Wind Profiles and Forests

Validation of Wind Resource Assessment Methodologies Including the Effects of Forests

1 2
Shiu-Yeung Hui and Anthony Crockford
M.Sc. Thesis Project at Technical University of Denmark
1 2
[email protected], [email protected]

Abstract

Wind turbines are often erected at sites with nearby forests and the effects of forests on the energy
production and turbine loading are not fully understood. It is critical to be able to predict these effects
to ensure the success of these wind energy projects. This project represents a validation of three wind
resource assessment methodologies (WAsP, WAsP Engineering, and WindSIM) using various models
to represent forests within the simulations.

These simulation programs use two parameters to describe forests: roughness length and
displacement height. An extensive literature review found many models for these parameters in use
within the micro-meteorological field. Several simple models exist based on the mean tree height (over
fifty combinations of models were found), while other models incorporate expressions for the canopy
density.

These models were tested by simulating the mean wind speed, energy production, turbulence
intensity, and wind shear at sites inside or near forests, and comparing the predictions to measured
data at those sites. Data was available from four Ris measurement campaigns in forests and from
two Siemens Wind Power windfarms.

The analysis shows that the use of some of the forestry models can significantly improve the
prediction of mean wind speed, annual energy production, turbulence intensity, and shear in sites
inside forests. For sites near forests, the models have negligible effects. It was found that WAsP and
WAsP Engineering were generally more accurate than WindSIM for these sites.

1. Introduction

The most important factor in selecting a wind turbine site and predicting the economic success of the
project is the wind resource. Understanding the wind resource is especially critical as the energy
available in the wind varies as the cube of the wind speed.

The wind resource can be modeled quite well for simple terrain (relatively flat, small vegetation) but
more wind farms are being erected in forested environments. The power production of these wind
farms is significantly affected by the wind over the forested areas. Unfortunately, many of the effects of
forests are not fully understood within the wind-energy industry or even the micro-meteorological
community.

It was hoped that forests can be adequately modeled, by modifying input parameters that are common
to many wind resource assessment methodologies roughness length and displacement height.
Meteorological experts have developed and published dozens of different methods to determine these
parameters. Unfortunately, this wealth of choices can lead to confusion in terms of the best model for
prediction of wind energy production.

This project will evaluate these methods, and validate their predictions using data from measurement
campaigns and actual wind farms in forests.

Theory
It is possible to use wind speed measurements at one height to determine the wind speed at other
heights (often the turbines hub height), based on a logarithmic profile that approximates how the wind
speed increases with increasing height above the ground.

This profile has been extensively validated for non-complex terrain. However, if the terrain is overly
steep or forested, the relationship is not so simple. A generic forest profile is depicted in Figure 1.
 Figure 1 - Forest Wind Profile [1]

This curve shows a distinct deviation from a standard logarithmic curve. Considering heights above
the canopy, the wind profile is essentially shifted upwards by d, the zero-plane displacement height.
This profile can be represented as:

Roughness Length
The roughness of a terrain is parameterized with the roughness length (z0), which is a measure of
terrain roughness as seen by the surface wind. Formally, z0 is the level at which the logarithmic wind
profile has zero wind speed.

Zero-Plane Displacement Height

For wind speed profiles over very rough terrain, such as forests, the concept of zero-plane
displacement height (often referred to simply as displacement height or d) is necessary, as shown in
Figure 1.

The most basic definition of displacement height is that it is level which would be obtained by flattening
out all the roughness elements into a smooth surface. More detailed analyses have determined that
the displacement height also represents the level at which surface drag acts on the roughness
elements (i.e. trees) [2]. As such, the displacement height is always smaller than the mean height of
the trees.

2. Models for Determining Roughness Length and Displacement Height

Both WAsP and WindSIM use the roughness length and displacement height as the primary
parameters to describe rough areas, such as forests. The micro-meteorological community has
extensively studied these parameters in areas with tall vegetation or forestry, yielding many different
methods to find them

Several simple models exist based on the mean tree height (over fifty combinations of models were
found), while other models incorporate expressions for the canopy density. These models can be
classified as the following types: simple relations to forest density [3], drag partitioning [4, 5], second-
order closure [6, 7], and momentum transfer [8].
Simple Models
After an initial analysis of the 57 possible combinations (outlined later), four models were highlighted
for further verification. Two of these models were based on the mean stand height, as seen in Table
1.

Table 1 Two models based on mean tree height

Roughness Displacement
Length Height Reference
Garratt [11],
Dolman [12]
Jarvis [13],
Hicks [14]

However, the mean stand height does not provide a complete picture of the forest, as it does not
describe the density of the forest. The other two models use different ways to represent the forest
density.

Drag Partitioning Model

Raupach [4, 9, 10] approximated the canopy with the frontal area index:

where h is the tree height, b is the tree breadth and D is the spacing between trees.

Raupachs model is based on the theory that the drag on forests is partly absorbed by the canopy and
partly by the ground beneath. This concept of drag partitioning was used to develop expressions for
z0 and d as functions of canopy height and frontal area index:

where is a free parameter, is the frontal area index, is the roughness sublayer influence

function and is the inverse of the approximate drag coefficient:

where is the substrate surface drag coefficient, is the roughness element drag coefficient.

For practical application, Raupach further simplified these expressions, by providing estimates of
several of the parameters:
Raupach verified these predictions over a range of field and wind-tunnel data, and found good results.
Similarly, several later studies have verified and implemented this model [13, 5, 15, 16, 17]. Due to its
simplicity, it is a recommended parameterization for WAsP [18].

Second-Order Closure Model

Another parameter is more commonly used to describe forest density: the leaf area index (LAI). It is
defined as the one-sided leaf area per unit land area. LAI can be determined experimentally, often
using a leaf area index meter. A recent study of LAI measurements [19] is available with about 1000
published estimates from 400 unique sites.

Shaw and Pereira [6] developed a numerical model for the aerodynamic roughness of a canopy, and
Choudhury and Monteith [7] later fitted simple functions curves to the curves obtained from this
second-order closure model:

for

for

This parameterization was based on the unnamed parameter X:

where is the mean drag coefficient (assumed to be uniform within the canopy, )
and LAI is the leaf area index.

This model also accounts for , the roughness length of the substrate (forest floor). Yang and Friedl
[20] approximate that for unvegetated surfaces and for vegetated
substrates, where is the mean height of the vegetation.

Using the Models

A roughness map must be created to represent the roughness length of terrain for use in WAsP or
WindSIM.

The displacement height can be used in wind prediction models in two ways. First, the height of any
measurement masts or wind turbines within forests can be reduced by d. This method is simple, but
does not account for forests near to the sites.

A more complex method is to build up a hill (of height d) at forested areas. Since WAsP uses a
simple flow model, it is necessary to build-up the slope around the hill to ensure it does not exceed
about 10 [21]. Several different methods were attempted to create this hill. The simplest method
used only the WAsP Map Editor, but created steep slopes. Using Surfer (commercial mapping
software) it was also possible to create a hill with smoother slopes (1:5) as recommended by the
WAsP team.

One final option is to download SRTM maps (Shuttle Radar Topography Mission). These maps are
available for free download online [22] and are based on a NASA measurement campaign, using a
space-based radar system to determine worldwide surface elevations. Over forested terrain, the radar
will have reflected off of the trees, rather than the true ground level. The exact height of this hill is
uncertain, but the method is simple and may account for the forest better than the original topography
map.

Each of these methods is illustrated in Figure 2.

 Figure 2 - Possible Terrain Modifications

3. Data

The project data was supplied by Ris National Laboratories and Siemens Wind Power. Two types of
sites are considered in this analysis inside forests, and near forests.

Measurements Inside Forests

Measurement campaigns have been carried out by Ris at three forests in Denmark: Sor, Corselitze,
and Ulfborg. Meteorological masts with multiple instruments were mounted inside the forests. For
prediction of the local wind climate at each site a reference site with low vegetation is used. All sites
are rather small forests.

Measurements Near Forests

A Ris measurement campaign also recorded the winds at a distance of 140m from the Sor forest.
As well, Siemens Wind Power provided data from two windfarms. Measurements from meteorological
masts situated within each site were used to generate the wind climate. Wind speed, turbulence
intensity and energy production data was available for three turbines at each site. Forests surround
each site and there are small forested areas within the windfarms.

4. Method

The analysis involved cross-prediction, from data at a reference site to data at other sites.
Measurements of wind speed, wind shear, turbulence intensity and/or energy production were
available at these sites to allow a comparison of predicted and measured values. Three software
packages were used:
1. WAsP to predict mean wind speed and energy production
2. WAsP Engineering to predict vertical wind shear and turbulence intensity
3. WindSIM to predict mean wind speed, energy production, wind shear and turbulence
intensity

After the initial literature review, 13 models are proposed for roughness length, as well as 16 models
for displacement height. In total, 57 scenarios (combined roughness lengths and displacement
heights) were initially considered. This required an excessive number of simulations, so only initial
tests were performed with all of them, before selecting only the best models for further analysis. Four
models were selected for further verification

Least Squares Method

A useful statistical method for the comparison of models is called the Least Squares Method. It can be
represented mathematically as:
 a minimum

Where is the measured data point, is the predicted data point and n is the number of data
points for comparison. More accurate predictions will yield a smaller sum. This method provides a
convenient way to compare the models, but only relative to each other. The calculated value can only
be compared to sums of other similar data.

5. Results and Conclusions

Inside Forests
The mean wind speed predictions were drastically improved by applying the various d-z0 models in
both WAsP and WindSIM. The four best models are explained in Section 2. Using these models, the
prediction error at the maximum met mast height was reduced by at least a factor of at three,
compared to the default model, often used for forests ( , ).

The Garratt-Dolman and Jarvis-Hicks models are simply dependent on the mean tree height, but they
performed well. This may be due to similar canopy densities amongst the forests in this analysis. The
Choudhury & Monteith and Raupach models both account for forest density, so they may be more
applicable to forests with less dense canopies.

The use of these models in WAsP Engineering led to improved accuracy in turbulence intensity and
wind shear predictions, while the terrain inputs offered no significant additional improvements. The
various models offered similar degrees of improvement. The WindSIM predictions were less accurate
and the use of different models showed negligible effects. With the use of d-z0 models in WAsP
Engineering, the turbulence intensity is still under-predicted, but the prediction error is generally
reduced from around 50% to 25%. There is some improvement in the predictions of wind shear (the
error is reduced from about 75% to 35%), but it is inconsistent; shear can be under- or over-predicted.
The WindSIM Forest Module produced poor results overall.

Figure 3 Prediction based on default settings and new models compared to measured values for
(left) wind speed and (right) turbulence intensity, inside forests

Near Forests
WAsP correctly predicted the mean wind speed to within about 5% for the sites near forests.
However, there was little difference between using the models and simply using the default
methodology. Similarly, no improvements were found with different terrain modifications. WindSIM
produced similar trends, but was slightly less accurate.

The annual energy production was accurately predicted to within 15%, but the various forestry models
offered no improvements compared to the default WAsP and WindSIM methodologies.
For turbulence intensity and wind shear, WAsP Engineering yielded more accurate predictions
compared to WindSIM. The use of different models in WAsP Engineering had little effect on both,
while the terrain inputs offered no significant additional improvements. The use of d-z0 models in
WindSIM show negligible effects, while the use of the Steep Hill model decreased accuracy for
turbulence intensity predictions. The prediction error in WAsP Engineering remained at around 25%
for turbulence intensity, and around 43% for shear, regardless of the models used.

Figure 4 - Prediction based on default settings and new models compared to measured values for
(left) wind speed and (right) turbulence intensity, near forests

6. Recommendations

Inside Forests
For sites inside the forest, an appropriate forestry model will greatly increase the accuracy of
predictions of mean wind speed, annual energy production, turbulence intensity, and wind shear. The
recommended models are given in Section 2. Since the accuracies of each of these models are so
similar with each other, the selection of a model should depend on the knowledge of the forest. If the
frontal area index ( ) or leaf area index (LAI) of the forest is known, the Raupach or Choudhury &
Monteith models can be used confidently. However, if only tree height information is available, the
Jarvis-Hicks or Garratt-Dolman models should be sufficient.

Terrain modification methods such as the Steep and Sloping Hill models have negligible or detrimental
effects and should be avoided. Instead, the displacement height should simply be subtracted from the
height of the turbines or masts inside the forest.

WindSIM can be used to predict wind speed and energy production, as it was slightly more accurate
than WAsP, however WAsP Engineering should be used for more accurate predictions of turbulence
intensity and wind shear. If using WindSIM, the Forest Module should not be used.

Near Forests
The default WAsP or WindSIM methodologies can be used for sites near forests, as the models
produced negligible differences in the predictions of mean wind speed or annual energy production.

Similarly, the effects of the d-z0 models are negligible in turbulence intensity and wind shear
prediction. However, WindSIM should not be used as the prediction errors are generally very high
(over 50%). The default WAsP Engineering methodology is recommended.
References

1. Gardiner B. Airflow Over Forests and Forest Gaps. BWEA Tree Workshop Forestry
Commission March 2004.
2. Jackson PS. On the displacement height in the logarithmic velocity profile. Journal of Fluid
Mechanics 1981; 111:1525.
3. Finnigan JJ, Belcher SE. Flow Over a Hill Covered with a Plant Canopy. Quarterly Journal of
the Royal Meteorological Society 2004; 130:1-29.
4. Raupach MR. Drag and drag partitioning on rough surfaces. Boundary-Layer Meteorology
1992, 60:375-395.
5. Verhoef A, McNaughton KG, Jacobs AFG. A parameterization of momentum roughness length
and displacement height for a wide range of canopy densities. Hydrology and Earth System
Sciences 1997; 1:8191.
6. Shaw RH, Pereira AR. Aerodynamic roughness of a plant canopy: a numerical experiment.
Agricultural Meteorology 1982; 26:51-65.
7. Choudhury B, Monteith J. A Four-Layer Model for the Heat Budget of Homogeneous Land
Surfaces. Quarterly Journal of the Royal Meteorological Society 1988; 114:373398.
8. Massman WJ. An Analytical One-dimensional Model of Momentum Transfer by Vegetation of
Arbitrary Structure. Boundary-Layer Meteorology 1997; 83:407421.
9. Raupach MR. Simplified expressions for vegetation roughness length and zero-plane
displacement height as functions of canopy height and area index (research note). Boundary-
Layer Meteorology 1994; 71:211216.
10. Raupach MR. Corrigenda. Boundary-Layer Meteorology 1995; 76:303304.
11. Garratt JR. The Atmospheric Boundary Layer; Cambridge University Press: 1992.
12. Baldocchi DD, Meyers TP. Turbulence Structure in a Deciduous Forest Canopy. Boundary-
Layer Meteorology 1988; 43:345-365
13. Irvine MR, Gardiner BA, Hill MK. The Evolution Of Turbulence Across A Forest Edge.
Boundary Layer Meteorology 1997; 84:467-496.
14. Hicks BB, Hyson P, Moore CJ. A Study of Eddy Fluxes over a Forest. Journal of Applied
Meteorology 1975; 14:58-66.
15. Schaudt KJ, Dickinson RE. An approach to deriving roughness length and zero-plane
displacement height from satellite data, prototyped with BOREAS data. Agricultural and Forest
Meteorology 2000; 104:143155.
16. Borak JS, Jasinski MF, Crago RD. Time Series Vegetation Aerodynamic Roughness Fields
Estimated From Modis Observations. Agricultural and Forest Meteorology 2005; 135:252-
268.
17. Jasinski MF, Borak J, Crago R. Bulk Surface Momentum Parameters for Satellite-Derived
Vegetation Fields. Agricultural and Forest Meteorology 2005; 133:5568.
18. Dellwik E, Landberg L, Jensen NO. WAsP in the Forest. Wind Energy 2005, 9:211-218.
19. Scurlock JMO, Asner GP, Gower ST. Worldwide Historical Estimates of Leaf Area Index,
1932-2000; Oak Ridge National Laboratory: Tennessee, 2001.
20. Yang R, Friedl MA. Determination of roughness lengths for heat and momentum over boreal
forests. Boundary-Layer Meteorology 2003; 107:581603.
21. Dellwik E, Jensen NO, Landberg L. Wind and Forests General Recommendations for Using
WAsP. BWEA Tree Workshop March 2004; Ris National Laboratory.
22. Shuttle Radar Topography Mission download site: ftp://e0srp01u.ecs.nasa.gov/srtm/version2