for a simple
valley.
At night drainage flow of cold air down the mountains dominates, while a valley wind
upslope of the sun heated slopes dominate in the late afternoon. The morning shows upslope
flow on the part of the slopes that is reached by the sun, while down slope drainage wind
prevails in the valley, not yet reached by the sun. The depth and strength of the flows can vary
from deep and strong catabatic flows down stream to high winds upslope, down to shallow
weak breezes in both directions. Obviously such flows can also exhibit sharp gradients in all
direction as well a remarkable unsteadiness. The detailed flow structure will depend both on
the solar input and also on synoptic pressure gradients across the landscape. Again, they
depend on the detailed orientation of the slopes of the landscape. Therefore measurements
must be recommended combined with modeling if detailed knowledge about such flows is
required.

46 DTU Wind Energy-E-Report-0029(EN)

Figure 28: Three dimensional picture of idealized local mountain circulations (a) at night, with
cold air drainage, (b) morning with mixed up-and down slope flows, dependent on sun-rise,
and (c) afternoon up-slop flow along the sun-heated slopes (Stull, 1988).

2.9 Boundary Layer Climatology for Wind energy

Until now we have discussed the processes controlling the different boundary layer parameters
and how the processes and parameters were forced by processes and conditions outside the
ABL. The resulting climate at any location develops from a specific local mix of the processes
and forcings described above, and must be determined either from measurements or integrated
modeling.
It is important to limit the scope of such a climate study, one starts by specifying the
focus, which is here offshore wind energy, meaning that one needs sufficient site specific
environmental data for wind resource estimation and for establish the external conditions for
the design or choice of turbine type. The data base can be established using existing data, new
in situ measurements or modeling- or combinations of all these. There will be considerations
about “Need to know and nice to know” and about the price and project duration for the
different solutions.
For wind resource estimations, data of wind speed and direction distributions at hub height
should be available for at least one year, since a year is the longest simple cycle in the climate.
Preferably are several years, to account for the known inter-annual variability. Nice to know
would be some stability information for comparison with models, since the offshore site is
likely to be in a coastal zone for some wind directions, where conditions are influenced by
the nearby land. However, pressure and temperature and humidity provide also the air density
influencing directly the wind resource.
Next we come to the effort to establish environmental data for design basis and turbine
details. The required statistics are summarized in Bredmose et al. (2012) and detailed in the
Design Standards (2007-11). For load estimation one needs to establish hub-height distribu-
tions of turbulence intensity and shear at the location and estimate extreme values of these
parameters see fig. 15.
For an offshore site, also the wave climate has to be established, both with respect to
wave height and wave direction. This is not to estimate the roughness, but to estimate the
wave loads and combined wind- wave loads on the wind turbine and foundation. The wind
and wave extreme load analysis are formulated in terms of return periods of 1 to 100 years,
utilizing Gumbel statistics (Bredmose et al., 2012), and in terms of joined distribution for
wind and waves.
Some of this information can best be obtained on site, and here the new advances of the
LIDAR technology (Peña et al. , 2012) have made it possible, and economically feasible,
to have in-situ measurements offshore at larger heights than before, corresponding to the
hub-heights today. But aside from the prize of offshore measuring stations, also the necessary
project duration for obtaining the long return period makes it necessary to use other data of
other types, either long term measurements from nearby sites, that can be model translated
to values at the site, or full scale model generated data. This has for some time been used
for wave data, and can now be used on wind and other meteorological data as well, with the

DTU Wind Energy-E-Report-0029(EN) 47

progress of mesoscale meteorological models to compute a sufficiently detailed climatology
based on a climate from reanalysis data ((Hahmann et al., 2012)). Some available global
reanalysis data bases are listed in fig. 29.

Figure 29: Different reanalysis sets, with start time, and horizontal grid size and time resolu-
tion. The vertical resolution corresponds to 20-30 levels.

Stability (temperature and humidity) and other possible data are here again “Nice to know”,
for the same reasons as for the wind resource estimation.

Finally we mention that it is quite normal that wind farm operators keep a meteorology
mast running when the farm has started to operate, both to support the monitoring of the
farm performance, and to support the service and maintenance function.

2.10 Summary
We have all most entirely avoided discussions of the different measuring systems in this pre-
sentation and concentrated on meteorological climatological aspects.

We have summarized knowledge about atmospheric boundary layers, starting with the
simplest ideal form, being statistically stationary and horizontally homogenous. We have
thereafter introduced aspect of reality into the picture, considering stability and unstability,
typical terrestrial boundary layers and marine boundary layers, and the important points of
instationarity and inhomogeneity.
Over land we have seen that stability changes have a clear diurnal cycle, the importance
of which becomes the smaller for the boundary layer structure, the higher the wind speed
and the larger the cloud cover. On the other hand the thermal properties become more and
more important the greater the height within the boundary layer. Over water stability is more
associated with advection of air masses than daily variation of insolation. On the other hand
high wind does not force the stability towards neutral to the same degree as for the terrestrial
boundary layers.
When the surface change, the boundary layer reacts by forming an internal boundary layer.
Orography changes, induce a pressure perturbation that reaches out in all directions within
a range of the same scale as the width the terrain change. The pressure perturbation also
perturbs the oncoming flow. Changes due to roughness and thermal changes on the other
hand diffuse into the flow as it moves across the surface. The transition zone for these changes
will typically be between 10 and 50 kilometers before a new boundary layer establishes itself
as a homogeneous boundary layer. The highest levels in the boundary layer reach equilibrium
the latest. The stable internal boundary layer over water for offshore flow takes the longest
fetch to reach equilibrium up to more than 100 km, meaning for example that an enclosed
sea as the Baltic Sea, with relatively cold water, must be considered entirely coastal.
We find that for complex terrain some general methodologies are possible, but one must
expect to have into invest more specific studies for projects within such areas, before having

48 DTU Wind Energy-E-Report-0029(EN)

sufficient information.
Finally, we compare the two regional seas of the region, the Baltic and the North Sea, in
the light of the discussion of the ABL in the rest of the manuscript. The main differences
are derived from the large scale features of the two seas, The North Sea region with an
additional Gulf Stream heat input, is relatively mild coastal maritime climate. This becomes
more continental as we move east along the Baltic shore, where the Baltic Sea additionally is
fed by colder input. We conclude that there is a high probability of transient stable internal
boundary layers over the Baltic Sea. It would be interesting to see it confirmed by independent
data and analyses, from the measuring station in the region. Wind-wise the southern Baltic
Sea seems very similar to the inner danish waters, but with a larger open high wind area in
the center, where winds get closer to the North Sea level.

Notation
A Coefficient in the resistance laws (6)
an entrainments coefficient in (10), area (15)
coefficient of proportionality in (17)
ABL, ASL Atmospheric Boundary Layer, Atmospheric Surface Layer
α Power in power law wind profile (9), Geostrophic angle (6)
B Another coefficient in the resistance laws (6)
β Charnock’s coefficient as function of wave age in (16)
C Surface heat capacity in (14), coefficient of proportionality in (17)
Cp The heat capacity at constant pressure for air
c Phase velocity of surface water waves (16)
δ Length scale lag-length, height of interfacial layer fig. 17
E Latent heat flux (14)
es : es = es (T ) saturated water vapor pressure in the atmosphere
η detailed surface height (13)
F unspecified function used in (11) and in deriving (10)
F Coriolis paramert (1)
G, U1G , U2G Geostrophic wind (3)
Soil energy flux (14)
g acceleration due to gravity (1)
H Height of hill (Fig 8), turbulent heat flux (14)
h height of ABL, and height of IBL (Figs. 11 and 20)
height of roughness element (15)
IBL Internal Boundary Layer
κ von Kármán constant in Monin-Obukhov turbulence description (6)
< psi(z/L) Stability function for profiles (8)
L Monin-Obukhov stability length scale (7)
width of the hill (fig. 21)
heat of evaporation
Γ The dry adiabatic lapse rate (2)
NR Net radiation (14)
ν, νθ Molecular viscosity and heat conductivity (13)
ρ density of air (1)
p air pressure (1)
q water vapor mixing ratio defined as density of water vapour/air density,
surface value of q (1)
q∗ Turbulence scale for q (7)
S Cross wind area of roughness element (15)
σ Standard deviation
T Temperature in K (1)
T∗ Turbulence scale for T or θ (7)
t time (1)
θ, θ0 Potential temperature in K, surface value of θ , (2)
Ui = u1 , u2 , u3 = u, v, w Three components of the wind velocity
Xi = x1 , x2 , x3 , = x, y, z Three spatial coordinates
z0 , z0T , z0q Roughness length for u, T and q

DTU Wind Energy-E-Report-0029(EN) 49

References
Andersen, O.J. and J. Løvseth (2010) Stability modifications of the Frøya wind spectrum. Journ. Wind
Engeneering Industrial Aerodynamics, 236-242

Bodyko, M.I. (1974) Climate and Life. International Geophysical Series, Vol. 18, Academic Press, New York
18:508

Bredmose, H., S.E.Larsen, D. Matha, A. Rettenmeier, E. Marino, L. Saettran (2012) D2.04: Collation of
offshore Wind Wave Dynamics. MARINET Report, EU-FP7 Grant no. 262552, 50p

Brutsaert, W.H. (1982) Exchange processes at the earth atmosphere interface Engineering Meteorology, Ed.
E. Platte, Elsevier, 319-369

Charnock H. (1955) Wind stress over a water surface. Quart. J. Roy. Meteorol. Soc. 81:639–640

Design Standards, (2007-2011) IEC 61400-3: Wind turbines- Part 3: Design of offshore wind turbines,
2009 DNV-OJ-J101. OFF SHORE STANDARD: Design of offshore wind turbine structures. DNV, 2007.
ABS(American Bureau of Shipping), Offshore Wind Turbine Installation, American Bureau of Shipping,
2010 ABS(American Bureau of Shipping), Design Standards for offshore wind farms, American Bureau of
Shipping, 2011 IEC 61400-1, Design requirements, 2008. IEC 61400-1, Design requirements-amendment1,
2010

Garreaud, R. D. and R. Munoz, 2005. The low level jet off the subtropicalwest coast of South America,
Structure and Variability. Monthly Weather Review 133:2246-2261

Gryning, S.E. and E. Batchvarova, (1990) Analytical model for the growth of the coastal internal boundary
layer during onshore flow. Quart. J. Roy. Met. Soc. 116:187-203

Gryning S.-E., Batchvarova E., Brümmer B., Jørgensen H., and Larsen S. (2007) On the extension of the
wind profile over homogeneous terrain beyond the surface layer. Bound.-Layer Meteorol. 124:251–268

Hahmann, A. N., J. Lange, A. Peña and C. Hasager The NORSEWInD numerical wind atlas for the South
Baltic DTU Wind Energy-E-Report-0011(EN)

Hasager, C.B.; Nielsen, N.,W.; Jensen, N.O.; Bøgh, E.; Christensen, J.H.; Dellwik, E.; Søgaard, G.. (2003)
Effective roughness calculated from satellite-derived land cover maps and hedge-information used in a
weather forecasting model Boundary-Layer Meteorology 109:227-254

Hunt, J.C.R. and J.E.Simpson, (1982) Atmospheric Boundary Layers over Non-Homogeneous terrain Engi-
neering meteorology (E.: E. Plate) Elsevier, Amsterdam 269-318

Håkanson, L. (1991) The Baltic Sea Environment Session 1: Physical Oceanography of the Baltics. The Baltic
University Programme, Uppsala University, Swedem, 35 p

Højstrup, J., (1982) Velocity spectra in the unstable planetary boundary layer J. Atmos. Sci. 39:2239-2248

Jackson, P.S. and J.R.C. Hunt, (1975) Turbulent wind flow over a low hill J. Roy. Met. Soc. 101:929–955

Jones, I S F, Y Volkov, Y Toba, S Larsen, N Huang, M Donelan (2001) Overview. Chapter 1 Wind stress over
the ocean ( Ed. I.S.F.Jones and Y.Toba). Cambridge University Press, 1 - 31.

Johnson, H K, H.J Vested, H. Hersbach, J Højstrup and S E Larsen (1999) The coupling between wind and
waves in the WAM model Journ. Atmos. Ocean. Tech 16(11):1780 - 1790

Kaimal, J.C., J.C. Wyngaard, Y. Izumi, and O.R. Cote, (1972) Spectral characteristics of surface-layer turbu-
lence Quart. J. Roy. Met. Soc., 563–589

Karagali, I,, A. Peña, M. Badger and C.B. Hasager (2012) Wind characteristics of the North Sea and the
Baltic Seas from QuickScatt satellite Wind Energy

Lange, B., S.E.Larsen, J. Højstrup and R. Barthelmie, (2004) The influence of thermal effects on the wind
speed profile of the coastal marine boundary layer Boundary-Lay. Meteorol. 112:587-617

Larsen, S.E. and Jensen, N.O. (1983) Summary and Interpretation of Some Danish Climate Statistics Risø-
R-399, 76 pp

Larsen, S.E., N. E. Tarp-Johansen, S. Frandsen and E. R. Jørgensen (1990) Södra Midsjöbanken Environmental
Data-Initial analysis Risø-I-2505. FP6 Integrated project DOWNVInd

Lettau H. H. (1962) Theoretical wind spirals in the boundary layer of a barotropic atmosphere. Beitr. Phys.
Atmos. 35:195–212

Makin, V. K. (2005) A note on the drag of the sea surface at hurricane winds Boundary Layer Meteorology
115:169-175

50 DTU Wind Energy-E-Report-0029(EN)

Mann, J. (1998) Wind field simulation Prob. Engng. Mech 14(4):269-282

Melas, D., (1998) The depth of the stably stratified internal boundary layer over the sea Geophysical Research
Letters 25(13):2261-2264

Mortensen, N. G. and E.L. Petersen (1997) Influence of topographical input data on the accuracy of wind
flow modeling in complex terrain European Wind Energy Conference, Dublin, Ireland

Olesen, H.R., S.E. Larsen, and J. Højstrup, (1984) Modelling velocity spectra in the lower part of the planetary
boundary layer Boundary-Layer Meteorol. 29:285–312

Peña A., Gryning S.-E., and Hasager C. B. (2008) Measurements and modelling of the wind speed profile in
the marine atmospheric boundary layer. Bound.-Layer Meteorol. 129:479–495

Peña A. and Gryning S.-E. (2008) Charnock’s roughness length model and non-dimensional wind profiles over
the sea. Bound.-Layer Meteorol. 128:191–203

Peña A., Mikkelsen T., Gryning S.-E., Hasager C. B., Hahmann A. N., Badger M., Karagali I., and Courtney
M. (2012) Offshore vertical wind shear. Final report on NORSEWInD’s work task 3.1 DTU Wind Energy-
E-Report-005(EN), DTU Wind Energy, Risø¸ Campus, Roskilde

Smedman, A-S., H. Bergström and B. Grisogono, (1997) Evolution of stable internal boundary layers over a
cold sea Journ. Geophys. Research 102:1091-1099

Stull R. B. (1988) An introduction to boundary layer meteorology, Kluwer Academic Publishers, 666 pp

Tennekes, H. and J.L. Lumley (1972) A first course in turbulence MIT, Cambridge, Mass. 300p.

Troen, I. and E.L. Petersen (1989) European Wind Atlas Risø National Laboratory, Roskilde, Denmark

Wang, H., R. Barthelmie, S.G. pryor, and H-G. Kim (2012) A new Turbulence model for offshore wind
standards Subm. Geophysical Researh

Zilitinkevich, S.S. (1972) On the determination of the height of the Ekman boundary layer Boundary-Layer
Meteorol. 3:141–145

Zilitinkevich, S.S. (1975) Resistance laws and prediction equations for the depth of the planetary boundary
layer J. Atmos. Sci. 32:741–752

Zilitinkevich, S. , I. Mamarella, A. Baklanov and S. Joffre (2009) The effect of Stratefication on the aerody-
namic roughness length Meteorological and Air quality models for urban areas. Springer Berlin

DTU Wind Energy-E-Report-0029(EN) 51

3 Atmospheric turbulence
Jakob Mann
DTU Wind Energy, Risø Campus, Roskilde, Denmark

3.1 Introduction
For many civil engineering structures, including wind turbines, dynamic wind loading caused by
the atmospheric turbulence is a serious concern for the designer. Gust loading on streamlines
bridge decks requires knowledge of the vertical wind field fluctuations not only in one point, but
also how the fluctuations are correlated in space Simiu and Scanlan (1996); Larose and Mann
(1998). Also the horizontal components may be of importance in bridge aerodynamics. For
dynamical load calculations on a wind turbine, for example at an off-shore location knowledge
of all three wind components and their spatial correlations are needed because the gusts are
‘sampled’ in a complicated way by the sweeping blades. Yet other structures such as tension
leg platforms used for extracting oil on deep waters are sensitive to slow variation in the
direction of the wind. Thus various engineering structures are sensitive to various components
of wind fluctuations at a wide range of frequencies and also to the spatial correlations of these
fluctuations.
The spatial structure of turbulence is also important in order to understand how remote
sensing instruments such as lidars measure in a turbulent flow fields. That is because the lidar’s
sampling volume is rather extended and thus very far from the almost point-like measurements
of a ultra-sonic anemometer. The description of how lidars measure turbulence may be found
in Mann et al. (2009) for a pulsed lidar, or in Sjöholm et al. (2009) for a continuous wave
(cw) lidar.
The purpose of this contribution is to model the spectral tensor of neutral atmospheric
surface layer turbulence. The spectral tensor contains all information on spectra, cross-spectra
and coherences, which usually are the input requested by wind engineers. We also want to
devise a general algorithm to simulate three-dimensional fields of all three components of the
wind velocity fluctuations. Such simulations are particular useful for time domain simulations
of gust loading of wind turbines and other structures.
In section 3.3 rapid distortion theory (RDT) is used to estimate the tendency of shear to
make turbulence anisotropic. RDT is a linearization of the Navier–Stokes equations and has
as such limited applicability. The influence of the non-linearity is modeled by postulating some
limit as to how much shear is allowed to make the turbulence anisotropic. This modelling uses
the concept of eddy lifetime. Despite the various assumptions and postulates the tensor model
only contains three adjustable parameters: a length scale describing the size of the energy
containing eddies, a non-dimensional number used in the parametrization of eddy lifetime,
and the third parameter is a measure of the energy dissipation.
These three parameters are estimated by comparing the model to measurements over the
sea in section 3.4. In section 3.5 the model is compared to various widely used wind engineering
spectral formulations. Finally, in section 3.7 the spectral tensor is used in a numerical algorithm
to simulate three-dimensional fields of all three components of the wind vector. This is done
by recasting the Fourier representation of the wind field in the discrete wave-vector space,
i.e. as a trigonometric series, where the statistics of the random coefficients are determined
by the spectral tensor. The method is considerably simpler, faster and in some aspects more
physical than many other currently used simulation algorithms. The method is now used in
bridge aerodynamics and in load calculations on wind turbines.
Much of the material presented here has previously been reported in Mann (1994, 1998),
and more details on many aspects may be found in these papers. Newer comparison with
neutral atmospheric data taken from Risø’s test station Høvsøre may be found in Peña et al.

52 DTU Wind Energy-E-Report-0029(EN)

(2010) and comparison under different atmospheric stabilities are under way.

3.2 Definitions
The atmospheric turbulent velocity field is denoted by ũ(x ), where x = (x, y, z) is a right-
handed coordinate system with the x-axis in the direction of the mean wind field and z as
the vertical axis. The fluctuations around the mean wind, u (x ) = (u1 , u2 , u3 ) = (u, v, w) =
ũ(x ) − (U (z), 0, 0), are assumed to be homogeneous in space, which is often the case in the
horizontal directions but is only a crude approximation in the vertical. Since turbulence over
the sea at high wind speeds is primarily shear-generated, the mean wind field is allowed to
vary as a function of z. Because of homogeneity, the covariance tensor
Rij (r ) = hui (x )uj (x + r )i (24)
is only a function of the separation vector r (h i denotes ensemble averaging).
We shall use Taylor’s frozen turbulence hypothesis to interpret time series as ‘space series’
and to serve as a ‘dispersion relation’ between frequency and wave number (Panofsky and
Dutton, 1984). Therefore, we can suppress the time argument in u .
We only aim at modelling the second order statistics of turbulence, such as variances, cross-
spectra, etc. For simulation purposes the velocity field is otherwise assumed to be Gaussian
(see section 3.7). It is still not clear how much influence the statistics of third order, such as
skewness, has on load calculations.
All second order statistics can be derived from the covariance tensor or its Fourier transform,
the spectral tensor: Z
1
Φij (k ) = Rij (r ) exp(−ik · r )dr , (25)
(2π)3
R R∞ R∞ R∞
where dr ≡ −∞ −∞ −∞ dr1 dr2 dr3 . The spectral tensor is the basis of the Fourier
simulation in section 3.7.
The stochastic velocity field can be represented in terms of a generalized stochastic Fourier-
Stieltjes integral: Z
u (x ) = eik ·x dZ (k ), (26)

where the integration is over all wave number space. The orthogonal process Z is connected
to the spectral tensor by
hdZi∗ (k )dZj (k )i = Φij (k )dk1 dk2 dk3 , (27)
∗
which is valid for infinitely small dki and where denotes complex conjugation (Batchelor,
1953).
Is it very difficult to measure the spectral tensor directly. Instead cross-spectra, defined as
Z ∞
1
χij (k1 , ∆y, ∆z) = Rij (x, ∆y, ∆z)e−ik1 x dx (28)
2π −∞
are often measured, say by two instruments separated by ∆y in the horizontal direction
perpendicular to the wind and ∆z in the vertical, and are used in practical applications. The
connection between the components of the spectral tensor and the cross-spectra is
Z ∞Z ∞
χij (k1 , ∆y, ∆z) = Φij (k )ei(k2 ∆y+k3 ∆z) dk2 dk3 . (29)
−∞ −∞
When the two indices i and j are the same and ∆y = ∆z = 0 Eq. (29) becomes the one-point
spectrum Fi (k1 ) = χii (k1 , 0, 0). This definition implies that spectra are two-sided, i.e. we get
the variance by integrating from −∞ to ∞. This convention is used throughout this chapter.
To distinguish between spectra as functions of wave number k1 (= 2πf /U ) and frequency
f we use F for the former and S for the latter, i.e. Si (f )df = Fi (k)dk . The coherence is
defined as
|χij (k1 , ∆y, ∆z)|2
cohij (k1 , ∆y, ∆x) = , (30)
Fi (k1 )Fj (k1 )
which can be interpreted as a normalized cross-spectrum.

DTU Wind Energy-E-Report-0029(EN) 53

3.3 Rapid distortion theory
The incompressible Navier-Stokes equation may be written as
Du 1
+ u · ∇ U = − ∇ p + non-lin. and viscous terms, (31)
Dt ρ
where p is the pressure, and D/Dt ≡ ∂/∂t + U · ∇ is the ‘average Lagrangian derivative.’
Assuming a linear shear (∇ U constant), taking the curl, and dropping the non-linear and
viscous terms we get
Dω
= Ω·∇ u +ω ·∇ U, (32)
Dt
where Ω and ω are the mean and the fluctuating part of the vorticity. It is not at all clear
that this linearization is permissible. For example, it can be shown that if the curl of Eq. (31)
is used to estimate the change in mean square vorticity the non-linear terms will dominate
the linear. However, Hunt and Carruthers (1990) argue that when used for the calculation
of the response of velocity fluctuations (u or Rij ) to a sudden application of a large scale
shearing or straining motion the linearization Eq. (32) is valid.

Figure 30: Interpretation of the interplay of shear and turbulence: Two differently oriented
eddies are followed over three successive times. Shear stretches (along the axis of rotation)
and speeds up the upper eddy while the lower eddy is compressed and slowed down.

Physically, the last term on the right hand side of Eq. (32) may be interpreted as the
stretching of vorticity by the mean shear (see Figure 30). The first term is a distortion of the
mean vorticity by velocity fluctuations.
In order to solve Eq. (32) we have to Fourier transform the equation. In order to do so, it
is important to notice that wave fronts are advected by the mean flow i.e.
dk
= −(∇ U )k . (33)
dt
The solution to this wave front advection equation is
k (t) = exp(−∇ U t)k0 (34)
where exp means the matrix exponential.
For a general linear U Eq. (32) does not have analytic solution. However, for many simple
situations such as unidirectional shear, non-rotational stretching or compression, etc. such
solutions exists (Townsend, 1980).

54 DTU Wind Energy-E-Report-0029(EN)

 To get the velocity field from the vorticity we shall express dZ in terms of dΩ , which is
the Fourier transform of ω defined in parallel to Eq. (26):
ω = ∇ × u ⇒ dΩ = ik × dZ ⇒ −ik × dΩ = k × (k × dZ). (35)
Because of the general identity A × (B × C ) = B(A · C ) − C (A · B) and that k · dZ = 0
we get
k × dΩ
−ik × dΩ = −k 2 dZ ⇒ dZ = i . (36)
k2
We shall re-derive (3.11) in Mann (1994), i.e. set up the equations of motion for
 
0 0 0
∇ U =  0 0 0 . (37)
dU
dz 0 0
In this case  
1 0 0
k (t) = exp(−∇ U t)k0 =  0 1 0  k0 , (38)
− dU
dz t 0 1
in accordance with (3.13) of Mann (1994), and Ω = (0, dU /dz , 0). The equations of motion
Eq. (32) becomes
 
dΩ3
Dk × dZ
= k2 dZ +  0  . (39)
Dβ
0
Taking the cross product with k and adding k̇ × (k × dZ) on both sides we get
Dk 2 dZ Dk Dk × dZ
− = × (k × dZ) + k ×
Dβ Dβ Dβ
 
0
Dk
= × (k × dZ) + k2 k × dZ +  k3  dΩ3 . (40)
Dβ
−k2
Writing this more explicitly we get
 2 
(k1 − k22 − k32 )dZ3 − 2k1 k3 dZ1
Dk 2 dZ
= 2k1 (k2 dZ3 − k3 dZ2 )  (41)
Dβ
0
and using Dk 2 /Dβ = −2k1 k3 from Eq. (38) this can be shown to be equivalent to (3.11) in
Mann (1994).
The differential equations Eq. (41) are easily solved given the initial conditions k (0) =
k0 = (k1 , k2 , k30 ) and dZ (k0 , 0). Instead of time, t, we shall use the non-dimensional time,
β, defined as
dU
β= t. (42)
dz
The solution to Eq. (41) is
 
1 0 ζ1
dZ (k , β) =  0 1 ζ2  dZ (k0 , 0), (43)
0 0 k02 /k 2
where    
k2 k2
ζ1 = C1 − C2 , ζ2 = C1 + C2 (44)
k1 k1
with
βk12 (k02 − 2k30
2
+ βk1 k30 )
C1 = (45)
k (k1 + k22 )
2 2

and " #
1
k2 k02 βk1 (k12 + k22 ) 2
C2 = 2 3 arctan . (46)
(k1 + k22 ) 2 k02 − k30 k1 β
Eqs. (38) and (43) give the temporal evolution of individual Fourier modes.

DTU Wind Energy-E-Report-0029(EN) 55

3.3.1 RDT and surface layer turbulence
In this section we first discuss the connection between RDT and stationary surface-layer
turbulence, then the key concept of eddy lifetime, and finally we combine the different parts to
obtain the spectral tensor model. The theory in the previous section describes how turbulence
react to a sudden and fast application of a linear shear. It is natural to ask what this has to
do with turbulence in the surface layer over the ocean.
If the initial conditions can be represented by the isotropic von Kármán tensor,
E(k)

Φij (k ) = δij k 2 − ki kj , (47)
4πk 4
with the energy spectrum
2 5 (Lk)4
E(k) = αε 3 L 3 17 , (48)
(1 + (Lk)2 ) 6
then the tensor Φij (k , t) will become more and more ‘anisotropic’ with time.
The linearization implied by RDT is unrealistic, and at some point (in time) the stretched
eddies will break up. We postulate that eddies of linear dimension ≈ |k |−1 (or more precisely
the Fourier modes) are stretched by the shear over a time which is proportional to their
lifetime. The lifetime τ is
1 2
τ (k) ∝ ε− 3 k − 3 (49)
pertaining, at least in the inertial subrange, to eddies with wave vector magnitude k = |k |
(Landau & Lifshitz 1987, § 33).
The basic postulate is that the stationary spectral tensor
Φij (k ) ≡ Φij (k , τ (k)) (50)
describes the surface layer turbulence well. The combination of RDT and scale dependent
eddy lifetimes has previously been used by Derbyshire and Hunt (1993).
Maxey (1982) has described a similar model with the exception that the lifetime τ was
assumed to be constant for all wavevectors. (τ dU /dz is called ‘the equilibrium value of the
effective distortion strain’ by Maxey.) Maxey’s model gives a reasonable, but not perfect,
2
description of the ratios between σu2 , σv2 , σw and huwi for turbulent shear flows. There are,
however, two grave drawbacks when the model of Maxey (1982) is used to calculate spectra:
− 35
1. The uw-cross-spectrum in the inertial subrange decays as k1 whereas Wyngaard &
− 73
Coté (1972) observe and give scaling arguments for k1 .
2. For typical values of the effective distortion strain the model predicts Fu /Fw ≈ 7 in the
inertial subrange whereas it should be Fu /Fw = 43 .
The models presented here do not suffer from these shortcomings.

3.3.2 Eddy lifetimes

At scales larger than the inertial subrange Eq. (49) is not necessarily valid. We construct an
alternative model for the ‘eddy lifetime’ assuming that the destruction of an eddy with size
k −1 is mainly due to eddies comparable to or smaller than k −1 . The characteristic velocity
R∞
1
of these eddies may be written as k E(p)dp 2 , and we simply assume the lifetime to be
proportional to the size k −1 divided by this velocity:
Z ∞ − 21
−1
τ (k) ∝ k E(p)dp
k
  − 21  2
2 1 17 4 k− 3 for k → ∞
∝ k− 3 2 F1 , ; ; −(kL)−2 ∝ (51)
3 6 3 k −1 for k → 0
where we have chosen E as the von Kármán energy spectrum Eq. (48) and where 2 F1 is the
hypergeometric function.

56 DTU Wind Energy-E-Report-0029(EN)

 103
Eq. (49)

τ (k) [Arbitrary units]

102 Eq. (51)
Eq. (52)
10 Eq. (53)

1 10
kL

Figure 31: Eddy lifetimes as functions of the magnitude of the wave vector. The lifetimes
given by Eq. (51) give the most realistic results.

Comte-Bellot and Corrsin (1971) give another lifetime model which has the right asymptotic
behaviour for k → ∞, the ‘coherence-destroying diffusion time’ :
Z ∞ − 21
τD (k) ∝ k −2 p−2 E(p)dp
k
  − 21  2
− 32 4 17 7 k− 3 for k → ∞
∝ k 2 F1 , ; ; −(kL)−2 ∝ (52)
3 6 3 k −2 for k → 0
which was constructed as the square of the eddy size divided by a k-dependent ‘turbulent
viscosity’.
Further, the inverse ‘eddy-damping rate’
( 2
3

− 12 k− 3 for k → ∞
τE (k) ∝ k E(k) ∝ 7 (53)
k− 2 for k → 0

is used by Lesieur (1987) in eddy-damped quasi-normal theories of turbulence as a character-

istic non-linear relaxation time.
All lifetime models are shown in Figure 31 normalized such that they coincide in the inertial
subrange. It turns out that σu2 becomes infinite using Eq. (52) or (53), while Eq. (49) and
(51) give reasonable results. It also turns out that the spectra calculated from Eq. (51) fit the
data better than Eq. (49) for which reason Eq. (51) is used in the rest of this chapter. Some
support for Eq. (51) may be found in Panofsky, Larko, Lipschutz, Stone, Bradley, Bowen and
Højstrup (1982) who measured eddy ‘response times’ of eddies in the neutral atmospheric
surface-layer. Also Kristensen and Kirkegaard (1987) were in their theoretical model of the
growth of a puff in a turbulent fluid compelled to use Eq. (51) rather than Eq. (52) or (53).
It is convenient to write Eq. (51) as
 −1   − 21
dU − 32 1 17 4 −2
τ (k) = Γ (kL) 2 F1 , ; ; −(kL) , (54)
dz 3 6 3
where Γ is a parameter to be determined. 1
It should be emphasized that at low wave numbers the assumptions made so far are not
valid. F.ex. the assumptions of linear shear is only valid over small distances, i.e. for large k.
Similarly, homogeneity is a dubious assumption for large vertical separations. Finally, despite
1 Keith Wilson has reformulated this expression in terms of the incomplete beta function.

DTU Wind Energy-E-Report-0029(EN) 57

talking about eddy lifetimes, there is no real modelling of the decay process, because there is
no equation describing the non-linear transfer of energy among various wave vectors.
In an attempt to relax the assumption of vertical homogeneity Mann (1994) modelled
the influence of the blocking of the surface in addition to shear. This gave slightly better
coherence predictions than the present model, but greatly complicated the mathematics and
had also other negative consequences.

3.3.3 The uniform shear model

To make a stationary model we use Eqs. (54) and (50) discussed in the beginning of this
section, i.e. we substitute t with τ given by Eq. (54). For the 33-component we get
k04 E(k0 ) 2
Φ33 (k ) = Φiso
33 (k0 ) = (k + k22 ), (55)
k4 4πk 4 1
where Φiso
33 refers to the isotropic von Kármán tensor and E to the energy spectrum Eq. (48).
The other components become
E(k0 ) 2

Φ11 (k ) = 4 k0 − k12 − 2k1 k30 ζ1 + (k12 + k22 )ζ12 (56)
4πk0
E(k0 ) 2

Φ22 (k ) = 4 k0 − k22 − 2k2 k30 ζ2 + (k12 + k22 )ζ22 (57)
4πk0
E(k0 )

Φ12 (k ) = −k1 k2 − k1 k30 ζ2 − k2 k30 ζ1 + (k12 + k22 )ζ1 ζ2 (58)
4πk04
E(k0 )

Φ13 (k ) = 2 2
−k1 k30 + (k12 + k22 )ζ1 (59)
4πk0 k
and
E(k0 )

Φ23 (k ) = 2 2
−k2 k30 + (k12 + k22 )ζ2 . (60)
4πk0 k
Eqs. (55)–(60) with Eq. (54) constitute the uniform shear model (US).
These equations have two differences from the expressions of Townsend (1976) for plane
shearing of homogeneous turbulence. The first is the elimination of time by Eq. (54) and
the second and related difference is that we do not use the turbulent viscosity of Townsend,
which would make the decay time for all eddies equal, independent of their sizes.

3.4 Fitting spectra to observations

First the uncertainties on estimated spectra are discussed. Theses are either caused by vari-
ations in atmospheric stability, which persists even at high wind speeds (> 16 m s−1 ) over
water, or by statistical variations. Secondly, the measured neutral spectra are fitted to the
spectral tensor model. Based on this fit the coherences are finally predicted and compared to
the measurements.

3.4.1 Uncertainties on spectra

Often spectra are averaged over, say, n consecutive frequencies or wave numbers to decrease
the random error of the estimate. Alternatively, the time series could be divided into n
segments of equal duration. Each segment is then Fourier transformed and the spectrum
determined as the average of the absolute square of these Fourier transforms. For either
definition the statistical uncertainty on spectral density F calculated from a stationary time
series is (under the assumption that the time series is long compared to the time scale of the
process)
σ(F ) 1
= 1 (61)
hF i n2

58 DTU Wind Energy-E-Report-0029(EN)

(Koopmans, 1974; Bendat and Piersol, 1986).
Figure 32 shows the result of an analysis of 14 two-hour time series from the Great Belt.
The series have mean speeds U between 16 and 20 m s−1 and the mean directions are within
a narrow range around south where there is an uninterrupted fetch over water for at least
20 km.
Assuming the stability to be neutral, the variation of spectral densities should obey Eq.
(61) and the standard deviation at the lowest wavenumbers should be around 25% and 5%
at k1 = 0.1 m−1 . The observed rms variations are clearly larger, at least 50% at the lowest
frequencies and maybe 20% at higher frequencies. Most noticeably, there are spectra with
only 10% of the spectral density of the others.
This variation is due to the stability of the atmosphere not being neutral. The case with
suppressed turbulence is slightly stable and has U = 16 m s−1 . From the point of view of
aerodynamic loads this may imply enhanced loads on a bridge deck. While the buffeting loads
are smaller the loads from vortex shedding can be much larger. Usually vortex shedding from
a bridge deck is suppressed or even destroyed by the turbulence in the atmosphere, but if
turbulence is absent as in a stably stratified atmosphere (e.g. warm air flowing out over a
cold sea) the vortex shedding might be strong. Stable stratification might also alter loads on
off-shore wind turbines because of increased shear.
Unstable stratification also alters the spectrum. Though none of the spectra from the Great
Belt are obtained under very unstable situations, an analysis of unstable, high-wind spectra
on the west coast of Norway indicate that the spectra are mainly enhanced (by more than
100%) at very low frequencies (f < 0.02 Hz). These might be relevant for various off-shore
production units (Mann, 1992).

4 × 10−4
k1 Fw (k1 )
U2

2 × 10−4

10−3 10−2 0.1 1

Wavenumber k1 (m−1 )

Figure 32: Spectra of w from the Great Belt Coherence Experiment. Mean wind speeds are
between 16 and 20 m s−1 and directions are in a narrow interval around the South. Dashed
spectra have slightly unstable stratification, gray have stable, and the thin have neutral.

3.4.2 Spectral fitting and prediction of coherences

In order to conduct simultaneous measurements of spectra and coherence over the sea a 70 m
high mast was erected 40 m from an existing mast on the easterly spit of Sprogø, an island in
the midst of the Great Belt separating the two Danish islands Funen and Zealand. A 15 m long
horizontal boom was mounted symmetrically at the top of the new mast so that the whole
construction has the form of a letter “T”. A Kaijo-Denki DAT-300 omni-directional sonic

DTU Wind Energy-E-Report-0029(EN) 59

 0.6

Spectral density (×103 ) k1 F (k1 )/U 2 0.4 Fu

Fv

0.2
Fw

ℑ(χuw )

ℜ(χuw )

0.001 0.01 0.1 1.

Wave number k1 [m−1 ]

Figure 33: Average u-, v-, w-, and cross-spectra of all the neutral runs present in Figure 32.
The ragged curves are measurements while the smooth are the model spectra. The model
has zero imaginary part of the cross-spectrum (quadrature spectrum).

Figure 34: The mast array on Sprogø viewed from SSE. The tiny dots at the top of the masts
are the omni-directional sonic anemometers.

60 DTU Wind Energy-E-Report-0029(EN)

anemometer was installed at each end of the boom and at the top of the old mast, providing
15.0, 32.5 and 47.5 m horizontal separations between the three co-linear instruments. The
mast array is shown in Figure 34. More details about the experiment including correction for
flow distortion by the sonic anemometers may be found in Mann et al. (1991).
The measured spectra shown in Figure 33 are an average of 16 neutral two hour runs
with wind speeds between 16 and 20 m s−1 . The smooth curves are model spectra derived
from the spectral tensor model with the parameters Γ = 3.2, L = 61 m, and αε2/3 /U 2 =
1.810−4 m−2/3 , which are taken from Mann (1994), who used fewer two hour runs but slightly
higher wind speeds.
uu vv ww
1.0
∆y = 15 m

0.5

1.0
∆y = 32.5 m
coh(k1 )

0.5

1.0
∆y = 47.5 m

0.5

1 2 1 2 1 2
k1 ∆y

Figure 35: The dots are measured coherences from the same set of data as used for Figure
33 for various horizontal separations ∆y and for all three velocity components. The lines are
the coherences predicted by the model.

These parameters are in turn used to predict the coherences as shown in Figure 35. As seen
from this figure the predictions agree well with the measurements except for the w coherence,
especially at the largest separation.

3.5 Model spectra over the ocean and flat land

Here we compare the tensor model of section 3.3.1 to spectra and coherences from the
literature. We will not give an exhaustive review of spectral models but select a few modern
models which the author believes is used in wind engineering. The purpose is to estimate
the parameters Γ , L and αε2/3 for a given mean wind speed U and height above the water
surface z.
The logarithmic mean wind profile defines the roughness length:
u∗
U (z) = ln(z/z0 ), (62)
κ
where u∗ ≡ (− huwi)1/2z→0 is the friction velocity and κ = 0.40 the von Kármán constant
(Landau and Lifshitz, 1987; Panofsky and Dutton, 1984).
ESDU International (1982) gives a slightly more accurate wind profile:
u∗
U (z) = (ln(z/z0 ) + 34.5f z/u∗) (63)
κ
with the Coriolis parameter f ≡ 2Ω sin φ, where Ω is the angular velocity in rad s−1 of the
Earth and φ the geographical latitude. The profile Eq. (63) is valid up to z = 300 m, below

DTU Wind Energy-E-Report-0029(EN) 61

 3

1000CDN
1

0
10 20 30
U (10 m) [m s−1 ]

Figure 36: The neutral drag coefficient CDN as a function of mean wind speed at z = 10 m.
The broad line is from Charnock’s relation Eqs. (64) and (62). The thin lines are empirical
relations from Geernaert (1987) and the dotted line is from NDP (1998), see Eq. (73).

30 m Eq. (62) is a good approximation to Eq. (63). Throughout this comparison we use
f = 10−4 s−1 .
Charnock (1955) argued that over the sea the roughness length is related to g = 9.8 m s−2
the acceleration due to gravity and the friction velocity by
u2∗
z0 = A (64)
g
where A, the Charnock constant, must be determined experimentally. On basis of an extensive
literature study of ocean data Garratt (1977) found that the best fit of Eq. (64) is A = 0.0144.
A slightly newer value is given by ESDU International (1982):
A = 0.0167, (65)
which will be used here. Over the ocean the neutral drag coefficient
 2
u∗
CDN = (66)
U (10 m)
increases monotonicly with U as can be seen by solving Eqs. (64) and (62). This is shown
in Figure 36 as a broad line together with several recent empirical relations. The figure gives
a good impression of the uncertainty in estimates of drag coefficients. Among the various
reasons for this variability are atmospheric stability, surface currents, ‘wave age’, length of
the fetch over water, and water depth (Garratt, 1977; Geernaert, 1987; Brown and Swail,
1991). The spectral density of velocity fluctuations is in general proportional to the drag
coefficient so the uncertainty of the former is probably of the same order of the latter.

3.5.1 Code and textbook spectra

Surface layer scaling is used in many spectral models, implying that length scales are propor-
tional to z and that variances are proportional to u2∗ . Therefore, it is convenient to normalize
the spectra with u2∗ and present them as functions of either n ≡ f z/U or k1 z. All spectra in
R −∞
this paper are ‘two-sided’ implying ∞ S(f )df is equal to the variance2 .
The spectra of Kaimal are (Kaimal et al., 1972; Kaimal and Finnigan, 1994)
f Su (f ) k1 Fu (k1 ) 52.5n
= = , (67)
u2∗ u2∗ (1 + 33n)5/3
f Sv (f ) 8.5n
2
= , (68)
u∗ (1 + 9.5n)5/3
2 The
R∞
so-called ‘one-sided’ spectra, where 0 S(f )df is equal to the variance, are probably more commonly
used.

62 DTU Wind Energy-E-Report-0029(EN)

and
f Sw (f ) 1.05n
2
= . (69)
u∗ 1 + 5.3n5/3
Kaimal’s spectra are based on measurements over flat homogeneous terrain in Kansas.

5
ESDU
u Kaimal
Simiu
Højstrup
NPD
1

0.5

0.1
5
v ESDU

Kaimal
k1 F (k1 ) [m2 s−2 ]

Simiu

0.5

0.1
5
w ESDU
Kaimal

Simiu

0.5

0.1
0.0001 0.001 0.01 0.1 1
−1
Wave number k1 [m ]

Figure 37: Comparison of spectral models. For the comparison z = 40 m and U = 40 m s−1
(over the sea) is chosen. For u (ESDU International, 1985), Eqs. (67), (70), (80), (76) are
used. For v and w (ESDU International, 1985), Eqs. (68) and (71), and (ESDU International,
1985), Eqs. (69) and (72), respectively. Eq. (63) together with Eq. (64) gives u∗ = 1.78 m
s−1 and z0 = 0.0054 m.

The spectra of Simiu and Scanlan (1996) have the same functional shapes as Kaimal’s but
the numerical constants are different:
f Su (f ) 100n
= , (70)
u2∗ (1 + 50n)5/3
f Sv (f ) 7.5n
2
= , (71)
u∗ (1 + 9.5n)5/3
and
f Sw (f ) 1.68n
= . (72)
u2∗ 1 + 10n5/3
Deviations from surface layer scaling are found in the model spectra from ESDU Interna-
tional (1985). Also the spectra of Norwegian Petroleum Directorate NDP (1998) and Højstrup,

DTU Wind Energy-E-Report-0029(EN) 63

Larsen and Madsen (1990) do not obey surface layer scaling, but they are only limited to
u-spectra.
The Engineering Science Data Unit (ESDU) wind profile, spectra and coherences (ESDU
International, 1982, 1985 and 1986) are derived from many sources from all over the world
during several decades. ESDU proposes that the turbulence intensities and length scales in
the surface layer are dependent on mean wind speed. The argument is that the boundary
layer depth increases with increasing wind speed implying larger scales of the turbulence. The
other models, relying on surface layer scaling do not contain any information on the boundary
layer depth and they contain no explicit reference to the mean wind speed. The equations of
ESDU are, compared to all other spectral models discussed here, by far the most complicated.
Therefore we shall not cite them explicitly. The most important input parameters are, as for
the other spectral models, the height above the surface z, and the mean wind speed at some
height. Of less important input is the Coriolis parameter which, as mentioned previously, is
taken to be f = 10−4 s−1 . The models we use are valid for the neutral atmosphere.

u
0.5
v
w
0.2
k1 F (k1 )/u2∗

0.1

0.05

0.02

0.1 1 10 100
k1 z

Figure 38: The ‘sheared spectral tensor’ of section 3.3.1 (curves with dots) fitted to the
models by Simiu and Scanlan Eqs. (70)–(72). The result is given by Eq. (83).

The u-spectrum of NDP (1998) applies to winds over oceans and assumes the drag coef-
ficient to be
CDN = 0.525 × 10−3 (1 + 0.15U10), (73)
√
see Figure 36. Integrating dU /dz = u∗ /(κz) = CDN U10 /(κz) Eq. (73) implies that
 z 
U (z) = U10 1 + C ln (74)
10 m
with
C = 0.0573(1 + 0.15U10 )1/2 (75)
where U10 has to be measured in meters per second. While discussing the NPD spectrum we
also assume the unit of z to be meter, f is Hz and Su is m2 s−2 Hz−1 . The spectral density
of the longitudinal wind component is
 2  
U10 z 0.45
160
10 10
Su (f ) =   3n
5 (76)
1 + f˜n

with
 z 2/3  U −3/4
10
f˜ = 172f (77)
10 10

64 DTU Wind Energy-E-Report-0029(EN)

and n = 0.468. This spectrum implies that the variance
2.75
U10
σu2 = 0.00309
0.217
(78)
z
will decrease with height and not constant as implied by surface layer scaling. Furthermore,
the integral length scale
1/4
length scale ∝ z 2/3 U10 (79)
will not be proportional with height but will grow somewhat slower and it will also increase a
little with wind speed. This is not consistent with surface layer scaling where it under neutral
conditions is constant with wind speed.
Højstrup et al. (1990) suggested that spectra at low frequencies do not obey surface
layer scaling because the low frequency part scales with the height of the boundary layer,
not z. To verify their model they used data selected for neutrality and high wind speeds
(11 < U < 23 m s−1 ) from both over sea and land sites in Denmark. The u-model is3
!
f Su (f ) 2.5nt 52.5n 1
= + (80)
u2∗ 1 + 2.2nt
5/3 (1 + 33n) 5/3 1 + 7.4(z/A) 2/3

where the ‘neutral length scale’ A = 3000 m and nt = f A/U . The second term in the
parenthesis is the Kaimal spectrum Eq. (67).
All spectral models are compared in Figure 37 for a specific choice of U and z. Generally,
ESDU has larger length scales compared to those by Kaimal and by Simiu & Scanlan, which
are similar. NPD and Højstrup support ESDU’s large u-scale. ESDU, though, has the most
peaked spectra and, at high wave numbers, slightly lower spectral densities. All spectra agree
fairly well at high wave numbers but have substantial scatter at low wave numbers.

3.6 Comparison with the spectral tensor model

u
2
v
1 w
k1 F (k1 ) [m2 s−2 ]

0.2

0.1

0.001 0.01 0.1 1

−1
Wave number k1 [m ]

Figure 39: Example with z = 40 m and U = 40 m s−1 of the fit of the spectral tensor model
(curves with dots) to the ESDU models.

Here we fit the spectral tensor of section 3.3.1 to models that describe all three component
spectra, namely the ones by Kaimal, Simiu & Scanlan and ESDU.
We obtain the parameters Γ , L and αε2/3 by making a simultaneous least squares fit to
the u-, v- and w-model spectra for wave numbers in the range 0.05 < k1 L < 100. For the
3 Højstrup, Larsen and Madsen (1990) also gives a model for the v spectrum, but it was never compared

with data, so it will not be discussed here.

DTU Wind Energy-E-Report-0029(EN) 65

Kaimal spectra we get
Γ = 3.9
L = 0.59z (81)
u2∗
αε2/3 = 3.2 2/3 ,
z
where the dependence on z is a consequence of surface layer scaling. For the Simiu & Scanlan
spectra, where the fit is shown in Figure 38, we get
Γ = 3.8
L = 0.79z (82)
u2∗
αε2/3 = 2.8
z 2/3
and for both models u∗ can be obtained from Figure 36.
200
300
180
120
80

100
40
z [m]
25
50
L [m]

30 15

20
10
15 20
25
10 30
40
10 50
U [m/s] 60
70
5
80

0.01 0.1 1 10
αε2/3 [m4/3 s−2 ]

5
U
60 70
80
50
4 40

30

3 25
Γ

20
2
10 m/s
1 15

100 200 300

z [m]

Figure 40: The parameters of the spectral tensor model derived from fits to the ESDU model
spectra for turbulence over the sea. Given U and z, all three parameters can be extracted
from these plots.

It is more complicated to get the parameters from the ESDU models because the spectra no
longer depend on U and z in a simple way. For each set {U, z}, a fit to the tensor model has

66 DTU Wind Energy-E-Report-0029(EN)

to be calculated. We do that on a mesh limited by 10 < U < 80 m s−1 , 5 < z < 300 m over
the sea. The result is shown in Figure 40. As an example of use of these graphs, suppose that
the parameters for U (z = 80 m) = 20 m s−1 are wanted. From the upper plot of Figure 40
we get L = 33 m and αε2/3 = 0.1 m4/3 s−2 . The lower plot gives Γ = 4.5.

Table 4: Parameters of the spectral tensor derived from different sources for U (40 m) =
40 m s−1 at sea.

Γ L [m] αε2/3 [m4/3 s−2 ]

Great Belt 3.2 35 0.79
Kaimal 3.9 24 0.86
Simiu 3.8 31 0.76
ESDU 4.5 66 0.62

Another example is shown in Table 4 where the Great Belt data from Mann (1994) are
extrapolated using neutral surface layer scaling to U (40 m) = 40 m s−1 . The spectral fit for
these values of U and z is shown in Figure 39.
Literature coherences and coherences derived from the spectral tensor by Eqs. (29) and
(30) are compared in Mann (1998). Generally, the agreement is good.

3.7 Wind field simulation

Having discussed the spectral tensor in relation to commonly used literature spectra we now
describe how to simulate a velocity field u (x ), which can be used for calculating loads on
engineering structures. We approximate the integral Eq. (26) by a discrete Fourier series:
X
ui (x ) = eik ·x Cij (k )nj (k ), (83)
k
P
where the l’th component of x is xl = n∆Ll with n = 1, ..., Nl . The symbol k de-
notes the sum over all wave vectors k with components ki = m2π/Li , with the integer
m = −Ni /2, ..., Ni /2, nj (k ) are independent Gaussian stochastic complex variables with
unit variance and Cij (k ) are coefficients to be determined. The great advantage of Eq. (83)
is that, once the coefficients are known, it can be evaluated very fast by the fast Fourier
transform (FFT) (Shinozuka and Deodatis, 1991).
Solving Eq. (83) we get approximately (Mann, 1998)
Z
1
Cij (k )nj (k ) = ui (x )e−ik ·x dx , (84)
V (B) B
R
where V (B) = L1 L2 L3 is the volume of B and B dx means integration over the box B.
From Eq. (84) it is easy to see that nj (k ) have to be Gaussian when ui (x ) is a Gaussian field.
Many authors relax this constraint and let nj (k ) have random phase but a fixed absolute
value (Shinozuka and Jan, 1972; Shinozuka and Deodatis, 1991, 1996). Using this approach
every sample will get exactly the same variance and, given a wavenumber (or vector), the
estimated power spectral density at this wavenumber will be the same for all realizations of
the same process. This might be advantageous in some situations, but it is in contrast to
power spectral density estimates of stationary time series which have 100% rms (Press et al.,
1992; Bendat and Piersol, 1986). The difference between the two approaches is discussed in
detail in Grigoriu (1993). In practice there is little difference and both models could be used.
However, the Gaussian approach is usually easier to analyze theoretically and we shall stick
to that here.

DTU Wind Energy-E-Report-0029(EN) 67

 To find the coefficients Cij (k ) we calculate the covariance tensor of Eq. (84) obtaining
∗
Cik (k )Cjk (k )
Z Z
1 ′
= hui (x )uj (x ′ )i eik ·x e−ik ·x dx dx ′ (85)
V 2 (B) B B
Z Z
1 ′
= 2
Rij (x − x ′ )1B (x )1B (x ′ )eik ·(x −x ) dx dx ′ ,
V (B)
where 1B (x ) = 1 if x ∈ B and 0 otherwise. Using the change of variables r = x − x ′ and
s = x + x ′ having the Jacobian |∂(r , s)/∂(x , x ′ )| = 8 we get
Z Z    
1 −ik ·r s +r s −r
Cik (k )Ckj (k ) = Rij (r )e 1B 1B dsdr (86)
8V 2 (B) 2 2
The inner integration can be carried out according to
 3
Z 
s +r
 
s −r
  Y

2(Ll − |rl |) for |rl | < Ll for all l
1B 1B ds = (87)
2 2 
 l=1
0 otherwise
so, using the convolution theorem and noting that the Fourier transform of L−|r| (for |r| < L
and else 0) is L2 sinc2 (kL/2), we get
Z 3
Y  
∗ ′ 2 (kl − kl′ )Ll
Cik (k )Cjk (k ) = Φij (k ) sinc dk ′ , (88)
2
l=1

where sinc x ≡ (sin x)/x. For Ll ≫ L, the sinc2 -function is ‘delta-function-like’, in the sense
that it vanishes away from kl much faster than any change in Φij , and the area beneath the
sinc2 -curve is 2π/Ll . Therefore, we get
∗ (2π)3
Cik (k )Cjk (k ) = Φij (k ). (89)
V (B)
The solution to Eq. (89) is
(2π)3/2 1/2
Cij (k ) = Aij (k ) = (∆k1 ∆k2 ∆k3 ) Aij (k ) (90)
V (B)1/2
with A∗ik Ajk = Φij and ∆kl = 2π/Ll . This result should be expected when comparing Eq.
(26) to (83).
Two problems occur by simulating a field by the Fourier series Eq. (83) with the coefficients
Eq. (90). The first is that for many applications the dimensions of the simulated box of
turbulence need not to be much larger than the length scale of the turbulence model L.
Therefore Eq. (89) may not be a good approximation to Eq. (88). The second problem is
that the simulated velocity field Eq. (83) is periodic in all three directions. Both problems
have been addressed in Mann (1998).
The algorithms above simulate a three-dimensional vector field on a three-dimensional
domain, but it can easily be modified to simulate one- or two-dimensional vectors in a 2- or
3-D domain (Mann, 1998). The algorithms are not needed for a one-dimensional domain, i.e.
simulation of wind fluctuations in one point as a function of time.
The implementation of the model includes three steps:
1. Evaluate the coefficients Cij (k ), either by Eq. (90) or a modification of this (Mann,
1998).
2. Simulate the Gaussian variable nj (k ) and multiply.
3. Calculate ui (x ) from Eq. (83) by FFT.
The time consumption in the first step is proportional to the total number of points N =
N1 N2 N3 in the simulation. The required time to perform the FFT is O(N log2 N ) (Press
et al., 1992). In practice, simulating a three-dimensional field, used for load calculations on
wind turbines, with millions of velocity vectors takes of the order of a few minutes on a
modern pc.

68 DTU Wind Energy-E-Report-0029(EN)

Notation
A Charnock constant
neutral length scale
B box of e.g. turbulence
cohij coherence or normalized cross-spectrum
cw continuous wave
Cij coefficients for a discrete Fourier series
CDN neutral drag coefficient
E(k) energy spectrum
ESDU engineering science data unit
f frequency
Coriolis parameter
F spectrum (function of wave number)
FFT fast Fourier transform
g gravitational acceleration
i tensor index
j tensor index
k wave number/component
k wave vector
L turbulence length scale or length of a volume
m integer number
n number of segments
normalized frequency
nj (k) Gaussian variable
nt normalized neutral frequency
N number of points
NPD Norwegian Petroleum Directorate
p pressure
r separation vector
Rij (r) covariance tensor
RDT rapid distortion theory
S spectrum (function of frequency)
t time
u longitudinal wind component
ũ(x ) turbulent velocity field
u∗ friction velocity
U mean wind speed
U mean flow
US uniform shear model
v transversal wind component
V (B) volume of the box B
w vertical wind component
x direction of the mean wind field
x coordinate system
y transversal direction
z vertical axis direction or height
z0 roughness length
Z orthogonal process
αε energy dissipation measure
β non-dimensional time
Γ anisotropy parameter
δ Kronecker delta
κ von Kármán constant
σX standard deviation of a variable X
τ eddy lifetime
τD coherence-destroying diffusion time
τE eddy-damping rate
φ geographical latitude
Φij (k) spectral tensor
χij cross-spectra
ω fluctuating part of the vorticity

DTU Wind Energy-E-Report-0029(EN) 69

 Ω angular velocity
Ω mean part of the vorticity
∂ partial derivate
∇ vector differential operator
hXi ensemble average of the variable X

References
Batchelor G. K. (1953) The theory of homogeneous turbulence, Cambridge University

Bendat J. S. and Piersol A. G. (1986) Random data: Analysis and Measurement Procedures, 2 edn, Wiley-
Interscience

Brown R. D. and Swail, V. R. (1991) Over-water gust factors. Ocean Engng. 18(4):363–394

Charnock H. (1955) Wind stress on a water surface. Q. J. Royal Meteorol. Soc. 81:639–640

Comte-Bellot G. and Corrsin S. (1971) Simple Eulerian time correlation of full- and narrow-band velocity
signals in grid generated, ‘isotropic’ turbulence. J. Fluid Mech. 48:273–337

Derbyshire S. H. and Hunt J. C. R. (1993) Structure of turbulence in stably stratified atmospheric boundary
layers; comparison of large eddy simulations and theoretical models. in S. D. Moobs and J. C. King (eds),
Waves and Turbulence in Stably Stratified Flows, Clarendon, Oxford, 23–59 pp

ESDU International (1982) Characteristics of wind speed in the lower layers of the atmosphere near the
ground: strong winds (neutral atmosphere), ESDU International, London

ESDU International (1985) Characteristics of atmospheric turbulence near the ground. Part II: single point
data for strong winds (neutral atmosphere), ESDU International, London

Garratt J. R. (1977) Review of drag coefficients over oceans and continents, Mon. Wea. Rev. 105:915–929

Geernaert G. L. (1987) On the importance of the drag coefficient in air-sea interactions. Dynamics Atmos.
Oceans 11:19–38

Grigoriu M. (1993) On the spectral representation method in simulation. Prob. Eng. Mech. 8:75–90

Højstrup J., Larsen S. E. and Madsen P. H. (1990) Power spectra of horizontal wind components in the neutral
atmospheric boundary layer. in N. O. Jensen, L. Kristensen and S. E. Larsen (eds), Ninth Symposium on
Turbulence and Diffusion, American Meteorological Society, 305–308 pp

Hunt J. C. R. and Carruthers D. J. (1990) Rapid distortion theory and the ‘problems’ of turbulence. J. Fluid
Mech. 212:497–532

Kaimal J. C. and Finnigan J. J. (1994) Atmospheric Boundary Layer Flows, Their Structure and Measurement,
Oxford University Press, New York

Kaimal J. C., Wyngaard J. C., Izumi Y. and Coté O. R. (1972) Spectral charcteristics of surface-layer turbu-
lence, Q. J. R. Meteorol. Soc. 98:563–589

Koopmans L. H. (1974) The Spectral Analysis of Time Series, Academic Press

Kristensen L. and Kirkegaard P. (1987) Puff kinematics, Technical Report R–548, Risø National Laboratory

Landau L. D. and Lifshitz E. M. (1987) Fluid Mechanics, Pergamon Press

Larose G. L. and Mann J. (1998) Gust loading on streamlined bridge decks, J. Fluids Structures 12(5):511–536

Lesieur M. (1987) Turbulence in fluids, Martinus Nijhoff Publishers

Mann J. (1992) Investigation of atmospheric low-frequency turbulence over the ocean, Technical Report
I–634(EN), Risø National Laboratoty, Roskilde

Mann J. (1994) The spatial structure of neutral atmospheric surface-layer turbulence. J. Fluid Mech. 273:141–
168

Mann J. (1998) Wind field simulation, Prob. Engng. Mech. 13(4):269–282

Mann J., Cariou J.-P., Courtney M. S., Parmentier R., Mikkelsen T., Wagner R., Lindelöw P., Sjöholm M.
and Enevoldsen K. (2009) Comparison of 3D turbulence measurements using three staring wind lidars and
a sonic anemometer. Meteorol. Z. 18(2):135–140

Mann J., Kristensen L. and Courtney M. S. (1991) The great belt coherence experiment – a study of atmo-
spheric turbulence over water, Technical Report R–596, Risø National Laboratory, Roskilde

70 DTU Wind Energy-E-Report-0029(EN)

Maxey M. R. (1982) Distortion of turbulence in flows with parallel streamlines. J. Fluid Mech. 124:261–282

NDP (1998) Acts, regulations and provisions for the petroleum activities, Vol. 2,, Norwegian Petroleum
Directorate

Panofsky H. A. and Dutton J. A. (1984) Atmospheric Turbulence, John Wiley & Sons, New York

Panofsky H. A., Larko D., Lipschutz R., Stone G., Bradley E. F., Bowen A. J. and Højstrup J. (1982) Spectra
of velocity components over complex terrain. Q. J. R. Meteorol. Soc. 108:215–230

Peña A., Gryning S.-E., Mann J. and Hasager C. B. (2010) Length scales of the netural wind profile over
homogeneous terrain J. Appl. Meteor. Climat. 49:792–806

Press W. H., Flannery B. P., Teukolsky S. A. and Vetterling W. T. (1992) Numerical Recipes, 2nd edn,
Cambridge University Press

Shinozuka M. and Deodatis G. (1991) Simulation of stochastic processes by spectral representation, Appl.
Mech. Rev. 44(4):191–203

Shinozuka M. and Deodatis G. (1996) Simulation of multi-dimensional gaussian stochastic fields by spectral
representation. Appl. Mech. Rev. 49(1):29–53

Shinozuka M. and Jan C.-M. (1972) Digital simulation of random processes and its applications. J. Sound
Vibration 25(1):111–128

Simiu E. and Scanlan R. H. (1996) Wind Effects on Structures, 3. ed., John Wiley & Sons

Sjöholm M., Mikkelsen T., Mann J., Enevoldsen K. and Courtney M. (2009) Time series analysis of continuous-
wave coherent Doppler lidar wind measurements. Meteorol. Z 18(3):281–287

Townsend A. A. (1976) The Structure of Turbulent Shear Flow, 2nd edn, Cambridge University Press

Townsend, A. A. (1980) The response of sheared turbulence to additional distortion. J. Fluid Mech. 98:171–
191

DTU Wind Energy-E-Report-0029(EN) 71

4 Introduction to continuous-wave
Doppler lidar
Mark Pitter, Chris Slinger and Michael Harris
Zephir Limited, The Old Barns, Fairoaks Farm,
Hollybush, Ledbury, HR8 1EU, U.K.

4.1 Introduction
Remote sensing offers the wind industry an attractive alternative or complement to the tra-
ditional methods for obtaining accurate wind measurements that involve the siting of tall
meteorological masts. Laser anemometry (lidar) has now demonstrated its capacity for re-
source assessment, wind turbine power curve measurement, and turbine mounted deployment
for advance wind speed detection. Widespread acceptance of lidar by the industry has re-
quired that this technique be extensively validated, aiming towards a certifiable and traceable
measurement standard and formal accreditation of lidar methods for different applications in
a range of terrain types. This chapter outlines the lidar measurement process and capabilities
specifically for the case of continuous wave (CW) systems.
Wind lidar systems were first demonstrated in the 1970’s (Jelalian, 1992) and have since
been applied to a wide variety of applications including aviation and meteorology. Although
applications to wind energy were explored in the 1980’s (Hardesty and Weber, 1987; Vaughan
and Forrester, 1989), the lidar systems that existed at that time were too large, complex
and expensive to achieve serious acceptance in the industry. The situation has now changed
dramatically, with rapid growth of the wind generation industry coinciding with development
of a new generation of lidars based on optical fibre and other components that emerged from
the telecommunications boom of the 1990’s. The first all-fibre lidars were demonstrated in
the late 1990’s, and a commercial prototype unit (ZephIR) was mounted on a turbine to
demonstrate wind speed detection in front of the rotor plane in early 2003. A demonstration
of ground-based wind profiling followed shortly afterwards. ZephIR is a CW coherent lidar
system, and this approach was selected as a means to combine simplicity with high sensitivity
at measurement ranges relevant to wind energy resource assessment, and hence achieve a
robust, reliable system at relatively low cost. Following this pioneering work, the pace of
development has accelerated, and lidars have increasingly become an established tool in the
wind industry.
Section 4.2 of this chapter provides an overview of lidar techniques and technology. Dif-
ferent types of lidar system are surveyed, and the generic physical principles underlying their
operation are reviewed. The specific case examined in detail here is that of wind profiling by a
ground-based conically-scanned continuous-wave lidar, which has rapidly become established
as a powerful tool in the wind energy industry, and is exemplified by the ZephIR lidar, initially
developed by QinetiQ, the Natural Power, and now by Zephir Limited.
A number of assumptions must be made in order to extract values of wind speed from raw
lidar data; these are reviewed in section 4.3. The necessary steps that are required during
the end-to-end measurement process in order to arrive at a value of wind speed are detailed
in section 4.4. It is important to understand the potential sources of error and uncertainty,
and these are reviewed and analysed in section 4.6. Section 4.7 examines the important
requirement for lidar calibration and traceability and 4.8 discusses the emerging area of turbine
mounted lidar. Finally, section 4.9 draws together some conclusions and a summary of the
future outlook for lidar in wind energy.

72 DTU Wind Energy-E-Report-0029(EN)

4.2 Basic principles of lidar operation and system descrip-
tion
4.2.1 Brief survey of lidar types
There are many different types of lidar (Jelalian, 1992) and these are capable of performing
a diverse range of tasks (e.g. 3D imaging and range finding, gas species detection, remote
measurement of vibrations). We concern ourselves here specifically with systems for the mea-
surement of wind speed in the atmosphere (Zak, 2003). Such systems fall into two broad
categories: coherent lidar and direct detection lidar. Coherent lidar measures Doppler shifts
by comparing the frequency of backscattered radiation to that of a reference beam via a light
beating process, whereas direct detection lidar (Chanin et al., 1989) performs its frequency-
shift measurements by passing the light through an optical filter, such as a Fabry-Perot etalon.
By operating in the ultra-violet, direct detection lidars can exploit molecular scattering pro-
cesses, guaranteeing signal returns even in very clean air where there is an absence of aerosols.
A further type of direct detection lidar that uses cross-correlation of the backscatter signal
measured at several points to determine wind speed and direction is also emerging, but this
chapter concentrates on systems that use the Doppler shift from aerosols or molecules.
Coherent wind lidar systems can be categorised according to their emission waveform
(pulsed or continuous), waveband (UV, visible, near-IR, far-IR), and their transmit/receive ge-
ometry (monostatic or bistatic). This chapter concentrates specifically on continuous-wave co-
herent monostatic lidar systems that operate in the telecommunications near-IR band around
1.55 µm (Karlsson et al., 2000). At this wavelength, reliable components including optical
fibre and circulators are readily available. Such systems can route the light within the lidar via
fibre cables (creating an “all-fibre” lidar’), rather than using mirrors to direct the beams in
free space. This confers an enormous design advantage, simplifying alignment and improving
robustness and stability. Pulsed all-fibre lidar has also been developed as reported in Pearson
et al. (2002) and is discussed in other chapters.

4.2.2 Principles underlying anemometry by coherent laser radar (CLR)

The principle by which coherent lidar measures the velocity of a target is simple: a beam of
coherent radiation illuminates the target, and a small fraction of the light is backscattered
into a receiver. Motion of the target along the beam direction leads to a change in the light’s
frequency via the Doppler shift: motion towards the lidar brings about a compression of the
wave and an increase in its frequency (a “blue shift”), while movement away stretches the
wave reducing its frequency (a ”red shift”). This frequency shift is accurately measured by
mixing the return signal with a portion of the original beam, and sensing the resulting beats
at the difference frequency on a photodetector. Like the Doppler effect, the beat phenomenon
is perhaps most familiar in the context of acoustics as, for example, when two closely (but
not identically) tuned guitar strings are simultaneously plucked.
The essential features are readily seen in the simplified generic CLR depicted in Figure 41.
In order to illustrate the concept this is drawn as a bistatic system, in which the transmit and
receive optics are separate and distinct. In practice a monostatic geometry is more usual, in
which the transmit and receive paths share common optics, since this is a much more stable
arrangement and allows for far simpler alignment.

4.2.3 Role of local oscillator and range selection by focus

The reference beam, or local oscillator (LO), plays a crucial role in the operation of a CLR
(Sonnenschein and Horrigan, 1971). Firstly, it defines the region of space from which light
must be scattered for efficient detection of the beat signal; radiation from other sources (e.g.
sunlight) is rejected both spatially and spectrally, so that CLR systems are usually completely
immune to the effect of background light. The LO also provides a stable reference frequency

DTU Wind Energy-E-Report-0029(EN) 73

Figure 41: Generic bistatic lidar system. A small fraction of the transmitted light is tapped off
by a beamsplitter to form a reference beam. This is superimposed at a second beamsplitter
with the weak return scattered from moving aerosols. The detector picks up the resulting
beat signal; this undergoes spectral analysis to determine aerosol velocity.

to allow very precise velocity determination; as a consequence the Doppler shift measurement
by a CLR system is inherently calibrated. Finally, the LO amplifies the signal via the beating
process to allow operation at a sensitivity that approaches the shot-noise (or quantum) limit.
This very high sensitivity permits the operation of CLR systems in an unseeded atmosphere,
relying only on detection of weak backscattering from natural aerosols.
CW systems are the simplest form of lidar possessing the advantage of reduced complexity,
and their performance can be tailored closely to the requirements of the wind industry. How-
ever, the overall trade-off between the pulsed and CW options for each specific application
must take into account a number of factors including sensitivity, cost, velocity resolution,
and maximum and minimum ranges. Unlike pulsed lidar systems, which use time of flight
to discriminate between returns from different ranges, a CW lidar achieves operation at a
given range by beam focusing. Wind profiling is achieved by continuously scanning the beam,
focusing in turn at a number of predetermined ranges. For each range, a circular scan is
typically used. The rapid sampling rate inherent to CW lidar permits 1-s “snapshots” of the
flow across the scan disk at each measurement range. Focusing of the lidar beam results in a
Lorentzian spatial weighting function along the beam axis, with the sensitivity peak located
at the beam waist (Sonnenschein and Horrigan, 1971; Karlsson et al., 2000). This function
has a half-width given by the Rayleigh range (the distance from the waist at which the beam
area has doubled).
The beam diameter at the waist increases linearly with range while the Rayleigh range
increases roughly as the square. Hence the effective probe volume varies as the 4th power of
the focus range, and this strong dependence has some implications for the signal statistics
at shorter ranges (Harris et al., 2001b). The minimum range for a CW lidar is very short
with detection possible in principle at zero range, whereas a pulsed system is blinded while
the pulse is leaving the transmitter leading to a minimum range of 10’s of metres, typically
around 40–50 m.

4.2.4 Doppler frequency analysis and signal processing

The stages of signal processing required for CLR wind signals are discussed in section 4.4.
The detector output, containing the beat signal information embedded in broadband noise, is
typically digitised by an analogue-to-digital converter (ADC). Spectral analysis (e.g. by fast
Fourier transform methods) leads to the generation of Doppler spectra. It is usually necessary
to average a number of these spectra in order to improve the signal-to-noise ratio (SNR),
after which the Doppler peak stands clearly above a flat shot-noise floor. A value for the line-
of-sight wind speed can then be computed via a velocity estimation algorithm. This might
calculate, for example, the peak, median or centroid value of the Doppler signal.

74 DTU Wind Energy-E-Report-0029(EN)

4.2.5 Wind profiling in conical scan mode
Since a single lidar measurement only provides the component of wind speed along the beam
direction, it is necessary to scan the direction of the beam in order to generate a measurement
of the wind speed vector. A conical or VAD (velocity-azimuth-display) scan pattern has been
widely used (Banakh et al., 1993) (see Figure 42); as the beam moves, it intercepts the wind
at different angles, thereby building up a series of measurements around a disk of air from
which the wind speed vector can be derived. In uniform flow, a plot of the measured line-of-
sight wind speed (VLOS ) versus scan azimuth angle (φ) takes the form of a cosine wave (or a
rectified cosine for a homodyne lidar system that cannot distinguish the sign of the Doppler
shift). The peak Doppler shifts correspond to measurements when the azimuth scan angle
aligns with the upwind and downwind directions. Doppler shifts close to zero are obtained
when the azimuth angle is perpendicular to the flow.

Figure 42: Conical scan pattern as used for lidar wind profiling. Left: ground based vertical
scanning. The cone half-angle (θ) is typically of order 30◦ . The lidar can operate successfully
even when part of its scan is obscured, e.g. by an adjacent met mast. To build up a wind
profile, the lidar operates in a repeating sequence during which all the heights are interrogated
in series. Right: one of several wind turbine mounted configurations, where the lidar scans
around a horizontal axis, usually pointing into the wind.

4.2.6 Pioneering a revolution: ZephIR lidar

Many different research groups have built and successfully deployed wind lidars over the
past 30 years. However, commercial lidar products have until very recently been available
from only a few companies. In 2003 the UK company QinetiQ (formerly the government-
funded establishment RSRE, later DRA then DERA), launched the first commercial all-fibre
lidar (“ZephIRTM ”) which exploited decades of research in the coherent lidar area. QinetiQ
began a programme to develop a commercial fibre-based lidar in 2001; the resulting ZephIR
product is now an established tool for wind profiling in the wind energy industry. Systems have
been deployed successfully around the world in several demanding applications that illustrate
the flexibility and robustness of the solution. Initial deployment of the ZephIR lidar (March
2003) was on the nacelle of a large (2.3 MW) wind turbine (Figure 43-left frame), remotely
measuring for the first time the wind speed up to 200 m in front of the blades (Harris et al.,
2006, 2007). The lidar consisted of a 19 inch rack unit containing laser source, detector and
signal processing computer, situated in the base of the tower and connected via over 100 m
of electrical and optical fibre cable to the transceiver head mounted on the top of the nacelle.
The lidar system was installed and was fully operational after just a few hours, thus allowing
a demonstration of advance warning of oncoming gusts and providing valuable experience in

DTU Wind Energy-E-Report-0029(EN) 75

practical deployment issues.

Figure 43: Stages of evolution of the ZephIR lidar. The upper left-hand picture (a) shows
the lidar head mounted on the nacelle of a Nordex N-90 wind turbine (March 2003). The
top central picture (b) shows a prototype ground-based wind profiler at the Risø wind energy
test site at Høvsøre, Denmark. The top right picture (c) shows an early ZephIR production
model deployed in the field. The lower pictures show a more recent dual mode ZephIR DM300
deployed on an offshore platform (d) and on a turbine nacelle (e).

The system achieved several weeks of successful operation. It was then converted into a
ground-based scanning unit for wind profiling (Figure 43-upper middle frame). The ground
based system was first trialled in December 2003, and subsequently deployed in numerous
campaigns in the UK, Europe, and other parts of the world. The experience gained through
these trials has built confidence in the robustness and reliability of the core ZephIR design. In
late 2004, work started on a production instrument (Figure 43-upper right frame), designed
to perform autonomous wind profiling measurements at heights up to 200 m (Smith et al.,
2006), primarily for site surveys at proposed wind farm sites. The technology was transferred
to Natural Power in 2007, and subsequent development resulted in the more integrated ZephIR
Z300 system (Figure 43-lower frames), and the dual mode DM300 which can be both turbine
and ground mounted. ZephIRs have logged more than 3.6 million hours of deployment (May
2013 figures) around the world and Zephir Limited was eventually established as a standalone
member of the Fred Olsen Limited group in 2012.

4.3 Lidar measurement process: Assumptions

The following sections discuss generic CW lidar considerations (most of which apply equally
to pulsed systems). Where appropriate, application to the ZephIR lidar is used to provide an
illustrative example.

76 DTU Wind Energy-E-Report-0029(EN)

4.3.1 Behaviour of scattering particles
The lidar signals from which wind speeds are derived originate via backscattering of the
beam by particles in the atmosphere. The exact constitution of these aerosols is generally
unknown, but they are normally assumed to consist of dust, organic matter (e.g. pollen),
soot, or water droplets. Knowledge of the aerosols’ make-up is not a requirement for lidar
wind speed measurement. The particles must provide sufficient backscatter signal for Doppler
analysis and their motion must faithfully follow that of the wind flow. The latter assumption
is very good, since viscous forces are dominant for such small particles. Larger particles, for
which this does not apply, will rapidly fall to ground. Raindrops or snowflakes provide a strong
contribution to the lidar signal. Their gravity induced downward motion can lead to an error
in the vertical component of wind speed (just one parameter of interest, and affected data
can be easily identified and filtered), but the important horizontal component will be correct.
A further excellent assumption is that the return signal is dominated by light generated by
single-scattering events. While it is possible for light to suffer multiple scattering in dense fog,
it is a valid assumption that any effect on the Doppler spectrum is almost always negligible.

4.3.2 Uniformity of flow and backscatter

A least-squares fitting to the azimuthal variation of line-of-sight wind speed allows the deriva-
tion of wind parameters from conical scan data. These parameters pertain to a significant
volume of atmosphere – the signal originates from a disk whose diameter commonly exceeds
100 m, and whose depth along the beam direction can be over 10 m. Except in situations
of strong shear, turbulence or highly complex terrain the wind speed is reasonably uniform
throughout this sampled volume, and the best fit wind parameters are used to indicate the
average values over the volume. In fact, ZephIR data itself can provide a straightforward check
on wind field uniformity since conical scanning provides measurements at many different az-
imuth angles; where the assumptions have broken down, measurements with less certainty
can be flagged.
The contribution to the lidar signal from different regions of the lidar probe volume is
weighted by the value of the atmospheric backscatter coefficient β(π) at each point. The
value of β(π) is typically constant to ∼ 10% throughout the probe volume (Banakh et al.,
1993) except in conditions that lead to stable mist layers, or when the lidar beam is incident
on a low cloud base.

4.3.3 Beam positional accuracy

Lidar scan angle and focus calibration are performed in the laboratory, and these must be
correctly maintained throughout a period of deployment in the field. Obviously errors in the
focus setting would result in wind speed measurement at the wrong height. Careful design
eliminates the risk of uncertainty in the beam focus: thermal expansion, which could change
the length of the transceiver telescope, can be compensated and the position of the focus
mechanism can be automatically checked to provide information on any malfunction.
The lidar must be correctly set up, with the vertical and azimuthal orientation adjusted
appropriately during installation. External to the lidar, it has been established that small-scale
refractive-index atmospheric fluctuations will have negligible effect on the propagation of the
lidar beam (Clifford and Wandzura, 1981; Lading et al., 1984).

4.3.4 Optical and electrical interference sources

The lidar identifies the presence of a wind signal when the power density in the Doppler
spectrum exceeds a threshold level. In the absence of any significant source of spurious signal,
the only mechanism that can lead to such detection events is the backscatter of Doppler-
shifted light into the lidar receiver. Optical interference is extremely unlikely - even when the

DTU Wind Energy-E-Report-0029(EN) 77

lidar points directly at the sun the spectral power density is insufficient to cause a problem,
and the interaction between two adjacent lidars can be neglected. Thorough electromagnetic
screening eliminates the risk of spurious spectral features caused by electrical interference for
any normal deployment.

4.3.5 Time-of-flight considerations

The round-trip time for light interrogating the atmosphere at a height of 100 m is less than 1
µs. On this timescale the ZephIR scanner moves the focused beam a distance of only 300 µm,
and the laser phase drifts by an insignificant amount. The polarisation state of the lidar output
is similarly frozen on this timescale, so no appreciable misalignment of loss of interference
efficiency is incurred.

4.4 End-to-end measurement process for CW Doppler li-

dar
4.4.1 Introduction
The wind parameter measurement process can be divided into a number of sequential steps.
This section describes the overall end-to-end process of the wind speed measurement process
for a CW coherent Doppler lidar wind profiler. Again, where appropriate, the ZephIR lidar is
used as an example.

4.4.2 Transmitter optics

The role of the transmitter is to provide a focused beam at a desired location. This location
can be moved around in space with a combination of (i) altering the focus range and (ii)
passing the beam through a scanning element such as a rotating prism (wedge). Wind profiling
lidars conveniently employ a conical scan with its axis aligned vertically; the cone half-angle is
commonly of order 30◦ (i.e. the beam elevation angle is ∼ 60◦ ). Some turbine mounted lidars
use lower cone angles, the optimum choice depending upon a variety of factors including the
mounting position on the turbine, the rotor diameter and the measurement ranges of most
interest.
In a monostatic CW system, a Doppler-shifted contribution to the signal is generated via
light scattering from any moving part of the atmosphere that the beam illuminates. The
contribution from any point is weighted by the square of the beam’s intensity at that point
(Harris et al., 2001a). Hence it can be shown that focusing of an ideal Gaussian beam
(Siegman, 1986) gives rise to a spatial sensitivity along the beam direction that depends on
the inverse of beam area. It follows that the sensitivity rises to a peak at the beam waist, and
falls symmetrically on either side. There is also a spatial dependence of sensitivity transverse
to the beam, but because the beam is very narrow this is of little interest and can be ignored.
To a good approximation the axial weighting function for a CW monostatic coherent lidar is
given by a Lorentzian function (Sonnenschein and Horrigan, 1971; Karlsson et al., 2000):
Γ/π
F = , (91)
∆2 + Γ2
where ∆ is the distance from the focus position along the beam direction, and Γ is the half-
width of the weighting function to the -3 dB point, i.e. 50% of peak sensitivity. Note that
F has been normalised such that it’s integral from −∞ to ∞ gives unity. To another good
approximation, Γ is given by:
λR2
Γ= , (92)
πA2
where λ is the laser wavelength, here assumed to be the telecommunications wavelength
λ ∼ 1.55 × 10−6 m, R is the distance of the beam focus from the lidar output lens, and A is

78 DTU Wind Energy-E-Report-0029(EN)

the beam radius at the output lens. The beam intensity profile is assumed to be an axially-
symmetric 2D Gaussian; A is calculated for the point at which the intensity has dropped
to 1/e2 of its value at the beam centre. For example, if A takes the value 28 mm (broadly
equivalent to the current production ZephIR) then, at a focus range R of 100 m, Γ has
a value of ∼ 5.5 m, or a probe length (to -3 dB points) of ∼ 11 m. Figure 44 shows the
behaviour of the theoretical sensitivity curves for the two example cases (A = 20 mm, 28 mm)
at several focus heights above ground level. In addition the theoretical curve corresponding
to one of the calibration ranges has been plotted, with experimental calibration data for
comparison. (Section 4.7 contains more detail of the calibration processes). The minimum
range is determined by the focusing capability of the transceiver optics, and for ZephIR it
takes the value 10 m.

Figure 44: Theoretical lidar sensitivity curves at focus heights 25, 50, 75 and 100 m for the
two cases listed above with A = 20 mm and 28 mm, corresponding to respectively the original
(red curve) and current (blue curve) ZephIR design. The peak is normalised to unity in each
case; the absolute peak value decreases as the inverse of height squared, so that the area under
each curve (representing the overall sensitivity) is always the same. This illustrates a useful
feature of focused CW coherent lidar that in uniform scattering, the signal-to-noise ratio is
independent of focus range. Data obtained in calibration measurements (black squares) at a
calibration range R = 68 m are in close agreement with the corresponding theoretical values
(dashed curve) at the equivalent height 58 m (=68 m× cos 30◦ )

4.4.3 Light scattering by aerosols

Coherent lidar measures the Doppler shift resulting from the component of target velocity
along the beam (or line-of-sight) direction. Motion of the target transverse to the beam
direction produces no net Doppler shift. Hence, for a lidar at (0,0,0) measuring at a specific
location (x, y, z) where wind components are (u, v, w), the lidar will detect a line-of-sight
velocity given by the dot product of the wind vector (u, v, w) and the unit vector along the
beam direction:  !
 x+y+z 
 
VLOS = (u + v + w) · p , (93)
 x2 + y 2 + z 2 
where VLOS is the component of aerosol speed along the line of sight (i.e. the beam direction),
and the modulus applies to homodyne systems that cannot distinguish the sign of the Doppler
shift.
In the backscattering geometry considered here, the scattered light experiences a Doppler

DTU Wind Energy-E-Report-0029(EN) 79

shift in frequency given by:
2VLOS ν 2VLOS
δν = = , (94)
c λ
where c is the speed of light (2.998 × 108 m s−1 ) and ν and λ are the laser frequency and
wavelength respectively.
Since the signal originates from a finite probe length, the overall return exhibits a spectrum
of frequencies. This results from the contributions from different velocities (with strengths
determined by the weighting function, Eq. 91) over all the space occupied by the lidar beam.
Note that in the absence of additional information it is not possible to identify from what range
each component of the spectrum has originated. Section 4.6 will outline how information from
additional focus ranges can be used to identify and reject spectral components originating
from strongly-scattering objects (e.g. clouds) situated outside the probe length.
For a CW coherent system, the time-averaged optical signal power, Ps , backscattered by
the aerosols into the receiver is given to a good approximation by:
Ps = πPT β(π)λ, (95)
where PT is the transmitted laser power. It is notable that Eq. (95) contains no dependence
on either the focus range or the system aperture size. With a value of 10−8 (m srad)−1 for
β(π) in clear boundary-layer air, a transmitted power PT ∼ 1 W and λ ∼ 1.5 µm, the received
power Ps derived from Eq. (95) is only of order 5 × 10−14 W emphasising the need for high
sensitivity.

4.4.4 Receiver optics

In a monostatic system, the backscattered light returns through the transmission optics (the
term transceiver is commonly used to denote this dual role). After entering the transceiver,
optical means are used to isolate the return light, and this is passed to the next stages of the
detection process.

4.4.5 Light beating

In coherent laser radar, the incoming Doppler-shifted radiation is optically mixed with a
reference LO beam. The mixing of two waves in this manner leads to the well-known “beat”
phenomenon in which the resulting amplitude oscillates at the difference frequency. In lidar,
the process conveniently “downmixes” the optical frequency of the Doppler shifted return at
∼ 2 × 1014 Hz to a more manageable signal in the MHz range. The efficiency of the beating
process is optimised when the signal and LO beams overlap perfectly in space (i.e. they occupy
identical spatial “modes”). This condition is ensured when both beams propagate in the same
single-mode optical fibre, assuming that they share the same polarisation state.
It is instructive to consider a simple classical description of the light beating process.
Superposition of a LO field ELO cos(ωLO t) and a stable signal field Es cos(ωs t) results in a
fluctuating detector output:
i(t) ∝ (ELO cos(ωLO t) + Es cos(ωs t))2 . (96)
This is conveniently separated into a “constant” term and a cross term oscillating at the
difference frequency:
2


i(t) ∝ ELO + Es2 + 2ELO Es cos |ωs − ωLO | t. (97)
Since the optical power of the local oscillator beam typically exceeds that of the signal beam
2
by many orders of magnitude, the first term is given by ELO to a very good approximation,
and after high pass filtering produces only quantum fluctuations which give rise to the shot
noise floor of the instrument (section 4.7). For a system in which there is no frequency shift
between the LO and transmitted beams, the measured Doppler shift is given simply by:
δν = 2π |ωs − ωLO | (98)

80 DTU Wind Energy-E-Report-0029(EN)

from which the value of VLOS is derived via Eq. (94). In practice a signal field originating from
atmospheric scattering exhibits fluctuations in both its amplitude and phase (or frequency).
The coherent detection process ensures that these properties are reproduced in the detector
output so that, in the limit of high SNR, its spectral analysis gives a correct representation
of the scattered light’s spectrum (Harris et al., 1994).
The coherent detection process described above is also commonly referred to as homodyne
or heterodyne detection. A rigorous quantum-mechanical theoretical treatment of the detec-
tion process is given in Loudon (2000). Note that although the detection process is described
as coherent, the backscattered radiation itself is incoherent in nature, meaning that its phase
is uncorrelated with that of either the transmitted beam or the local oscillator. The phase
and intensity are typically subject to random fluctuations on a timescale that is related to the
inverse of the signal bandwidth (see section 4.9).

4.4.6 Photodetection
The beat signal is detected by directing the optically-mixed beam onto a photodetector
which measures fluctuations in the light’s intensity. In the telecommunications wavelength
band around 1.55 µm, reliable photodiodes are readily available that are well suited to this
purpose. The photodiode converts the incident photons into photoelectrons, which generate
a measurable current (or voltage) that is normally amplified before digitisation. There are
generally four contributions to the output of the photodetector module:
• Dark noise – the intrinsic wideband noise floor generated by the detector and amplifier
combination in the absence of any incident light. Dark noise is due to the random
generation of electrons and holes within the depletion region of the photodetector device
that are then swept by the photodetector’s electric field.
• Photon shot noise (Bleaney and Bleaney, 1976) (sometimes called quantum noise) –
the random generation of photoelectrons by the incident LO beam leads to a wideband,
spectrally flat (white) Gaussian noise source. The shot noise power spectral density
increases in proportion to the optical power of the LO beam.
• Laser relative intensity noise (RIN) – intensity fluctuations that are in excess of shot
noise, caused for example by relaxation oscillation of the laser output (Siegman, 1986).
For a RIN-dominated noise floor, the power spectral density increases as the square of
LO power. Such oscillation is typically at relatively low frequency, peaking below 1 MHz,
and hence only affects the sensitivity of the lidar at low line-of-sight wind speeds around
1 m s−1 . In some systems it is possible to cancel the RIN by use of a dual-channel
balanced detector.
• Beat term resulting from the wind signal – this is the contribution that contains the
information on Doppler shifts from which the wind speed is derived. Its power spectral
density increases in proportion both to the LO power and the signal power.
The requirements for the detector are high quantum efficiency, sufficient bandwidth to
cope with the maximum Doppler frequencies of interest, and for the shot noise contribution
to significantly exceed that of dark noise. This latter requirement depends on a combination
of the detector’s intrinsic noise floor and the optical saturation threshold.

4.4.7 Fourier analysis and lidar sensitivity

In order to extract the Doppler frequency information, it is necessary to perform a spectral
analysis of the detector output. This is conveniently done digitally; an example of a typical
signal processing procedure is described below and illustrated in Figure 45. An ADC with a
sampling rate of 100 MHz permits spectral analysis up to a maximum frequency of 50 MHz,
corresponding to a wind speed VLOS of ∼ 38.8 m s−1 for an upwardly pointing 30◦ scan (Eq.

DTU Wind Energy-E-Report-0029(EN) 81

(95), with λ = 1.55 µm). An analogue low-pass filter with a cut-off frequency of 50 MHz,
inserted between the detector and ADC, eliminates aliasing. Spectra are calculated by digital
Fourier transform (DFT) methods; a 512 point DFT gives rise to 256 points in the output
spectrum with a bin width of ∼ 200 kHz, corresponding to a line-of-sight velocity range of
∼ 0.15 m s−1 . Each DFT represents ∼ 5 µs of data; successive DFTs are then calculated,
and the resulting “voltage” spectra are squared in order to generate a power spectrum. These
power spectra are then averaged to find a mean spectrum for the averaging period. The
random fluctuation in the shot noise floor of the spectrum reduces as the square root of
the number of averages: the sensitivity increases by this same factor. For 4000 averages, the
measurement time amounts to ∼ 20 ms (or a data rate of ∼ 50 Hz). This requires that the
processing is capable of 100% duty cycle, which is achieved in ZephIR with a fast Fourier
transform (FFT) block within a field-programmable gate array (FPGA). It has been shown
that a standard PC with no additional duties to perform can achieve a similar performance.
It is possible to accommodate reasonable variations in any of the above parameters (sample
rate, DFT size and number of averages) and maintain the 100% duty cycle.

Figure 45: Stages in typical lidar signal processing: DFT analysis is carried out by a computer
integrated into the lidar system. As an example, 4000 individual spectra might be averaged
to achieve high sensitivity and measurable returns even in very clear air. This entire process
takes only 20 ms, giving ∼ 50 measurements of line-of-sight wind velocity per second

The width of the Doppler spectrum is determined by three elements:

• Instrumental width: this corresponds closely to the ∼ 200 kHz bin width mentioned
above.
• Transit-time broadening: during the conical scan, the beam passes through the aerosol
particles in a timescale of ∼ 10 − 15 µs, independent of the lidar focus setting. The
corresponding broadening is again of order 200 kHz.
• Turbulence broadening: the probing of a significant volume results in a range of Doppler
shifts from parts of the atmosphere that are moving at different speeds (see section
4.3). In general, this contribution increases with turbulence and shear, and occasionally
there is more than one peak in the spectrum as a result. There is potential for using this
broadening to measure and characterise turbulence at a fundamental level.
The turbulence broadening dominates except under conditions of very uniform airflow. High
system sensitivity is of crucial importance for a wind lidar reliant on weak backscatter from

82 DTU Wind Energy-E-Report-0029(EN)

the atmosphere. The SNR4 for a wind speed measurement by a CW CLR is given by:
ηPs
SNR = . (99)
(hc/λ) ∆ν [1 + D(ν) + R(ν)]
Here η is an efficiency term incorporating optical losses and photodetector sensitivity (typically
η ∼ 0.5, approaching the value 1.0 only for a “perfect” system), Ps is the input signal power,
as defined in Eq. (95) and hc/λ is the light quantum energy, of order 1.3 × 10−19 J. The
signal bandwidth ∆ν is determined by the three contributions listed above, and the term inside
the square brackets denotes the various noise sources listed in section 4.7. D(ν) and R(ν)
represent the power spectral density (at frequency ν) from dark noise and RIN respectively in
units of the power spectral density of the local oscillator shot noise. Ideally D(ν) and R(ν)
should both be ≪ 1 over the range of Doppler frequencies of principal interest, so that the
shot noise is the dominant noise source.
The SNR as defined here is the power spectral density at the Doppler peak divided by that
in the surrounding noise floor. The averaging of many spectra (described in the following
sections) ensures that good performance can be obtained even when the SNR is well below
unity. For example, an SNR of 0.1 will easily exceed a 5σ threshold level (see next section) for
an average of 4000 spectra. From the above it is possible to derive an approximate value of
β(π)min ∼ 10−9 (m srad)−1 for the minimum detectable backscatter, assuming a transmitted
intensity 1 W and a 20 ms measurement time.

4.4.8 Velocity estimation

From the preceding sections it is apparent that each measurement of line-of-sight wind speed,
obtained over a timescale of ∼ 20 ms, generates a Doppler spectrum consisting of one or
more peaks of variable width, superimposed on a noise floor that is predominantly white, but
which may have spectral features originating from RIN and dark noise sources. This section
outlines steps that can be followed to derive an appropriate estimate of the wind speed.
First, the noise floor is “whitened” so that each spectral bin contains the same mean noise
level, achieved by dividing the power value in each bin of the spectrum by a previously-
measured value for the same bin obtained with the shutter closed. A flat threshold is then
applied at a pre-determined level above the mean noise; see Figure 45. A suitable and conser-
vative choice for the threshold is 5 standard deviations (5σ) above the mean noise level. In the
absence of any wind signal (e.g. with the output of the lidar blocked) such a setting will give
rise to negligible occurrences in which the noise alone exceeds threshold. It follows that any
bin whose level exceeds the threshold is deemed to contain a valid contribution to the wind
signal. For each 20 ms measurement, the wind spectrum is reconstructed by subtracting the
mean noise contribution from the contents of each bin that exceeds threshold, and applying
a small re-correction for any distortion resulting from the noise whitening. In order to proceed
to the next stage, a single velocity value is derived from the resulting spectrum. A number of
options are available, including peak and median values; a common solution is to calculate
the mean (or centroid) value hVLOS i.
A series of these values of mean line-of-sight wind speed is generated as the ZephIR lidar
performs a conical scan. Wind parameters are usually calculated from data obtained from a
single revolution of the scanner. With a rotation time of ∼ 1 s, up to 50 line-of-sight values
are available for the next stage, in which a least-squares fitting algorithm is applied. Data is
generally generated for a 1-s, single scan rotation as this is appropriate for most applications
e.g. gust detection for turbine control or mechanical load mitigation.
4 In the lidar community, this is commonly, and more properly, referred to as the carrier-to-noise ratio

(CNR)

DTU Wind Energy-E-Report-0029(EN) 83

4.5 Ground based, vertical scan configuration wind field
parameter determination
4.5.1 Least-squares fitting routine
The data that are fed to the fitting routine consist of up to 150 pairs of values of hVLOS i and
azimuth angle φ. In conditions of uniform wind flow, this gives rise to a rectified cosine wave
of the form:
hVLOS i = |a cos(φ − b) + c| . (100)
The derivation of this function is straightforward and can be found in a number of publications,
e.g. Banakh et al. (1993). The peaks of the function correspond to the azimuth angle aligned
parallel or anti-parallel to the wind direction. The function passes through zero when the
azimuth angle is perpendicular to wind bearing since there is no component of velocity along
the line of sight. The data are also conveniently displayed on a polar plot (Figure 46), which
provides information at a glance on the speed, direction and vertical wind component. A
standard least-squares fitting routine provides the best estimates of the values of the three
floating parameters (a, b and c).

Figure 46: Wind lidar output for a ground-based, vertical scan ZephIR, illustrating many of
the features of a wind profile measurement. This example has been obtained at a height
150 m above ground level, one of several heights being probed in sequence. The lower trace
shows 147 individual line-of-sight wind speed values, obtained over a total period of 3 seconds
(plotted as white squares against azimuth scan angle). The same data, along with the least-
squares fit in red, are displayed above in polar coordinates on the figure-of-eight plot showing
the wind bearing to lie slightly to the East of North. The wind parameters, derived from the
fit, appear in the table on the right; the horizontal wind speed at this height is determined
to be 9.1 m s−1 . The plot on the upper left shows one of the spectra from which each point
on the other graphs is derived.

The high level of redundancy in the fitting process is advantageous and can be used to
identify non-uniform flow. The root mean square deviation of the points from the optimum
solution gives an indication of the quality of fit, and this can be related to the value of
turbulent kinetic energy (TKE; see Wagner et al. (2009)). More work is needed to establish
a full understanding of the turbulence information available from lidar signals (Banakh et al.,
1999). Note that information on turbulence is also available from the spectral widths of the

84 DTU Wind Energy-E-Report-0029(EN)

individual line-of-sight wind speed measurements, but this is not currently used to evaluate
turbulence parameters.

4.5.2 Parameter extraction

The wind parameters for each measurement period are extracted from the best fit as follows,
where (θ is the scan cone half angle of order 30◦ ):
Horizontal speed VH = a/ sin θ,
Vertical speed w = −c/ cos θ, (101)
Bearing B = b, or b ± 180◦ .
Where there is an ambiguity in the sign of the Doppler shift, there are two equally valid best-
fit solutions corresponding to values of b separated by 180◦ . The correct choice is usually
easily made by choosing the solution that lies closest to a conventional measurement from a
met station situated close to ground. Conventionally, a wind profiling lidar incorporates such
a station that performs these (and other) measurements and feeds the results to the analysis
software.
The 1–second wind parameter values are stored internally for subsequent analysis; they can
also undergo further processing for extraction of average values.

4.5.3 Data averaging

It is a common requirement to calculate 10–minute averaged wind data for compatibility
with industry standards. This is most easily achieved by calculation of the arithmetic mean
(“scalar average”) of the individual values of VH , w and B that have been obtained during
the required period. A vector average is also possible in which the resultant of the individual
measurements is calculated over each 10 minute period. In practice the results from the
two methods differ negligibly in reasonably stable conditions. In accordance with industry
standards, ZephIR computes a scalar average for VH and w, and a vector average for B.
When a CW lidar is operating as a wind profiler it is necessary to measure each height
in series. Hence, at any given height the wind is not monitored continuously. Instead, an
individual measurement (taking 1 to 3 s to obtain) is followed by a period of order 7–20 s
during which the lidar is focused at other heights. Since this sampling is carried out randomly
with respect to any behaviour of the wind, this duty cycle of order 15% has negligible impact
on the validity of the resulting 10–minute averaged values. Also the typically large scan area
ensures the beam samples a much higher fraction of the overall turbulent fluctuations.

4.6 Uncertainty analysis

4.6.1 Rain/snow/cloud, solid objects
In general the Doppler shift measured by coherent laser radar is very accurate. This is apparent
from Eq. (95) as long as the laser wavelength remains stable and the signal processing has
been correctly performed - both good assumptions in practice. The laser wavelength is defined
by the manufacturer’s specification to within ±1 nm of the nominal wavelength (1565 nm).
So the contribution to velocity uncertainty from wavelength variation is 1/1565 = ±0.07%.
The Doppler frequency spectra are calculated in a dedicated DSP board with a manufacturer’s
specification of clock stability to within ±50 ppm. The clock stability is directly proportional
to uncertainty in wind velocity and therefore the uncertainty due to this potential source of
error is again small at ±0.05% The values of hVLOS i that are derived from the centroids of
the spectra can be measured to considerably better than a bin width. Confirming the above
instrumental considerations, the line-of-sight velocity calibration was experimentally verified
in a recent wind-tunnel trial (Pedersen et al., 2012). A modified ZephIR 300 configured to

DTU Wind Energy-E-Report-0029(EN) 85

stare directly along the flow reported measurements in very good agreement with a reference
pitot tube for a wide range of wind speeds from 5-75 m s−1 . (Figure 47),

Figure 47: ZephIR lidar wind speed correlation with instrumented wind tunnel pitot tube.
Courtesy of LM Wind Power, DTU Wind Energy, and NKT Photonics.

A greater source of error arises from uncertainty about what provides the scattering from
which the Doppler shift is derived. The scattering is assumed to originate from atmospheric
particles moving at the same speed as the wind and positioned close to the focus of the
lidar beam (section 4.3). An obvious example where this breaks down is when the beam
intersects a solid object (e.g. a bird) that is moving at a different speed from the wind
giving a measurement which could be in error. However, in such a case the value of hVLOS i
so derived will stand out as clearly anomalous on the polar plot (Figure 46). The presence
of such points will be diluted by approximately 50 correct values of hVLOS i obtained from
uncontaminated parts of the atmosphere, and their inclusion should not introduce any bias.
A further safeguard against these erroneous points is provided by a simple “outlier removal”
algorithm. This identifies points that lie anomalously far from the best fit solution to Eq.
(100) and eliminates them. The least-squares routine is then rerun on this slightly reduced
set of hVLOS i, data pairs. The presence of precipitation within the probe volume leads to a
different source of uncertainty. The downward motion of rain and snow inevitably leads to
some error in the vertical component of wind speed. However, the presence of rain and snow is
normally easily identified from the measurement process (for example by detecting activation
of a rain sensor), and the resulting values of vertical wind eliminated from the data. Other
wind parameters are unaffected and can still be correctly inferred.

4.6.2 Cloud effects

CW laser wind profilers focus the beam in order to measure wind speed at a given height. This
technique has the advantage of uniformly high sensitivity independent of the measurement
range, and of very small probe volumes at lower ranges where detailed investigation of shear
or accurate prediction of high turbulence wind fields is important. However, the signals do
require correct processing when the beam impacts a cloud base at higher altitude since the
contribution to the Doppler signal from cloud provides an additional contribution to that from
the aerosols at the desired height.

86 DTU Wind Energy-E-Report-0029(EN)

 A general approach to mitigating this problem needs first to identify the presence of a
cloud return and then remove its contribution from the Doppler spectra. Cloud returns have
a number of characteristics that allow them to be distinguished from aerosol returns:
• Their velocity usually higher
• Their spectral width usually narrower
• For horizontally scanned lidars, only a section (generally the upper part) of the circular
scan is generally affected by cloud
• The power in Doppler peak has clear dependence on lidar focus; the power is maximised
when the lidar beam is focused close to the height of the cloud base
• The Doppler spectrum from cloud is independent of focus range
The latter two characteristics are highly dependable and form the basis for identification
and elimination of spurious cloud returns. The general strategy for removal of cloud signals
is outlined in the following steps (and illustrated in Figure 48):
1. Routinely run the lidar at an additional greater height (e.g. 800 m - essentially a colli-
mated beam output) immediately before or after the maximum height of interest (say
150 m in this example).
2. For each azimuth angle around scan at 150 m, identify the 800 m (“cloud”) spectrum
obtained at the closest value of azimuth angle.
3. Apply test conditions to the 150 m spectra to determine whether any cloud signal is
present in the spectral data, then if necessary apply cloud removal algorithm.
4. Run standard thresholding and centroiding routines on resulting “clean” spectra and fit
to the rectified cosine wave (Eq. (100)) to obtain wind parameters.
A cloud removal algorithm based on this approach is implemented in ZephIR; this has been
extensively tested in a number of locations, and its effectiveness demonstrated by correlation
analysis against calibrated tall masts. During the 800 m (“wind profile”) scan, background
measurements are taken to quantify the specific cloud return and any cloud effect is then
removed from the processed data.
In general, lidars of various types of design will all have difficulty measuring in very low
cloud and fog scenarios where the light emitted from the lidar is unable to reach all the
heights of interest due to absorption in the atmosphere. While this atmospheric condition
mostly occurs during low wind speed periods, it is important that these periods be identified.
In the majority of cases they are removed by filtering methods.
Trials of a ZephIR unit at Risø DTU’s test site at Høvsøre (Courtney and Gottschall, 2010)
took place in long periods of low cloud and hence provide a demonstration of the performance
in challenging cloud conditions. The data set summarised in Table 5 below was taken during a
period of 4 weeks in October and November 2009. Cloud height was measured using a ceilome-
ter; 25% of data was obtained with the cloud base below 300 m, and 43% obtained with the
cloud base below 600 m. A more recent independent evaluation of a ZephIR 300 system in simi-
lar conditions is available at: http://www.zephirlidar.com/sites/yourwindlidar.com/files/ZephIR-
301-Evaluation-at-Risoe-Test-Site.pdf.
The results of the trial (Table 5) indicate a good agreement between lidar and mast at
all heights from 40 up to 116 m. Filtering has been applied to remove sectors prone to the
influence of turbine wakes, and speeds below 4 m s−1 , to ensure measurement within the
calibration range of the mast cups.

DTU Wind Energy-E-Report-0029(EN) 87

Figure 48: Cloud removal for a vertical scan CW lidar. The left plot shows the lidar conical
scan focused at a typical height above ground level. The Lorentzian sensitivity curve is also
shown; a spurious return is generated when the far wing of this curve intersects a strongly
scattering low cloud layer. The right plot shows the aerosol (red) and cloud (purple) returns
as the lidar is focused at various heights - the level of cloud contamination increases with
focus height. The cloud signal is easily identified from the 800 m focus, and these data are
then used to eliminate the cloud return at the measurement heights.

Table 5: Results of correlation analysis of 10-minute averaged horizontal wind speed for a
ZephIR 300 trial at Høvsøre, Denmark in March 2011. Gradients, m, (forced through the
origin) and coefficients of determination, R2, of unity would imply perfect agreement be-
tween the lidar and the mast-mounted cup anemometers. (It should be noted that the slopes
very close to unity are slightly fortuitous, since the cup anemometer measurements have
uncertainties at least of order ±1%, due to calibration and mounting/shadowing effects).

Height AGL [m] Slope m R2

116 0.993 0.977
100 0.987 0.988
80 0.984 0.992
60 0.990 0.992
40 1.007 0.992

4.6.3 System positioning accuracy

Correct alignment ensures the risks are low, but errors in aligning the lidar during set-up will
have an impact on the measurement of wind bearing (if the lidar is rotated from its correct
orientation) and vertical wind speed (if the lidar is tilted, so that the axis of its conical scan
is not precisely vertical). For a small tilt angle δ, the error in vertical wind speed w will vary

88 DTU Wind Energy-E-Report-0029(EN)

from ±VH sin δ (if the tilt is towards or away from the direction of the wind) to zero (if the
tilt is perpendicular to the wind). Any bias on VH is negligible to first order.

4.6.4 Probe volume effects and operation at greater heights

As discussed in section 4.2, the lidar samples the motion of air from a finite volume, centred
on the beam waist at the focus. Clearly there is minimal risk of bias while all the air within the
probe volume moves at the same speed. However, for vertical scan lidars, there is usually some
degree of shear across the sample region. For a linear shear this leads to spectral broadening
of the returns, but no overall bias. A strong non-linear shear profile across the probe volume
is required to induce any bias of significance. In practice such conditions are rare, in particular
for measurement heights around hub height and below where the probe length is relatively
small.
Most lidar comparisons have taken place beside masts of heights around 100 m or less.
However, in early 2009 a study took place in Iowa, USA against a 200 m mast in flat terrain.
The results showed high correlation (Table 6, taken from Barker (2009)) even at the greater
heights examined (150 and 200 m), which approach the expected maximum operating range
for focused CW lidar.

Table 6: Results of a comparative trial of a ZephIR lidar against a very tall mast, equipped
with two types of cups at each height. The data indicate that the extended probe length at
greater heights did not result in excessive bias or errors. [1]: Forced through the origin; [2]:
Only hourly average containing 6 valid 10 min measurements are compared

NRG IceFree3 NRG MAX#40C

Ten min Hourly Ten min Hourly
average average[2] average average[2]
Height R2 Slope[1] R2 Slope[1] R2 Slope[1] R2 Slope[1]
AGL [m]
193 0.984 0.987 0.987 0.987 0.982 0.993 0.988 0.992
157 0.982 1.006 0.988 1.005 0.984 1.000 0.989 1.000

4.6.5 Flow uniformity and complex terrain

Because only line-of-sight wind components are measured, a single ground-based lidar unit
inevitably provides an incomplete picture of the 3D vector flow, regardless of the scan pattern
employed. Firstly, this ”cyclops” LOS velocity determination at any one probe point is unable
to disambiguate the full wind vector information, merely measuring one component. The
full vector at a given point can only be measured by the provision of three (or more) lidar
units positioned on the ground at an appropriate separation distance (comparable to the
measurement height for best accuracy), such as the Windscanner system under development
by Risø DTU, web address below: http://www.risoe.dtu.dk/research/sustainable_
energy/wind_energy/projects/vea_wind_scanner.aspx?sc_lang=en/
Secondly, whilst a given scan pattern can provide more information about the wind flow,
certain assumptions, for example uniformity of flow across the probed area, linear or log-
arithmic vertical shears, are often reasonably made. However, in complex terrain, the flow
undergoes stable and unstable non-uniformities, and the figure-of-eight plot (Figure 46):) can
distort systematically for a given wind direction, reflecting the speeding up and slowing down
in certain regions of the scan. The ZephIR lidar provides some information about the flow
non-uniformity, with up to 50 points per second being interrogated around the scan disk.
In the presence of non-uniformity in flow (section 4.2), a lidar measurement can indicate a
wind speed different to that from a point measurement by a mast-mounted cup anemometer.

DTU Wind Energy-E-Report-0029(EN) 89

Lidar data has been successfully combined with the output from flow-modelling software,
using both linear models (Bingöl et al., 2008, 2009; Bingöl, 2010) and computational fluid
dynamics, CFD (Harris et al., 2010; Pitter et al., 2012). This pragmatic approach generates
measurements equivalent to a “point-in-space” sensor by using the results of flow modelling
to adjust the measured lidar wind speed. This topic will is dealt with elsewhere in this lecture
series, examining possible improvement of lidar resource assessment capability in complex
terrain.

4.6.6 Dependence on backscatter level

Under conditions of high backscatter, the spectrum provides an accurate measure of the
distribution of line-of-sight velocities within the probe volume, weighted according to Eq.
(91). As the backscattering strength drops (usually associated with increased air clarity) this
has a similar effect to raising the detection threshold, and will lead to elimination from the
spectrum of weaker components of velocity. The impact of the system noise floor on the
detailed spectral shape will also be increased. The centroid values hVLOS i will be unbiased
and independent of threshold level when the spectrum is symmetrical. However, for a skewed
(asymmetric) spectrum the precise value of hVLOS i will be sensitive to the threshold. Hence
a small difference in measured wind speed is possible between two measurements under
conditions that are identical in every way apart from the level of backscatter. However, there
is no evidence from comparisons so far to suggest that this leads in practice to a significant
discrepancy.
A further possibility to be considered is the effect of saturation (by very strong scattering
returns from thick cloud) of the lidar detector, electronics or signal processing. In the event
that the input signal exceeds these limits, the spectrum will become distorted, possibly fea-
turing higher harmonic components of the true Doppler frequencies. In practice, the range of
inputs to the ADC can be tailored to accommodate the highest levels of backscatter that will
reasonably be encountered, eliminating the risk of bias.

4.6.7 Beam obscuration and attenuation

Lidar can operate successfully even when part of its scan is obscured. This confers great
flexibility so that the system can easily be located adjacent to masts, buildings or in forests.
Stationary objects pose no major problem other than the loss of wind measurements from the
relevant obscured sector of the scan. Slowly moving objects can also easily be filtered, based
on the magnitude of their Doppler shift.
In the above cases, the fit to Eq. (100) will no longer contain data over the full 360◦ range of
φ. Laboratory experiments on moving belt targets have indicated that accurate measurements
are obtained even when over half of the scan is obscured. Large errors in the least-squares
fitting process become possible as the obscuration increases yet further; such conditions are
identified and a null result returned.

4.6.8 Wind direction

For ground based, vertically scanning ZephIR, the two best-fit solutions ZephIR obtains to Eq.
(100) give values of wind direction that are 180◦ apart. Selection between the two options is
made with reference to the measurement of wind direction from a ground-based anemometer.
This needs to be in disagreement by over 90◦ with the direction at the chosen height for the
incorrect choice to be made. While such a directional shear (veer) is conceivable in highly
complex terrain and at very low wind speed, it is much less likely in the reasonably uniform
conditions of interest for wind energy applications, and the wind direction selection can be
propagated upwards from measurements at several heights. In the event of the wrong choice
being made, leading to a wind direction that is in error by 180◦ , the value of vertical component
of the wind w will have the wrong sign. In other words, an updraught will be wrongly identified

90 DTU Wind Energy-E-Report-0029(EN)

as a down draught (of the same absolute speed) and vice versa.

4.7 Calibration, validation and traceability

For historical reasons, the clearest demonstration of validity is provided by direct side-by-side
comparisons between the lidar system and a fully instrumented IEC-compliant meteorological
mast of suitable height. Rigorous comparisons must be carried out with great care to avoid
a number of problems associated with cup anemometers (Kristensen, 1999). These are well
known and include the following:
• Shadowing of the cup anemometer by the mast from certain directions
• Cup over-speeding in turbulence and sensitivity to any vertical wind component
• Cup icing
• Valid cup anemometer calibration
• Topographic effects leading to non-uniform flow across the area occupied by mast and
lidar scan (including turbine wakes)
A lidar/mast comparison is commonly used to provide a validation of lidar performance,
and examples of such checks were provided by the results in Tables 5 and 6. The lidar can
then be used as a traceable reference for comparison with other units.
Lidar systems are normally calibrated in the laboratory before shipping. Routine checks on
the calibration of units on their return to base provide confidence of long-term stability. As an
example, the calibration process undertaken for a ZephIR lidar is outlined below. This consists
of three stages:
1. Velocity and direction check against a calibrated moving belt. The process provides a
direct check of laser wavelength and scanner cone angle, each of which affects the velocity
calibration (via Eqs. (94) and (100) respectively).
2. A focus range check is carried out with a moving target located at precise distances from
the lidar. The closed loop positioning system ensures no drifts over time. An example of
the output data from a focus calibration test was plotted in Figure 44 (section 4.4).
3. Finally, each unit undergoes an outdoor test to measure wind speed at several heights
using an industry-certified, 92 m, meteorological mast. Figure 49 shows an example
correlation plot of 10 min average horizontal wind speed, obtained over a period of 7
days.
Each of the three tests above gives information on the sensitivity of the unit; for deployments
in “clean” air, it is important to ensure this aspect of performance is fully optimised and has
not deteriorated, or there is a risk of reduced data availability.
It is important that no adjustments are performed during validation trials, or afterwards
for as long as the lidar remains a traceable reference unit. The certification process outlined
above has been defined in collaboration with industry experts including Garrad Hassan and
provides the traceability that is a key element of formal energy prediction reports used by the
financial community.
In addition to its velocity measurements being closely traceable to primary units of time
and length, lidars offer a potential advantage for accurate shear profiling, for both speed
and direction, in that the same instrument is used to make the measurements at all heights.
By contrast, a mast relies on consistent calibration of the full set of cups and vanes; any
differences in calibration of the individual instruments will lead to uncertainty and error in
the shear assessment. An example of the difficulty in calibrating instruments such as cup
anemometers that rely on relatively complex, non-linear physical interactions is illustrated in
Figure 50, where calibration results from four high quality cups are compared.

DTU Wind Energy-E-Report-0029(EN) 91

Figure 49: A ground based ZephIR 300 system compared to the industry-certified met mast at
Pershore, demonstrating strong correlation and a gradient very close to unity. In this example,
comparison was carried out at a height of 70.5 m. From (Rutherford et al., 2012).

Figure 50: Graph showing results from routine re-calibration of four cups to be used at an
accredited met mast site. The vertical axis shows differences when the same cup was calibrated
at two independent wind tunnel standards facilities. The tunnel-to-tunnel calibrations of the
same cups show variability of the order of 1%.

4.8 Turbine mounted continuous wave lidar

An application that has recently generated much interest is the use of remote sensors, such
as lidar, to measure the wind field ahead of an operating wind turbine generator. Since, for a
correctly yawed turbine, the lidar is always pointing into the incident wind field; this can be very
useful for power curve measurement as no sectors need to be eliminated for turbine shadow,
as is the case when using a fixed meteorological mast and cup anemometers (of course sectors
affected by the wakes of adjacent turbines will still need to be filtered). Providing advance
wind data to the turbine control system also has many potential advantages such as reducing
stress loads by warning of incoming gusts or optimising rotor pitch control. The lidar can

92 DTU Wind Energy-E-Report-0029(EN)

Table 7: Combined results from 28 ZephIR 300 units. The mean and standard deviation of
the mast comparison parameters, gradient and R2 , were calculated from the first batch of 28
ZephIR 300 units. These results confirm the consistency of the lidars’ performance.

Gradient R2 Laser sensitivity

Height [m]
Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.
91 1.0039 0.0072 0.9894 0.0059
70 1.0033 0.0072 0.9928 0.0059
1.0350 0.0893
45 1.005 0.0050 0.9924 0.0050
20 0.9967 0.0045 0.9925 0.0048

be mounted on the nacelle roof, within the rotor spinner (Figure 51) or built into the rotor
blades.

Figure 51: Continuous wave dual mode ZephIR 300 lidars mounted on a nacelle roof (left)
and within a wind turbine spinner (right).

The high sensitivity and hence fast data rates of circular scanning CW lidar make it very
well suited to turbine mounted applications. By adopting different signal processing and data
analysis strategies; hub height wind speed, direction, shear exponent and turbulence can be
measured, or speed, direction and turbulence at discrete heights across the rotor can calculated
(or both). The latter technique can be used to generate rotor equivalent wind data which has
been shown to produce more precise and representative wind turbine power curves, especially
for large rotor diameters (Wagner et al., 2008; Wharton and Lundquist, 2012).By measuring
the inclination of the lidar in real-time, the circular scan CW approach allows the effects of
nacelle motion on both line-of-sight velocity and measurement height to be negated.

4.8.1 Least-squares fitting routine for horizontal scanning (turbine mounted) oper-
ation
The use of a CW lidar for turbine mounted applications is fundamentally a quite different
arrangement when compared to a ground based, vertical scanning configuration. Unlike the
latter, the scan axis is approximately horizontal, and the lidar is almost always predominantly
staring into the wind. A consequence of this is that the polar plot (of the measured line-of-
sight wind speeds as a function of scan angle) is no longer a figure-of-eight shape, but instead
takes on a more circular appearance (Figure 52).

DTU Wind Energy-E-Report-0029(EN) 93

 Important quantities of interest for turbine relevant wind field determination include hub
height horizontal wind speeds, the vertical wind shear, and the yaw misalignment. The latter
is an angular measurement of the difference between the horizontal direction of the lidar scan
axis and the wind direction, and it is useful for yaw control of the turbine or calibration of
wind turbine nacelle vanes. Horizontal shear and wind inflow angles are also of interest.
As before, a wind model can be constructed. This must take into account the mounting
geometry on the turbine e.g. (Mikkelsen et al., 2010; Angelou et al., 2010). A least squares
fit of the measured wind field can be performed to extract the parameters of interest.
One of the attractive features of the CW lidar’s circular scan pattern is that it samples the
wind field around the full range of rotation of the turbines rotor. Typically 50 line-of-sight
measurements are obtained over one circular scan in 1 second (i.e. 20 ms sample rate). This
dense sampling of the wind field around the rotor disk can give valuable preview data to
allow feed-forward control for both collective and individual pitch control of the blades of the
turbine.

Figure 52: An example of visualisation and analysis of data from a turbine-mounted ZephIR.
Left: polar plot of raw data, showing line-of-sight wind speeds with scan angle. The radial
axis is the LOS speed. The breadth and structure of plotted distribution gives an indication of
the spatial turbulence within the scan volume e.g. ground induced shear and turbulence can
be seen in the lower range of angles. Low level wind jets and wakes from other turbines can
also be detected in this manner. Centre: real time analysis of the received polar plot, showing
centroids of the received line-of-sight spectra (red dots) and fitted wind parameters (indicated
by the green curve). The central red dots are turbine blade returns and are automatically
filtered prior to fitting. Right: reference data and wind characteristics calculated from the fit.

An example of filtering that can be required is for the case of turbine blades. For a turbine
mounted CW lidar, situated on the roof of a turbine’s nacelle, and scanning upwind through
the turbine blades, the lidar must contend not only with quasi-periodic blocking of the beam,
but also strong Doppler returns from the blades themselves. Although the intensity of the
back reflected laser signals can be very high from these blades (typically 50 times higher than
the wind returns), this can help distinguish them from the line of sight Doppler returns from
the incoming wind. Additionally, the relatively slow, near perpendicular path of the blade
surfaces means that the Doppler shifts are relatively low frequency (giving Doppler returns
corresponding to typically < 2 m s−1 ). Hence efficient blade rejection filters, which remove
these signals from the wind field fitting process, are simple to implement. However, blade
effects do reduce the number of data points around the scan, and for this reason, hub (or
spinner) mounted CW lidars can have some advantages.

4.8.2 Turbine mounted CW lidar for wind turbine power curve measurement
Continuous wave lidar have now been deployed in a number of wind turbine power curve
measurement campaigns. Some results from two of these are reproduced in this section. The
first campaign discussed here was organised by ROMO Wind and carried out in flat terrain

94 DTU Wind Energy-E-Report-0029(EN)

in Eastern Jutland between January and April, 2012 (Slinger et al., 2013). The lidar, a dual
mode ZephIR 300, was mounted on the nacelle roof of a NEG-Micon 2 MW wind turbine
with a 72 m rotor diameter. There was no meteorological mast data from the site - the trial
was intended to measure relative, rather than absolute power curves, in order to demonstrate
and quantify turbine performance improvements after turbine tuning.
Lidar data from the first part of the trial, before wind turbine tuning was carried out,
identified a turbine yaw error of between 14◦ and 16◦ (Figure 53). This was remedied by a
nacelle vane recalibration before the second phase of the trial was undertaken. The effects on
the lidar measured power curve (at a measurement range of 180 m or 2.5 rotor diameters)
are clearly visible (Figure 53). ROMO Wind estimated an improved annual energy production
(AEP) of approximately 5% after yaw recalibration.

Figure 53: (a) Measured turbine yaw misalignment before and after nacelle vane recalibration.
(b)Wind turbine power curves measured by nacelle mounted lidar before and after nacelle vane
recalibration at a measurement range of 180 m.

Measurements were also made at a series of other ranges between 10 m and 180 m and
are shown in Figure 54). While the measurements at close range ”flatter” the power curve
due to the rotor induction effect, they demonstrate that high quality power curves can be
obtained very close to the rotor. While these would have to be corrected to be quantitative,
they are very representative of the true wind field incident on the turbine rotor and can be
expected to be immune to say, the effects of complex terrain, as compared to more remote
measurements.

Figure 54: (a) time series of wind speed measurements taken at a series of measurement
ranges. The measurements are highly correlated, but rotor induction (blockage) effect is
clearly visible.(b) Series of wind turbine power curves generated from measurements at several
ranges. The upper curve is from the closest range (10 m), the lower curves were measured at
30, 50, 100 and 180 m respectively.

The second study reported here took place on a DTU test turbine situated at Roskilde,

DTU Wind Energy-E-Report-0029(EN) 95

Denmark. The turbine was a 500 kW Nordtank turbine with a rotor diameter of 41 m and a
hub height of 36 m. A meteorological mast was situated at a distance of approximately 2.2
rotor diameters, so absolute as well as relative measurements could be carried out. The met
mast was equipped with cup anemometers at heights of 18, 27, 36, 45 and 54 m above ground
level, spanning the entire swept area of the rotor. Figure 55. below shows a comparison of
hub height wind speed measured by the lidar (at a range of 2.2 rotor diameters) and the met
mast, when the wind direction was in a sector that was unaffected by shadowing or wake
effects from adjacent turbines. It can be seen that the measurements demonstrate a high
level of correlation, particular when the sloping nature of the terrain at this site is taken into
account.

Figure 55: Correlation plot of hub height wind speeds measured by a nacelle mounted CW lidar
and a pair of cup anenometers on a 500 kW Nordtank wind turbine at Roskilde, Denmark.

The lidar data was then reprocessed to measure wind speed and direction at the height of
each of the cups listed above, taking full account of any nacelle motion on both measurement
height and line-of-sight velocity. Turbine SCADA power output data was corrected for air
density and temperature and used to generate rotor equivalent power curves from both the
met mast and the lidar data (Figure 56). This preliminary result indicates that turbine mounted
CW lidars are capable of measuring power curves of comparable quality to met masts, but
with the advantage of easy redeployment. These are believed to be the first reported rotor
equivalent power curve measurements using a commercially available nacelle mounted lidar.
Subjects of current research in the turbine mounted lidar area include feed-forward control
for stress load reduction and short range (i.e. less than 2 rotor diameters) wind speed mea-
surement for turbines situated in more complex terrain where long range measurements are
less representative of the actual wind field incident on the turbine rotor.

4.9 Summary, state of the art, and future developments

Coherent monostatic CW lidar is a method capable of rapid wind speed measurement at
relatively short ranges (all the way from 10 m to 200 m) and hence is well suited to several

96 DTU Wind Energy-E-Report-0029(EN)

Figure 56: Normalized power curves using rotor-equivalent wind speeds measured by a nacelle
mounted CW lidar (blue) and mast mounted cups (red) at heights of 18, 27, 36, 45 and 54
m above ground level. Wind speed bin size was 0.5 m s−1 and the error bars indicate the
standard deviation of the scatter in the measured data.

requirements in the field of wind energy. Examination of the measurement process reveals
that the basic acquisition of line-of-sight Doppler spectra is a well-established method with
little scope for gross errors and miscalibration. The subsequent steps required to convert these
spectra into a profile of wind speed are more complex, however, and their validity relies on a
number of well-established assumptions. Much work has been performed to test the validity
of the assumptions outlined in section 4.3, and to understand the uncertainties and other
issues discussed in section 4.6.
Complex terrain remains a topic of great interest as it becomes increasingly necessary to
explore less ideal locations as potential wind farm sites. In such sites the horizontal wind
speed deduced by conically-scanned lidar can be subject to differences in comparison to that
measured by co-located cup anemometers when the flow is non-uniform across the lidar mea-
surement disk. A method has recently been developed in which the impact of inhomogeneous
flow at complex flow sites is examined using computational fluid dynamics (CFD modelling to
predict the bias that will be experienced by a lidar in comparison to a conventional met mast
equipped with cup anemometers. Similar percentage changes in wind speed as measured by
a mast are shown to occur if the mast were to be moved by ±50 m from its original location.
This suggests a methodology for resource assessment in complex terrain in which lidar is
used in combination with CFD modelling in order to (i) adjust the lidar data for the impact
of non-uniform flow and (ii) investigate the wind variations across the site that are a major
source of uncertainty for current techniques.
Lidar offers some potential advantages in turbine power curve measurement. The measure-
ment over an extended volume may give a more representative estimate of the wind energy
content of the air interacting with the blades, and the ability to re-position the lidar quickly
is clearly advantageous. A study reported by Wagner et al. (2008) has shown that exploiting
the lidar wind profile data can reduce the scatter of points in a measured power curve. In
another recent study (Cayla, 2010) a ZephIR lidar gave an almost identical power curve to
an IEC-instrumented power performance mast. The scatter of the points in the power curve
obtained using the ZephIR data at hub height was somewhat lower than that for the mast.
This result needs further investigation and possibly is a consequence of the more effective
sampling of the wind around the scan disk. It follows, interestingly, that remote sensing equip-
ment that agrees perfectly with the mast would therefore have provided higher scatter in the
power curve than ZephIR!
The extraction of turbulence data relevant to the wind industry from lidar signals is an area
that will benefit from further research and verification through field comparisons. Turbulence

DTU Wind Energy-E-Report-0029(EN) 97

can manifest itself as gusts, eddies, and fluctuations in wind speed. It is important in wind
energy applications to characterise the levels of turbulence encountered at a specific site
location. A commonly-used basic measure of turbulence is turbulence intensity (TI). ZephIR
calculates the turbulence intensity that a conventional cup would have obtained at the same
measurement height by analysing the variation in individual wind speed values during a 10-
minute averaging period. This value of TI is automatically logged in the output data. The
calculation takes into account the difference between point measurements obtained from a
cup anemometer, and spatially-averaged lidar data where a volume is interrogated (Barker
et al, 2012).ZephIR’s measurements of turbulence have been investigated in a number of
independent studies against calibrated met masts in flat, offshore and complex terrain, and
at different heights above ground (Wagner et al., 2009).
Resource assessment in maritime locations is becoming increasingly relevant as offshore
wind farms assume greater importance. The cost of installing an offshore tall mast is very
high, so remote sensing may prove particularly advantageous in such locations. ZephIR lidars
have been involved in successful trials on several offshore platforms in the North Sea (e.g.
Peña et al. (2009)), the Baltic, and around the lakes and coasts of North America. A floating
lidar platform offers an exciting future concept; an early attempt to develop a ZephIR system
on a buoy (SeaZephIR) took place in 2004/5. After a redesign, the system took to the water
off S Norway in 2009. A world-first demonstration trial took place over a period of several
weeks in late 2009, involving one ZephIR unit stationed on land, with the floating SeaZephIR
unit positioned 800 m out to sea. The wind speeds measured by the two ZephIR units showed
excellent correlation, with differences in mean wind of ∼ 1% or less at all heights over a 3
week test period (see Table 8, from Wiggins (2009)). In this trial there was no attempt to
compensate for the platform motion; it may be necessary in very severe conditions to use
measurements of the 6 degrees of freedom (3 rotational and 3 translational) that can in
principle distort the lidar measurement. The low impact of the motion observed in trials so
far may be a consequence of the high stability of the buoy combined with the very fast 50 Hz
measurement rate for the ZephIR lidar, which allows a snapshot of the wind around a 360◦
disk to be obtained in 1 s. Further development of SeaZephIR is ongoing.

Table 8: Correlation analysis from the first SeaZephIR trial in 2009: the table shows gradient
(m) and coefficient of determination R2 for plots of 10-minute wind speed for SeaZephIR on
a floating platform versus those measured by a second ZephIR unit positioned 800 m away
on land

Height AGL [m] Slope m R2

120 0.993 0.972
90 0.998 0.970
60 1.004 0.968
30 0.990 0.954
10 0.984 0.953

Forward-looking turbine mounted lidar, either on the nacelle or in the hub, is another ex-
citing lidar development. Applications include turbine power curve measurement, energy yield
optimisation (e.g. by reducing turbine yaw misalignment) and gust and fatigue load reduction
allowing longer turbine lives and/or turbine build cost reduction. As already remarked, CW
lidar seems particularly well suited to this type of application, owing to its high sensitivity
(high average photon flux), high sample rate (50 Hz), and scan path that probes the wind
around the rotation path of the rotor. Another of its benefits is its flexibility in terms of
turbine mounting. In addition to nacelle roof mounting, it is, to date, the only class of lidar
that has been installed in a rotor hub (or spinner). Interest in the concept has increased
significantly since the world-first proof-of-principle demonstration of turbine-mounted lidar in
2003 (Harris et al., 2006, 2007), with several groups currently working towards evaluating

98 DTU Wind Energy-E-Report-0029(EN)

the concept. Developments include incorporation of a conical-scanning ZephIR lidar in the
spinner of a large turbine (Mikkelsen et al., 2010; Angelou et al., 2010) giving an unobscured
view of the approaching wind. More recent still has been research and development activity
funded by the Danish High Technology Fund (DHTF) studying implementation of CW turbine
blade mounted lidar. Here the concept is to have a ZephIR base unit installed in a turbine
hub, connected by fibre optics to small, fixed focus telescopes mounted on the blades of the
turbine. The rotation of the blades naturally allows scanning of the wind field around the
blade path, and this approach holds some promise for blade pitch optimisation, for example.
Initial experiments in a wind tunnel (Figure 57) (Pedersen et al., 2012) have confirmed the
potential of the approach and turbine trials are currently underway.

Figure 57: CW lidar experiments in a wind tunnel, prior to blade lidar deployment experiments.
Left: wind tunnel schematic. Centre: CW lidar twin telescopes. Right: External view of the
wind tunnel, showing the ZephIR 300 base unit.

In connection with turbine mounted lidars, significant recent efforts in the industry have
focussed on quantifying their potential benefits, as well as looking at the optimum lidar
configurations to use. For CW lidars, the cone scan angle, the number of ranges to scan over
(if indeed more than one is required), scan rates and the LOS processing algorithms are all
being investigated. Recent results in the literature have included:
1. Conical scan CW lidar was used to determine yaw alignment of a lidar (Kragh et al.,
2013) and demonstrated the ability to achieve a sub 4◦ yaw error over a 2 hour period,
even during periods of high turbulence.
2. Simulations examining the ability of turbine mounted lidar for accurate yaw alignment
(Kragh and Hansen, 2011) indicated yield, at below rated power, could be raised by 1%
to 5%.
3. A study reported in Schlipf et al. (2011) comparing conventional nacelle based wind vane
with lidar yaw alignment control, indicated that the yearly energy output of a 5 MW
turbine could be enhanced by ˜ 2% using the lidar.
4. Schlipf and Kuhn (2008) modelled the benefits of a nacelle mounted lidar for feed-forward
control, in particular turbine speed control. The study found reductions in standard
deviations of 91%, 90% and 71% for rotor speed, tower fore-aft moment and blade root
flap moment for gusts. For turbulent airflows, the reductions in standard deviations were
77%, 32% and 17% respectively.
5. Simley et al. (2011) simulated a conical scan CW lidar and showed accurate yaw align-
ment should be possible. Even in highly turbulent airflow, a precision of a few degrees
was achievable. The same paper also showed that RMS wind speed measurement errors
were lower for a CW system than a pulsed system for ranges ¡125m
6. Simulations using lidar feed-forward control (Laks et al, 2011) showed turbine fatigue
load reductions of approximately 20%.

DTU Wind Energy-E-Report-0029(EN) 99

 7. A recent study by Rogers et al. (2012) analysed a variety of scenarios that could be
addressed by turbine mounted lidar, including retrofitting lidar to existing turbines, larger
rotors and taller towers. Benefits of turbine mounted lidar included a 6 year life extension
and 30% increase in total energy production (when a lidar was retrofitted to a 2.5 MW
turbine); an increase in permitted rotor area of 6% and an associated energy output
increase of 4% (larger rotor on 5 MW turbine); a 3% energy output increase from a
greater allowable tower height, achieved through reduced fatigue loads (again on a 5
MW turbine). The same study also estimated an achievable increase in energy output
due to optimisation of lidar control alone to be just 0.6%.
Clearly, turbine mounted lidars have an important role to play in reducing costs of energy
generated by wind turbines. This application is discussed in more detail and in broader scope
in other lectures.

Acknowledgements
The authors are grateful for the support and enthusiasm of their colleagues and collaborators.
Without them, wind power lidar technology could not have evolved to its current advanced
state.

Notation
a floating parameter for the fit of the line-of-sight velocity
A beam radius at the output lens
ADC analogue-to-digital converter
b floating parameter for the fit of the line-of-sight velocity
B wind bearing
c speed of light
floating parameter for the fit of the line-of-sight velocity
CFD computational fluid dynamics
CLR coherent laser radar
CNR carrier-to-noise ratio
CW continuous wave
DFT digital Fourier transform
D(ν) power spectral density from dark noise
ELO LO field
Es stable signal field
FFT fast Fourier transform
FPGA field-programmable gate array
h Planck constant
i fluctuating detector power output
IR infrared
LO local oscillator
m slope of the linear regression
Ps time-average optical signal power
PT transmitted laser power
R distance of the beam focus from the lidar output lens
R2 coefficient of determination
RIN laser relative intensive noise
D(ν power spectral density from RIN
SNR signal-to-noise ratio
t time variable
TI turbulence intensity
TKE turbulent kinetic energy
u wind speed component in the x-direction
v wind speed component in the y-direction
VAD velocity-azimuth-display
VH horizontal wind speed
VLOS line-of-sight wind speed
w wind speed component in the z-direction

100 DTU Wind Energy-E-Report-0029(EN)

 x horizontal position in longitudinal direction
y horizontal position in transverse direction
z position perpendicular to the horizontal plane
β atmospheric backscatter coefficient
Γ half-width of the lidar’s peak sensitivity
δ lidar’s tilt angle
δν Doppler shift in frequency
∆ target distance from the focus position along the beam direction
η lidar efficiency
θ lidar’s cone half-angle
λ laser wavelength
ν laser frequency
φ lidar’s azimuth angle
σ standard deviation
ωLO local oscillator frequency
ωs stable signal frequency
hXi ensemble average of a variable X

References
Angelou N., Mikkelsen T., Hansen K. H., Sjöholm M., and Harris M. (2010) LIDAR Wind Speed Measurements
from a Rotating Spinner: SpinnerEx 2009. Project Upwind and Risø report 34: Risø-R-1741(EN)

Banakh V. A., Smalikho I. N., Köpp F., and Werner C. (1993) Representativeness of wind measurements with
a CW Doppler lidar in the atmospheric boundary layer. Appl. Opt. 34:2055–2067

Banakh V. A., Smalikho I. N., Köpp F., and Werner C. (1999) Measurements of turbulent energy dissipation
rate with a CW Doppler lidar in the atmospheric boundary layer. J. Atmos. Oceanic Tech. 16:1044–1061

Barker W. (2009) Analysis of ZephIR data from Mason City, Iowa. Natural Power internal report

Barker W., Pitter M., Burin de Roziers E., Harris M., and Scullion R. (2012) Can lidars measure turbulence.
EWEA conference PO74

Bingöl F., Mann J., and Foussekis D. (2008) Lidar error estimation with WAsP engineering. IOP Conf. Series:
Earth and Environ. Sci. 1:012058

Bingöl F., Mann J., and Foussekis D. (2009) Conically scanning lidar error in complex terrain. Meteorol. Z.
18:189–195

Bingöl F. (2010) Complex terrain and wind lidars. Risø-PhD-52(EN), Roskilde

Bleaney B. I. and Bleaney B. (1976) Electricity and magnetism, Oxford University Press, Section 23.4

Cayla M. (2010) Comparison of ZephIR measurements against cup anemometry and power curve measure-
ments. Natural Power internal report (http://www.zephirlidar.com/resources/publications)

Chanin M. L., Gariner A., Hauchecorne A., and Portneuve J. (1989) A Doppler lidar for measuring winds in
the middle atmosphere. Geophys. Res. Lett. 16:1273–1276

Clifford S. F. and Wandzura S. (1981) Monostatic heterodyne lidar performance: the effect of the turbulent
atmosphere. Appl. Opt. 20:514–516

Courtney M. and Gottschall J. (2010) ZephIR 145 validation test. Risø report for Natural Power

Hardesty R. M. and Weber B. F. (1987) Lidar measurement of turbulence encountered by horizontal-axis wind
turbines. J. Atmos. Oceanic Tech. 67:191–203

Harris M., Pearson G. N., Hill C. A., and Vaughan J. M. (1994) Higher moments of scattered light fields by
heterodyne analysis. Appl. Opt. 33:7226–7230

Harris M., Constant G., and C Ward (2001) Continuous-wave bistatic laser Doppler wind sensor. Appl. Opt.
40:1501–1506

Harris M., Pearson G. N., Ridley K. D., Karlsson C. J., Olsson F. A., and Letalick D. (2001) Single-particle
laser Doppler anemometry at 1.55 µm. Appl. Opt. 40:969–973

Harris M., Hand M., and Wright A. (2006) Lidar for turbine control. Tech. Report NREL/TP-500-39154

Harris M., Bryce D. J., Coffey A. S., Smith D. A., Birkemeyer J., and Knopf U. (2007) Advance measurements
of gusts by laser anemometry. J. Wind Eng. Ind. Aerodyn. 95:1637–1647

DTU Wind Energy-E-Report-0029(EN) 101

Harris M., Locker I., Douglas N., Girault R., Abiven C., and Brady O. (2010) Validated adjustment of remote
sensing bias in complex terrain using CFD. European Wind Energy Conf., Warsaw

Jelalian A. V. (1992) Laser radar systems, Artech House, Boston

Karlsson C. J., Olsson F. A., Letalick D., and Harris M. (2000) All-fiber multifunction CW 1.55 micron
coherent laser radar for range, speed, vibration and wind measurements. Appl. Opt. 39:3716–3726

Kristensen L. (1999) The perennial cup anemometer. Wind Energy 2:59–75

Kragh K. A., and Hansen M. H (2011) Individual Pitch Control Based on Local and Upstream Inflow Measure-
ments. Proceedings of 49th AIAA Aerospace Sciences, American Institute of Aeronautics and Astronautics,
AIAA 2011–813, Orlando, Florida

Kragh K. A., Hansen M. H, and Mikkelsen T. (2013) Precision and shortcomings of yaw error estimation
using spinner-based light detection and ranging. Wind Energy 16:353–366

Lading L., Hanson S., and Skov Jensen A. (1984) Diffraction-limited lidars: the impact of refractive turbulence.
Appl. Opt. 23:2492-2497

Laks J., Pao L. Y., Simley E., Wright A., Kelley N., and Jonkman B. (2011) Model Predictive Control Using
Preview Measurements From LIDAR. Proceedings of 49th AIAA Aerospace Sciences, American Institute
of Aeronautics and Astronautics, AIAA 2011–813, Orlando, Florida.

Loudon R. (2000) The quantum theory of light, Oxford University Press, 3rd edition

Mikkelsen T., Hansen K., Angelou N., Sjöholm M., Harris M., Hadley P., Scullion R., Ellis G., and Vives G.
(2010) Lidar wind speed measurements from a rotating spinner. European Wind Energy Conf., Warsaw

Pearson G. N., Roberts P. J., Eacock J. R., and Harris M (2002) Analysis of the performance of a coherent
pulsed fiber lidar for aerosol backscatter applications. Appl. Opt. 41:6442-6450

Pedersen A. T., Montes B. F., Pedersen J. E., Harris M., and Mikkelsen T. (2012) Demonstration of short-
range wind lidar in a high-performance wind tunnel, EWEA conference PO78, Copenhagen

Peña A., Hasager C. B., Gryning S.-E., Courtney M., Antoniou I., Mikkelsen T. (2009) Offshore wind profiling
using light detection and ranging measurements. Wind Energy 12:105-124

Pitter M., Abiven C., Harris M., Barker W., and Brady O. (2012) Lidar and computational fluid dynamics for
resource assessment in complex terrain. EWEA conference PO80, Copenhagen.

Rogers T., Byme A., McCoy T., and Briggs K. (2012) Expected impacts on cost of energy through lidar based
wind turbine control. EWEA 2012.

A Rutherford A, M Harris M, W Barker W, E Burin de Roziers E., R Pitter M., R Scullion R., and C Slinger
C. (2012) Lidar calibration and performance validation process. AWEA 2012, Atlanta.

Schlipf D., and Kuhn M. (2008) Prospects of a collective pitch control by means of predictive disturbance
compensation assisted by wind speed measurements. German Wind Energy Conference (DEWEK 2008),
pp. 1–4

Schlipf D., Anger J., Kapp S., Bischoff O., Hofsäß M., Rettenmeier A., and Kühn M. (2011) Prospects of
optimization of energy production by lidar assisted control of wind turbines Proceedings of European Wind
Energy Conference and Exhibition, EWEA 2011.

Siegman A. E. (1986) Lasers, University Science Books, Mill Valley

Simley E., Pao L. Y., Frehlich R., Jonkman B., and Kelley N (2011) Analysis of Wind Speed Measurements
using Continuous Wave LIDAR for Wind Turbine Control. Proceedings of 49th AIAA Aerospace Sciences,
American Institute of Aeronautics and Astronautics, AIAA 2011-263, Orlando, Florida

Slinger C., Leak M., Pitter M., and Harris M. (2013) Relative Power Curve Measurements Using Turbine
Mounted, Continuous Wave Lidar. EWEA 2013.

Smith D. A., Harris M., Coffey A. S., Mikkelsen T., Jørgensen H. E., and Mann J. (2006) Wind lidar evaluation
at the Danish wind test site in Høvsøre. Wind Energy 9:87–93

Sonnenschein C. M. and Horrigan F. A. (1971) Signal-to-noise relationships for coaxial systems that heterodyne
backscatter from the atmosphere. Appl. Opt. 10:1600–1604

Vaughan J. M. and Forrester P. A. (1989) Laser Doppler velocimetry applied to the measurement of local
and global wind. Wind Eng. 13:1–15

Wagner R., Jørgensen H. E., Paulsen U. S., Larsen T. J., Antoniou I., and Thesbjerg L. (2008) Remote sensing
used for power curves. IOP Conf. Series: Earth and Environ. Sci. 1:012059

102 DTU Wind Energy-E-Report-0029(EN)

Wagner R., Mikkelsen T., and Courtney M. (2009) Investigation of turbulence measurements with a
continuous-wave conically scanning lidar. Risø-R-1682(EN), Roskilde

S Wharton S., and Lundquist J. K. (2012) Atmospheric stability affects wind turbine power collection. Envi-
ronmental Research Letters textbf7: 014005.

Wiggins J. (2009) Unpublished internal report. Natural Power

Zak J. A. (2003) Atmospheric boundary layer sensors for application in a wake vortex advisory system.
NASA/CR-2003-212175

DTU Wind Energy-E-Report-0029(EN) 103

5 Pulsed lidars
Jean-Pierre Cariou
Leosphere, Orsay, FR

This section complements the description of the measurement process for a Doppler lidar
for wind speed and direction determination, by focusing more on the pulsed lidar technology,
and particularly on the WINDCUBETM Doppler pulsed lidar, one of the most accurate remote
sensing devices available at the present time in the wind industry.

5.1 Introduction
There is a pressing need for good wind-speed measurements at greater heights and at more
site locations to assess the availability of the resource in terms of power production and to
identify any frequently occurring atmospheric structural characteristics that may impact the
operational reliability and lifetime of wind turbines and their components. To measure the
wind field up to the height of new generation wind turbines and at any location of interest,
remote sensors are needed to complement masts.
Different technologies are in use in this field, among them pulsed lidars (LIght Detection
And Ranging). The underlying principle of pulsed lidar measurement of wind and aerosols
is the use of optical heterodyne (coherent) detection, in which laser pulses are transmitted
into the atmosphere and scattered off of naturally-occurring small dust particles (aerosols)
entrained in the ambient flow field (Frehlich et al., 1994; Huffaker and Hardestry, 1996; Soreide
et al., 1997; Frehlich et al., 1998). Even though the measurement principle is well known and
similar to pulsed radars, pulsed lidars have only been used in wind energy site assessment
only since 2008. Their recent introduction is mainly due to laser revolution coming from fiber
telecommunication development in the late 1990s.
In this section, we describe the architecture of pulsed lidars, based on the WINDCUBETM
lidar developed by LEOSPHERE and ONERA, the French Aerospace Lab. We define the differ-
ent modules of a pulsed lidar, their specific functions and the individual level of uncertainties
they might bring to the lidar wind speed retrieval. We also show the differences between
pulsed and continuous wave (CW) lidars, used as well for wind sensing. We eventually give
the lidar equation in pulsed mode, giving the relationship between parameters, and we focus
on range and speed accuracy and resolution.

Figure 58: Leosphere Windcube7, first version (left) and V2 (right).

104 DTU Wind Energy-E-Report-0029(EN)

5.2 End-to-end description of pulsed lidar measurement
process
5.2.1 Architecture of pulsed lidars
Figure 59 illustrates the general set up of a pulsed lidar. The following paragraphs provide
more details of the key hardware components and explain the requirements and trade-offs on
solutions for the main parts of the lidar

Figure 59: Pulsed lidar set up.

Laser source A pulsed lidar needs a continuous wave laser, called master oscillator (MO) to
generate the local oscillator (LO) beam and a pulsed laser to generate the powerful transmitted
pulse. The frequency offset between the two sources need to be stable with time to allow an
unbiased measurement of the Doppler shift.
The master oscillator provides the laser wavelength, the laser linewidth, the laser intensity
noise and the state of polarization. Each of these parameters has to be well known and stable
to guarantee the lidar performance. The required CW power is at least some milliwatts.
The pulsed laser delivers cyclic pulses of high energy. The pulse duration is some hundreds
of nanoseconds, that determines the length of the pulse in the atmosphere and so the spatial
resolution.

Table 10: Spatial resolution versus pulse duration

Pulse duration [ns] Pulse length [m] Minimum spatial resolution [m]
200 60 30
400 120 60
800 240 120

The pulse repetition frequency (PRF) is as high as possible, but cannot exceed a maximum
value PRFmax . To avoid ambiguity between return signals, the time between pulses (1/PRF)
must be longer than the round trip time of flight of the pulse to the greatest height to be
measured Zmax , PRFmax = c/(2Zmax), where c is the speed of light.

Table 11: PRFmax versus lidar range

PRF [KHz] Maximum range [m]

10 15000
20 7500
50 3000

DTU Wind Energy-E-Report-0029(EN) 105

 Two technologies are used for the pulsed laser. In the first one, called MOPA (master oscil-
lator power amplifier), pulses are emitted from the MO by use of an optical pulse modulator.
Resulting low power pulses are amplified into a single pass high bandwidth optical amplifier to
generate high power pulses, having the same duration, frequency and polarization state as the
incoming pulses. In the second scheme, the MO is used as a seeder in a Q-Switched pulsed
laser. Cavity frequencies have to be matched to allow the seed frequency to be amplified.
Pulse duration and PRF depend on the cavity parameters. First solution allows higher PRF
but lower energy/pulse than the second one. Both deliver an equivalent average power.
Laser wavelength is an important parameter. Following the recent improvement of solid
state lasers in late 90s, the near Infrared spectrum (1.4–2.2 µm) is widely used for operational
wind lidars. Efficient technology, good Doppler sensitivity, fiber architecture, eye safety and
good atmospheric transmission are the main reasons (Cariou et al., 2006).

Circulator Figure 60 describes the circulator and the different elements which compose
it. The function of the circulator is to transmit the laser pulse from the laser (1) to the
telescope (2) and to direct the backscattered light from the telescope (2) to the receiver
(3). Power handling, transmission efficiency and isolation between (1) and (3) ports are
critical parameters. In coherent pulsed lidars, most circulators use polarization to perform
this function. Polarization is rotated in the telescope thanks to a quarter wave plate. The
transmitted polarization is circular.

Figure 60: Circulator.

Telescope The telescope magnifies the laser beam in order to reduce its divergence in
the far field, and focuses the beam at any distance. The larger the beam, the smaller the
divergence and the better the signal-to-noise ratio (SNR) at long range. Since the lidar SNR
is inversely proportional to the beam area, beam diameter must be minimized over the global
measurement range to ensure maximum efficiency.
The telescope can use reflective or refractive optics, perfectly corrected from geometrical
aberrations. In order to lose less than 3 dB on the detection efficiency, wavefront distortion
on the global roundtrip optical path has to be less than λ/4 RMS, including components and
atmospheric distortion.
Atmospheric turbulence as well creates wave distortion that degrades heterodyne efficiency.
To keep this distortion negligible, the telescope aperture has to be smaller than the coherence
diameter of the beam expressed by:
do = 2.4e10−8 λ6/5 Z −3/5 Cn−6/5 . (102)
Considering maximum values of index structure constant (Cn2
= 10 −13
m −2/3
) and propaga-
tion over about Z = 1 km, this limits the size of the telescope to about 10 cm at ground
level.

106 DTU Wind Energy-E-Report-0029(EN)

Scanner Coherent lidar measure the radial component of the wind, i.e. the projection of
the wind vector on the line of sight (LOS). To provide two or three components of the vector,
the beam has to be directed in two or three independent different directions (more details
in Section 2.6). The scanner can move the entire telescope, for example Mitsubishi (Ando
et al., 2008), or only the beam. Some manufacturers prefer skipping the scanner and duplicate
the telescope to suppress all moving parts from the lidar (for example Catch The Wind or
Leosphere WindcubeV2).
To perform the vertical wind profile, beams are directed upwards along a cone around the
zenithal direction. For that, a single rotating prism or dual flat mirrors are used (see Figure
61).

Figure 61: Prism scanner and dual flat mirror scanner.

To measure the wind speed in any direction, and display 2D or 3D wind maps, a more
flexible scanner is needed. Double flat mirror scanners are mostly used (CTI, Halophotonics,
Leosphere), while double prism scanners, like those in Risø DTUs Windscanners (Mikkelsen,
2008), offer a more original and compact solution. A global hemispherical field of view can
be obtained with both solutions.

5.2.2 Differences between pulsed vs. continuous wave lidars

In CW lidars, light is continuously transmitted to the atmosphere. Assuming atmospheric
parameters do not vary, the backscattered signal average power is constant, coming from all
distances at the same time. The distance weighting function is defined by the beam aperture
and the focus distance (see chapter on CW lidars from Mike Harris). Spectral bandwidth is
limited by atmospheric turbulence if laser line width is narrow. The size of the range gate
(contributing to the signal) increases as the square of the distance. The range gate can be
small and well defined at short distance but is always too large when the range exceeds few
hundreds of meters. Table 12 summarizes the main differences between CW and pulsed lidars:
In pulsed lidars, short pulses are transmitted to the atmosphere, illuminating at each instant
only a limited part of the line of sight. Therefore, backscattered signal arriving onto the
detector at a time only comes from a given range of distance. The time delay between the
pulse start and the measurement time informs on the distance of the analyzed zone. The
range gate length is always the same, at short and long range.
Pulsed LIDARs, as their name implies, emit regularly spaced emissions of highly collimated
light energy for a specified period of time (pulse length). Precision timing circuits then isolate
the returned signals to a period of time that corresponds to a specified segment of radial
distance along the beam called a range gate. The backscattered signals contained within
each gate are then processed to derive the radial velocities along the path of the LIDAR
beam.

DTU Wind Energy-E-Report-0029(EN) 107

 Table 12: Main differences between CW and pulsed lidars

CW lidar Pulsed lidar

Velocity accuracy limited by coherence limited by the
time of the atmosphere pulse duration
Range gate determined by focus, constant,
increases as R2 around cτ /2
Number of range gates less than 10, more than 100
sequentially addressed simultaneously addressed
limited by SNR at long range
Sensitivity to targets out of focus high no
Maximum range few hundreds of meters some kilometers
Laser source 1000 mW laser 10 mW MO
+ 200 mW power amplifier
Linear polarization not necessary mandatory

5.2.3 Signal processing

After the pulse has left by the laser, the detector starts to collect the backscattered signal
from the successive range gates. It first crosses the telescope optics and provides a zero
Doppler signal, used as a marker for the zero distance. Even if the stray light is small thanks
to optimized coatings (10−6 order), this signal is always larger than the light backscattered
by the atmosphere (10−12 order).
Each layer of atmosphere then backscatters light to the lidar. The power is proportional to
the backscattering atmospheric coefficient β while the frequency shift is proportional to the
radial velocity. For each pulse, the collected signal contains the total wind speed information
on the LOS.
However, both the signal and the noise fluctuate from pulse to pulse and it is necessary
to average signals to get a good estimation of the spectral content. Because of the short
wavelength (104 less than radars and sodars), signal phase changes quickly with particle
motion, atmospheric turbulence and small laser spectral drifts. This is useless to average time
series. To improve SNR on the Doppler spectrum, successive spectra corresponding to the
same range gate are summed. SNR increases as the square root of the number of average
pulses N. This computation is performed for all range gates.
To sum up, the successive steps for the signal processing are for every LOS (Figure 62):
• Break the time series into gates
• Compute the spectrum for each gate
• Average spectra for same range gate from different pulses
• Find frequency peak for each gate to find Doppler shift and convert to radial velocity
• Reconstruct wind vector for each gate with radial velocity on the different LOS.

5.2.4 Coherent detection

Basic principles of coherent detection are the same than for CW lidars (see chapter on CW
lidars from Mike Harris). The return signal mixes with local-oscillator creating the beat sig-
nal. The electronic signal on the detector contains the same amplitude, frequency and phase
information as the optical signal, but is frequency downshifted to allow detection with con-
ventional high speed detectors. So the Doppler shift, which is small in comparison to the
optical frequency ν can be measured in base band. To allow both negative and positive shifts

108 DTU Wind Energy-E-Report-0029(EN)

 Figure 62: Radial wind velocity retrieval process.

to be measured, an offset frequency Fi (intermediate frequency) is added on one arm of the

interferometer.
Fdop = (ν + Fi )LO − (ν + Fdop )signal = Fi + Fdop . (103)
An important advantage of coherent detection is that it can be limited by signal photon
noise, if some conditions are fulfilled. First condition is that the amplitude and phase match
between signal beam and the LO beam must be perfect. The second condition is that temporal
coherence is optimum, i.e the spectral width of the main oscillator must be narrower than
the spectral width of the electronic signal. The third condition is that polarization state must
be the same on the LO and the signal. System and component limitations lead however to a
loss in heterodyne efficiency. Frehlich and Kavaya (1991) demonstrated that the heterodyne
efficiency is limited to 40% by spatial coherence for a perfect Doppler lidar using a circular
aperture and a Gaussian beam. A good actual operational lidar heterodyne efficiency is more
than 20%.

5.2.5 Lidar equation

The lidar equation gives the expected signal power returning from the atmosphere within the
range gate to be analyzed. The signal power can then be compared to the noise power in
order to determine the range of the lidar. The total optical power Pr (z) reflected back in the
receiver telescope from the range gate at Z is:
cτ
Pr (Z) = Ppeak Tinst Tatm βπ (Z) Ω, (104)
2
where Ppeak is the transmitted pulse peak power, Tinst is the instrumental round trip trans-
mission and Tatm is the atmospheric round trip transmission, expressed as:
Z Z !
Tatm = exp −2 α(x)dx . (105)
0

Tatm = exp(−2αZ) if the atmosphere is homogeneous and βpi (Z) is the backscattering
coefficient of the atmosphere at a distance Z, τ is the full-width-half-maximum (FWHM)
pulsed duration and Ω is the reception solid angle:
πσ 2
Ω= , (106)
Z2
where σ is the efficient telescope aperture radius and α and βπ are roughly proportional since
they both depend on aerosol concentration in the atmosphere.
However, because of the limited heterodyne efficiency, only a part of Pr (Z) is efficient for
the coherent detection. The lost part of Pr (Z) comes from phase and polarization mismatches.
An interesting way to estimate the detection antenna diagram is to propagate back in the
atmosphere the LO, and to compute the overlap integral of the signal and LO along the LOS.

DTU Wind Energy-E-Report-0029(EN) 109

This defines the transverse and longitudinal efficiency of the lidar (BPLO theory (Siegman,
1966)). The efficient signal power incoming onto the detector is:
cτ
Ps (Z) = Ppeak Tinst Tatm βπ (Z) λI(Z), (107)
2
where I(Z) is a Lorentzian function including Ω and Z depending focus function:
∆Z
I(Z) = , (108)
(Z − Zo )2 + ∆Z 2
with Zo = Ft /(1 + (Ft /Zr )2 ), Zr = πσ 2 /λ, ∆Z = Zo Ft /Zr being Zo the distance where
I(Z) is maximum, i.e. maximum SNR, Zr the Rayleigh distance, Ft the instrumental geomet-
rical focus distance (wave curvature radius at telescope) and ∆Z half of FWHM geometric
depth of focus.
Figure 63 shows different plots of the Lorentzian function I(Z) for different values of
the focus distance Ft . Short focus improves signal power at short range whereas long focus
averages the power along the distance, leading to a more constant signal on the different
range gates, but with a lower value.

Figure 63: Variations of I(Z) with focus distance Ft = 40, 80, 120, 160, 200, 240 m respec-
tively.

The current power on the detector can then be derived using the same equations as in CW
mode,
< i2het >= 2ηhet S 2 PLO Ps (Z), (109)
where PLO is the local oscillator power, ηhet the heterodyne efficiency, which depends on
phase, amplitude and polarization matching and S the detector sensitivity.
R R ∗
2
R R Ad Es (x, y)E
R LO (x, y)dxdy
R
ηhet = 2 2 , (110)
Ad Es (x, y)dxdy Ad ELO (x, y)dxdy

< i2het >= ηhet Tinst Tatm S 2 βπ (Z)PLO Ppeak τ cI(Z). (111)
The lidar equation shows that the signal power is proportional to pulse energy Ppeak τ and
proportional to LO power. LO amplifies the signal allowing it to be detected over the detector
noise.

5.2.6 Spectral processing MLE

A maximum likelihood estimator (MLE) based on the likelihood of the Fourier transform
of the signal is used as spectral processing. This estimator assumes an uncorrelated Fourier

110 DTU Wind Energy-E-Report-0029(EN)

transform in order to use data obtained from the accumulated spectrum. The estimator is
slightly different from the likelihood of the spectrum traditionally used for spectral maximum
likelihood estimators but still shows the same efficiency.
Signal spectrum is calculated using a temporal model such as the Feuilleté model (Cariou
et al., 2006) and thus takes into account all the FFT algorithm disturbing effects such as the
spectral leakage, which must be carefully characterized in the case of a pulsed atmospheric
lidar.

5.2.7 Wind vector reconstruction

Pulsed lidars provide radial wind components on different lines of sight at different altitudes.
In an ideal case, and to mimic local sensors such as cup or sonic anemometers, beams intersect
at the point of interest within a small volume. This is the goal of the Windscanner project
with three lidars. In an operational situation, only one lidar is available. To reconstruct the
3D components of the wind vector, some assumptions are then necessary.
• Horizontal homogeneity: the three components of the wind are the same for the different
points of the disc at a given altitude. The numerous measurement campaigns have proven
that this assumption is valid on flat terrains and offshore, but not perfect on complex
terrains (hills, mountains, forest boarders)
• Temporal variations are slower than the inter-beam distance divided by the horizontal
wind speed. This time increases with altitude and matches the conical geometry.
• Wind slowly varies within a range gate. Wind dispersion lowers the SNR and provides a
bias if the shear is non linear.
The scanning configuration can be either velocity azimuth display (VAD) or Doppler beam
swinging (DBS). VAD uses information from a continuous scan in a part or total cone angle
and is mostly used in CW lidars. DBS is used in pulsed lidars to average more information
on the LOS. Since VAD is described in the “Remote Sensing QinetiQ Lidar Measurement
Report”, we focus here on DBS reconstruction.
First is important to define an orthogonal frame. The orthogonal frame of the WINDCUBE
is described in Figure 64:

Figure 64: Orthogonal frame of the WINDCUBE for retrieving the wind speed components

DTU Wind Energy-E-Report-0029(EN) 111

 Suppose the lidar probes the atmosphere with three beams in the three directions North,
East, and Zenith. Then the three measured LOS velocities V ri are described as following (θ
being the angle between Zenith and North and East, the so-called cone-angle):
V rN = u sin θ + w cos θ, (112)
V rE = v sin θ + w cos θ, (113)
V rZ = w. (114)
So the u, v and w wind speed components can be retrieved as,
V rN − V rZ cos θ
u = (115)
sin θ
V rE − V rZ cos θ
v = (116)
sin θ
w = V rZ . (117)
Supposing the lidar probes the atmosphere at four different locations, the LOS being East,
West, North, South, the system of wind equations will be,
V rN − V rS
u = (118)
2 sin θ
V rE − V rW
v = (119)
2 sin θ
V rN + V rS + V rE + V rW
w = . (120)
4 cos θ
Horizontal wind speed Vh and wind direction Dir are then retrieved as following:
p
Vh = u2 + v 2 (121)
Dir = mod(360 + atan2(v, u), 360). (122)
If the North beam of the lidar is offset from the geographical North with an angle α, the wind
direction is:
Dir = mod(360 + α + atan2(v, u), 360). (123)
With three beams, the solution is unique. The vertical component w is perfectly determined
if the one of the three axis is accurately vertical. No check of assumptions mentioned above
is possible. With four beams, the additional equation allows wind homogeneity to be checked
and skip undesired values.
Cone-angle θ is a trade-off between lidar velocity resolution and atmosphere homogeneity.
The smaller is θ, the better is the wind homogeneity but worse is the projection of the wind
vector on every beam. Boquet et al. (2010a) demonstrated that best θ values are between
15◦ and 30◦ . Even in complex terrains, in general wind non homogeneity condition, no better
estimation is obtained when reducing the cone angle (Boquet et al., 2009).

5.2.8 Fiber lidars

Before early 2000s, LIDAR systems were based on solid-state laser technologies that do not
meet operational requirements for remote site wind assessment due to high power consump-
tion, size, weight, reliability, and life cycle cost. It was therefore the purpose of Leosphere,
thanks to a partnership with French Aerospace Lab ONERA to introduce a unique fiber laser
technology geared for the wind industry requirements, enabling efficient realization of compact
wind Doppler lidar systems.
Fiber lidars use fiber amplifiers and coherent detection and fiber architecture based on
mainstream telecommunication components. Fiber amplifiers use codoped Erbium Ytterbium
silica fibers to amplify with a large bandwidth low power pulses cut out of a CW laser at 1.5 µm
(MOPA configuration). The electrical to optical efficiency of 1.5 µm fiber laser sources is of
the order of 10%, thus allowing low electrical consumption.

112 DTU Wind Energy-E-Report-0029(EN)

 This wavelength is also the most favorable for eye-safe lidar designs: the eye-safety laser
energy limitation being high, the laser power can be increased with little constraints on the
lidar operation or design. One advantage of the IR fiber technology is its reliability. It is
now well established that a fiber architecture is easy to adjust and mechanically reliable in
a vibrating environment. The other advantages of fiber architectures are their compactness
and flexibility in terms of installation. The lidar can be split up into subsystems spatially
far apart and linked together using fiber optics. The new technologies of large-mode-area
(LMA) fibers enable high peak power generation without nonlinear effects, while maintaining
a good spatial mode and polarization state. The average power exceeds several watts and high
PRF compensates efficiently the relative low pulse energy. Moreover, the MOPA architecture
flexibility in terms of pulse duration allows fulfilling a large panel of requirements, either with
high spatial resolution or long range.

5.3 Lidar performances

5.3.1 Noise
In coherent detection, noise sources come principally from 3 origins:
< i2SN > = 2eS PLO B LO shot noise (124)
< i2N EP > = 2eS N EP B Detector noise (125)
< i2RIN > = 2
(S PLO ) 10 RIN/10
B RIN noise, (126)
where N EP is the noise equivalent power density, B is the detection bandwidth and,
< i2n >=< i2SN > + < i2N EP > + < i2RIN > . (127)
For optimum detection, LO shot noise must be the main noise contributor. When other
sources are negligible, CNR (Carrier to Noise Ratio), describing the signal to noise ratio on
the carrier frequency is,
< i2het > S
CNR = 2
= ηhet Tinst Tatm βπ (Z)Ppeak τ c I(Z). (128)
< in > 2eB

5.3.2 Best Focus

Focus distance can be adjusted in order to optimize CNR over the measurement range. Figure
65 shows the simulated CNR variation versus focus distance, expressed as the wave radius of
curvature at the instrument exit (beam radius being 11 mm@1/e2 at the lens).

Figure 65: Variation of CNR vs distance (Altitude/cos(θ-dev)) for different beam radius of
curvature

Best focus corresponds to maximizing the data availability at all altitudes, from H = 40
m (Z = 46 m) to H = 200 m (Z = 230 m). Practically, it corresponds to balancing and

DTU Wind Energy-E-Report-0029(EN) 113

maximizing the CNR for H = 40 m and H = 200 m. For the Windcube WLS7, Fopt=120 m
(Hopt=104 m), corresponding to a beam curvature of 200 m at lens (black dot line).
For a beam radius of curvature of 200 m, the beam diameter slowly varies along the
propagation in the range of interest. Beam radius (1/e2 ) is σ = 11 mm at lens, 9 mm at 46
m, 7 mm at 120 m and 10 mm at 230 m, giving a maximum of 2 dB difference along the
total range. (CNR is proportional to 1/σ 2 ).
Maximum focus distance corresponds to half of the Rayleigh distance Zr :
Zr πσ 2
Zmax = = , (129)
2 2λ
where Zmax = 123 m for the WINDCUBE7 v2 parameters. This configuration minimizes the
beam diameter variation from Z = 0 to Z = Zr .
Lindelöw (2007) demonstrated a velocity error coming from the unbalanced velocity weight-
ing function due to a variation of CNR within the range gate. In the case of the WINDCUBE7
v2, the difference of CNR is always less than 0.5 dB/range gate, leading to a maximum bias
in the velocity of around 0.06 m s−1 under a vertical linear wind shear of 0.02 m s−1 /m.

5.3.3 Distance range and resolution

Because of the footprint of the laser pulse on the line of sight, range resolution is limited.
Moreover, during the measurement time τm , the pulse has moved further, enlarging the
range resolution. Figure 66 describes the pulse space and time propagation and the portion
of atmosphere illuminated during a time window analysis of length τm .

Figure 66: Pulse propagation and width of range gate represented on a time-distance plot

Velocity measurement at one point Vd (Z) depends on the velocity V r(R) at close points
and on the range weighting function RW F (R),
Z ∞
V d(Z) = RW F (R)V r(R)dR. (130)
−∞

For a pulsed lidar, Banakh and Smalikho (1994) gave an analytical equation for RWF, when
the pulse is Gaussian (FWHM = τ ) and range gate is flat (width τm ), as the convolution
between the pulse power profile and the range gate profile:
" √ √ ! √ √ !#
1 4 ln 2 ln 2 4 ln 2 ln 2
RW F (Z) = erf (Z − Zo ) + − erf (Z − Zo ) − .
τm c cτ τ cτ τ
(131)
Range resolution is defined as the FWHM of the function RW F which is roughly:
cτ
∆z1 = √ m . (132)
2erf ln 2τm /τ

114 DTU Wind Energy-E-Report-0029(EN)

Eq. (132) is adapted for collimated system but does not give accurate results for focused lidars,
i.e. focusing the laser beam leads to better resolution near the focusing point. Moreover, this
method is only valid for a Gaussian pulse and a flat measurement window.
In order to make this calculation more general, Lindelöw (2007) proposed multiplying the
focusing efficiency by the convolution of the pulse and the range gate profile:
RW F (Z) = ηf oc (Z) (P ulse F F Twindow ) (Z), (133)
where ηf oc (Z) is the focusing efficiency,
  −1
1 1
ηf oc (Z) = 1 + Zr2 − . (134)
Z Zf oc
These methods can be applied to the current WINDCUBE7 v2 pulse shape and range gate
profile. The “impulse response” FWHM is calculated to be 27 m. The altitude resolution is
therefore around 23.8 m (LOS zenithal deviation angle is 28◦ ). Lindelöw’s RWF is 26.3 m,
i.e. ∼ 23.2 m altitude resolution.

5.3.4 Velocity range and resolution

Velocity range ambiguity Radial velocity is proportional to Doppler frequency shift. Ve-
locity range is then determined by frequency range. The intermediate frequency Fi used in
most pulsed lidars allows the measurement of both positive and negative shifts. Maximum
downshift is limited by the value of Fi , since no negative frequency can be measured. Max-
imum upshift is limited by the Nyquist frequency, half of the sampling frequency Fs . Figure
67 describes the spectrum and velocity ranges.

Figure 67: Lidar spectrum and velocity range

Because of spectral extent of the signal and low frequency noise, this range is in fact a bit
smaller. For example, with λ = 1.55 µm, Fi = 68 MHz and Fs = 250 MHz, the practical
horizontal velocity range is [−50 m s−1 , +50 m s−1 ]. A passband filter cancels outband
Doppler shifts to avoid Doppler ambiguities.

Velocity resolution Velocity precision depends on both atmospheric parameters and lidar
parameters. Both are broadening the Doppler spectrum and hence limit the frequency es-
timation. Atmospheric parameters are wind gradient within the range gate and turbulence.
Lidar parameters are pulse duration CNR and number N of average spectra. The smaller the
pulse duration, the smaller the range gate and the velocity dispersion but larger the frequency
spectrum.
Eq. (135) gives the minimum velocity resolution as a function of relative parameters. It is
called Cramer Rao lower bound (CRLB),
√ √
2λ 1 + CNR
σvcrlb = √ . (135)
2τ N CNR
Eq. (135) assumes an infinite correlation time of the signal. In an actual lidar, σv is limited
by the finite correlation time τc , the smallest value between the pulse duration and the

DTU Wind Energy-E-Report-0029(EN) 115

correlation time of the atmosphere,
λ
σvsat = √ . (136)
2 N 4πτc
The global velocity resolution is then:
q
σv = 2
σvcrlb 2 .
+ σvsat (137)
Figure 68 illustrates the variation of σv with CNR for N = 100 and N = 10000.

Figure 68: Cramer Rao boundary for 100 averaged spectra (dark red) and 10000 averaged
spectra (light red). CNR is measured in narrow band (B = 1/τm )

For low CNR values, σv decreases as 1/CNR. Doubling the pulse energy divides by 2
the velocity resolution. For high CNR values, σv is constant. Spectral broadening comes from
speckle fluctuations in√the signal. The only way to reduce σv is to increase N . For intermediate
CNR, σv varies as 1/ CNR. It is therefore equivalent to increase pulse energy or number of
pulses. In this region, σv depends only on the average laser power. Using a low energy, high
pulse repetition rate (PRF) laser is then equivalent to using a high energy, low PRF laser,
assuming the average power is constant. This is one for the reasons of the recent raise of
high PRF fiber lasers for pulsed Doppler lidars.

Range ambiguity The rate at which pulses are transmitted, the PRF limits the range over
which heights can be unambiguously determined. To avoid ambiguity between return signals,
the inter pulse period (IPP=1/PRF) must be longer than the round trip time of flight of the
pulse to the greatest height. For example, to measure without ambiguity up to 5 km, the
PRF needs to be less than 30 KHz,
c
PRFmax = . (138)
2Zmax

5.3.5 Time-bandwidth tradeoffs

Spatial resolution is proportional to pulse duration. The shorter the pulse, the smaller the
resolution. Velocity resolution is proportional to spectrum width and is smaller when the
spectrum is narrow. Because the spectrum width is inversely proportional to the pulse duration,
range resolution and velocity resolution are also inversely proportional.

116 DTU Wind Energy-E-Report-0029(EN)

5.3.6 Existing systems and actual performances
In 2013, only a few pulsed lidars are available. Table 13 summarizes the characteristics of
commercial ones, which can be used in the wind industry, meteorology or airport safety. All
are based on coherent detection. In 2011, Pentalum introduced a new pulsed lidar, based
on backscatter correlation on two adjacent beams and direct detection. This technique is
currently under evaluation.

Table 13: Comparison of commercially available pulsed lidars

System Wave Range/ Data Sample pulse PRF Beam

length accuracy update volume duration/ [kHz] config.
[µm] [m]/[m s−1 ] [s] length energy
[m] [ns]/[µJ]
Leosphere 1.54 40–200/0.1 1/10 alt 20 175/10 30 Five
WindCube7 beams DBS
WindCube 1.54 100–6000/0.3 1 75 400/100 10 LOS
200S mapping
WindCube 1.54 200–12000/0.5 1 150 800/200 10 LOS
400S mapping
LMCT 1.6 400–15000/1 0.1 80 300/2000 0.75 LOS
Windtracer mapping
Mitsubishi 1.5 100–1500/NA 1 90 600/6.5 1 Scanning
head
Sgurr 1.5 40–250/0.1 0.1–30 24 150/10 20 LOS
Gallion G250 mapping
G4000 1.5 80–4000/NA NA 30 NA NA LOS
mapping

5.3.7 Validation of measurements

Since the instrumental sources of uncertainty are now well identified and the range of deviation
they might incur in the wind speed measurement are well estimated, it is necessary to compare
the measurement of a new lidar unit against a tall mast equipped with traditional anemometry
or against a well-known and validated lidar in a double phase verification/validation.
Validation is the process of ensuing that a WINDCUBE measures wind speed characteristics
in conformity with what a reference instrument would give. Today, validation is done against
well-known traditional anemometry, like a mast equipped with calibrated cup anemometers.
This instrument comparison introduces additional uncertainties, not related to the inner per-
formances of each individual instrument but related to the differences in the measurement
process between the two instruments. These differences in the measurement process might
incur differences in measured values, which will be site and time specific. Even if these are
small, it is important to closely define the range of variation that might occur on the validation
site.
With more than 140 field deployments worldwide for the past 4 years, the Windcube has
been extensively tested in different conditions and under a diversity of climates. Several authors
have reported very good measurement accuracy with reference to calibrated cup anemometers
in good operating conditions (Oldroyd et al., 2009; Faghani et al., 2009; Jaynes, 2009).

DTU Wind Energy-E-Report-0029(EN) 117

Figure 69: Comparison of Windcube 10-min averages with calibrated mast at Risø Høvsøre

5.4 Conclusions and perspectives

This chapter summarizes the principles of operation, the performances and critical parameters
of Doppler pulsed wind lidars, such as the WINDCUBE lidar from LEOSPHERE. It takes into
account the different steps of the velocity measurement process. This detailed process analysis
provides the backbone for a pulsed lidar development, optimization and issues related to serial
production.
More generally, pulsed and CW lidars have shown great reliability and accuracy in the
measurement of wind characteristics such as horizontal and vertical wind speed and wind
direction at various heights on flat terrains and offshore (Courtney et al., 2009; Westerhellweg
et al., 2010). The high data recovery rate up to above blade top ensures a high quality analysis
of the wind conditions available on the project site, leading to an optimized layout design and
the choice of the suitable turbines.
Regarding future improvements, the measurement of additional wind parameters like tur-
bulence intensities, kinetic fluxes or inflow-angles is under investigation at the present time
and should be available in the near future.
One of the remaining challenges is also reaching high accuracy in complex terrains where
flow distortion occurs and impairs the lidar wind components retrieval from measured radial
velocities. Since lidar radial velocities are still very accurate even on complex and rough
terrains, a methodology using CFD modeling has been recently developed to avoid taking
the flow homogeneity assumption (Boquet et al., 2009; Bingöl et al., 2009; Boquet et al.,
2010a). Even though new, this methodology has already shown good results on sites of various
complexities. The measurement of additional wind parameters like turbulence intensities,
kinetic fluxes or inflow angles is under investigation at the present time and should be available
in the near future.
To sum up, lidar anemometry has already proven its great utility in the development of
wind farm projects, as an instrument allowing considerable financial gains through a better
understanding of the wind conditions at a site and therefore reducing the capital risk of
the investors (Boquet et al., 2010b), but also for operational power curves measurements

118 DTU Wind Energy-E-Report-0029(EN)

(Gottshall et al., 2010; Albers et al., 2010). With the general trend to develop larger wind
turbines and wider wind farms, the lidar technology is elected to be more and more widely
used.

DTU Wind Energy-E-Report-0029(EN) 119

Notation
B detector bandwidth
c speed of light
Cn index structure constant
CNR carrier-to-noise ratio
CRLB Cramer Rao lower bound
CW continuous wave
do coherence diameter of the beam
DBS Doppler beam swinging
FFT fast Fourier transform
Fi intermediate frequency
Fs sampling frequency
Ft instrumental geometrical focus distance (wave curvature radius at telescope)
FWHM full width half maximum
< i2NEP > detector noise
< i2RIN > rin noise
< i2SN > shot noise
I(Z) Lorentzian function
IPP inter pulse period
LMA large mode area
LOS line of sight
LO local oscillator
MLE maximum likelihood estimator
MO master oscillator
MOPA master oscillator power amplifier
N EP noise equivalent power density
PLO local oscillator power
Ppeak transmitted pulse peak power
Pr total optical power
PRF pulse repetition frequency
PRFmax maximum pulse repetition frequency
N number of average pulses
R focus distance
RW F (Z) range weighting function
S detector sensitivity
SNR signal to noise ratio
Tatm atmospheric round trip transmission
Tinst instrumental round trip transmission
u north-south wind speed component
v east-west wind speed component
VAD velocity azimuth display
Vd (Z) velocity measurement at one point
V ri LOS velocities, i = N North, i = E East, i = S South, i = W West and i = Z Zenith
w vertical wind speed component
Z height
Zmax maximum height to be measured
Zo distance where I(Z) is maximum, i.e. signal to noise is maximum
Zr Rayleigh distance
β backscattering atmospheric coefficient
∆Z half of FWHM geometric depth of focus
ηf oc focusing efficiency
ηhet heterodyne efficiency
θ lidar cone angle
λ laser wavelength
ν optical frequency
σ efficient telescope aperture radius
σv global velocity resolution
σvcrlb Cramer Rao lower bound of velocity resolution
σvsat smallest value of the velocity resolution
τ pulse duration
τc finite correlation time
τm measurement time

120 DTU Wind Energy-E-Report-0029(EN)

References
Albers A., Janssen A. W. and Mander J. (2010) Hot to gain acceptance for lidar measurements. DEWEK

Ando T., Kameyama S. and Hirano Y. (2008) All-fiber coherent Doppler lidar technologies at Mitsubishi
Electric Corporation. IOP Conf. Series: Earth and Environ. Science 1:012011

Banakh V. and Smalikho I. (1994) Estimation of the turbulence energy dissipation rate from the pulsed
Doppler lidar data. J. Atmos. Oceanic Technol. 10

Bingöl F., Mann J. and Foussekis D. (2009) Conically scanning lidar error in complex terrian. Meteorol. Z.
18:1–7

Boquet M., Parmentier R. and Cariou J. P. (2009) Analysis and optimization of pulsed Doppler lidar in complex
terrain. EWEC, Marseille

Boquet M., Parmentier R. and Cariou J. P. (2010) Correction of pulsed wind lidars bias in complex terrain.
ISARS

Boquet M., Callard P., Neve N. and Osler E. G. (2010) Return on investment of a lidar remote sensing device.
DEWI Magazin 37:56–61

Cariou J.-P., Augere B. and Valla M. (2006) Laser source requirements for coherent lidars based on fiber
technology. Comptes Rendus Physique 7:213–223

Courtney M., Wagner R. and Lindelöw P. (2009) Commercial lidar profilers for wind energy. A comparative
guide. EWEC, Marseille

Faghani D. et al. (2009) Remote sensing: practical use for wind power project. AWEA Wind Resource &
Project Energy Assessment Workshop, Portland

Frehlich R. G. and Kavaya M. J. (1991) Coherent laser radar performance for general atmospheric refractive
turbulence. Appl. Opt. 30

Frehlich R. G., Hannon S. M., Henderson S. W. (1993) Performance of a 2-coherent Doppler lidar for wind
measurements. J. Atmos. Oceanic. Technol. 11:1517–

Frehlich R. G., Hannon S. M. and Henderson S. W. (1998) Coherent Doppler lidar measurements of wind
field statistics. Bound.-Layer Meteorol. 86:233–256

Gottshall J., Courtney M. S., Lindelöw P. and Albers A. (2010) Classification of lidar profilers: How to introduce
lidars to power performance testing. EWEC, Brussels

Huffaker R. M. and Hardestry R. M. (1996) Remote sensing of Atmospheric wind velocities using solid state
and CO2 coherent laser systems. Proc. IEEE 84

Jaynes D. (2009) Lidar validation and recommendations for wind resource assessments. Wind Power Conf.,
Chicago

Lindelöw P. (2007) Effective sample volume fiber based coherent lidars for remote sensing of wind. PhD thesis
from Ørsted-DTU, Lyngby

Mikkelsen T. (2008) WindScanner: a facility for wind and turbulence measurements around large wind turbines.
In Nordic Conf. on: Global Challenges-Regional Opportunities: How can research infrastracture and Science
support Nordic competiveness?, Stockholm

Oldroyd A. et al. (2009) Testing and calibration of various lidar remote sensing devices for a 2 year offshore
measurement campaign. EWEC, Marseille

Siegman A. E. (1966) The antenna properties of optical heterodyne receivers. Appl. Optics 5:1588–1594

Soreide D., Boque R. K., Seidel J. and Ehernberger L. J. (1997) The use of a lidar forward looking turbulence
sensor for mixed-compression inlet unstart avoidance and gross weight reduction on a high speed civil
transport Nasa Technical memorandum

Westerhellweg A., Beeken A., Canadillas B., Neumann T. (2010) One year of lidar measurements at Fino1-
platform: comparison and verification to met-mast data. DEWEK

DTU Wind Energy-E-Report-0029(EN) 121

6 Remote sensing for the derivation of
the mixing-layer height and detection of
low-level jets
Stefan Emeis
Institute of Meteorology and Climate Research,
Atmospheric Environmental Research Division (IMK-IFU),
Karlsruhe Institute of Technology, Garmisch-Partenkirchen, Germany

6.1 Introduction
This chapter gives an overview of the derivation of mixing-layer height (MLH) and the de-
tection of low-level jets (LLJs) by surface-based remote sensing instruments such as sodar,
lidar, ceilometer and RASS. The detection of vertical profiles indicating the structure of the
atmospheric boundary layer (ABL) is one of the principal tasks of experimental boundary-
layer research. MLH has become an important input parameter for the description of wind
profiles above the surface layer (Gryning et al., 2007; Peña et al., 2010). LLJs are secondary
wind maxima which occur at the top of a stable boundary layer. Over land, they are observed
at several hundred metres above ground at night-time. Over the sea, they are found with
offshore winds when warmer air flows over cooler water at the top of the shallow stable inter-
nal boundary layer at any time of the day. These shallow internal boundary layers are often
considerably less than 100 m deep.
The next section describes methods to detect the mixing-layer height, while Section 6.3
briefly mentions methods to capture the boundary-layer height. Section 6.4 gives more infor-
mation on low-level jets. Acoustic (sodar) and optical sounding techniques (lidar) have got a
broad coverage elsewhere in this volume. RASS techniques are still quite unusual in the assess-
ment of wind resources. Therefore, a subsection on technical details of this instrumentation
has been added in section 6.2. A more complete survey of remote sensing instrumentation
is given in Emeis (2010), an overview of applications of ground-based remote sensing is pre-
sented in Emeis (2011). The full scope of wind energy meteorology is presented in Emeis
(2012).

6.2 Mixing-layer height

We must distinguish between the mixing-layer height, MLH (see section 6.2) and the boundary-
layer height, zi (see section 6.3). The boundary-layer height is the height up to which the
influence of the presence of the lower surface is detectable. The mixing-layer height is the
height up to which atmospheric properties (such as wind speed and turbulence) or substances
originating from the surface are dispersed by turbulent vertical mixing processes. The mixing-
layer – if it is present at all – is a part of the ABL. The mixing-layer height is usually shallower
than the boundary-layer, but it fills the whole ABL in deep convective boundary layers.
Sometimes the terms mixed-layer height or mixing height are used as well for MLH, but we
will stick here to the most common term mixing-layer height. The mixing-layer height is the
height up to which atmospheric properties or substances originating from the Earth’s surface
or formed within this layer are dispersed almost uniformly over the entire depth of this layer by
turbulent vertical mixing processes. Therefore, the existence and the height of a mixing layer
can either be analyzed from a detection of the presence of the mixing process, i.e. turbulence,
or from the verification that a given conservative atmospheric variable is distributed evenly
over the full height range of the well-mixed layer. The level of turbulence can for instance be

122 DTU Wind Energy-E-Report-0029(EN)

derived from fluctuations of the wind components or from temperature fluctuations. Suitable
conservative atmospheric variables for the identification of the mixing layer and its height are,
e.g., potential temperature, specific humidity or aerosol particle concentrations. Temperature
can be measured with RASS, temperature gradients may be assessed from sodar data and
aerosol concentration data from lidar and ceilometer measurements. Therefore, these instru-
ments offer the opportunity to detect mixing-layer heights. A suitable remote sensing method
for deriving humidity profiles with high vertical resolution (comparable to sodar, RASS and
lidar) is still missing.
A well-known overview of methods to determine MLH from in-situ measurements and
surface-based remote sensing had been given by Seibert et al. (2000). Since then considerable
development has taken place, especially concerning the usage of optical surface-based remote
sensing methods (see the review paper by Emeis et al. (2008)) and RASS. Updates are given
in Emeis (2011) and Emeis (2012).
Optical methods for MLH detection may be used to illustrate this recent progress. Seibert
et al. (2000) still classified lidar methods as expensive, not eye-save, with a high lowest range
gate, limited range resolution, and sometimes subject to ambiguous interpretation. This has
changed drastically in the last ten years when and smaller Doppler wind lidars have been built
and the simpler non-Doppler ceilometers have been discovered to be a nearly ideal sounding
instrument for the detection of the vertical structure of the boundary layer. Progress has been
made in the field of acoustic sounding as well. Similarly, algorithms for the determination of
MLH from vertical profiles of the acoustic backscatter intensity as described in Beyrich (1997)
and Seibert et al. (2000) have been enhanced by using further variables available from sodar
measurements such as the wind speed and the variance of the vertical velocity component
(Asimakopoulos et al., 2004; Emeis and Türk, 2004). Such enhancements had been named
as possible methods in Beyrich (1995) and Seibert et al. (2000) but obviously no example
was available at that time.
A variety of different algorithms have been developed by which the MLH is derived from
ground-based remote sensing data (see Table 14 for a short overview). We will mainly con-
centrate on acoustic and optical remote sensing because electro-magnetic remote sensing
(by RADAR and wind profiler) has too high lowest range gates for a good coverage of shal-
low MLH. The disadvantage of a too high lowest range gate can partly be circumvented by
slantwise profiling or conical scanning if the assumption of horizontal homogeneity can be
made.

6.2.1 Acoustic detection methods (Sodar)

Acoustic methods either analyze the acoustic backscatter intensity, or, if Doppler shifts in the
backscattered pulses can be analyzed, features of vertical profiles of the wind components
and its variances as well. The acoustic backscatter intensity is proportional to small-scale
fluctuations in atmospheric temperature (usually generated by turbulence) or by stronger
vertical temperature gradients. The latter feature may be an indication for the presence of
temperature inversions, which can often be found at the top of the mixing layer.
Beyrich (1997) listed possible analyses which can mainly be made from acoustic backscatter
intensities measured by a sodar. Later, Asimakopoulos et al. (2004) summarized three different
methods to derive MLH from sodar data: (1) the horizontal wind speed method (HWS), (2)
the acoustic received echo method (ARE), and (3) the vertical wind variance method (VWV).
We will mainly follow this classification here and finally add a fourth method, the enhanced
ARE method (EARE).
Figure 70, showing an acoustic sounding taken in an Alpine valley, gives an impression what
wealth of detailed vertical information can be derived from acoustic boundary-layer sounding.
The left-hand frame displays the acoustic backscatter intensity and the right-hand frame the
wind direction as time-height sections over one day (from midnight to midnight) and over
a height range of 700 m. The depicted wintry situation from a day in January exhibits a
multiple layering of the air in that valley due to the very stable thermal stratification of the

DTU Wind Energy-E-Report-0029(EN) 123

Table 14: Overview of methods using ground-based remote sensing for the derivation of the
mixing-layer height mentioned in this chapter (see right most column to find its placement
in section 6.2).

method short description which section?

acoustic ARE analysis of acoustic 6.2.1
backscatter intensity
acoustic HWS analysis of wind 6.2.1
speed profiles
acoustic VWV analysis of vertical 6.2.1
wind variance profiles
acoustic EARE analysis of backscatter and 6.2.1
vertical wind variance profiles
optical threshold detection of a given 6.2.2
backscatter intensity threshold
optical gradient analysis of backscatter 6.2.2
intensity profiles
optical idealized backscatter analysis of backscatter 6.2.2
intensity profiles
optical wavelet analysis of backscatter 6.2.2
intensity profiles
optical variance analysis of backscatter 6.2.2
intensity profiles
acoustic/electro-magnetic RASS 6.2.3
acoustic/electro-magnetic sodar-RASS and 6.2.3
windprofiler-RASS
acoustic/electro-magnetic/in situ sodar-RASS plus 6.2.3
surface heat flux data
acoustic/electro-magnetic sodar plus windprofiler 6.2.4
acoustic/optical sodar plus ceilometer 6.2.4

Figure 70: Sample time-height cross-section from acoustic sounding with a sodar. Left: acous-
tic backscatter intensity, right: horizontal wind direction. Thin black lines demark inversions.

valley air over a snow-covered valley floor. The multiple layering originated from an interlacing
of down-valley (wind direction around 190◦) and down-slope (wind direction around 230◦ )
flows. The layers are separated by temperature inversions and each higher layer is potentially
warmer than the next lower layer. They persisted nearly the whole day because no vertical
mixing took place in the stably stratified valley atmosphere.

124 DTU Wind Energy-E-Report-0029(EN)

Acoustic received echo (ARE) method The ARE method is the most basic method of
determining MLH from acoustic remote sensing. Most of the methods listed in Beyrich (1997)
belong to this method. The method does not require an analysis of the Doppler shift of the
backscattered signals. The method makes use of the assumption that turbulence is larger
in the mixing layer than in the atmosphere above, and that this turbulence is depicted in
the intensity of the acoustic backscatter. MLH is analyzed either from the maximum negative
slope or from the changing curvature of the vertical profile of the acoustic backscatter intensity
or it is analyzed from the height where the backscatter intensity decreases below a certain
pre-specified threshold value.

Horizontal wind speed (HWS) method The HWS method requires the analysis of the
Doppler shift of the backscattered acoustic signals. The algorithm is based on the analysis
of the shape of hourly-averaged vertical wind speed profiles using the assumption that wind
speed and wind direction are almost constant within the mixing layer but approach gradually
towards the geostrophic values above the mixing layer. Beyrich (1997) listed this method in
his Tab. 2 but did not discuss it further. The applicability of the method is probably limited
to the well-developed convective boundary layers (CBL) due to the underlying assumptions.
Such CBLs are often higher than the maximum range of a sodar. Even if the CBL height is
within the range of the sodar the algorithm for the analysis of the Doppler shift often fails
above the inversion topping of the CBL due to too low signal-to-noise ratios. Today, small
Doppler wind lidars are available to derive wind speed and direction profiles through the whole
depth of the boundary layer. This facilitates the application of the HWS method.

Vertical wind variance (VWV) method The VWV method is also working only for CBLs.
It is based on the vertical profile of the variance of the vertical velocity component σw . In a
CBL σw reaches a maximum in a height azi . Typical values for a are between 0.35 and 0.40.
Thus, in principle, this is an extrapolation method. It has been tried for sodar measurements
because it permits a detection of MLH up to heights which are 2.5 times above the limited
maximum range (usually between 500 and 1000 m) of the sodar. Beyrich (1997) classified
this method as not reliable. A related method, which is based on power spectra of the vertical
velocity component, is integrated in the commercial evaluation software of certain sodars
(Contini et al., 2009). The application of the VWV method is now also been facilitated by
the easy availability of small Doppler wind lidars.

Enhanced acoustic received echo (EARE) method The EARE algorithm has been pro-
posed by Emeis and Türk (2004) and Emeis et al. (2007). The method is an enhancement of
the ARE method in two ways. Firstly, it includes further variables into the MLH algorithm that
are available from Doppler-sodars. The benefits of the additional usage of the variance of the
vertical velocity component have been demonstrated by Emeis and Türk (2004). Secondly, it
determines not only MLH from sodar measurements but also the heights of additional lifted
inversions. Especially in orographically complex terrain, the vertical structure of the ABL can
be very complicated. Emeis et al. (2007) have shown that several persistent inversions one
above the other which form in deep Alpine valleys can be detected from sodar measurements
(Fig. 70).
EARE determines three different types of heights based on acoustic backscatter intensity
and the variance of the vertical velocity component. Because the horizontal wind information
above the inversion is not regularly available from sodar measurements, horizontal wind data
have not been included into this scheme. In the following a letter “H” and an attached
number will denote certain derived heights which are related to inversions and the MLH;
while the variable z is used to denote the normal vertical coordinate. The EARE algorithm
detects:
• the height (H1) of a turbulent layer characterised by high acoustic backscatter intensities
R(z) due to thermal fluctuations (therefore having a high variance of the vertical velocity
component σw ),

DTU Wind Energy-E-Report-0029(EN) 125

 • several lifted inversions (H2n ) characterized by secondary maxima of acoustic backscat-
ter due to a sharp increase of temperature with height and simultaneously low σw (like
those depicted in the left-hand frame of Fig. 70), and
• the height of a surface-based stable layer (H3) characterised by high backscatter inten-
sities due to a large mean vertical temperature gradient starting directly at the ground
and having a low variance of the vertical velocity component.
The height H1 corresponds to a sharp decrease ∂R/∂z < DR1 of the acoustic backscatter
intensity R(z) below a threshold value Rc with height z usually indicating the top of a
turbulent layer:

H1 = z, if (R(z) < Rc and R(z + 1) < R(z) + zDR1

and R(z + 2) < R(z) + 2zDR1 ). (139)
Rc = 88 dB and DR1 = −0.16 dB m−1 have proven to be meaningful values in the
above mentioned studies. Rc is somewhat arbitrary because the received acoustic backscatter
intensities from a sodar cannot be absolutely calibrated. An absolute calibration would require
the knowledge of temperature and humidity distributions along the sound paths for a precise
calculation of the sound attenuation in the air. DR1 is, at least for smaller vertical distances,
independent from the absolute value of Rc . An application-dependent fine-tuning of Rc and
DR1 may be necessary.
Elevated inversions are diagnosed from secondary maxima of the backscatter intensity that
are not related to high turbulence intensities. For elevated inversions increase in backscat-
ter intensity below a certain height z = H2 and a decrease above is stipulated while the
turbulence intensity is low:

H2n = z, if (∂R/∂z|z+1 < −DR2 and ∂R/∂z|z−1 > DR2

and σw < 0.70 m s−1 ) (140)
for n = 1, ..., N . In Emeis et al. (2007) N was chosen to be five. A threshold value
DR2 = 0.08 dB m−1 has proven suitable. But again, an application-dependent tuning may
be advisable.
The determination of the height of the stable surface layer H3 is started if the backscatter
intensity in the lowest range gates is above 105 dB while σw is smaller than 0.3 m s−1 . The
top of the stable layer H3 is at the height where either the backscatter intensity sinks below
105 dB or σw increases above 0.3 m s−1 ,

H3 = z, if (R(z) > 105 dB and R(z + 1) < 105 dB and σw (z) < 0.3m s−1 ) or
if (σw (z) < 0.3 m s−1 and σw (z + 1) > 0.3 m s−1 and R(z) > 105 dB).(141)
The σw values used in Eqs. (140) and (141) have been determined by optimizing the
automatic application of the detection algorithm. In doing so it turned out that no lifted
inversions occurred with a variance σw higher than 0.7 m s−1 and that the variance σw in
nocturnal stable surface layers was always below 0.3 m s−1 . The first σw threshold made it
possible to distinguish between inversions and elevated layers of enhanced turbulence. The
latter σw threshold made it possible to differentiate between nocturnal stable surface layers
and daytime super-adiabatic surface layers although both types of surface layers yield more
or less the same level of backscatter intensity. Finally MLH from the acoustic remote sensing
is determined as the minimum of H1, H21 , and H3:

MLHac = min (H1, H21 , H3) . (142)

126 DTU Wind Energy-E-Report-0029(EN)

Figure 71: Sample time-height cross-section from optical sounding with a ceilometer. Left:
optical backscatter intensity, right: vertical derivative of this backscatter intensity. Dots mark
mixing-layer height derived from a gradient algorithm.

6.2.2 Optical detection methods

Usually the aerosol content of the mixing layer is higher than in the atmospheric layer above,
because the emission sources for aerosol are in most cases on the ground. Aerosol formation
from precursors mainly takes place near the surface as well. Making the assumption that the
vertical aerosol distribution adapts rapidly to the changing thermal structure of the boundary
layer, MLH can be determined from the analysis of the vertical aerosol distribution. This also
includes the assumption that the vertical aerosol distribution is not dominated by horizontally
advected aerosol plumes or layers. The heights of the near surface aerosol layers (H4n ) can
be analysed from the optical vertical backscatter profile obtained from optical remote sensing.
Several methods have been developed, the most prominent of these being: (1) the threshold
method, (2) the gradient or derivative method, (3) the idealised gradient method, (4) the
wavelet method, and (5) the variance method. In addition, the horizontal wind speed method
and the vertical wind variance method mentioned in section 6.2.1 above are available to derive
the vertical structure of the boundary layer from Doppler wind lidar data.
The application of optical remote sensing for MLH determination has focussed on the use
of ceilometers in recent years. Ceilometers can be regarded as a small lidar. They are simpler
and they have a much lower lowest range gate than lidars. For the detection of MLH below
150 to 200 m a ceilometer with one optical axis for the emitted and the received beam should
be used. Due to the thin light beams the overlap of the emitted and received beam from
a ceilometer with two parallel optical axes can be insufficient in this height range. Further
on, Doppler shifts are not analyzed by ceilometers. Therefore, in contrast to acoustic remote
sensing with Doppler-sodars, additional variables in addition to the backscatter intensity are
not available from ceilometers for the design of determination schemes for MLH. Thus the
schemes listed below all resemble to the ARE methods for acoustic remote sensing.
Figure 71 shows a sample measurement with a mono-axial ceilometer. The left-hand frame
displays the optical backscatter intensity and the right-hand frame the negative vertical deriva-
tive of this intensity as time-height sections over one day (from midnight to midnight) and
over a height range of 2000 m. The data was received on a clear day in spring and the ver-
tical structure of the ABL was dominated by surface heating due to incoming solar radiation
during daytime and radiative surface cooling during night-time. In the morning hours until
about 0900 LST a shallow stable nocturnal surface layer with a depth of about 200 m and
a residual layer with a depth of about 1200 to 1400 m can be distinguished. From 0900
LST onwards the evolution of a daytime convective boundary layer with a maximum depth of
about 1400 m can be clearly seen. The dots in both frames of Fig. 71 indicate the mixing-layer
height determined with the gradient method described below. The right-hand frame in Fig.
71 demonstrates that the analysed MLH values indeed coincide with maxima of the negative
vertical gradient of the optical backscatter intensity.

DTU Wind Energy-E-Report-0029(EN) 127

Threshold method Melfi et al. (1985) and Boers et al. (1988) used simple signal threshold
values, though this method suffers from the need to define them appropriately (Sicard et al.,
2006). H4 is defined here as the height within the vertical profile of the optical backscatter
intensity where the backscatter intensity first exceeds a given threshold when coming down-
ward from the free unpolluted troposphere. The determination of several heights H4n would
require the definition of several thresholds which probably cannot be done a priory to the
analysis. Therefore this will always lead to a subjective analysis of MLH. The left-hand frame
in Fig. 71 shows that the threshold value cannot be kept constant during the diurnal evolution
of the boundary layer in order to get a result which is comparable to the one from the gradient
method applied in Fig. 71.

Gradient or derivative methods Hayden et al. (1997) and Flamant et al. (1997) proposed
to use the largest negative peak of the first derivative of the optical attenuated backscatter
intensity (B(z)) for the detection of H4 from LIDAR data (height of gradient minimum
H4GM ):

H4GM = min(∂B(z)/∂z). (143)

The right-hand frame of Fig. 71 demonstrates that this is a very meaningful assumption.
Likewise Wulfmeyer (1999) used the first minimum of the slope to detect the top of a convec-
tive boundary layer from DIAL data. Münkel and Räsänen (2004) and Schäfer et al. (2004,
2005) applied the gradient method to ceilometer data. Menut et al. (1999) took the minimum
of the second derivative of B(z) as the indication for MLH:

H4IPM = min(∂ 2 B(z)/∂z 2 ). (144)

This method is called inflection point method (IPM). It usually gives slightly lower values
for H4 than the gradient method in Eq. (143). A further approach was suggested by Senff
et al. (1996). They looked for the largest negative gradient in the logarithm of the backscatter
intensity (height of logarithmic gradient minimum H4LGM ):

H4LGM = min(∂ ln B(z)/∂z). (145)

This approach usually gives the largest value for H4. According to Sicard et al. (2006)
H4IPM from Eq. (144) is closest to the MLH derived from radiosonde ascents via the Richard-
son method. The other two algorithms in Eqs. (143) and (145) give slightly higher values.
In Emeis et al. (2007) the gradient method in Eq. (143) has been further refined and
extended to enable the calculation of up to n = 5 lifted inversions. This algorithm, which
has also been used for the MLH analysis shown in Fig. 71, is described in the following.
Prior to the determination of gradient minima the overlap and range corrected attenuated
backscatter profiles have to be averaged over time and height to suppress noise generated
artefacts. Therefore the H4 values are determined in a two-step procedure. Between 140 and
500 m height sliding averaging is done over 15 min and a height interval ∆h of 80 m. In
the layer between 500 and 2000 m ∆h for vertical averaging is extended to 160 m. Two
additional parameters have been introduced to further reduce the number of false hits. The
minimum accepted attenuated backscatter intensity Bmin right below a lifted inversion is set
to 200 × 10−9 m−1 srad−1 in the lower layer and 250 × 10−9 m−1 srad−1 in the upper layer.
Additionally the vertical gradient value ∂B/∂zmax of a lifted inversion must be more negative
than 0.30 × 10−9 m−2 srad−1 in the lower layer and more negative than −0.60 × 10−9 m−2
srad−1 in the upper layer.
If B(z) denotes the measured attenuated backscatter intensity in the height z above ground
averaged over time and height and ∆h is the height averaging interval, then the gradient
∂B/∂z in the height z is calculated as
∂B/∂z|z = (B(z + ∆h/2) − B(z − ∆h/2)) /∆h. (146)

128 DTU Wind Energy-E-Report-0029(EN)

 A gradient minimum is characterized by a change of sign from minus to plus of the second
derivative of B(z). The height interval under examination is searched from bottom to top for
these gradient minima H4n .
The second derivative of B(z) in the height z is


∂ 2 B/∂z 2|z = ∂B/∂z|z+∆h/2 − ∂B/∂z|z−∆h/2 /∆h. (147)
There is a gradient minimum H4n in the height z if the second derivative of B(z) one
range gate below z is not positive, if the second derivative of B(z) in the height z is positive,
and if the false hit conditions mentioned above are fulfilled:

H4n = z, if ∂ 2 B/∂z 2|z−1 ≤ 0 and ∂ 2 B/∂z 2 |z > 0 and B(z − ∆h/2) ≥ Bmin
and ∂B/∂z|z ≤ ∂B/∂zmax for n = 1, ..., 5. (148)
The MLH from optical remote sensing is taken as the lowest height H4n :

MLHop = H41 . (149)

Idealised backscatter method A parallel development by Eresmaa et al. (2006) using an

idealised backscatter profile, originally described by Steyn et al. (1999), is also an extension of
the gradient method. MLH is not determined from the observed backscatter profile, but from
an idealised backscatter profile fitted to the observed profile. The robustness of this technique
is founded on utilising the whole backscatter profile rather than just the portion surrounding
the top of the mixing layer. In this method an idealized backscattering profile Bi (z) is fitted
to measured profile by the formula

Bi (z) = ((Bm + Bu ) /2 − (Bm − Bu ) /2) erf((z − h)/∆h) (150)

where Bm is the mean mixing layer backscatter, Bu is the mean backscatter in air above the
mixing layer and ∆h is related to the thickness of the entrainment layer capping the ABL in
convective conditions. Two new parameters A1 and A2 are defined so that A1 = (Bm +Bu )/2
and A2 = (Bm − Bu )/2. The value of A1 is kept constant during the fitting procedure. A
good estimation of A1 based on an initial order-of-magnitude guess for the MLH is crucial
for the quality of the result.

Wavelet method A Wavelet method has been developed for the automatic determination
of mixing layer height from backscatter profiles of an LD-40 ceilometer by de Haij et al.
(2006). Before that wavelet transforms have been applied in recent studies for MLH deter-
mination from LIDAR observations (Cohn and Angevine, 2000; Davis et al., 2000; Brooks,
2003; Wulfmeyer and Janjić, 2005). The most important advantage of wavelet methods is the
decomposition of the signal in both altitude as well as vertical spatial scale of the structures
in the backscatter signal.
The Wavelet algorithm in de Haij et al. (2006) is applied to the 10 minute averaged range
and overlap corrected backscatter profile B(z) within a vertical domain of 90–3000 m. For
each averaged profile the top of two significant aerosol layers are detected in order to detect
MLH as well as the top of a secondary aerosol layer, like e.g. an advected aerosol layer or
the residual layer. This Wavelet MLH method uses the scale averaged power spectrum profile
WB (z) of the wavelet transform with 24 dilations between 15 and 360 m and step size 15 m.
The top of the first layer, H41 , is detected at the first range gate at which the scale averaged
power spectrum WB (z) shows a local maximum, exceeding a threshold value of 0.1. This
threshold value is empirically chosen, based on the analysis of several cases with both well
pronounced and less clearly pronounced mixing layer tops. H42 is optionally determined in
the height range between H41 and the upper boundary of detection. A valid H42 is detected
at the level with the strongest local maximum of WB (z) provided that this maximum is larger
than the WB (z) of H41 . MLH is set equal to H41 .

DTU Wind Energy-E-Report-0029(EN) 129

Figure 72: Sample time-height cross-section from a potential temperature sounding with
RASS.

However, problems with this method arise e.g. in case of multiple (well defined) aerosol
layers, which renders the selection of the correct mixing layer top ambiguous. Furthermore,
in spring and summer the detection of the MLH for deep (convective) boundary layers often
fails. This is mostly due to the high variability of the aerosol backscatter signal with height
which limits the range for MLH estimation in those conditions (de Haij et al., 2006).

Variance method At the top of the CBL we have entrainment of clear air masses from
the free troposphere into the ABL. The entrainment process is temporarily variable and leads
locally to considerable fluctuations in the aerosol concentration. Therefore the maximum in
the vertical profile of the variance of the optical backscatter intensity can be an indicator for
an entrainment layer on top of a CBL (Hooper and Eloranta, 1986; Piironen and Eloranta,
1995). The method is called variance centroid method in Menut et al. (1999). The variance
method for the CBL height is also described in Lammert and Bösenberg (2006). Due to
the assumptions made this method is suitable for daytime convective boundary layers only.
An elucidating comparison between the gradient method and the variance method can be
found in Martucci et al. (2004) although they used a Nd:YAG LIDAR at 532 nm instead of
a ceilometer and thus suffered from a high lowest range gate in the order of 300 m.

6.2.3 RASS
The acoustic and optical methods for MLH determination, which have been described in the
sections above, are all indirect methods that try to infer the mixing-layer height from other
variables which usually adapt to the vertical structure of the ABL. The only direct and key
variable for the analysis of the presence of a mixing layer is the vertical profile of virtual
temperature. Temperature profiles can directly be measured with a radio-acoustic sounding
system (RASS). Fig. 72 shows an example. We start here with a short description of the
available RASS methods.

Instrumentation A radio-acoustic sounding system (RASS) operates acoustic and electro-

magnetic sounding simultaneously (Marshall et al., 1972). This instrument is able to detect
acoustic shock fronts of the acoustic pulses and to determine their propagation speed from the
Doppler shift of the backscattered electro-magnetic waves. This propagation speed is equal
to the speed of sound which in turn is a known function of air temperature and humidity. To
different types of RASS have been realised (Engelbart and Bange, 2002): a Bragg-RASS and

130 DTU Wind Energy-E-Report-0029(EN)

Figure 73: Bragg-related acoustic (below) and electro-magnetic (above) frequencies for RASS.

a Doppler-RASS.

Bragg-(windprofiler) RASS A Bragg-RASS (or windprofiler-RASS) is basically a windpro-

filer with an additional acoustic emitter. When the Bragg condition is fulfilled (Fig. 73), i.e.
the wavelength of the sound waves λa is half the one of the electro-magnetic waves λe , then
there is optimal backscatter of the electro-magnetic waves from the acoustic waves (Fig. 74).
The electro-magnetic signal is emitted at a fixed frequency, but the emitted sound signal is
a chirp signal with varying frequency fa . From the sound wave length λa,B at which opti-
mal backscatter occurs the propagation speed of the sound signal can be determined via the
following dispersion relation:

ca = λa,B fa /2. (151)

For a VHF windprofiler operating at 50 MHz a sound frequency of about 100 Hz is used,
for a UHF windprofiler operating at 1 GHz a sound frequency around 2 kHz is most suitable
to fulfil the Bragg condition. Because the attenuation of sound waves in the atmosphere is
strongly frequency dependent, a UHF RASS can detect temperature profiles up to about 1.5
km height whereas a VHF RASS can observe temperature profiles throughout the troposphere.

Doppler-(sodar) RASS A Doppler-RASS (or sodar-RASS) is a sodar with an additional

electro-magnetic emitter and receiver (Fig. 75) operating at a frequency fe,0 . From the
Doppler shift ∆fe of the electro-magnetic radiation which is backscattered at the density
fluctuations caused by the sound waves the propagation speed ca of the sound waves is
determined:

ca = −c∆fe /(2fe,0 ) (152)

where c denotes the speed of light. A Doppler-RASS like a Bragg-RASS also emits a chirp
sound signal in order to assure that the Bragg condition is optimally met due to the varying
temperature over the entire height range.
The so determined propagation speed ca is a sum of the speed of sound cs and of the
vertical movement of the air w within which the sound waves propagate:

ca = cs + w. (153)

DTU Wind Energy-E-Report-0029(EN) 131

 Figure 74: Bragg condition for RASS sounding.

Figure 75: Sodar-RASS. The acoustic antenna is in the middle, the electro-magnetic antennas
to the left and right.

The vertical air speed component w can be determined separately from the Doppler shift
of the backscattered electro-magnetic clear-air signal when operating a Bragg-RASS or from
the Doppler shift of the backscattered acoustic signal when operating a Doppler-RASS. Using
the definition of the acoustic temperature the height profile of cs can then be converted in a
height profile of the acoustic temperature Ta . For many purposes this acoustic temperature
is a sufficiently accurate approximation of the virtual air temperature. The maximum range
of a sodar-RASS is usually less than thousand metres so that such an instrument covers the
lower part of the boundary layer. Due to the high vertical resolution and the low minimum
range of about 30 to 40 m, which is comparable to the abilities of a sodar, a sodar-RASS is
well suited for the detection of shallow nocturnal boundary layers.

MLH detection algorithms MLH can be determined from the lowest height where the
vertical profile of potential temperature increases with height indicating stable thermal strat-
ification of the air. The great advantage of RASS measurements is that the magnitude of

132 DTU Wind Energy-E-Report-0029(EN)

stability (inversion strength) can be assessed quantitatively which is not possible from the
acoustic and optical sounding devices described before. Fig. 72 displays a time-height cross-
section of potential temperature over three days starting at midnight for a height range of
540 m. On the afternoons of the second and the third day well-mixed boundary layers formed,
which can be deduced from the vertically constant potential temperature. New surface layers
form on the evenings of all three days at about 1800 LST. The depth of these new surface
layers increase during the night to about 200 to 300 m. Above these nocturnal surface layers
low-level jets (see below) form, indirectly visible from the white areas indicating missing data
in Fig. 72. Stronger winds like those in low-level jets blow the sound pulses from the RASS out
of the focus of the electro-magnetic antenna and hence lead to a failure of the temperature
detection.
Ideally, thermal stratification of air should be analyzed from the virtual potential temper-
ature (Θv = Θ(1 + 0.609q), where q is specific humidity) in order to include the effects
of the vertical moisture distribution on the atmospheric stability. Unfortunately, no active
remote sensing device for the determination of high-resolution moisture profiles is available.
Therefore, the acoustic potential temperature (Θa = Θ(1 + 0.513q)), which actually is the
temperature that is delivered by a RASS, is often used as a substitute. This is sufficient for
cold and dry environments, but somewhat underestimates the virtual potential temperature in
humid and warm environments. In case of larger vertical moisture gradients and small vertical
temperature gradients this can lead to a switch in stability from stable to unstable or vice
versa.
Engelbart and Bange (2002) have analyzed the possible advantages of the deployment of
two RASS instruments, a sodar-RASS and a high-UHF windprofiler-RASS, to derive boundary-
layer parameters. With these instruments, in principle, MLH can either be determined from
the temperature profiles or from the electro-magnetic backscatter intensity. The latter de-
pends on temperature and moisture fluctuations in the atmosphere. The derivation of MLH
from the temperature profile requires a good vertical resolution of the profile which is mainly
available only from the sodar-RASS. But even if the inversion layer at the top of the bound-
ary layer is thick enough, due to the high attenuation of sound waves in the atmosphere,
also the 1290 MHz-windprofiler-RASS used by Engelbart and Bange (2002) can measure the
temperature profile only up to about 1 km. Therefore, in the case of a deeper CBL, MLH was
determined from a secondary maximum of the electro-magnetic backscatter intensity which
marks the occurrence of the entrainment zone at the CBL top. Thus, with this instrument
combination the whole diurnal cycle of MHL is ideally monitored by interpreting the tem-
perature profile from the sodar-RASS at night-time and by analyzing the electro-magnetic
backscatter intensity profile from the windprofiler-RASS during daytime.
Hennemuth and Kirtzel (2008) have recently developed a method that uses data from a
sodar-RASS and surface heat flux data. MLH is primarily detected from the acoustic backscat-
ter intensity received by the sodar part of the sodar-RASS and verified from the temperature
profile obtained from the RASS part of the instrument. Surface heat flux data and statistical
evaluations complement this rather complicated scheme. The surface heat flux is used to
identify situations with unstable stratification. In this respect this observable takes over an
analogous role as the σw in the EARE algorithm (see above). The results have been tested
against radiosonde soundings. The coincidence was good in most cases except for a very low
MLH at or even below the first range gate of the sodar and the RASS.

6.2.4 Further techniques

Beyrich and Görsdorf (1995) have reported on the simultaneous usage of a sodar and a wind
profiler for the determination of MLH. For the sodar data the ARE method was used. From
the wind profiler data MLH was likewise determined from the height of the elevated signal
intensity maximum (Angevine et al., 1994; Grimsdell and Angevine, 1998; White et al., 1999).
Good agreement between both algorithms was found for evolving CBLs. The vertical ranges
of the two instruments (50 to 800 m for the sodar and 200 to 3000 m for the wind profiler)

DTU Wind Energy-E-Report-0029(EN) 133

allowed following the complete diurnal cycle of MLH.

6.2.5 Comparison of acoustic and optical MLH detection algorithms

There is an interesting difference between the schemes for the determination of MLH from
acoustic and optical backscatter intensities which should be noted carefully. While the acous-
tic backscatter intensity itself is taken for the detection of H1 and H3 (see Eqs. (139) and
(141)) and the first derivative of this backscatter intensity for the determination of H2 (see
Eq. (140)), the first and the second derivative of the optical backscatter intensity (but not the
optical backscatter intensity itself) is used to determine H4 (see Eq. (143)). This discrepancy
in the processing of the two backscatter intensities is due to the different scattering processes
for acoustic and optical waves: Acoustic waves are scattered at atmospheric refractivity gra-
dients and thus at temperature gradients (Neff and Coulter, 1986) while optical waves are
scattered at small particles. Therefore the optical backscatter intensity is proportional to the
aerosol concentration itself. The MLH on the other hand, which we desire to derive from
these backscatter intensities, is in both cases found in heights where we have vertical gradi-
ents of the temperature and of the aerosol concentration. Therefore, in principle, the vertical
distribution of the acoustic backscatter intensity should look very much alike the vertical
distribution of the vertical gradient of the optical backscatter intensity.
Simultaneous measurements with different remote sensing devices have mainly been made
in order to evaluate one remote sensing method against the other (Devara et al., 1995).
But one could also think of combining the results two or more remote sensing devices for
determining the structure of the ABL. The analysis of the sodar data and the ceilometer data
can be combined to one single piece of MLH information by forming the minimum from Eqs.
(142) and (149):

MLH = min (MLHac , MLHop ) . (154)

6.3 Boundary-layer height

Often, the boundary layer consists of more layers than just the mixing layer. For example,
at night, a residual layer may persist over a newly formed near-surface stable surface layer.
The deployment of sophisticated lidars (Bösenberg and Linné, 2002) may be a choice to
detect such features as well as the combined deployment of a sodar and a ceilometer. Such a
combination of parallel measurements of the vertical structure of the atmospheric boundary
layer by a ceilometer and a sodar is described in Emeis and Schäfer (2006). Fig. 76, which is
taken from this study, shows a daytime convective boundary layer, shallow nocturnal surface
layers in the morning and the evening, and a residual layer above the nocturnal surface layers.
The ceilometer detects the overall boundary layer height (blue triangles) whose height is partly
modified by large-scale sinking motion in the anticyclone dominating the weather during the
measurement period. Stable nocturnal surface layers with a depth of a few hundred metres can
be detected underneath the black squares derived from the sodar soundings. The convective
boundary layer during daytime fills the full depth of the boundary layer. This combination
offers the same advantages as the combination of sodar and windprofiler presented in Beyrich
and Görsdorf (1995). First results from a combined deployment of a RASS and a ceilometer
are given in Emeis et al. (2009).
In such combined remote sensing measurements a sodar better detects the near-surface
features such as nocturnal stable layers (a RASS instead of a sodar directly delivers the near-
surface temperature profile) while the ceilometer is able to follow the diurnal variation of the
daytime convective boundary layer and the top of the whole boundary layer. The residual
layer then becomes visible as the layer between the new nocturnal surface layer below and
the top of the boundary layer above.

134 DTU Wind Energy-E-Report-0029(EN)

Figure 76: Combined sounding with a sodar (black squares and red asterisks) and a ceilometer
(blue triangles) giving a complete view of the diurnal variation of the vertical structure of the
ABL.

Figure 77: Example of a sodar observation of a nocturnal low-level jet.

6.4 Low-level jets

Sometimes vertical wind profiles exhibit secondary maxima in heights which are relevant
for today’s large wind energy converters, because the vertical wind shear underneath these
maxima may be enhanced considerably. Figure 77 shows an example for a nocturnal low-level
jet over land.

6.4.1 Formation
The formation of low-level jets requires a temporal or spatial change in the thermal stability
of the atmospheric boundary layer which leads to a sudden change between two different
equilibria of forces. The flow must transit from an unstable or neutral condition where fric-
tion, pressure-gradient and Coriolis forces balance each other to a stable condition where only
pressure-gradient and Coriolis force balance each other (see Fig. 78). The sudden disappear-

DTU Wind Energy-E-Report-0029(EN) 135

Figure 78: Balance of forces before (upper frame) and after (lower frame) the onset of a
low-level jet. The red arrows indicate the changes leading to the low-level jet.

ance of the retarding friction in the equilibrium of forces leads to an inertial oscillation of
the horizontal wind vector. Wind speed shoots to much higher values and the increased wind
speed leads to a stronger Coriolis force which provokes a turning of the wind vector as well.
The relevant equations for the wind components u and v at the moment of the disappearance
of the frictional force read:
∂u
= −f (v − vg ) (155)
∂t
∂v
= f (u − ug ) (156)
∂t
where f is the Coriolis parameter and ug and vg are the components of the geostrophic wind.
The terms on the left-hand side involve a dependence on time. Therefore, the analytical
solution of Eqs. 155 and 156 describes an oscillation with time, t:
u − ug = Dv sin f t + Du cos f t (157)
v − vg = Dv cos f t − Du sin f t (158)
where Du and Dv are the ageostrophic wind components at the beginning of the oscillation
in the moment when the retarding frictional force vanishes.
In the temporal domain this corresponds to a sudden change from an unstable daytime
convective boundary layer to a nocturnal stable boundary layer. This requires clear skies
in order to have rapid changes in thermal stratification but still non-vanishing horizontal
synoptic pressure gradients. Therefore, nocturnal low-level jets usually appear at the edges of
high-pressure systems (see shaded area in Fig. 79).
In the spatial domain this corresponds to a sudden transition of the flow from a surface
which is warmer than the air temperature to a smooth surface which is colder than the air
temperature. This may happen when the flow crosses the coast line from warm land to a
colder ocean surface or from bare land to snow or ice-covered surfaces.

6.4.2 Frequency and properties of low-level jets

It was mentioned in the preceding subchapter that the occurrence of nocturnal low-level
jets depends on certain synoptic weather conditions. Therefore, it can be expected that the
frequency of occurrence is linked to the appearance of certain weather or circulation types.
For Central Europe the “Grosswetterlagen” (large-scale weather types) have proven to give
a good classification of the weather situation (Gerstengarbe et al., 1999). Figure 80 shows

136 DTU Wind Energy-E-Report-0029(EN)

 Figure 79: Favorite regions (shaded area) for the formation of nocturnal low-level jets.

Figure 80: Annual frequency of low-level jet events ordered by large-scale synoptic weather
patterns (Grosswetterlagen). From two years of sodar data for Hannover, Germany.

the frequency of occurrence of low-level jets over Northern Germany as function of these 29
large-scale weather types. The two most relevant types (the two left-most columns in Fig. 80)
are a high-pressure bridge over Central Europe (type “BM”) and a high-pressure area over
the British Isles (type “HB”). All in all a low-level jet appeared in 23% of all nights
Figure 80 showed the frequency of occurrence of a low-level jet as function of the weather
type. The relevance of a certain weather type for the formation of a low-level jet can be
assessed when comparing the frequency of low-level jet occurrence with the overall frequency
of occurrence of the respective weather type. Figure 81 has been produced by dividing the fre-
quencies shown in Fig. 80 by the occurrence frequency of the respective weather types during
the same observation period. There are two weather types where the occurrence frequency
is identical to the occurrence frequency of the low-level jets during this weather type. This
means that in every night when this weather type prevailed a low-level jet was observed. This
is indicated by a low-level jet efficiency of 1.0 in Fig. 81. Small deviations from unity are due
to the limited sample size evaluated for this purpose. These two weather types are “HNFA”
and “HFZ” which are both related to high-pressure systems to the North of the investigation
site
Such a high efficiency for forming a low-level jet allows for a quite certain forecast of the
occurrence of a low-level jet. Once such weather types are forecasted a low-level jet will form
with a very high probability. The values given in Fig. 81 can be used to give the low-level jet

DTU Wind Energy-E-Report-0029(EN) 137

Figure 81: Efficiency of a large-scale synoptic weather pattern (Grosswetterlage) to form a
low-level jet. From two years of sodar data for Hannover, Germany.

Figure 82: Correlation between the height of the core of a low-level jet and its maximum wind
speed over Northern Germany (from two years of sodar measurements).

formation probability for Northern Germany for each of the weather types. For other areas
the investigation has to be repeated with local low-level jet data.
The height of the core of a low-level jet and its maximum wind speed seems to be correlated.
Figure 82 shows a correlation analysis from two years of sodar measurements in Northern
Germany. Weaker jet cores may appear at heights between 150 m and 200 m above ground
while stronger events (20 m s−1 core wind speed) usually appear at about 400 m with
considerable scatter. Typical wind shear values for the layer between the surface and 160 m
above ground vary between 0.04 and 0.10 s−1 . An analysis of daily wind speed amplitudes
at 160 m above ground from these sodar data shows that amplitudes of up to 14 m s−1 are
possible due to low-level jet events.

6.5 Summary
Wind resources depend on the large-scale weather conditions as well as on the local vertical
structure of the atmospheric boundary layer. Ground-based remote sensing is now a viable
technique to monitor the vertical structure of the atmospheric boundary layer. Three different
techniques are presently available: acoustic sounding (sodars), optical sounding (wind lidars
and ceilometers) and the combination of acoustic and electro-magnetic sounding (RASS).

138 DTU Wind Energy-E-Report-0029(EN)

Direct detection of MLH from acoustic backscatter intensities is limited to the order of about
1 km due to the rather high attenuation of sound waves in the atmosphere. In contrast,
optical remote sensing offers much larger height ranges of at least several kilometres, because
the attenuation of light waves in the atmosphere is small unless there is fog, clouds or
heavy precipitation. This now offers the possibility to introduce information on boundary
layer structure such as mixing-layer height and the frequency of occurrences of low-level jets
into the monitoring and description of wind resources.

Notation
a factor of proportionality
A1 (BM + BU ) /2
A2 (BM − BU ) /2
ABL atmospheric boundary layer
ARE acoustic received echo (method)
Bmin threshold value for B(z)
BM mixing-layer mean of B(z)
BU value of B(z) above the mixing layer
B(z) optical backscatter intensity
Bi (z) idealized optical backscatter intensity
∂B/∂zmax threshold value for the vertical derivate of B(z)
c speed of light
ca RASS speed of sound (= cs + w)
cs true speed of sound
CBL convective boundary layer
Du initial ageostrophic wind component
Dv initial ageostrophic wind component
DR1 threshold value for the vertical gradient of R(z)
DR2 threshold value for the vertical gradient of R(z)
EARE enhanced acoustic received echo (method)
f Coriolis parameter
fa acoustic frequency
fe electro-magnetic frequency
Hn analysed height (n is a one-digit number)
H4GM height of minimum of vertical gradient of B(z)
H4IPM height of minimum of second vertical derivate of B(z)
H4LGM height of minimum of logarithmic vertical gradient of B(z)
HWS horizontal wind speed (method)
MLH mixing-layer height
MLHac mixing-layer height from acoustic sounding
MLHop mixing-layer height from optical sounding
q specific humidity
Rc threshold value for R(z)
R(z) acoustic backscatter intensity
RASS radio-acoustic sounding system
t time
Ta acoustic temperature
u horizontal wind component (Eastwards)
ug horizontal component of geostrophic wind (Eastwards)
v horizontal wind component (Northwards)
vg horizontal component of geostrophic wind (Northwards)
VWV vertical wind variance (method)
w vertical wind component
WB (z) scale averaged power spectrum profile
z height above ground
zi boundary-layer height
∆fe Doppler shift of electro-magnetic frequency
∆h height interval
Θ potential temperature
Θa acoustic potential temperature
Θv virtual potential temperature
λa wavelength of sound wave

DTU Wind Energy-E-Report-0029(EN) 139

 λa,B wavelength of sound wave fulfilling the Bragg condition
λe wavelength of electro-magnetic wave
σw standard deviation of the vertical wind component

References
Angevine W., White A. B., and Avery S. K. (1994) Boundary layer depth and entrainment zone characterization
with a boundary layer profiler. Bound.-Layer Meteorol. 68: 375–385

Asimakopoulos D. N., Helmis C. G., and Michopoulos J (2004) Evaluation of SODAR methods for the
determination of the atmospheric boundary layer mixing height. Meteorol. Atmos. Phys. 85: 85–92

Beyrich F. (1995) Mixing height estimation in the convective boundary layer using sodar data. Bound.-Layer
Meteorol. 74: 1–18

Beyrich F. (1997) Mixing height estimation from sodar data - a critical discussion. Atmosph. Environ. 31:
3941–3954

Beyrich F. and Görsdorf U. (1995) Composing the diurnal cycle of mixing height from simultaneous SODAR
and Wind profiler measurements. Bound.-Layer Meteorol. 76: 387–394

Boers R., Spinhirne J. D., and Hart W. D. (1988) Lidar Observations of the Fine-Scale Variability of Marine
Stratocumulus Clouds. J. Appl. Meteorol. 27: 797–810

Bösenberg J. and Linné H. (2002) Laser remote sensing of the planetary boundary layer. Meteorol. Z. 11:
233–240

Brooks I. M. (2003) Finding boundary layer top: application of a wavelet covariance transform to lidar backscat-
ter profiles. J. Atmos. Oceanic Technol. 20: 1092–1105

Cohn S. A. and Angevine W. M. (2000) Boundary Layer Height and Entrainment Zone Thickness Measured
by Lidars and Wind-Profiling Radars. J. Appl. Meteorol. 39: 1233–1247

Contini D., Cava D., Martano P., Donateo A., and Grasso F. M. (2009) Comparison of indirect methods for
the estimation of Boundary Layer height over flat-terrain in a coastal site. Meteorol. Z. 18: 309–320

Davis K. J., Gamage N., Hagelberg C. R., Kiemle C., Lenschow D. H., and Sullivan P. P. (2000) An objective
method for deriving atmospheric structure from airborne lidar observations. J. Atmos. Oceanic Technol.
17: 1455–1468

de Haij M., Wauben W., and Klein Baltink H. (2006) Determination of mixing layer height from ceilometer
backscatter profiles. In: Slusser J. R., Schäfer K., Comerón A. (eds) Remote Sensing of Clouds and the
Atmosphere XI. Proc. SPIE Vol 6362: paper 63620R.

Devara P. C. S., Ernest Ray P., Murthy B. S., Pandithurai G., Sharma S., and Vernekar K. G. (1995)
Intercomparison of Nocturnal Lower-Atmospheric Structure Observed with LIDAR and SODAR Techniques
at Pune, India. J. Appl. Meteorol. 34: 1375–1383

Emeis S. (2010) Measurement methods in atmospheric sciences. In situ and remote. Vol. 1 of Quantifying
the Environment. Borntraeger, Stuttgart, XIV+257 pp., 103 Figs, 28 Tab. ISBN 978-3-443-01066-9.

Emeis S. (2011) Surface-Based Remote Sensing of the Atmospheric Boundary Layer. Series: Atmospheric and
Oceanographic Sciences Library, Vol. 40. Springer Heidelberg etc., X+174 pp. 114 illus., 57 in color., H/C.
ISBN: 978-90-481-9339-4, e-ISBN 978-90-481-9340-0, ISSN 1383-8601, DOI: 10.1007/978-90-481-9340-0

Emeis S. (2012) Wind Energy Meteorology - Atmospheric Physics for Wind Power Generation. Springer
Heidelberg etc., XIV-196 pp. ISBN 978-3-642-30522-1, e-book: ISBN 978-3-642-30523-8

Emeis S. and Schäfer K. (2006) Remote sensing methods to investigate boundary-layer structures relevant to
air pollution in cities. Bound.-Layer Meteorol. 121: 377–385

Emeis S. and Türk M. (2004) Frequency distributions of the mixing height over an urban area from SODAR
data. Meteorol. Z. 13: 361–367

Emeis S., Jahn C., Münkel C., Münsterer C., and Schäfer K (2007) Multiple atmospheric layering and mixing-
layer height in the Inn valley observed by remote sensing. Meteorol. Z. 16: 415–424

Emeis S., Schäfer K., and Münkel C (2008) Surface-based remote sensing of the mixing-layer height - a review.
Meteorol. Z. 17: 621–630

Emeis S., Schäfer K., and Münkel C (2009) Observation of the structure of the urban boundary layer with
different ceilometers and validation by RASS data. Meteorol. Z. 18: 149–154

140 DTU Wind Energy-E-Report-0029(EN)

Engelbart D. A. M. and Bange J. (2002) Determination of boundary-layer parameters using wind profiler/RASS
and sodar/RASS in the frame of the LITFASS project. Theor. Appl. Climatol. 73: 53–65

Eresmaa N., Karppinen A., Joffre S. M., Räsänen J., and Talvitie H. (2006) Mixing height determination by
ceilometer. Atmos. Chem. Phys. 6: 1485–1493

Flamant C., Pelon J., Flamant P. H., and Durand P. (1997) Lidar determination of the entrainement zone
thickness at the top of the unstable marin atmospheric boundary-layer. Bound.-Layer Meteorol. 83: 247–284

Gerstengarbe F.-W., Werner P. C., and Rüge U. (Eds.) (1999) Katalog der Großwetterla-
gen Europas (1881 - 1998). Nach Paul Hess und Helmuth Brezowsky. 5th edition. Ger-
man Meteorological Service, Potsdam/Offenbach a. M. (available from: http://www.deutscher-
wetterdienst.de/lexikon/download.php?file=Grosswetterlage.pdf or http://www.pik-potsdam.de/ uw-
erner/gwl/gwl.pdf)

Grimsdell A. W. and Angevine W. M. (1998) Convective Boundary Layer Height Measurement with Wind
Profilers and Comparison to Cloud Base. J. Atmos. Oceanic Technol. 15: 1331–1338

Gryning S.-E., Batchvarova E., Brümmer B., Jørgensen H., and Larsen S. (2007) On the extension of the
wind profile over homogeneous terrain beyond the surface layer. Bound.-Layer Meteorol. 124:251–268

Hayden K. L., Anlauf K. G., Hoff R. M., Strapp J. W., Bottenheim J. W., Wiebe H. A., Froude F. A., Martin
J. B., Steyn D. G., and McKendry I. G. (1997) The Vertical Chemical and Meteorological Structure of the
Boundary Layer in the Lower Fraser Valley during Pacific ’93. J. Atmos. Environ. 31: 2089–2105

Hennemuth B. and Kirtzel H.-J. (2008) Towards operational determination of boundary layer height using
sodar/RASS soundings and surface heat flux data. Meteorol. Z. 17: 283–296

Hooper W. P. and Eloranta E. (1986) Lidar measurements of wind in the planetary boundary layer: the method,
accuracy and results from joint measurements with radiosonde and kytoon. J. Clim. Appl. Meteorol. 25:
990–1001

Lammert A. and Bösenberg J. (2006) Determination of the Convective Boundary-Layer Height with Laser
Remote Sensing. Bound.-Layer Meteorol. 119: 159–170

Marshall J. M., Peterson A. M., and Barnes A. A. (1972) Combined radar-acoustic sounding system. Appl.
Opt. 11: 108–112

Martucci G., Srivastava M. K., Mitev V., Matthey R., Frioud M., and Richner H. (2004) Comparison of lidar
methods to determine the Aerosol Mixed Layer top. In: Schäfer K., Comeron A., Carleer M., Picard R. H.
(eds.): Remote Sensing of Clouds and the Atmosphere VIII. Proc of SPIE 5235: 447–456

Melfi S. H., Spinhirne J. D., Chou S. H., and Palm S. P. (1985) Lidar Observation of the Vertically Organized
Convection in the Planetary Boundary Layer Over the Ocean. J. Clim. Appl. Meteorol. 24: 806–821

Menut L., Flamant C., Pelon J., and Flamant P. H. (1999) Urban Boundary-Layer Height Determination from
Lidar Measurements Over the Paris Area. Appl. Opt. 38: 945–954

Münkel C. and Räsänen J. (2004) New optical concept for commercial lidar ceilometers scanning the boundary
layer. Proc. SPIE Vol 5571: 364–374

Neff W. D. and Coulter R. L. (1986) Acoustic remote sensing. In: Lenschow D. H. (ed) Probing the Atmospheric
Boundary Layer. Amer. Meteorol. Soc., Boston, MA, 201–239

Peña A., Gryning S.-E., and Hasager C. B. (2010) Comparing mixing-length models of the diabatic wind
profile over homogeneous terrain. Theor. Appl. Climatol. 100:325–335

Piironen A. K. and Eloranta E. W. (1995) Convective boundary layer depths and cloud geometrical properties
obtained from volume imaging lidar data. J. Geophys. Res. 100: 25569–25576

Schäfer K., Emeis S. M., Rauch A., Münkel C., and Vogt S. (2004) Determination of mixing-layer heights
from ceilometer data. In: Schäfer K., Comeron A. T., Carleer M. R., Picard R. H., and Sifakis N. (eds.):
Remote Sensing of Clouds and the Atmosphere IX. Proc. SPIE Vol 5571: 248–259

Schäfer K., Emeis S., Junkermann W., and Münkel C. (2005) Evaluation of mixing layer height monitoring
by ceilometer with SODAR and microlight aircraft measurements. In: Schäfer K., Comeron A. T., Slusser
J. R., Picard R. H., Carleer M. R., and Sifakis N. (eds) Remote Sensing of Clouds and the Atmosphere X.
Proc. SPIE Vol 5979: 59791I-1–59791I-11

Seibert P., Beyrich F., Gryning S.-E., Joffre S., Rasmussen A., and Tercier P. (2000) Review and intercompar-
ison of operational methods for the determination of the mixing height. Atmosph. Environ. 34: 1001-1-027

DTU Wind Energy-E-Report-0029(EN) 141

Senff C., Bösenberg J., Peters G., and Schaberl T. (1996) Remote Sesing of Turbulent Ozone Fluxes and
the Ozone Budget in the Convective Boundary Layer with DIAL and Radar-RASS: A Case Study. Contrib.
Atmos. Phys. 69: 161–176

Sicard M., Pérez C., Rocadenbosch F., Baldasano J. M., and Garcı́a-Vizcaino D. (2006) Mixed-Layer Depth
Determination in the Barcelona Coastal Area From Regular Lidar Measurements: Methods, Results and
Limitations. Bound.-Layer Meteorol. 119: 135–157

Steyn D. G., Baldi M., and Hoff R. M. (1999) The detection of mixed layer depth and entrainment zone
thickness from lidar backscatter profiles. J. Atmos. Ocean Technol. 16: 953–959

White A. B., Senff C. J., and Banta R. M. (1999) A Comparison of Mixing Depths Observed by Ground-Based
Wind Profilers and an Airborne Lidar. J. Atmos. Oceanic Technol. 16: 584–590

Wulfmeyer V. (1999) Investigation of turbulent processes in the lower troposphere with water-vapor DIAL
and Radar-RASS. J. Atmos. Sci. 56: 1055–1076

Wulfmeyer V. and Janjić T. (2005) 24-h observations of the marine boundary layer using ship-borne NOAA
High-Resolution Doppler Lidar. J. Appl. Meteorol. 44: 1723–1744

142 DTU Wind Energy-E-Report-0029(EN)

7 What can remote sensing contribute
to power curve measurements?
Rozenn Wagner
DTU Wind Energy, Risø Campus, Roskilde, Denmark

7.1 Introduction
Power performance measurement is central to the wind industry since it forms the basis for
the power production warranty of the wind turbine. The power curve measurement has to
be realised according to the IEC 61400-12-1 standard. The power curve is obtained with
10-min mean power output from the turbine plotted against simultaneous 10-min average
wind speeds. The standard requires the wind speed to be measured by a cup anemometer
mounted on top of a mast having the same height as the turbine hub and located at a distance
equivalent to 2.5 rotor diameters from the turbine.
Such a plot usually shows a significant spread of values and not a uniquely defined function.
The origin of the scatter can mainly be grouped into three categories: the wind turbine com-
ponents characteristics, sensor error and the wind characteristics. Within the last group, the
current standard only requires the wind speed at hub height and the air density measurement.
However, other wind characteristics can influence the power production like the variation of
the wind speed with height (i.e. wind speed shear). The influence of wind speed shear on
the power performance was shown in several studies: some based on aerodynamic simulations
(Antoniou, 2009; Wagner et al., 2009) others based on measurements (Elliot and Cadogan,
1990; Sumner an Masson, 2006).
A major issue is to experimentally evaluate the wind speed shear. The wind speed profile
is usually assumed to follow one of the standard models such as the logarithmic or power law
profiles. However, these models are valid for some particular meteorological conditions, and
therefore, cannot represent all the profiles experienced by a wind turbine. Measurements are
then a better option but are also challenging. Indeed characterising the wind speed profile
in front of the rotor of a multi-MW wind turbine requires measurements of wind speed at
several heights, including some above hub height, i.e. typically above 100 m. Remote sensing
instruments such as lidar or sodar then appear as a very attractive solution.
This chapter starts with a description of the influence of the wind speed shear on the power
performance of a multi-MW turbine. The challenge of describing the wind speed profile is
then discussed followed by a description of an experiment using a lidar for its characterisation.
This is followed by the introduction of the definition of an equivalent wind speed taking the
wind shear into account resulting in an improvement of the power performance measurement.
Finally, some recommendations about remote sensing instruments are given to successfully
apply this method.

7.2 Power performance and wind shear

7.2.1 Shear and aerodynamics
In order to see the effect of the wind speed shear on a wind turbine, aerodynamic simulations
were carried out for two inflow cases:
1. constant wind speed profile (same wind speed everywhere) with 8 m s−1
2. power law profile with 8 m s−1 at hub height and a shear exponent of 0.5

DTU Wind Energy-E-Report-0029(EN) 143

The power law profile is defined as:
 α
z
u = uhub , (159)
zhub
where u is the wind speed at height z, zhub the hub height, uhub the wind speed at that
height and α the shear exponent. Both profiles are shown in Figure 83.
The model used was HAWC2Aero. The modeled turbine was a Siemens 3.6 MW with a
rotor diameter of 107 m and a hub height of 80 m.The wind speed is assumed horizontally
homogeneous (i.e. the wind speed is the same everywhere on each horizontal plane). In order
to emphasize the effect of wind speed shear, the simulations were carried out with laminar
inflow, the tower shadow was turned off and the tilt angle of the rotor was set to 0◦ .

height @mD
140

120

100

80

60

40
wind speed @msD
4 6 8 10
Figure 83: Wind profiles used as input for the wind speed shear aerodynamic investigation.
Black curve: no shear; grey curve: power law profile with shear exponent of 0.5. Both profiles
have the same wind speed at hub height

Figure 84 shows the free wind speed (i.e. the absolute wind speed in absence of a turbine)
seen by a point at a radius of 30 m from the rotor centre, rotating at the same speed as
the rotor as a function of time for the 2 inflow cases. Whereas in a uniform flow the blade is
subjected to a constant wind speed, in a sheared flow, the point is exposed to large variations
of wind speed even though the inflow is laminar. The variation of the wind speed seen by this
rotating point in time is only due to the fact that it is rotating within a non uniform flow
(wind speed varying with height).
Figure 85 shows the variations of the free wind speed seen by the same rotating point as
function of the azimuth position (0◦ = downwards). The point experiences the hub height
wind speed (same as uniform inflow) when it is horizontal (±90◦ ), lower wind speed when it
is downward (0◦ ) and higher wind speed when it is upward (180◦ ).
A rotating blade does not experience the free wind speed because of the induction from the
drag of the rotor. In reality, a rotating blade is directly subjected to the relative wind speed w
(i.e. the speed of the wind passing over the airfoil relative to the rotating blade) and the angle
of attack φ (i.e. the angle between the blade chord line and the relative wind speed) with the
effects of the induced speed included. The variations of these two parameters as function of θ
are shown in Figure 86. As these two parameters directly depend on the free wind speed, they
vary with the azimuth angle in a sheared inflow, whereas they remain constant in a uniform
inflow.
The relative speed and the angle of attack are derived from the rotor speed and the induced
velocity. Therefore, they depend on the way that the induction is modeled and it is difficult

144 DTU Wind Energy-E-Report-0029(EN)

 u y @msD

9.0

8.5

8.0

7.5

7.0

time@sD
120 140 160 180 200

Figure 84: Time series of free wind speed seen from a rotating point, positioned at a radius
of 30 m, rotating at rotor rotational speed (no induced velocity). Black curve: no shear; grey
curve: power law profile with shear exponent of 0.5

u y @msD

9.0

8.5

8.0

7.5

7.0

Θ @°D
-150 -100 -50 50 100 150

Figure 85: Free wind speed seen from a rotating point, positioned at a radius of 3 0m, rotating
at rotor rotational speed, as function of the azimuth angle θ. Black curve: no shear; grey curve:
power law profile with shear exponent of 0.5

to evaluate their variations due to a non uniform flow in a simple way. However, some basic
considerations (ignoring the induction) can give a basic insight to the variation of the relative
speed and the angle of attack as the blade rotates. In case of uniform inflow, the free wind
speed is the same at any point of the swept rotor area. Therefore, the angle of attack and
the relative speed are the same at any azimuthal position (see Figure 87).
In case of sheared inflow, the free wind speed depends on the position of the blade. When
the blade is horizontal, the free wind speed is the speed at hub height and the speed triangle
is the same as in Figure 87. Below hub height, the wind speed is lower than the hub height
speed, see Figure 88 (left). Consequently w and φ are lower than those at hub height. Above
hub height, the wind speed is higher than the hub height wind speed. Consequently w and φ
are higher than those at hub height, see Figure 88 (right). The variations in φ and w cause a
variation of the local lift and drag as the blade rotates, which finally results in the variation of
the local tangential force contributing to the wind turbine power (see Figure 89). For a given
φ, local lift dFL and local drag dFD , the local tangential force dFT is given by (Manwell

DTU Wind Energy-E-Report-0029(EN) 145

 angle of attack @°D Vrel @msD
45.0

4.0 44.5

3.5 44.0

3.0 43.5

2.5 43.0

Θ @°D Θ @°D
-150 -100 -50 50 100 150 -150 -100 -50 0 50 100 150

Figure 86: phi (left) and w (including induction) (right) as a function of θ, seen from a point
at radius r = 30 m on a rotating blade. Black curve: no shear; grey curve: power law profile
with shear exponent of 0.5

rotation
rW
Φ
u
w

Figure 87: Speed triangle for a blade element at radius r. rΩ is the blade speed and w
corresponds to the sum of the pitch angle, the twist angle and φ. As the twist angle is
constant for a given position on the blade and the pitch angle is 0◦ for wind speeds below
rated speed, φ represents the variation of the angle of attack

et al., 2002):
dFT = dFL sin φ − dFD cos φ. (160)
As the wind speed increases with height (e.g. in the case of the power law profile), the
amplitude of the variations of the free wind speed seen by a rotating point increases with the
radius (not shown here). The local tangential force consequently varies with the radius too.
As the torque results from the integral of the tangential force over the whole rotor, it thus
depends on the wind speed profile.

7.2.2 Consequences on the power production

A series of cases were simulated with theoretical wind speed shear defined from the power law
in Eq. (159), with −0.1 < α < 0.5 and 5 m s−1 < uhub < 10 m s−1 . The relative variations in
power (defined as the percentage difference between the power outputs obtained with a shear
inflow and an uniform inflow) are shown in Figure 90. According to the simulations results, the
power output obtained with shear inflow is generally smaller than the power output obtained
with an uniform inflow. Moreover, it decreases as the shear exponent increases except at 5 m
s−1 where the power output reaches a minimum for α = 0.2 and increases for larger shear
exponents, even exceeding the power output from uniform inflow with α = 0.5.
The first difference between a sheared and an uniform inflow is the kinetic energy flux. In
case of horizontally homogeneous inflow, the kinetic energy flux can be expressed by:
Z H+R
KEprofile = 0.5ρu3c (z − (zhub − R)) dz, (161)
H−R

146 DTU Wind Energy-E-Report-0029(EN)

 Bottom of rotor Top of rotor
rotation rotation
rW rW
Φ
Φ
u
w
u
w

Figure 88: Speed triangles at the top and bottom of the rotor showing the effect of wind
speed shear. The speed triangle at hub height is shown with dashed arrows

dFT @kNmD
0.18

0.16

0.14

0.12

0.10

0.08

0.06

Θ @°D
-150 -100 -50 50 100 150

Figure 89: Local tangential force seen from a point r = 30 m on a rotating blade as function
of θ. Black curve: no shear; Gray curve: power law profile with shear exponent of 0.5

where ρ is the air density, R the rotor radius and c is the chord (of the circle defined by the
rotor swept area) as function of z which varies between the bottom and the top of the rotor:
p
c = 2 2Rz − z 2 . (162)
In order to compare to the power output variations, Figure 90 also shows the difference
between the kinetic energy flux for a power law profile and a constant profile, normalised with
the power obtained with a constant profile KEprofile − KEhub /KEhub (see gray dashed line)
5
. Figure 90 shows two other interesting results:
1. The kinetic energy flux varies with shear exponent.
2. The power output of the turbine does not follow the same trend as the kinetic energy
flux.
Despite the high uncertainty in the modeled power output for a sheared inflow, the results
highlight that the influence of the shear on the power performance of a turbine can be seen
as the combination of two effects:
• The variation in kinetic energy flux (power input).
• The ability of the turbine to extract the energy from the wind, which depends on the
details of the design and the control strategy of the turbine.
5 The kinetic energy flux is not an output of HAWC2Aero, it has been here estimated and for power law

profiles, the normalized difference does not depend on the wind speed

DTU Wind Energy-E-Report-0029(EN) 147

 @%D
10
æ Power 5 ms
à Power 6 ms
ì Power 7 ms
æ
ò Power 8 ms
æ
5 ô Power 9 ms
O Power 10 ms
æ
à æ KE
ì
æ
ò
ô
ç
æ
ì
ç
ò
ô
æ
à
æ Α
-0.2 -0.1 0.1
æ
ç
0.2
æ 0.3 0.4 0.5
ô
ò æ
ì ç
ô
à ò ç ç ç
ô ô ô
ì ò
à ò ò
ì
-5 æ à ì ì
à
à
æ
æ

-10

Figure 90: Normalised difference in power output and kinetic energy flux between shear and
uniform case as function of the shear exponent, for various wind speed at hub height

These results clearly indicate that wind speed profiles encountered by the turbine during a
power performance measurement should be known and taken into account.

7.3 Wind speed profiles

Within the power performance evaluation context, the wind speed shear is often described
by α obtained from the assumption of a power law profile (Eq. (159)). This procedure was
applied to profiles measured by a high meteorological mast located at the test site for large
wind turbines, at Høvsøre, on the west coast of Denmark with zhub = 80 m and z = 40 m
to determine α. For some cases, this model fits the measured profile very well, but it cannot
represent all kinds of profiles observed at Høvsøre. Figure 91 shows two examples of measured
profiles and their corresponding modelled profiles.
The distribution of the error made by such an approximation is shown in Figure 92. We
define the error as the difference between the wind speed at 116.5 m (top of the mast)
estimated by the power law model and the speed measured by the cup anemometer. Over a
year of measurements, for a large wind sector 60◦ –300◦ degrees (with predominant wind from
west), 7% of the profiles show a wind speed error at 116.5 m larger than 10%. We should
keep in mind that all the anemometers are mounted on a boom except the top anemometer,
and this can induce an error in the profile extrapolation to the top (116.5 m).
As shown by the simulations presented in section 7.2, such an error in the wind speed profile
can significantly affect the power curve. The shear exponent from wind speed measurements
at two heights is not acceptable for this application. Therefore it is important to measure the
wind speeds at several heights below and above hub height. For this purpose remote sensing
instruments such as lidar and sodar are highly relevant since in many cases they can measure
up to 200 m with the required degree of accuracy.
An experimental campaign using a lidar to measure the wind speed profile in front of
a multi-MW turbine showed the importance of measuring the complete profile for power
performance. In our experiment, the lidar measured the wind speed at 9 uniformly distributed
heights covering 90% of the vertical rotor diameter. Each wind speed profile measured by the
lidar was fitted to a power law profile in order to find the most representative shear exponent
for this profile (αf it ). The fit is forced through the point of coordinate (uhub , zhub ):
 αf it
z
uf it (z) = uhub . (163)
zhub
In order to separate the profiles for which the power law assumption was acceptable, we

148 DTU Wind Energy-E-Report-0029(EN)

 1.6 1.6

æ æ
1.4 1.4
model: Α= 0.33 model: Α= 0.4
mast measurement æ mast measurement æ
Height  HeightHub

Height  HeightHub
1.2 1.2

1.0 æ 1.0 æ

0.8 0.8
æ æ

0.6 0.6
æ æ

0.4 0.4
0.6 0.7 0.8 0.9 1.0 1.1 1.2 0.6 0.7 0.8 0.9 1.0 1.1 1.2
U  Urated U  Urated

Figure 91: Two examples of profiles and their fit to the power law model (using the wind
speed values at 40 and 80 m). The model fits very well the measured profile on the left, but
it does not work for the profile on the right

10
% Hprofiles numberL

8
6
4
2
0
-0.4 -0.2 0.0 0.2 0.4
wsp error at 1.45 hub height
Figure 92: Distribution of the error made on the wind speed at 116.5 m height when assuming
a power law profile with a shear exponent estimated with the wind speeds at 40 and 80 m

evaluated the goodness of fit with the residual sum of squares RSS, defined as:
X 2
RSS = (uf it (zi ) − ui ) , (164)
i

where uf it is the fit function defined in Eq. (163) and ui the wind speed measured by the
lidar at height zi (i = 1 to 9). Figure 93 shows two examples of measured profiles with their
shear exponents and RSSs.
The profiles were then divided into two groups according to the RSS:
• Group 1: RSS ≤ 0.1 – the profiles from this group have a shape close to a power law
profile and are referred to as the power law profiles.
• Group 2: RSS > 0.1 – the profiles from this group have a shape that cannot be well
represented by a power law profile and are referred to as the non power law profiles.

DTU Wind Energy-E-Report-0029(EN) 149