NAVAL

POSTGRADUATE
SCHOOL
MONTEREY, CALIFORNIA

THESIS

ATMOSPHERIC EFFECTS ON RADIO FREQUENCY (RF)

WAVE PROPAGATION IN A HUMID, NEAR-SURFACE
ENVIRONMENT

by

Samuel P. Mason

March 2010

Thesis Co-Advisors: Peter S. Guest

Andreas K. Goroch

Approved for public release; distribution is unlimited

 REPORT DOCUMENTATION PAGE Form Approved OMB No. 0704-0188
Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instruction,
searching existing data sources, gathering and maintaining the data needed, and completing and reviewing the collection of information. Send
comments regarding this burden estimate or any other aspect of this collection of information, including suggestions for reducing this burden, to
Washington headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington, VA
22202-4302, and to the Office of Management and Budget, Paperwork Reduction Project (0704-0188) Washington DC 20503.
1. AGENCY USE ONLY (Leave blank) 2. REPORT DATE
3. REPORT TYPE AND DATES COVERED
March 2010 Master’s Thesis
4. TITLE AND SUBTITLE: Atmospheric Effects on Radio Frequency (RF) 5. FUNDING NUMBERS
Wave Propagation in a Humid, Near-Surface Environment
6. AUTHOR(S) Samuel P. Mason
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION
Naval Postgraduate School REPORT NUMBER
Monterey, CA 93943-5000
9. SPONSORING /MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSORING/MONITORING
N/A AGENCY REPORT NUMBER

11. SUPPLEMENTARY NOTES The views expressed in this thesis are those of the author and do not reflect the official policy
or position of the Department of Defense or the U.S. Government. IRB Protocol number ________________.
12a. DISTRIBUTION / AVAILABILITY STATEMENT 12b. DISTRIBUTION CODE
Approved for public release; distribution is unlimited A
13. ABSTRACT (maximum 200 words)
Currently, the meteorological and physical phenomena associated with the various dynamic processes in the very near
surface environment (for example, within the surface layer), are poorly understood. By properly characterizing what
is happening in the real world, there is potential for obtaining an empirical formula that correlates well with real world
data, and thus can be used as a means of quantifying these physical processes. This, in turn, can be used to more
accurately model the effects of the atmosphere on RF waves. This thesis is an analysis of the propagation loss
measurements taken from the Near Earth Propagation-6 (NEP-6), Panama City, FL, experiment in Aug 2009, where
propagation loss was measured at 1768 MHz within a few wavelengths (≈ 0.5 meters) of the surface. The results
support and extend the near-surface, short range RF propagation conclusions drawn by Merrill et al. (2004). In
particular, we focus on a novel technique that takes advantage of tidal sea level variation to continuously vary antenna
height above the surface. Results confirm a strong dependence of propagation loss on antenna height similar to
Merrill et al.’s (2004) observations.

14. SUBJECT TERMS: Electromagnetic propagation, electromagnetic scattering, Ground- 15. NUMBER OF
wave propagation, Mathematical techniques, Variance reduction. PAGES
87
16. PRICE CODE
17. SECURITY 18. SECURITY 19. SECURITY 20. LIMITATION OF
CLASSIFICATION OF CLASSIFICATION OF THIS CLASSIFICATION OF ABSTRACT
REPORT PAGE ABSTRACT
Unclassified Unclassified Unclassified UU
NSN 7540-01-280-5500 Standard Form 298 (Rev. 2-89)
Prescribed by ANSI Std. 239-18

i
THIS PAGE INTENTIONALLY LEFT BLANK

ii
 Approved for public release; distribution is unlimited

ATMOSPHERIC EFFECTS ON RADIO FREQUENCY (RF) WAVE

PROPAGATION IN A HUMID, NEAR-SURFACE ENVIRONMENT

Samuel P. Mason
Lieutenant, United States Navy
B.S., United States Naval Academy, 2002

Submitted in partial fulfillment of the

requirements for the degree of

MASTER OF SCIENCE IN METEOROLOGY

from the

NAVAL POSTGRADUATE SCHOOL

March 2010

Author: Samuel P. Mason

Approved by: Peter S. Guest

Co-Advisor

Andreas K. Goroch
Co-Advisor

Phillip Durkee
Chairman, Department of Meteorology

iii
THIS PAGE INTENTIONALLY LEFT BLANK

iv
 ABSTRACT

Currently, the meteorological and physical phenomena associated with the various
dynamic processes in the very near surface environment (for example, within the surface
layer), are poorly understood. By properly characterizing what is happening in the real
world, there is potential for obtaining an empirical formula that correlates well with real
world data, and thus can be used as a means of quantifying these physical processes.
This, in turn, can be used to more accurately model the effects of the atmosphere on RF
waves. This thesis is an analysis of the propagation loss measurements taken from the
Near Earth Propagation-6 (NEP-6), Panama City, FL, experiment in Aug 2009, where
propagation loss was measured at 1768 MHz within a few wavelengths (≈ 0.5 meters) of
the surface. The results support and extend the near-surface, short range RF propagation
conclusions drawn by Merrill et al. (2004). In particular, we focus on a novel technique
that takes advantage of tidal sea level variation to continuously vary antenna height above
the surface. Results confirm a strong dependence of propagation loss on antenna height
similar to Merrill et al. (2004) observations.

v
THIS PAGE INTENTIONALLY LEFT BLANK

vi
 TABLE OF CONTENTS

I. INTRODUCTION........................................................................................................1
A. MOTIVATION ................................................................................................1
B. BACKGROUND ..............................................................................................1
1. Electromagnetic Waves .......................................................................1
2. Atmospheric Effects on Propagation .................................................2
3. Characterization of Water Surface ....................................................6
4. Other Factors Affecting Transmission Loss......................................7
II. EXPERIMENT ..........................................................................................................11
A. INTRODUCTION..........................................................................................11
B. METHODOLOGY ........................................................................................11
C. DISTRIBUTED SENSOR SETUP ...............................................................13
D. EQUIPMENT.................................................................................................14
III. DATA PROCESSING ...............................................................................................17
A. SUMMARY ....................................................................................................17
B. RF DATA PROCESSING.............................................................................17
C. METEOROLOGICAL DATA PROCESSING ..........................................20
D. OCEANOGRAPHIC DATA PROCESSING..............................................25
IV. DATA ANALYSIS .....................................................................................................27
A. SUMMARY OF METHODOLOGY............................................................27
B. RAW DATA PREVIEW ...............................................................................27
C. SUB-PLOT COMPARISON OF VARIABLES..........................................30
D. VARIANCE REDUCTION TECHNIQUES...............................................32
1. Linear Trend-Line Correlation ........................................................32
2. Best Fit Sine Wave Correlation ........................................................38
E. PHYSICAL GEOMETRIES ASSOCIATION............................................40
F. MODELED DATA COMPARISON............................................................45
1. Introduction........................................................................................45
2. Theory .................................................................................................45
3. Model Summary.................................................................................49
V. CONCLUSION ..........................................................................................................53
A. SUMMARY OF EXPERIMENT..................................................................53
B. KEY RESULTS AND CONCLUSIONS......................................................53
C. APPLICABILITY TO DOD OPERATIONS..............................................53
D. AREAS FOR FURTHER RESEARCH.......................................................54
APPENDIX.............................................................................................................................57
A. SUPPLEMENTARY BACKGROUND INFORMATION ........................57
1. Atmospheric Boundary Layer ..........................................................57
2. Turbulence and Flux..........................................................................57
B. SCATTER PLOTS OF RESIDUALS ..........................................................60
LIST OF REFERENCES ......................................................................................................67
vii
INITIAL DISTRIBUTION LIST .........................................................................................69

viii
 LIST OF FIGURES

Figure 1. Schematic representation of EM waves (From Kidder and Vonder Haar et

al. [and elsewhere], 1995 ).................................................................................2
Figure 2. Refraction via Snell’s Law, where θ1/θ2 are incident and refracted angles
(respectively), and n1/n2 are the indices of refraction for each medium
(Rinehart, 1997). ................................................................................................4
Figure 3. Scattering regimes with respect to wavelength (λ), particle cross-sectional
size r, and size parameter (χ) (Kidder et al., 1995)............................................5
Figure 4. Forms of scattering (After Kidder et al., 1995). ................................................6
Figure 5. Elevated trapping layer, which enables extended EM wave propagation. ........9
Figure 6. Refraction categories (SPAWAR Systems Center, 2007). Center ..................10
Figure 7. Alligator Bayou, NSWC-PC............................................................................12
Figure 8. Northern transmit site on floating dock, NSWC-PC. ......................................12
Figure 9. PC layout of data logger network, from Wong (2009). ...................................14
Figure 10. Transmit site M1 (foreground) and MNPS (background). Towers were
identified as follows: M1 (NRL tower #1, Tx site), M2 (NRL tower #2, Rx
site), and MNPS (NPS tower, Tx site). ..............................................................15
Figure 11. Representative sample of raw (blue) and smoothed (green) RF data. .............18
Figure 12. Plot of χ2 difference between raw data and 8th order polynomial, justifying
selection of polynomial order. .........................................................................19
Figure 13. Sub-plot comparison of RF loss to smoothed wind speed...............................22
Figure 14. Sub-plot comparison of RF loss to q................................................................23
Figure 15. Sub-plot comparison of RF loss to referenced M. ...........................................24
Figure 16. Time-series comparison of NOAA and NRL tidal data...................................25
Figure 17. Time-series plot of salinity values (blue) and trend line (green). ....................26
Figure 18. Hourly segment sample of RF loss. .................................................................28
Figure 19. Three-day time-series of RF net loss depicting times of missing data. ...........29
Figure 20. Scatter plot of RH (raw with outliers) at each level to the functionally
dependent residuals of RF loss vs. tides. .........................................................31
Figure 21. Scatter plot of RF loss to NRL derived MSL tides with linear trend line. ......33
Figure 22. Sub-plot comparison of net loss and ε1. ...........................................................34
Figure 23. Time-series sub-plot of ε1 to solar irradiance. .................................................35
Figure 24. Scatter plot of ε1 to solar irradiance. ................................................................35
Figure 25. ε1 to ε2 sub-plot comparison and overlay plot..................................................36
Figure 26. Best fit sin wave correlation via datacorr6.m with mean relative error
between fit and trend line (red) and mean relative error between mean line
and trend line....................................................................................................38
Figure 27. Trigonometric relationship between direct and surface scatter rays from
transmitter to receiver. (Red is the actual ray, green is the virtual reflected
ray.) ..................................................................................................................40
Figure 28. Mean sea level height plot of tides, Rx antenna and stationary dock. .............41
Figure 29. Direct (slant) path to surface scattered ray path length....................................42

ix
Figure 30. Path length difference (top) corresponding to phase difference (below) at
1.78 GHz. .........................................................................................................43
Figure 31. Illustration of destructive interference. Blue curve is zero phase, green is
the phase shifted waveform, and red is the sum of the two. ............................44
Figure 32. RF loss to phase difference time-series. ..........................................................45
Figure 33. Direct path ray and reflected path actual (in red) and virtual (in green)
rays...................................................................................................................48
Figure 34. Empirical and modeled (scaled) RF loss (See Equation 27) to antenna
height................................................................................................................50
Figure 35. Modeled RSSI characteristics. .........................................................................51
Figure 36. Scatter plot of tidal height to net RF loss (dB) in blue and linear trend-line
(gray). All subsequent plot plots utilize the residuals from this calculation...60
Figure 37. Scatter plot of calculated residuals to water surface skin temperature (˚C).
The top plot is from sensor #KT18.85-I, the bottom from sensor
#KT18.85-II. ....................................................................................................60
Figure 38. Scatter plot of calculated residuals to sampled atmospheric pressure
(mbar)...............................................................................................................61
Figure 39. Scatter plot of residuals to air temperature (˚C) at each of the six levels. .......61
Figure 40. Scatter plot of residuals to solar irradiance (W m-2).......................................62
Figure 41. Scatter plot of calculated residuals to submerged water temperature (˚C) at
each of the three levels, where #1 was 4.0 cm, #2 was 8.0 cm, and #3 was
18.0 cm below the surface. ..............................................................................62
Figure 42. Scatter plot of calculated residuals to relative humidity (%) at each of the
six measured levels. #1 was the lowest (11.5 cm), up to #6 which was the
highest (247.0 cm) above the water surface.....................................................63
Figure 43. Scatter plot of calculated, normalized residuals to raw wind speed (top)
and the vector mean of wind speed (bottom). Both plots are in units of m
s-1.....................................................................................................................63
Figure 44. Scatter plot of normalized residuals to the vector mean of wind direction
(˚True) ..............................................................................................................64
Figure 45. Scatter plot of calculated, normalized residuals to specific humidity (g g-
1), at each of the six levels, and referenced to standard atmosphere
(1013.25 mbar).................................................................................................64
Figure 46. Scatter plot of normalized residuals to calculated modified refractivity (M-
units), at each of the six observed levels..........................................................65
Figure 47. Scatter plot of calculated, normalized residuals to the gradient of modified
refractivity (M-units m-1) ................................................................................65

x
 LIST OF TABLES

Table 1. Equipment listing, where Locations depicts possible areas in which units
were deployed. Towers were identified as follows: M1 (NRL tower #1,
Tx site), M2 (NRL tower #2, Rx site), and MNPS (NPS tower, Tx site). ..........16
Table 2. Tower sensor profiles listing, towers were identified as follows: Tower1
(NRL tower #1, Tx site), Tower2 (NRL tower #2, Rx site), and TowerNPS
(NPS tower, Tx site). .......................................................................................21
Table 3. Table of known transmitter interruptions. .......................................................29
Table 4. Summary of variance reduction where ε is the resultant vector after
removing the variance attributed to that variable, Variance is the raw value
for each ε vector and % Variance Reduction is the overall percent
reduction from the original variance................................................................37

xi
THIS PAGE INTENTIONALLY LEFT BLANK

xii
 LIST OF ACRONYMS AND ABBREVIATIONS

ABL Air Boundary Layer

CPU Central Processing Unit
dB Decibel
dBm Decibels referenced to one milliwatt
DC Direct Current
DoD Department of Defense
d/dz Gradient in Height
e Saturation Vapor Pressure
EM/EO Electromagnetic/Electro-Optical
GPS Global Positioning System
IED Improvised Explosive Device
M Modified Index of Refraction
n Index of Refraction
N Refractivity
NEP-6 Near-Earth Propagation (6th in series)
NOAA National Oceanic and Atmospheric Administration
NPS Naval Postgraduate School
NRL Naval Research Laboratory
NSWC-PC Naval Surface Warfare Center – Panama City
NTP Network Time Protocol
PMCC Pearson Product-Moment Correlation
PVC Polyvinyl Chloride
q Specific Humidity
RF Radio Frequency
RH Relative Humidity
RSSI Received Signal Strength Indication
ρ Air density
T Temperature

xiii
THIS PAGE INTENTIONALLY LEFT BLANK

xiv
 ACKNOWLEDGMENTS

First and foremost, I would like to thank the Lord for his grace, support and
peace. Additionally, this thesis would not have been possible without the guidance and
solid support of several individuals. My thanks and sincere gratitude go to Dr. Peter
Guest and Dr. Andreas Goroch. These fine gentlemen are exacting scientists and
consummate professionals, and it was by their patient guidance, thought provoking ideas,
and gentle corrections that this research project came to fruition. Thank you as well to
Mr. Dick Lind for his efforts in designing and building the NPS measurement suite, in
addition to his daily quips that always brought a smile to my face.
Next, I would like to thank Mr. Robert Wert, Evan Worthington, and Vincent
Wong in the Tactical Electronics Warfare Division at Naval Research Laboratory,
Washington DC for their organizing the tests and providing advice and support in the
collection and analysis of the propagation measurements. It was through them that I
learned what field research really was.
Any worthwhile endeavor requires not only intellectual, but also financial
support. I would like to thank the NPS JIEDDO foundation for their considerable and
ongoing commitment to furthering this area of research and in so doing, making this
project feasible.
Finally, I extend my love and appreciation to my dearest wife, whose constant
support, gentle understanding, and quiet willingness to simply listen to both my rants of
frustration, as well as my giddy declarations of success, were integral in my ability to
complete this work.

xv
THIS PAGE INTENTIONALLY LEFT BLANK

xvi
 I. INTRODUCTION

A. MOTIVATION

Currently, the effects of meteorological and physical phenomena associated with

the various dynamic and thermodynamic processes in the very near surface environment
(i.e., within the lowest 10% of the surface layer), is poorly understood. By properly
characterizing what is happening in the real atmosphere, we can obtain an understanding
of the physical processes involved in radio frequency (RF) transmissions, with the goal of
obtaining empirical formulas relating these processes. These, in turn can be used to more
accurately model RF propagation at short ranges and near the surface. This study is a
crucial step to more reliable communications with unmanned vehicles, wireless networks
and a more effective means of detecting, jamming, and/or pre-detonating mines and IEDs
in various environments.

B. BACKGROUND

1. Electromagnetic Waves

Electromagnetic radiation (EM) consists of alternating electric and magnetic

fields (Figure 1). The electric field vector is perpendicular to the magnetic field vector,
and the direction of propagation is perpendicular to both. The fundamental unit of
frequency, ν, is the hertz (Hz), rather one cycle per second and is related to wavelength λ,
by

ν = cλ
, (1)

Where, c is the speed with which electromagnetic radiation travels in a vacuum (2.997 x
108 m s-1) and is known as the speed of light. In the atmosphere, it travels slightly more
slowly due to interaction with air molecules.

1
 Figure 1. Schematic representation of EM waves (From Kidder and Vonder Haar et
al. [and elsewhere], 1995 ).

The standard description of a plane wave electric field is

eikR
E (k , R) = A
kR
(2)

where k is the wave number (2π/λ) and R is the scalar distance between the transmitted
and received signal. The wavelength is related to the frequency by the speed of the
signal, which in this case is the speed of light. The power transmitted by the electric field
is

P= E
2

2
e ikR
=
kR
(3)

Radiation is often specified by its wavelength, λ, which is the distance between

crests of the electric or magnetic fields as depicted in Figure 1. The electromagnetic
spectrum spans a wide range of wavelengths from the ultraviolet to the microwave
region. This thesis focuses on the 1.78 GHz frequency.

2. Atmospheric Effects on Propagation

An EM signal propagating through a medium consisting of atoms and molecules

(i.e., not a vacuum), will interact with that medium through various processes, among
them absorption and scattering of EM energy. The efficiency of the interaction is

2
related to the amount of signal absorption. Changes in propagation velocity due to this
interaction results in a change in the wave’s direction, called refraction and is described
by its index of refraction (Bean et al., 1966).

It is important to note that only minor changes in refractive index are necessary to
cause a significant change in energy propagation. By using the relationship between
refractivity (defined below) and refractive index, n can be derived in terms of total
pressure, temperature, and water vapor concentration, as shown in Equation (4).
Variations of temperature and moisture in the propagation path cause local refraction of
the signal, resulting in signal loss and increase of noise (Thayer, 1974).

⎛P⎞ ⎛e⎞ ⎛ e ⎞
n = 77.6 x 10−6 ⎜ ⎟ – 5.6 x 10−6 ⎜ ⎟ + 3.73 x 10−1 ⎜ 2 ⎟ + 1 (4)
⎝T⎠ ⎝T ⎠ ⎝T ⎠
where n: index of refraction
T: temperature (°K)
P: localized atmospheric pressure (mbar)
e : water vapor pressure (mbar)
Since n ≈ 1 in the atmosphere the only portion of this parameter that truly matters
is tied to the fourth, fifth and sixth decimal places. To alleviate the use of exceedingly
small numbers, scientists have found it useful to define another parameter, called
refractivity (N) and is related to n by (n – 1) x 106, or similarly:

⎛P⎞ ⎛e⎞ ⎛ e ⎞
N = 77.6 ⎜ ⎟ − 5.6 ⎜ ⎟ + 3.73 ×105 ⎜ 2 ⎟ (5)
⎝T ⎠ ⎝T ⎠ ⎝T ⎠
where N: refractivity
T: temperature (°K)
P: localized atmospheric pressure (mbar)
e: water vapor pressure (mbar)
Note the strong variability due to humidity (vapor pressure). On average, refractivity
decreases with altitude, but at a gradually decreasing rate (Rinehart, 1997).

It is often more convenient to represent EM radiation as rays of energy instead of

waves. Rays are lines along which waves travel and are drawn perpendicular to each
wave front. Snell’s Law (also known as Descartes’ law) is a formula used to describe the

3
relationship between the angles of incidence and refraction of rays passing through a
boundary between two different isotropic media:

n1sin(θ1) = n2sin(θ2), (6)

where θ1, θ2 are the incoming and outgoing incident angles (respectively), and n1, n2 are
the indices of refraction of two isotropic materials. These indices of refraction of the
media are used to represent the factor by which a wave’s speed decreases when traveling
through a refractive medium (Medwin, 2000). Snell’s law quantifies these speed changes
in the form of angular change, rather, the degree of “bending” of the ray. This bending is
one component which affects signal strength. Additionally, by knowing the refractive
index in the atmosphere at each level, one can predict the path the EM waves will follow
(Rinehart, 1997). See Figure 2.

Figure 2. Refraction via Snell’s Law, where θ1/θ2 are incident and refracted angles
(respectively), and n1/n2 are the indices of refraction for each medium
(Rinehart, 1997).

Finally, scattering of EM radiation also plays a part in the total of atmospheric

effects. Scattering is a function of several factors: particle shape, size, index of
refraction, wavelength (λ) and viewing geometry. In 1908, Mie applied Maxwell’s
equations (which describe EM radiation), to the case of a plane electromagnetic wave

4
incident on a sphere. Mie showed that for a spherical scatterer, the scattered radiation is a
function of only viewing angle, θ, and the size parameter defined as:

2π r
χ≡ , (7)
λ

where r is the radius of the sphere.

This size parameter can be used to divide scattering into three regimes. Mie
scattering occurs for size parameters in the range of 0.1–50 in which the wavelength of
the radiation and the circumference of the particle are comparable. Rays undergoing Mie
scattering are strongly scattered forward and backward, respectively. For χ greater than
about 50, the scatterer is large in comparison to the wavelength of the radiation and
results in Geometric optics, widely observed in the occurrence of rainbows following a
storm. Geometrically scattered rays are very strongly oriented forward. Finally,
Rayleigh scattering results when the size parameter is much smaller than the given
wavelength and thus is largely insensitive to particle shape (Kidder et al., 1995). See
Figure 3.

Figure 3. Scattering regimes with respect to wavelength (λ), particle cross-sectional

size r, and size parameter (χ) (Kidder et al., 1995).

5
 3. Characterization of Water Surface

Regardless of the regime, how RF radiation is reflected or absorbed is a

complicated process and approximations are often used in place of rigorous calculations.
These approximations of reflectance properties include Lambertian and specular
reflectance. Lambertian, also known as isotropic reflectance, occurs when rays are
equally reflected in all directions, e.g., fog. Specular, or “mirror-like” reflectance occurs
when rays are scattered in a specific direction, such as over a smooth body of water.
Depending on the surface material type, index of refraction (n), incident angle and
roughness, diffuse scattering may also occur (Kidder et al., 1995).

Figure 4. Forms of scattering (After Kidder et al., 1995).

The importance of scattering is quantified by σo, the scattering cross section, and
is most well known for its use in the radar equation. There are two primary mechanisms
which contribute to σo, namely specular reflection (as mentioned previously), as well as
resonant or Bragg scatter. Surface waves that are near the backscatter radiation length,
Lb,

λ
Lb = (8)
2sin(θ )

6
when subjected to coherent EM/EO waves greater than 20° off nadir (θ), will be Bragg
scattered. This simply means that reflected radiation from one wave reinforces (or
cancels) that from the successive wave (Kidder et al., 1995, p. 246).

Another important aspect of the water surface is roughness. Surface roughness

occurs when wind induces waves of all wavelengths on the sea surface. Currently, the
best indicator available of surface roughness can be found in sea state tables that relate
wave height to wind speed. However, it is unlikely that this manner of quantifying
surface roughness will play a part in the near surface environment. Other contributors to
surface roughness include currents, tides, precipitation and turbulence.

4. Other Factors Affecting Transmission Loss

Ultimately, we seek to link signal propagation loss to the micrometeorology

associated with various processes occurring within the near Earth environment. A
number of those traditional sources of transmission loss were addressed in the preceding
sections. However, there exist other significant propagation mechanisms, and their
affects must also be taken into account.

An atmosphere at rest produces no variations in amplitude, phase shift or angle-

of-arrival. While this is not the environment in which we work, fortunately, the myriad
effects on RF wave propagation in the free atmosphere are well documented. One such
affect is attenuation by molecular gaseous absorption; however, it is only noted here, as
there is very little molecular absorption of RF energy over short distances.

In contrast to the above, the propagation loss due to the diffusion (spreading) of
the EM waves’ energy over distance from the source possesses tremendous influence
over an RF waves’ propagation behavior and strength. The two major geometries utilized
⎛1⎞
to model this spreading effect are spherical, in which intensity (I) is proportional to ⎜ 2 ⎟
⎝r ⎠
⎛1⎞
and cylindrical, in which I ∝ ⎜ ⎟ , where r is the radial distance from the source
⎝r⎠
(Medwin, 2000). Since power is the square of intensity, the 4th power drop off is shown
in spherical spreading.
7
 Refractive losses, specifically those due to changes in wave propagation speed
through a medium, can result in either a loss or a gain of signal strength. Constructive
interference results in an increase in signal strength, while destructive interference can
reduce or obliterate entirely received signal strength. These effects are due to multipath,
which is the propagation phenomenon resulting when an RF signal reaches the receiving
antenna by two or more paths. Causes of multipath include atmospheric ducting,
ionospheric reflection and refraction, and reflection from terrestrial objects, such as
mountains, buildings or ocean waves (FS 1037C, 1996). Additionally, the weight of the
influence of multipath depends upon frequency, path length (or phase) difference and the
reflection coefficient of the reflecting surface. This element was crucial to this thesis and
can be best described as the relative “sum” of the direct path and scatter path ray, where
signal strength is increased when phase difference is very small (i.e., nearly zero) and
wave propagation is reduced when phase difference is large (i.e., maximum destructive
interference).

As well as the aforementioned convergence, divergence and multipath effects,

waves propagating through a medium comprised of a particular type of gradient can bend
downward more than the Earth’s curvature and propagate well over the horizons.
Dubbed a trapping layer, this phenomenon is one of several classes of refractivity.
Before investigating these further, one must first be able to clearly identify those regions
where this propagation pattern is most likely to occur. When the gradient of N is less
than -0.157 m-1, an atmospheric duct can likely be found. Unfortunately, this is rather
difficult to ascertain visually from traditional N vs. z diagrams. This problem was solved
with the introduction of the modified refractivity parameter, M.

p e e
M = 77.6 − 5.6 + 375000 2 + 0.157 z (9)
T T T
Where: T is temperature (°K)
p is atmospheric pressure (mbar)
e is partial water vapor pressure (mbar)
z is height above surface (meters)

8
 Modified refractivity is used to determine how rays are bent relative to the Earth’s
curvature and assist in determining regions of ducting. This is illustrated by way of
Figure 5, an adaptation from Helvey et al. (1983), in which the areas of negative gradient
contain a high likelihood of a duct. Thus, ducts, which can be surface based or elevated,
are simply atmospheric layers that have formed when M decreases with height. Radar or
other EM waves can be “trapped” within them and may result in greater energy
propagation than normally expected.

Figure 5. Elevated trapping layer, which enables extended EM wave propagation.

In addition to ducting, EM waves can also be refracted in different ways and their
behavior forms the basis of the classes of refractivity. When rays are sub-refracted, their
propagation distance to the horizon has been decreased and occur when dM/dz > 157
(unitless). Conversely, super-refracted rays occur when 78 > dM/dz > 0 and distance to
the horizon will increase. Finally, normal refracted rays occur for 157 > dM/dz > 78, and
distance of propagation is unaffected outside of typical propagation loss terms (see Figure
6)

9
Figure 6. Refraction categories (SPAWAR Systems Center, 2007). Center

10
 II. EXPERIMENT

A. INTRODUCTION

The Panama City Near-Earth Propagation Test (NEP-6) was designed to build
upon previous experimentation conducted by Naval Research Laboratories (NRL) under
their Near-Earth Propagation (NEP) program. This research effort investigated the
unique radio frequency (RF) phenomena that occur within the lowest two to three meters
above the Earth’s surface. The results of this experiment form the subject of this thesis.
The purpose of the NEP-6 was to measure the propagation of radio frequency (RF)
signals close to the earth’s surface but in an environment characterized by high
atmospheric humidity and surfaces of high liquid water content.

While understanding RF effects is certainly worthwhile from a basic scientific

standpoint, the underlying reason behind this endeavor is to address a recently identified
threat to maritime defense and harbor areas in particular.

B. METHODOLOGY

The NEP-6 program collected meteorological and radio frequency data in the
open atmosphere under a variety of environmental conditions. The test was hosted by the
Naval Surface Warfare Center in Panama City Beach, FL, with data being collected in
two separate segments over a period of time from 21 August 2009 to 02 September 2009.
Part one investigated RF propagation over land adjacent to a body of water—in this case,
the shore of St. Andrew’s Bay. Part two focused on RF wave propagation strictly over a
body of water and is the focus of this analysis. Data for this portion were collected from
29 August 2009 until completion of the experiment. The meteorological and RF
transmit/receive equipment were situated over a small inlet of the Ovauahannah River,
known locally as “Alligator Bayou.” The bayou is connected to the Gulf of Mexico and
as such water level varies diurnally due to the tides (Figure 7).

11
 Transmit

Receive

Figure 7. Alligator Bayou, NSWC-PC.

Figure 8. Northern transmit site on floating dock, NSWC-PC.

12
 The signal variation was measured in the 225 MHz, 760 MHz, and 1786 MHz RF
spectral regions with omni-directional disc cone and horn transmit and receive antennas
placed near the surface of the water. The investigation was comprised of two test sites: a
transmit station located to the North on a floating dock (Figure 8), and a receive site
placed 73 meters to the South on a permanent (non-floating) dock (Figure 7). Both sites
included a full meteorological tower while the transmit site contained additional IR
temperature and Campbell water temperature probes, as well as a 3-D sonic anemometer,
pyranometer, and LI-COR open path gas analyzer for moisture, temperature and CO2
flux. Furthermore, a few salinity measurements were obtained manually at both sites.
This configuration enabled the collection of data necessary to study the effects of surface
roughness, temperature and humidity gradient, turbulence and tidal and salinity variation
on changes in the net loss of RF energy.

C. DISTRIBUTED SENSOR SETUP

Over the course of four days, multiple data collection runs were conducted at
periods of predominantly twelve and twenty-four hours in length. Individual nodes were
established at both the North (Tx) and South (Rx) positions, where each sites’ laptop
computer collected and time-stamped RF data internally. In addition to the RF data
logging, meteorological data were stored on a separate Campbell data logger. All of this
information was then synchronized with GPS time using an NTP time server
broadcasting over a secure wireless network. Finally, all recorded data were downloaded
from the data logger to a master computer on a daily basis. This was done to prevent
potential data loss due to failure of the Campbell data logger’s flash media cards. Even
with these safeguards in place, there were some brief periods of no data collection

13
 Figure 9. PC layout of data logger network, from Wong (2009).

D. EQUIPMENT

The meteorological towers measured temperature and relative humidity with

Campbell Scientific DC215-L temperature/RH probes installed on a logarithmic vertical
grid from 3 cm above the water to a max height of 258 cm (see Figure 10).

14
 Figure 10. Transmit site M1 (foreground) and MNPS (background). Towers were
identified as follows: M1 (NRL tower #1, Tx site), M2 (NRL tower #2, Rx site),
and MNPS (NPS tower, Tx site).

The probes were aspirated with a low-flow fan to maintain a flow of ≈2 m s-1 past
the sensor. Analog wind speed and direction were measured with a Campbell 05106-L
(RM Young) marine wind monitor. Water radiative skin temperature was measured with
a Heimann KT18.85-II passive infrared sensor. Solar radiation was measured by a Kipp
& Zonen CNR-1 pyranometer. Lastly, the Campbell sonic anemometer (CSAT3) and LI-
COR (CS7500) were extended 2.5 meters over the water from an aluminum pole that was
secured to the stationary dock next to the transmit site. Additional equipment used is
listed in Table 1.

15
Table 1. Equipment listing, where Locations depicts possible areas in which units
were deployed. Towers were identified as follows: M1 (NRL tower #1, Tx
site), M2 (NRL tower #2, Rx site), and MNPS (NPS tower, Tx site).

16
 III. DATA PROCESSING

A. SUMMARY

The data that were collected during the Panama City Near-Earth Propagation Test
(NEP-6) consisted of very high resolution (200 Hz) radio-frequency (RF) signals,
turbulence and flux measurements, as well as reduced resolution (1 Hz) meteorological
data. Sea surface level was measured every 6 minutes. Additionally, water samples for
salinity measurements were obtained manually on a sporadic basis, but typically once
each during the morning and evening time-frame. Due to the wide range of data
resolutions, it became necessary to first establish a baseline resolution in the time series
(in this case, 1 Hz), and then interpolate the coarse data sets to higher frequencies before
filtering, or similarly, average the higher resolution data sets to match the desired 1
second time scale.

B. RF DATA PROCESSING

The RF data set consisted of transmit and receive power levels in dBm for both
the 225 MHz and 1.78 GHz frequencies. This study focused on the 1.78 GHz portion as
this is within the frequency band typical of most peer-to-peer wireless networks,
particularly those with defense applications (Merrill et al., 2004, p. 36). Because both
transmit and receive data were collected at a rate of 200 Hz and the length of the time
intervals at which data were amassed ranged from a minimum of 12 hours to a 28-hour
maximum, the data sets were necessarily large (on the order of 1 terabyte). Due to the
difficulty of processing large files, these data were recorded in roughly one hour
increments and stored in binary format to flash media. Each day, the LoggerNet protocol
downloaded to a master computer where it was subsequently converted to an application-
specific (.dat) data file.

The first step in processing these data was to convert them from the proprietary .dat
format to MatLab™ compatible .mat format. This was accomplished using a script file
developed by Mr. Vincent Wong of Naval Research Laboratories, Washington, DC.
Once converted, each hour was plotted over time to provide a rough estimate of transmit,
17
receive, and net loss levels. Both transmit and receive data were recorded as dBm, which
is the power ratio in decibels (dB) of the measured power referenced to one milliwatt
(mW). This proved to be the most convenient unit of measure of absolute power because
of its capability to express both very large and very small values in a relatively short
form. Net propagation loss was the parameter selected to indicate the attenuation of an
RF wave propagated in the given environment. Net propagation loss is the difference
between transmitted power and received power. These calculations yielded the net loss
vector in dB, a dimensionless unit that quantified the ratio between these two values. At
this point, obvious outliers that were greater than 2 standard deviations from the mean
were removed using basic MatLab™ indexing commands. Next, the hourly data files
underwent initial smoothing to accommodate the expected shot and thermal noise from
the transmitting RF diode. This noise is clearly depicted in Figure 11 as a 0.5 dBm
fluctuation (in blue) following the received power trend.

Figure 11. Representative sample of raw (blue) and smoothed (green) RF data.

18
 Upon completion of this initial smoothing, the loss vector from each hourly data set
was filtered (using various window sizes), and then sampled to achieve an intermediate
resolution of 10 Hz. Window size was selected manually in order to minimize side lobe
influence on the running mean and consisted of starting with a very wide filter and then
gradually reducing the width until the desired resolution was obtained. This yielded a vector
approximately 24 million data points in length and was necessary due to computational
constraints. Data treatment continued by first calculating the best fit polynomial trend line (in
this case, an 8th order), subtracting out this trend and then filtering the output over various,
relatively large window sizes. Higher-order polynomial trend lines did provide slightly lower
χ2 error values, but as Figure 12 depicts, the relative reduction in error for each subsequent
increase in order was insignificant beyond the 8th order. This de-trending technique ensured
maximum filtration of noise from the actual data, while maintaining the dominant diurnal
oscillation, as well as to prevent those time periods in which data were missing from overly
influencing the filtered output.

Figure 12. Plot of χ2 difference between raw data and 8th order polynomial, justifying
selection of polynomial order.

19
 The polynomial trend line was added back in and the data were interpolated onto
a 1 Hz resolution time series, using interp1 (a built-in function utilizing linear
interpolation). All filtering calculations used the MatLab™ function filter, which
utilizes rectangular amplitude weighting to generate a simple running mean of the
surrounding data points, as follows: yn = Σck xn-k + Σdjyn-j , where d=zero, c=1/(window
order), and k goes from 1 to window width. Window order was linear, and widths were
selected manually based on completeness of data and resolution.

C. METEOROLOGICAL DATA PROCESSING

Meteorological data were collected at 200 Hz, 1 Hz and greater time scales—each
frequency class was treated differently. The purpose was to present the greatest amount
of information that could reasonably be assimilated and compared to the RF data series.
The 200 Hz turbulence and flux data were processed and analyzed in a similar manner to
the RF data series, however some effort was made in reducing the amount of information
lost to filtering and sampling as this data set was not nearly as robust as the one
previously discussed due to the “averaged” nature of these measurements. Data loss was
reduced by limiting the number of iterations of filter/sampling as well as fitting of
polynomial curves. This allowed the use of a “noise reduced” (or low pass filtered) data
set that still contained the important variations, while removing the higher frequency
components that were not of interest. The 1 Hz data were derived from measurements on
two full towers and one partial meteorological tower as detailed in Table 2 (next page).

20
 Table 2. Tower sensor profiles listing, towers were identified as follows: Tower1
(NRL tower #1, Tx site), Tower2 (NRL tower #2, Rx site), and TowerNPS
(NPS tower, Tx site).

On the transmit side of the experiment layout, the floating dock supported the RF
transmitters, two temperature and relative humidity (RH) probe suites situated over a
nominally logarithmic scale (from 3.0 cm to 258.0 cm), as well as various meteorological
equipment sensing wind and rainfall. Temperature and RH was smoothed using filter
with a rectangular window size of 1000 data points, and then sub-plotted against RF loss
for comparison and analysis. For derived meteorological variables such as specific
humidity (q), modified index of refraction (M), and gradients, measured data were
initially used and then the output was filtered and interpolated in a manner similar to
temperature and RH.

Wind data were integral to the analysis and were processed as follows. Raw wind
speed and directional data were first converted to component vector form (u,v) using the
MatLab™ script uv.m. Noise and aberrant data were removed from the data set by way
of two mathematical methods: the first was filtration of a simple scalar mean and the

21
second utilized the individual u and v components to calculate a vector mean. Wind
direction was averaged using a unit vector technique to avoid the problem of averaging
around North, e.g., preventing 359˚ and 001˚ from averaging to an obviously incorrect
180˚. Using the component vectors safeguarded against this and ensured a truthful
representation of the dominant winds experienced at the transmit site over the course of
the three day trial. Both resulted in similar time-series plots (Figure 13); however, the
calculated amplitude in the vector mean method was reduced by a factor of 0.2.

Figure 13. Sub-plot comparison of RF loss to smoothed wind speed.

22
 Figure 14. Sub-plot comparison of RF loss to q.

Specific humidity (q) and saturation vapor pressure (e) were first calculated using
raw values for temperature, relative humidity, and pressure, with subsequent output being
filtered and then sub-plotted against RF loss. Window size was selected based on
maximum smoothing of the time series, while still minimizing data loss. Additionally, to
aid in data assimilation and pattern recognition, q, T, M, and RH were also plotted
against an arbitrary reference level (see Figure 15). This was simply the centermost
measurement level for any given data set.

23
 Modified index of refraction was calculated using the mod_refract.m
function by Bean (1966), which is actually identical to Equation (13) with the exception
of the last decimal of accuracy. A separate script for this equation was not written simply
because the built in function was sufficient for our observations. Also, due to the close
ranges and low heights, Earth’s curvature was not important for this experiment; a base
height of 0 meters was assumed for z, thus in all actuality this calculation provided N,
refractivity—a difference which did not affect data analysis.

Figure 15. Sub-plot comparison of RF loss to referenced M.

Finally, internal water temperature and surface skin water temperature values
were gathered via submerged thermistor probes and two passive IR sensors. Processing
of this data was the same as in the case of air temperature and relative humidity.

24
D. OCEANOGRAPHIC DATA PROCESSING

Tidal data had the largest apparent effect on interference and attenuation in the
transmitted RF wave. This information was obtained from two sources: the first and
primary data set was of actual tides measured on site by means of a SR50A Sonic
Ranging Sensor that was situated on the NRL tower. According to the Campbell
Scientific data sheet (2009), this sensor emits a 250 mA ultrasonic pulse that enables
height measurements with an accuracy of ± 1 cm.

NRL vs. NOAA Tidal Variation

7:09 PM 6:18 AM 7:08 PM 6:18 AM 7:07 PM 6:19 AM
2.5

2
MSL Height of Tide (Meters)

1.5

0.5

NRL Tidal Data

NOAA Tidal Data
0
08/29 08/30 08/31
Time (Days - CDT)

Figure 16. Time-series comparison of NOAA and NRL tidal data.

A secondary data set was drawn from a NOAA tidal observation station based in
Panama City, FL (Station ID: 8729108), which was located approximately 5 n mi.
Southeast of the NEP-6 experiment site. As one would expect, the range of diurnal
oscillation of the more sheltered NRL tidal data was much reduced when compared to the
NOAA tides (which were collected at the inlet to St. Andrews Bay), and also possessed a
lower (by 0.299 meters) mean magnitude (Figure 16). Due to their low resolution, both

25
data sets were interpolated onto a one second time grid in order to match the RF data
series, although filtering was unnecessary given their relative clarity and lack of noise.

Salinity samples were collected at both the transmitter and receiver sites, and
generally in the morning. Each sample was tested on an “AutoSal” 8400B Laboratory
Salinometer which has a measurement accuracy of < ± 0.002 Equivalent PSU according
to the Guildline Instruments data sheet (2002), and is obtained via calculation and
substitution in the Bennet equation. It was difficult to garner any sort of understanding
from this data set due to its sporadic nature and low number of data points recorded (blue
circles above).

Salinity Variation 29Aug - 1Sept

40

35

30
Salinity (Equiv. PSU)

25

20

15

10

0
08/29 08/30 08/31
CDT Time (Days)

Figure 17. Time-series plot of salinity values (blue) and trend line (green).

However, some effort was made (using the spline command) in fitting a
recognizable curve (green line) to this sparse data set in order to enable broad comparison to
both RF and meteorological data (see Figure 17). Akin to the aforementioned data
processing techniques discussed, salinity values were interpolated onto a 1 Hz grid and then
sub-plotted against the other measured variables. This result proved to be less than perfect
since spline only calculates a cubic spline interpolation of the data and is limited by the
size of the data vector. Nonetheless, it did provide the basic insight that salinity fluctuated
approximately in time to tidal influence (as expected), and though we suspect it may have
been related to the observed lag between RF loss and tides, it is unlikely that salinity played a
direct role in RF propagation loss as shown by the single variable correlation calculations.

26
 IV. DATA ANALYSIS

A. SUMMARY OF METHODOLOGY

The primary goal of the analysis was to identify the variables that were most
correlated with RF loss variations. The variance of the RF data that can be explained
(i.e., correlated to) by a particular environmental variable can be subtracted out, leaving
residuals that can then be further correlated with other variables. Whatever variance that
remains after testing with all variables is ideally a small amount and represents unknown
effects. This study utilized a multi-tiered process of statistical and data processing
techniques as follows: a basic rough data preview, a smoothed data sub-plot comparison
via visual inspection, linear (through scatter plot trends) and sine wave correlation, and
finally, physical geometry association and subsequent modeling of the data.

B. RAW DATA PREVIEW

One of the more difficult challenges in working with so much measured

information was in determining what fraction of the data were vital, and which were
extraneous and therefore unneeded. While in the end the size and scope of the data sets
were reduced by way of the previously discussed methods, it was still considered
essential to provide an overview of the information as presented in its raw form. The
difficulty in this was that data were collected at 200, 1, and 1/360th Hz, and that each
individual data set was not necessarily complete, hence, the need for interpolation of the
longer (200 Hz) and shorter (1/360th Hz) data sets to a common time-series. The
reference level of 1 Hz was selected as because all of the meteorological data were
collected at this rate, it provided sufficient resolution for the variables considered, and did
not require excessive CPU processing time.

Computational constraints were accommodated by merely loading small slices of

the data (typically hourly but, for some lower resolution files, longer times were used) at
a time (Figure 18, next page). Two-hundred Hz data—particularly the RF signal files—
were loaded, plotted, and saved in hourly segments via an automated loop.

27
 Figure 18. Hourly segment sample of RF loss.

Visual inspection of these graphics provided the following insights:

a) High frequency variation on the order of 0.1 to 0.5 dBm, was noted on all
RF plots. This is comprised of both shot noise and thermal noise emitted
by the RF diode. These levels are consistent with expected shot noise
(typically very small) and the contribution from thermal noise. The
magnitude of thermal noise will increase as environmental temperatures
rise, hence the pseudo-correlation to time of day.

b) The two- to three-minute oscillations apparent in the transmit plot were likely
due to slight power draw fluctuations caused by the air conditioner that was
used to cool the transmitter. Internal environmental control was imperative in
order to protect the transmitter, preserve the positive signal to noise ratio, and
finally to protect the wireless radios that were transmitting data between the
two sites. Additionally, some fluctuation in the power supply was expected
due to the limited number of available receptacles.

28
 c) Known power outages and time periods in which the transmitter was
switched off for safety reasons (as shown in Table 3) were correlated to
empty intervals in the data (Figure 19).

Table 3. Table of known transmitter interruptions.

Figure 19. Three-day time-series of RF net loss depicting times of missing data.

While visual inspection of hourly data did permit depiction of fine scale
variability, it proved unwieldy for observing the dominant trends in the data. A three-day
plot of the RF net loss time series was finally realized thru exploitation of the sampled

29
version, which had been filtered in such a manner as to reduce the associated RF noise,
yet still allow the variance of the “unknown” drivers to have an effect. Moreover, by
superimposing the sunrise and sunset times onto the three-day plot, it became obvious
that the periods of greatest and least loss in received power occurred on a nearly perfect
24-hour cycle, and corresponded to sunrise and sunset. It was also noted that while the
RF loss vector seemed to follow this dominant 24-hour sinusoidal trend, there were also
smaller scale variations that could not be explained exclusively by the above three
observations. This predominant variability would ultimately be explained via the
analysis as follows.

C. SUB-PLOT COMPARISON OF VARIABLES

Proper smoothing and filtering of the meteorological and oceanographic variables

before they were compared to the RF loss vector were of utmost importance to this
portion of the analysis. Akin to the raw data analysis section, computational expense
(i.e., time) had to be considered, and thus either hourly segments or a course resolution of
three-day samples were used. Initially, each meteorological variable was simply sub-
plotted against the net loss vector on the same time series grid. This yielded an excess of
figures to review, but very few variables that clearly stood out as having a definitive
affect. Tidal fluctuation, solar radiation, and sampled air pressure seemed to have the
greatest visual correlation, while temperature and relative humidity provided relatively
high r-values.

The r-values for each of the above data sets were calculated using the MatLab™
command corrcoef, which is based on the Pearson product-moment correlation
coefficient (PMCC). By definition, this is the covariance of the two variables divided by
the product of their standard deviations:

(10)

where ρX,Y is the population correlation coefficient. By substituting the estimated

covariances and variances based on the sample, the above gives the sample correlation
coefficient, commonly denoted r:
30
 (11)

where Xi, Yi are paired sample data, X , Y are the sample means, and n is the number of
data points in the sample (Cohen, 1988).

In summary, the PMCC r-values provided guidance as to the best direction to

follow when sifting through the numerous meteorological and oceanographic variables.
The data that showed the most promise were scatter plotted against both the smoothed RF
loss data, as well as the calculated residuals of the loss data. The intent was to find a
linear relationship between the dependent and independent variables. Efforts were
initially thwarted due to outliers that were missed during the smoothing process, as
illustrated in Figure 20 (next page) however these extraneous data points were removed
in subsequent iterations and correlation processing was much improved.

Figure 20. Scatter plot of RH (raw with outliers) at each level to the functionally
dependent residuals of RF loss vs. tides.
31
 The greatest advancement at this juncture was in removal of data outliers,
confirmation of certain variables that would not have an impact on RF loss, and finally,
by the positive and linear relationship between propagation net loss and tidal height.

D. VARIANCE REDUCTION TECHNIQUES

Upon completion of scatter plotting of the residuals to dependent variables, we

found that antennae height change as a result of tidal flow had the largest correlation with
(i.e., explained the most) variance detected in the RF loss data vector. In order to
quantify the level of influence this and other variables had, two techniques were
employed:

1) single variable correlation via scatter plots, and

2) best fit sine waves.

Unsurprisingly, both yielded similar results, however for reasons of efficiency, the former
method was more prevalently employed.

1. Linear Trend-Line Correlation

To correlate via scatter plots first the variance ( σ 02 ) of the data set was calculated

using the MatLab™ command var, which served as a test bench mark for subsequent
iterations. This command uses the standard definition for variance,

(12)

where E[·] is the expected value or mean, X is the random variable and μ is the sample
mean. Hence, the variance is a measure of the amount of variation within the values of
that variable (Cohen, 1988). Next, smooth RF data were plotted against one of the
dependent variables and a linear trend line was extracted (Figure 21). This basic formula
(y=mx+b) was then evaluated at each data point and subtracted from the original loss
vector. The resulting vector of new loss values was dubbed epsilon 1 (ε1) and possessed
an updated variance as calculated using the above command, var.

32
 Figure 21. Scatter plot of RF loss to NRL derived MSL tides with linear trend line.

If by way of application of Equation (13) the variance was reduced, then that variable
was considered to have had an effect and the problem became simply one of finding the
variable of greatest consequence.

(13)

In this manner, the preliminary variance of 0.4676 dB was immediately reduced

(by 74.5%) by removing the effect of antennae height fluctuation (Figure 22). It should
be noted that the tidal data used was taken from the onsite NRL tidal sensor and not from
the NOAA observation station.

33
 Figure 22. Sub-plot comparison of net loss and ε1.

Figures 23 through 25 depict pictorially the processing procedure. Note the still
oscillatory nature of the ε1 vector and rapid decrease in “noise” amplitude. The first was
a characteristic that persevered through several cycles of this process while amplitude
degenerated more slowly with each subsequent application. The raw variance of ε1 was
0.1090 dB.

Solar radiation displayed a strong correlation (PMCC r-value of ≈1.0) that was
difficult to ascertain visually, yet still produced statistically significant variance
reduction.

34
 Figure 23. Time-series sub-plot of ε1 to solar irradiance.

Scatter plotting ε1 against solar irradiance illustrated this linkage:

Figure 24. Scatter plot of ε1 to solar irradiance.

35
 Ultimately, ε2 reduced overall variance to 0.0917 (dB) and compared very closely
to ε1.

ε1

1
"Noise" (dB)
Remaining

-1

08/29 08/30 08/31

Time (Days - CDT)

ε2

1
"Noise" (dB)
Remaining

-1

08/29 08/30 08/31

Time (Days - CDT)

ε1 and ε2 Overlay

1
"Noise" (dB)
Remaining

0
ε1
ε2
-1

08/29 08/30 08/31

Time (Days - CDT)

Figure 25. ε1 to ε2 sub-plot comparison and overlay plot.

36
 Subsequent iterations were similar in nature and resulting single variable variance
reduction is summarized in Table 4. Note that Air Temperature and RH produced exactly
the same degree of variance reduction.

Table 4. Summary of variance reduction where ε is the resultant vector after

removing the variance attributed to that variable, Variance is the raw value
for each ε vector and % Variance Reduction is the overall percent reduction
from the original variance.

This technique worked well primarily due to its ease of use and reasonably solid
statistical significance based on the large sample size and outlier testing via scatter plots.
However, this still does not fully answer the question of variability in the data set simply
because it deals only in the single variable domain and the independence requirements
may not have been met. This variance is illustrated in the first three variables used to
reduce variance, in that they all follow a roughly diurnal cycle.

37
 2. Best Fit Sine Wave Correlation

The alternate method of fitting a sinusoid to the data set was effective because of
the cyclic nature of the majority of the measured variables. It also possessed the added
benefit of extrapolating between holes in the data set by following the dominant trend
(thus making it easier for the eye to see), and in being rather outlier resistance…although
at times these benefits were outweighed by the relative difficultly and tediousness of the
process.

Tides v. Sine Comparison

MSL Tidal Variation (m.)

3
Tidetrend
2 Sin Fit
Meanline

0
08/29 08/30 08/31
Time (Days - CDT)
Relative Height Difference (m.)

Mean Error Comparison

3

γ: 0.249822 γ(mean): 0.567338

2

0
08/29 08/30 08/31
Time (Days - CDT)

Figure 26. Best fit sin wave correlation via datacorr6.m with mean relative error
between fit and trend line (red) and mean relative error between mean line and
trend line.

First, a dependent variable (tides, air temperature, etc.) was plotted against the RF
loss vector. If a similar trend was noted in the dependent variable, a sine wave was fitted
to the data (as shown in Figure 26). To enable this fitting process, a MatLab™ script
datacorr6.m was used to minimize the relative error between the modeled variable

38
and the actual variable in the following manner. A generic sine wave (sinefit) was
constructed with variables for frequency, amplitude and phase shift:

sinfit = a + b *sin( f * t + φ ) , (14)

where a = centerline of sine wave
b = amplitude of the wave
t = time-series
f = frequency of operation
φ = phase shift
By manipulating these variables manually, the generic waveform was constructed
to closely resemble the dependent variable. Relative error between the actual and
modeled waveform was summarized in the “Mean Relative Error,” γ in Equation (15):

( Xdep − Xmod )
γ= , (15)
Xdep
where γ = mean relative error
Xdep = dependent variable
Xmod = modeled waveform

Additionally, the relative error between the measured data and a mean line of said
data ( Xdep ) were used to provide reference as to the level of improvement the sine wave
fit provided over a simple average:

( Xdep − Xdep )
γ mean = . (16)
Xdep

This process yielded the coefficients necessary to evaluated the modeled function
and subtract it from the original function, thereby reducing overall variance. Just as in
the linear correlation method, this new epsilon vector possessed an updated variance that
was ideally less than the original.

39
E. PHYSICAL GEOMETRIES ASSOCIATION

The correlation techniques described above provided some insight to the base
cause of the RF loss wave fluctuation. Unfortunately, the magnitude of variance
reduction by these techniques was ultimately limited and further application yielded
results of dubious quality and validity.

Figure 27. Trigonometric relationship between direct and surface scatter rays from
transmitter to receiver. (Red is the actual ray, green is the virtual reflected ray.)

In order to explain the residual variance due to tidal influence (See Table 4),
attention was turned to the physical relationship between direct path and single reflection
scatter-path rays (Figure 27). The direct path length between the transmitter and
receiving unit was measured via a laser range finder and found to be 73 meters. This
rough slant path measurement was obtained from the receiving site, (approximately 2.667
meters above the water) to the transmitter, which was situated approximately 0.3028
meters above the water. Due to the change in tidal height above mean sea level (recall,

40
the transmitter was mounted to a floating dock), this dimension contained an unknown
level of error. Nonetheless, because the angles involved were so minute we made the
following approximation, sin(α) ≈ α. This allowed a negligible baseline distance
measurement error and therefore usable as the base distance in all of the trigonometric
calculations.

Height Variation of Various

3
Tidal Variation and Height Above Waterline (Meters)

2.5
Tides
RxHt
Dock Height
2

1.5

0.5
08/28 08/29 08/30 08/31 09/01
Time (Days - CDT)

Figure 28. Mean sea level height plot of tides, Rx antenna and stationary dock.

First, a benchmark slant path distance between the RF equipment was attained by way
of the Pythagorean Theorem, and modified by the changing height above water due to
tides (as depicted in Figure 28). Not surprisingly, the path lengths, when plotted over
time, were unremarkably similar in nature to the base tidal fluctuation, though 180˚ out of
phase. This phase issue arises from the simple logic that low tide correlates to the highest
antenna height and vice versa for high tide.

Next, these direct path lengths were compared to the single bounce-path lengths,
which were calculated using Snell’s Law (angle of incidence equals the angle of
reflection). Based on the assumption that the strongest portion of the waveform would
41
scatter at this approximate midpoint of the waterline separation distance, the total bounce
path ray length was simply the hypotenuse of the mirror image triangle formed by the
aforementioned law. Thus, total bounce-path length was:

(17)

where BL is the baseline path length (waterline), and RxHt & TxHt are the heights of
receiver and transmitter respectively.

Understandably, these results were also sinusoidal in nature but when plotted
against the standard 1 Hz time-series, the mean vertical height displacement was only
1.685 cm and the maximum path length difference between the two was 1.64 cm (Figure
29).

Path Length Variation

73.04
Slantpath
Bouncepath
73.035

73.03
Path Length (Meters)

73.025

73.02

73.015

73.01

73.005
08/29 08/30 08/31
Time (Days - CDT)

Figure 29. Direct (slant) path to surface scattered ray path length.

Given our 1.78 GHz operating frequency, which corresponds to a wavelength (λ) of
16.85 cm, the oscillations in RF net loss could not be due to complex sub-beamwidth
interactions of antennae lobe patterns. This is because the direct ray path distance was

42
very small, the target receiver did not move, and the transmit antennae’s beam shape was
uniform for all observed parts of the antenna pattern (3dB drop off was 40˚).
Additionally, in 1997, Fabry et al. conducted a similar study using S-band microwaves
(≈3.0 GHz) that lends credence to the above assumptions. In particular, he found that in
cases of fixed, point-like targets, the phase of the target was relatively insensitive to sub-
beamwidth inaccuracy in antennae pointing (as opposed to complex interference of
antennae lobe patterns), but very sensitive to small path length differences.

Nonetheless, what Fabry and our results do depict is that even with very slight
path length differences, significant phase difference variation results (Figure 30).

Path Length Difference

0.02
Path diff.
0.018
Difference (m.)

0.016

0.014

0.012

0.01
08/28 08/29 08/30 08/31 09/01
Time (Days - CDT)

Phase difference (°)

Angular Phase Difference (°)

40
Phase diff.
35

30

25

20
08/28 08/29 08/30 08/31 09/01
Time (Days - CDT)

Figure 30. Path length difference (top) corresponding to phase difference (below) at
1.78 GHz.

Figure 31 is provided to illustrate the magnitude of interaction between two

identical but phase shifted waves. Each sub-plot is composed of two generic sine waves
of frequency 1.78 GHz, and unit amplitude. The top plot has been phase shifted by the
minimum difference (24.31˚) as shown in Figure 30, while the second plot depicts the

43
maximum (36.03˚). Relative increase/decrease in amplitude of the resultant wave is
provided in the text and clearly shows that minute angular differences, particularly when
paired with the theoretical reflection coefficient of an ideal glossy surface (R(θ)= -1),
result in drastic changes in amplitude. In this case, the reflection coefficient accounts for
the 180˚ phase shift incurred by the ray scattering off of the surface of the water.

Min Phase Difference (24.31° + 180°)

2

-1
% Decrease: -89.6102
-2
-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4

Max Phase Difference (36.03° + 180°)

2

-1
% Decrease: -28.1485
-2
-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4

Figure 31. Illustration of destructive interference. Blue curve is zero phase, green is
the phase shifted waveform, and red is the sum of the two.

Thus, in summary, when tidal height was at a maximum, antenna height was at a
minimum, which resulted in minimal phase difference and therefore maximum
destructive interference (as shown in Figure 32). Note as well the “inversion” of the RF
Loss vector, which was calculated as PRx-PTx, as opposed to preceding versions. This
provided the same information but presented in a manner more intuitive at this stage of
the analysis.

44
 Net RF Loss (Rx-Tx) - All Days

-43
-43.5
-44
Net RF Loss (dB)

-44.5

-45
-45.5

-46
-46.5 RF Loss (Filtered)
-47 Trendline
Detrended Loss
-47.5
08/29 08/30 08/31
Time (Days - CDT)

Phase difference (°)

36
Phase diff.
34
Angular Phase Difference (°)

32

30

28

26

24
08/29 08/30 08/31
Time (Days - CDT)

Figure 32. RF loss to phase difference time-series.

F. MODELED DATA COMPARISON

1. Introduction

The focus of most prior modeling and empirical propagation studies has been of
ranges greater than 200 meters. There exists an extensive body of literature on long-
range propagation at frequencies up to approximately 2 GHz. In 2004, Dr. W. M. Merrill
and his cohort extended this research to the RF ground wave by considering lower
antenna heights and shorter ranges for a variety of outdoor locations. This study
capitalized on his empirical and modeled results by first confirming his low elevation
data subset findings, and then broadening the scope to include the ranges associated with
the NEP-6 experiment.

2. Theory

The propagation of an RF signal is governed by Maxwell’s Equations. The

electric field near the surface is given by the solutions to (Wait, 1998)

45
 1
E=− ∇ × ∇ × Ur , (18)
kn 2

where k is wave number, n is index of refraction and Ur is the electric field.

Traditionally, the far-field (Fraunhofer) zone is simply described as that region of radius
r, in which r >> λ. For the operating frequency used, λ=16.85 cm, which places the
receiver well within the far-field. In this region, the solution to the above second order
partial differential equation is the following plane wave:

eikr
E = E0 , (19)
kr

where E0 is a constant, r is the direct path ray and k is wave number. In this
development, we shall only consider the case of a transmitter and receiver operating near
the surface of a perfect reflector (see Figure 33). Following the standard plane wave
combination, the electric field at the receiver is:

eikr eiks
E = E0 + R (θ ) E0 , (20)
kr ks

where r is the direct path and s is the reflected path. For grazing angles near zero (θ 0) ,
the reflection coefficient, R(θ) = –1, which is consistent with an ideal lossy surface.
Taking advantage of the symmetry afforded by Snell’s Law, the path length is determined
by the direct path ray and the respective heights of transmitter (h1) and receiver (h2):

s = r 2 + (h1 + h2 ) 2 . (21)

Henceforth we will refer to the total height, h=h1+h2. Near the surface, r is much greater
than h and thus Equation (21) can be represented as a binomial series expansion, here to
the fourth order (Stewart, 2008):

⎡ 1 ⎛ h ⎞2 1 ⎛ h ⎞4 ⎤
s ≈ r ⎢1 + ⎜ ⎟ − ⎜ ⎟ ⎥ , (22)
⎣⎢ 2 ⎝ r ⎠ 8 ⎝ r ⎠ ⎦⎥

or for the inverse, since that will be needed later:

46
 1 1 ⎡ 1 ⎛ h ⎞ 3⎛ h ⎞ ⎤
2 4

= ⎢1 − ⎜ ⎟ + ⎜ ⎟ ⎥. (23)
s r ⎢⎣ 2 ⎝ r ⎠ 8 ⎝ r ⎠ ⎥⎦

The expansion of the exponential is bit more complicated. From Equation (20), we
utilize a Taylor series expansion of the reflected signal s about the direct path r,

⎡ 1 ⎛ h ⎞2 1 ⎛ h ⎞4 ⎤
ik ⎢ ⎜ ⎟ − ⎜ ⎟ ⎥
⎢⎣ 2 ⎝ r ⎠ 8 ⎝ r ⎠ ⎥⎦
eiks = e e
ikr

⎡ ik ⎛ h ⎞ 2 ik (k − 1) ⎛ h ⎞ 4 ⎤ (24)
≈ e ⎢1 + ⎜ ⎟ +
ikr
⎜ ⎟ ⎥
⎢⎣ 2 ⎝ r ⎠ 8 ⎝ r ⎠ ⎥⎦

The effective field is the sum of the direct and reflected fields, as follows:

E = Edir + REref

eikr ⎡⎛ h ⎞2 ik − 1 ⎛ h ⎞4 ⎛ 3 6
⎞ ⎛ h ⎞ ik ⎤ (25)
≈ ⎢⎜ ⎟ + ⎜ ⎟ ⎜ + ik ( k − 3) +
⎟ ⎜ ⎟ (4 − k ) ⎥
kr ⎢⎣⎝ r ⎠ 2 ⎝ r ⎠ ⎝8 ⎠ ⎝ r ⎠ 16 ⎥⎦

Since power is simply the electric field times its conjugate (after neglecting 5th order
terms and higher), we find the power difference between the direct signal and the
reflected signal to be,

P = E × E*
1− k 2 ⎛ h ⎞ ,
4
(26)
= ⎜ ⎟
4k 2 r 2 ⎝ r ⎠

rather, a fourth-power drop off with height (Goroch, personal correspondence, 2010).

At frequencies from 900 MHz to 5.8 GHz, the dominant propagation effects can
be modeled well with ray-tracing analogies, a mindset well suited to the analysis of our
data. Due to the protected nature of the harbor in which this experiment was conducted,
contributions from surface roughness scattering or lateral waves at the vegetation
interface along the water’s edge were not considered substantial. As expected due to the
above derivation, and similar to Merrill’s (2004) findings, near a finite-conductivity

47
surface, a fourth-power propagation falloff with height above the scattering surface was
empirically observed, although unlike Merrill, this study utilized only vertical
polarization.

This falloff can also be explained pictorially by way of a ray-tracing analogy in

which destructive interference between the direct-path ray and surface bounce ray
resulted in varying degrees of received signal strength (See Figure 33).

Figure 33. Direct path ray and reflected path actual (in red) and virtual (in green) rays.

Merrill’s simple two-ray model, as presented in Equation (20), provided the basis
equations for modeling the expected RF Loss in our particular environment. However,
because this study did not observe significant surface roughness, nor was it measured
empirically, Equation (20) can be simplified to Equation (27). This generalization also
takes into account the expansion of Equation (20) for circumstances in which the ratio of

antenna height to path distance is very small, rather h 1 .

r

The following frequency/polarization-independent form of Equation (27), was

used in the modeling portion of this experiment (Merrill et al., 2004, p. 38):

(27)
where: PRx = Received power (dBm)
PTx = Transmitted power (dBm)
hr = height of receiver (m.)
ht = height of transmitter (m.)
r = ray path-length (m.)

48
Excepting a few scaling constants, Equation (27) is identical to the derived Equation (26).
They are both included to depict the linkage between the theory behind RF propagation
and practical application of said formula.

3. Model Summary

Merrill’s investigation of propagation loss in his Quantifying Short-Range

Surface-to-Surface Communications Links (2004) made use of multiple discrete antenna
heights and path lengths to build an empirical data set, yet his modeled data set varied
continuously. The NEP-6 experiment is notable in that we took advantage of the
continuously varying antenna height as a function of tides, thereby yielding a much larger
empirical data set. This novel approach provided a continuous change of the antenna
height which would be more difficult to accomplish over land. Modeled data were
calculated via Equation (27) and sub-plotted against actual net RF loss. This plot clearly
depicted the aforementioned height dependency in both versions of the net loss data set.
Furthermore, after properly scaling the modeled data, these two vectors were then
overlaid on the same axis (Figure 34).

Scaling was accomplished by first multiplying the modeled loss by the mean of
the actual RF loss vector, and then dividing the mean of the modeled RF loss vector.
This provided the fourth power drop-off in both empirical and modeled data in the
overlay plot.

49
 Compiled Antenna Height to RF Loss
-43.5

-44

-44.5
RF Loss (dB)

-45

-45.5
Actual Loss
(Scaled) Modeled Loss
-46

-46.5
1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
Antennae Height above Water (m)

Figure 34. Empirical and modeled (scaled) RF loss (See Equation 27)
to antenna height.

From theory, we have mathematically shown that the net propagation loss of the
1.78 GHz wave depends on the height of the antenna above the water surface. As
illustrated in Figure 34, the near-surface empirical and modeled data exhibit the very
same behavior of waveforms from much higher up in the atmosphere, which is correctly
described by classic wave propagation equations. Additionally, the calculated RSSI of
the modeled waveform portrays this same tendency, particularly with respect to tidal
variation and path length difference (Figure 35).

50
 In summary, the overall result is a definite dependency of propagation loss to
antenna height in both the real-world data, as well as the modeled data.

Received Signal Strength

-42

-42.5
RSSI (dBm)

-43

-43.5

-44
08/29 08/30 08/31
Time (Days - CDT)
NRL Tower, Tidal Variation 29Aug - 31Aug
MSL Tidal Variation (Meters)

1.4
Tidal Data
1.2 Tidal Trendline
1

0.8

08/29 08/30 08/31

Time (Days - CDT)
Phase difference (°)
Angular Phase Difference (°)

40
Phase diff.
35

30

25

20
08/29 08/30 08/31
Time (Days - CDT)

Figure 35. Modeled RSSI characteristics.

51
THIS PAGE INTENTIONALLY LEFT BLANK

52
 V. CONCLUSION

A. SUMMARY OF EXPERIMENT

In this study, a 1.78 GHz waveform was emitted at a very low altitude over a
protected body of water and the received signal power was measured. Additionally,
various meteorological and oceanographic variables were sensed and preserved in a self-
contained wireless data logging network. These data were then used to first calculate net
propagation loss of the RF wave, and then in subsequent iterations, to correlate potential
cause of the loss to the measured environmental parameters.

B. KEY RESULTS AND CONCLUSIONS

While a certain number of the collected environmental variables did play a small
role in the resultant cyclic behavior of the net loss vector (e.g., solar irradiance), the vast
majority (i.e., temperature, humidity, and wind), were not correlated to RF propagation
loss at a statistically valid level of significance. Tidal variation was the predominant
environmental parameter to have an effect, and yet even this was merely the underlying
cause for antenna height variation. In summary, the overall result of this experiment was
that in the given near-surface environment, meteorological variance did not seem to have
a concrete impact on propagation loss. Additionally, we found that antenna height played
a direct role in propagation loss; a result currently accepted in EM/EO waveform
propagation theory, and one that we discovered to still be pertinent in the near surface
environment for the test range used. Finally, these results were confirmed both
mathematically and via modeling using appropriate propagation loss calculations.

C. APPLICABILITY TO DOD OPERATIONS

It is certainly accepted that the exploitation of unmanned and autonomous

vehicles, regardless of the medium in which they are used, can only increase in future
DoD operations. However, with their application come certain limitations and
constraints, most notably the need for reliable communication. While this experiment

53
explored a very narrow subset of potential operating conditions, it still provides valuable
insight as to the expected behavior of these systems.

Many U.S. forces utilize RF and optical propagation programs, such as AREPS
and TAWS, to predict waveform behavior in a given mission environment. These
programs are excellent provided the user does not require predicted waveform behavior
near the surface or transmitting antennae, where the model quickly falls apart. With the
inclusion of these remarkably simple yet pertinent results, the area of coverage of the
above propagation models could be extended to include the near-surface. This is critical
to today’s militaries that operate complex systems using RF energy in this near Earth
environment.

Maritime security, particularly that portion related to harbor security, is a rapidly

growing field that has spawned numerous commercial and governmental agencies
dedicated to defeating the threats pervasive in this complex environment. The use of
unmanned surface and undersea vehicles could revolutionize this industry by reducing
the workload as well as removing personnel from dangerous areas. Additional avenues
might also include simple security networks comprised of video, audio and other
pertinent environmentally sensed values (air temperature, humidity, wind speed, etc.) that
would be transmitted via EM energy to a central repository. However, to make this
realizable, communication and control must be assured. From this research, we find that
much of this problem could likely be solved by sufficient power output and placement of
the transmitting antenna.

Finally, this question of command and control naturally extends into overland and
aerial vehicles, with much headway being made in the mid- and long-range arenas.
However, by their very nature, these vehicles operate at much higher altitudes where
sufficient antenna height is assured. For very short range, or very low altitude/cluttered
environment scenarios, the results of this experiment can certainly be applied.

D. AREAS FOR FURTHER RESEARCH

Classic theory dictates that RF wave propagation above the surface layer is
dependent on antenna height. While this study has clearly shown that this characteristic
54
extends into the very near surface environment, it does so only at a given frequency and
path distance. Merrill et al. produced similar results over a variety of path lengths,
frequencies and environments, but with much fewer data points per trial. In order to
make valid generalizations of this RF propagation phenomenon, it is necessary to
duplicate the above experiment but over a much longer time frame, with a wider range of
environmental conditions, and over a wider set of operating frequencies and antenna
heights.

Additional analysis of the current data set would also be beneficial as this study
only reviewed the 1.78 GHz frequency set. It would be interesting to see if the same
results were obtained from the 225 MHz data and if so, the magnitude of difference
between the two. Dr. Christopher Anderson of the United States Naval Academy
conducted a similar propagation experiment concurrent with NEP-6 that was driven by
different goals. Nonetheless, the data that he collected could be useful when compared to
the current data set.

All of the above recommendations lead to ultimately establishing a generalization

of RF propagation in the near surface environment. Once this has been completed, it then
becomes necessary to include these results in both current (i.e., AREPS) and future wave
propagation models, thereby improving our ability to predict RF wave behavior and thus
expand the reliability and security of remotely controlled unmanned vehicles, and
wireless security networks.

55
THIS PAGE INTENTIONALLY LEFT BLANK

56
 APPENDIX

A. SUPPLEMENTARY BACKGROUND INFORMATION

1. Atmospheric Boundary Layer

The atmospheric boundary layer (ABL) is the part of the troposphere that is
directly influenced by the presence of the Earth’s surface and responds to surface forcing
with a time scale of about one hour or less. Over the oceans, the boundary layer depth
typically varies slowly in space and time. Under certain conditions, such as on a clear
calm day, diurnal variations in sea surface temperature may be observed because of the
solar insolation and little wind-generated turbulent mixing in the upper ocean. However,
such diurnal variation is much weaker in comparison to that in the atmospheric boundary
layer. Thus, temporal variation in surface thermal forcing is predominantly a result of
atmospheric temperature variability. Other sources of surface forcing in the atmospheric
boundary layer include frictional drag, evaporation and transpiration, heat transfer,
pollutant emission (in case of scalar concentrations), and terrain (or surface water wave)
induced flow modifications. These sources of forcing both directly and indirectly affect
EM wave propagation via changes in the index of refraction and surface reflections due
to turbulence (surface roughness) and turbulent fluxes (Stull, 1988).

As shown by Stull (1988), there are multiple layers to the ABL, but for the
purposes of this experiment, we will be focusing on the surface layer. This layer is
comprised of the lowermost 10% of the ABL in which the log wind profile is valid.
Vertical variations of the mean wind and temperature (i.e., mean profiles), are determined
by the structure of the roughness elements (in this case, waves, sea sprays, and swells).

2. Turbulence and Flux

Turbulence, which is the mechanism for mixing, is not a feature of fluids but of
fluid flows. The governing equations for turbulent flow is the same in all fluids, whether
they are liquids or gases and if the Reynolds number (the ratio of inertial force to viscous
force) is large enough, the major characteristics of turbulent flows are not controlled by
the molecular properties of the fluid in which the turbulence occurs. Turbulence is one of
57
the principal unsolved problems in physics today due to its nonlinear and random nature.
Since the Navier-Stokes equations of motion are nonlinear and a deterministic solution
has yet to be found, each individual flow pattern has certain unique characteristics
associated with its initial and boundary conditions. Thus, while frustrating and currently
incomplete, the study of turbulence has led to the following basic properties: irregular,
diffusive, three-dimensional vorticity fluctuations, and dissipative (Stull, 1988).

If turbulence is a feature of a mixing flow, eddies are the vehicle by which it

occurs. The largest are called energy containing eddies and are typically the depth of the
boundary layer in an unstable atmosphere. In a stable atmosphere, energy-containing
eddies are defined by the vertical distance of the ABL from the surface. These large
eddies deform and stretch, thereby creating smaller eddies, and so on to the smallest
turbulent eddies characterized by the Kolmogorov microscale, η (See Equation 28).
Eddies smaller than η are called dissipation eddies, where turbulent kinetic energy is
dissipated via molecular viscosity.
1
⎛ν 3 ⎞ 4
η =⎜ ⎟ (28)
⎝ε ⎠
where η: size of the smallest eddy
ν: viscosity (m2/s)
ε: dissipation rate (m2/s3)
Flux, in all its various forms, is particularly important when discussing the effects
of turbulence on its environment. There are many methods for diagnosing flux, but all
make use of Taylor’s Hypothesis (that turbulent eddies are frozen in time) and most are
dependent on the stability of the atmosphere. In neutral stability conditions, the flux
profile takes the following gradient relationship:

∂u u*
= (29)
∂z κ z
∂u
where ∂z : shear
u* : frictional velocity
κ : von Karman’s constant
z : height or level
58
 ∂u
This relationship shows that the vertical shear of the mean wind, , is inversely
∂z
proportional to height, z. In non-neutral conditions, the effects of a stability parameter
must be considered as shown in the following:

κ z ∂u z
=φm ( ) (30)
u* ∂z L

z
where φ m ( ) : momentum stability parameter (q, h are similar), although in the near
L
surface, stability is typically not a factor as (z/L) is rapidly approaching zero.

This study will be utilizing the log wind profile for a neutral surface layer (no
stability effects), with an important modification. It is possible to obtain frictional
velocity, u* directly from measurements, but this requires an accurate value for roughness
height, z0 (the base of the surface layer). Typically, this value is calculated via
Charnock’s Relationship over the water. Alternatively, in the surface layer, momentum
flux can be obtained by taking measurements of mean wind (likewise for moisture, q* ,
and heat, h*) at two levels and make use of the following integrated flux profile
relationship for neutral conditions.

u ( z2 ) − u ( z1 )
u* = κ (31)
⎛z ⎞
ln ⎜ 2 ⎟
⎝ z1 ⎠

where z1 and z2 are the respective measurement levels.

59
B. SCATTER PLOTS OF RESIDUALS

Figure 36. Scatter plot of tidal height to net RF loss (dB) in blue and linear trend-line
(gray). All subsequent plot plots utilize the residuals from this calculation.

Figure 37. Scatter plot of calculated residuals to water surface skin temperature (˚C).
The top plot is from sensor #KT18.85-I, the bottom from sensor #KT18.85-II.
60
Figure 38. Scatter plot of calculated residuals to sampled atmospheric pressure (mbar).

Figure 39. Scatter plot of residuals to air temperature (˚C) at each of the six levels.

61
 Figure 40. Scatter plot of residuals to solar irradiance (W m-2).

Figure 41. Scatter plot of calculated residuals to submerged water temperature (˚C) at
each of the three levels, where #1 was 4.0 cm, #2 was 8.0 cm, and #3 was 18.0 cm
below the surface.
62
 Figure 42. Scatter plot of calculated residuals to relative humidity (%) at each of the
six measured levels. #1 was the lowest (11.5 cm), up to #6 which was the highest
(247.0 cm) above the water surface.

Figure 43. Scatter plot of calculated, normalized residuals to raw wind speed (top) and
the vector mean of wind speed (bottom). Both plots are in units of m s-1.
63
 Figure 44. Scatter plot of normalized residuals to the vector mean of wind direction
(˚True)

Figure 45. Scatter plot of calculated, normalized residuals to specific humidity (g g-1),
at each of the six levels, and referenced to standard atmosphere (1013.25 mbar)
64
Figure 46. Scatter plot of normalized residuals to calculated modified refractivity (M-
units), at each of the six observed levels.

Figure 47. Scatter plot of calculated, normalized residuals to the gradient of modified
refractivity (M-units m-1)
65
THIS PAGE INTENTIONALLY LEFT BLANK

66
 LIST OF REFERENCES

Bean, B. R., and E. J. Dutton, 1968: Radio Meteorology, Dover Publications, Inc., New
York, 435 pp.

Campbell Scientific, SR50A Sonic Ranging Sensor Data Sheet, 2009.

Cohen, J., 1988: Statistical Power Analysis for the Behavioral Sciences, Lawrence
Erlbaum Associates, New Jersey, 567 pp.

SPAWAR Systems Center, 2007: Refractivity Classes. Space and Naval Warfare Systems
Center, Atmospheric Propagation Branch, 2858, San Diego, CA. pp. various.

Fabry, F., Frush, C., Zawadzki, I., and Kilambi, A., 1997: On the extraction of near-
surface index of refraction using radar phase measurements from ground targets.
Journal of Atmospheric and Oceanic Technology, 14, 4, 978–987.

Federal Standard 1037C. Telecommunications: Glossary of Telecommunication Terms,

August 1996. [Available at: http://www.its.bldrdoc.gov/fs-1037/] Accessed
September 2009.

Helvey, R. A., and J. S. Rosenthal, 1983: “Guide for Inferring Refractive Conditions
from Synoptic Parameters,” Technical Report, Pacific Missile Test Center, 36 pp.

Guildline Instruments Limited, AutoSal Data Sheet, 2002. [Available at:

http://www.bodc.ac.uk/data/documents/nodb/pdf/Buildline8400BDatasheet.pdf]
Accessed January 2010.

Kidder, S. Q., and S. H. Vonder Haar, 1995: Satellite Meteorology: An Introduction,

Academic Press, New York, 466 pp.

Medwin, H., 2005: Sounds in the Sea: From Ocean Acoustics to Acoustical
Oceanography, Cambridge University Press, New York, 643 pp.

Merrill, W., Liu, H., Leong, J., Sohrabi, K., and Pottie, G., “Quantifying Short-Range
Surface-to-Surface Communications Links,” IEEE Antennas and Propagation
Magazine, 46, 3, June 2004, pp. 36–46.

Rinehart, R. E., 1997: Radar for Meteorologists, 3rd Edition, Rinehart Publications,
Grand Forks, 418 pp.

Stewart, J. 2008: Multivariable Calculus: Early Transcendentals, Brooks/Cole

Publishing Company, New York, 1185 pp.

Stull, R., 1988: An Introduction to Boundary Layer Meteorology, Kluwer Academic

Publishers, Dordrecht, Netherlands, 667 pp.
67
Thayer, G. D., 1974: “An Improved Equation for the Radio Refractive Index of Air.”
Radio Science, Vol. 9, No. 10, pp. 803–807.

Wait, J. R., 1998: “The Ancient and Modern History of EM Ground-wave Propagation,”
IEEE Antennas and Propagation Magazine, 40, pp. 7–24.

Wilcox, R. R., 2005: Introduction to Robust Estimation and Hypothesis Testing,

Academic Press, New York, 588 pp.

Wong, V., “Overland PC Network Layout,” Naval Research Laboratories, unpublished

materials, Washington DC (2009)

68
 INITIAL DISTRIBUTION LIST

1. Defense Technical Information Center

Ft. Belvoir, Virginia

2. Dudley Knox Library

Naval Postgraduate School
Monterey, California

3. Dr. Peter Guest

Naval Postgraduate School
Monterey, California

4. Dr. Andreas Goroch

Marine Meteorology Division
Naval Research Laboratory
Monterey, California

5. Mr. Robert Wert

Tactical Electronics Warfare Division
Naval Research Laboratory
Washington, DC

69