ABSTRACT 
 
 
 
 
Title of Document: THIN-FILM PHOTOVOLTAICS UNDER 
EXTREME WIND LOADING DUE TO 
DOWNBURSTS IN THE WASHINGTON D.C. 
AREA 
  
  
 Kristen Sara Markham, M.S., 2010  
  
Directed By: Associate Professor, Yunfeng Zhang, 
Department of Civil and Environmental 
Engineering 
 
 
Extreme wind loading on buildings can be caused by a variety of different weather 
phenomenon, including straight-line wind-inducing events known as downbursts. 
With maximum wind gusts up to 168 mph, downbursts have the potential to cause 
significant damage to modern infrastructure, comparable to that of the more 
commonly-known tornado or hurricane. Among the many variables that affect the 
extent of damage to infrastructure from such events, the performance of a building is 
largely dependent on two factors ? (a) the magnitude of the loads induced on a 
building, and (b) the strength of the building components resisting these loads. The 
goal of this research is to characterize the downburst-induced horizontal wind loads 
on a building fa?ade of a given region, as well as the strength and behavior of a green 
building material used in the fa?ade of buildings ? known as thin-film building 
integrated photovoltaics (BIPVs).  With downburst data collected from the 
Washington D.C.-Baltimore metropolitan area (WBMA), a failure probability model 
is derived for BIPVs specific to this region. 
  
 
 
 
 
 
 
 
 
 
 
 
 
 
THIN-FILM PHOTOVOLTAIC PANELS UNDER EXTREME WIND LOADING 
DUE TO DOWNBURSTS IN THE WASHINGTON D.C. AREA 
 
 
 
 
By 
 
 
Kristen Sara Markham 
 
 
 
 
 
Thesis submitted to the Faculty of the Graduate School of the 
University of Maryland, College Park, in partial fulfillment 
of the requirements for the degree of 
Masters of Science  
2010 
 
 
 
 
 
 
Advisory Committee: 
Associate Professor Yunfeng Zhang, Chair 
Professor Bilal Ayyub 
Senior Research Professor Ed Link 
 
 
  
 
 
 
 
 
 
? Copyright by 
Kristen Sara Markham 
2010 
 
ii 
 
 
ACKNOWLEDGEMENTS 
I would like to first gratefully acknowledge Dr. Yunfeng Zhang, my advisor, for his 
guidance, support, advice and extensive patience during my graduate career at the 
University of Maryland. I thank him for showing me and helping me learn how to 
conduct research, write technical papers, give presentations, and for the opportunities 
he provided to me while working together. I would also like to thank Dr. Deborah 
Goodings, Alan Santos, and Dr. Dimitrios Goulias for encouraging me to attend 
graduate school, and making me realize the importance of an engineering graduate 
degree. Without your encouragement, I would not be where I am today.   
I also thank Dr. Dimitrious Goulias for allowing me the opportunity to do what I love, 
teach and work with students, while completing my masters degree. I thank Maggie 
Sharkey and Ehsan Aramoon, my fellow TAs, for their help and support, and for 
working with me over the years. To the organization Engineers Without Borders, I 
thank you for giving my engineering career meaning and motivation for the past 
nearly 6 years, and for the opportunity to lead the organization as it grows and 
expands. Both teaching and EWB have been instrumental in shaping my engineering 
career. I thank Dr. William Fourney and Kevin Calabro for encouraging me to do 
what makes me happy. 
I also deeply appreciate Dr. Bilal Ayyub and Dr. Ed Link for serving on the 
examination committee at the defense of my thesis. My appreciation also extends to 
the Civil Engineering department administration staff; without their support, none of 
this would have been possible.  
My love and appreciation extends to my wonderful family ? my mom, dad and sister 
who have provided motivation, support and encouragement when I needed it most.  
Also I would like to express my thanks to my wonderful friends, roommates, and 
boyfriend for their moral support, understanding, and encouragement.   
 
iii 
 
 
TABLE OF CONTENTS 
Acknowledgements ....................................................................................................... ii 
Table of Contents ......................................................................................................... iii 
List of Tables ............................................................................................................... iii 
List of Figures .............................................................................................................. iv 
Executive Summary ...................................................................................................... 1 
Chapter 1: Introduction and Motivation ...................................................................... 3 
Chapter 2: Downbursts ................................................................................................ 8 
2.1 Definition and Origin of Downburst Concept ............................................ 9 
2.2 Downburst Formation ............................................................................... 10 
2.3 Microbursts and Macrobursts ................................................................... 10 
2.4 Wet and Dry Downbursts.......................................................................... 11 
2.5 Washington D.C.-Baltimore Metropolitan Area (WBMA) Downbursts .. 14 
2.5.1 WBMA Data Coverage Limits ....................................................... 14 
2.5.2 The Convective Season of the WBMA .......................................... 15 
2.6 MethodologY of Downburst Data Collection and Analysis ..................... 15 
2.6.1 A Review of Downburst Studies Conducted in the United States . 16 
2.6.1.1 Long-term, Sensor Network Studies .......................................... 16 
2.6.1.2 Retroactive Surface Data Studies............................................... 18 
2.6.1.3 Computer Modeling Based Studies ........................................... 20 
2.7 WBMA Downburst Data Collection Methodology .................................. 20 
2.7.1 Datasets Used for Downburst Information Collection ................... 20 
2.7.1.1 Storm Event Database ? Data Collection Methodology: ........... 21 
2.7.1.2 News Search via Lexis Nexis ? Data Collection Methodology: 27 
2.7.1.3 Accuracy of Data Sources .......................................................... 27 
2.7.1.4 Methodology Comparison to ASCE 7-05 .................................. 28 
Chapter 3: Downburst Data and Distribution ............................................................ 36 
3.1 Summary of Data: ..................................................................................... 36 
3.2 Elimination of Extraneous Data Entries: .................................................. 36 
3.3 Method of Verification ............................................................................. 43 
3.4 Downbursts Categorized by Wind Speed ................................................. 43 
3.5 Frequency of Occurrence of Downbursts ................................................. 44 
3.5.1 Downburst Wind Distribution Functions ....................................... 44 
3.5.1.1 Extreme Value Distributions...................................................... 44 
3.5.2 Determination of Parameters to Fit Data to Distribution ............... 46 
3.5.3 Goodness-of-Fit .............................................................................. 46 
3.5.4 Final Downburst Distribution ......................................................... 49 
3.6 Downburst Design Wind Speed Model .................................................... 50 
3.6.1 Assumptions: .................................................................................. 50 
3.6.2 Downburst Model: .......................................................................... 50 
 
iv 
 
 
3.6.3 Discussion on Strike Factor: ........................................................... 51 
3.6.3.1 Method 1: Point Probability ....................................................... 51 
3.6.3.2 Method 2: Urban Area Ratio...................................................... 54 
3.6.3.3 Method 3: Conservative Assumption......................................... 54 
3.7 50-year and 100-year Downburst Design Wind Speed ............................. 55 
3.7.1 Conversion to Wind Pressure ......................................................... 59 
Chapter 4: Building Integrated Photovoltaics (BIPV) Under Uniform Loading ...... 64 
4.1 The Origin of Photovoltaics ...................................................................... 65 
4.2 Development of Thin-Film Photovoltaic (TFPV) Technologies .............. 66 
4.3 BIPV: Combining TFPVs and Architecture ............................................. 67 
4.3.1 Placement of BIPVs on Buildings: ................................................. 68 
4.4 Types of TFPV Panels used in BIPV applications and their Material 
Properties ............................................................................................................. 69 
4.4.1 Amorphous Silicon (a-Si) (and Micromorphous Silicon) .............. 70 
4.4.2 Copper-Indium-(Gallium)-Diselenide (CI(G)S) ............................ 72 
Origin: 72 
Efficiency: ................................................................................................. 72 
Structure and Material Properties: ............................................................ 72 
4.4.3 Cadmium Telleride (CdTe) ............................................................ 73 
Origin: 73 
Efficiency: ................................................................................................. 73 
Structure and Material Properties: ............................................................ 74 
4.5 Material Layers within PV Panels: ........................................................... 75 
4.5.1 Glass ............................................................................................... 75 
4.5.1.1 Annealed/Float Glass ................................................................. 76 
4.5.1.2 Heat Strengthened (HS) Glass ................................................... 77 
4.5.1.3 Fully Tempered/Toughened (FT) Glass..................................... 77 
4.5.2 Laminated glass (LG) ..................................................................... 79 
4.5.3 Laminated glass model for TFPV panels ....................................... 79 
4.6 TFPV Panel Manufactuers Data ............................................................... 80 
Assumed Glass Thickness in TFPV Panels ................................................... 78 
4.6.1 Assumed Glass Type in TFPV Panels ............................................ 78 
4.6.2 Assumed Geometric Properties ...................................................... 78 
4.7 Laminated glass Failure MOdel ................................................................ 79 
4.7.1 Glass Failure Prediction Model (GFPM) ....................................... 80 
4.7.2 Effect of Plate Geometry on Glass Strength .................................. 82 
4.7.3 Glass Strength Dependence on Load Duration .............................. 83 
4.7.4 Strength of Weathered Glass .......................................................... 84 
4.7.5 Failure Probability of Annealed Laminated Glass ......................... 84 
4.7.5.1 Conversion of monolithic glass to laminated glass strength ...... 84 
4.7.5.2 BIPV Panel Sizes Chosen for use in GFPM .............................. 85 
4.7.5.3 Probability of Failure Calculations ............................................ 86 
4.7.5.4 Model Verification ..................................................................... 86 
 
v 
 
 
Chapter 5: Downburst Design Pressure and TFPV Load Resistance ....................... 98 
5.1 ASCE 7-05 and DOWNBURST DESIGN WIND LOADS ..................... 98 
5.2 European IEC 61646 Mechanical Load Requirements and GFPM 
predicted load resistance ................................................................................... 101 
5.3 Downburst design wind loads and GFPM predicted load resistance ...... 103 
Conclusions and future work .................................................................................... 105 
Appendix A ............................................................................................................... 107 
A.1  Extreme Value Distribution Types ......................................................... 107 
A.1.1 Type I ?Extreme Value Distribution ............................................ 107 
A.1.2 Type III ?Extreme Value Distribution ......................................... 107 
A.2 Distribution Parameter Estimation Methods ........................................... 108 
A.2.1 Method of Moments ..................................................................... 108 
A.2.2 Maximum Likelihood Estimation ................................................ 109 
A.3 Goodness-of-Fit Tests: ............................................................................ 110 
A.3.1 Anderson-Darling (A-D) Test: ..................................................... 110 
A.3.2 Kolmogorov-Smirnov (K-S) Test ................................................ 112 
A.4 Poisson Pulse Process and Upcross Rate: ............................................... 113 
References   ............................................................................................................... 117 
 
ii 
 
 
 
iii 
 
 
LIST OF TABLES 
Table 2.1: Dry and Wet Downburst Characteristics (adapted from Rose 2010) ........ 12 
Table 2.2: The Beaufort Scale - Wind Speed Estimates (adapted from SPC 2010) ... 26 
Table 3.1: Downbursts in the WBMA (in order by date) 1996-2009 ......................... 38 
Table 3.2: Mean and Standard Deviation of Downburst Wind Speeds ...................... 48 
Table 3.3: Type I Extreme Value Distribution Parameters by Method of Moments .. 48 
Table 3.4: Type I Extreme Value Distribution Parameters by Maximum Likelihood 
Estimation ....................................................................................................... 48 
Table 3.5: Type III Extreme Value Distribution Parameters by Method of Moments 48 
Table 3.6: Type III Extreme Value Distribution Parameters by Maximum Likelihood 
Estimation ....................................................................................................... 48 
Table 3.7: Parameter Values for Determining Downburst Design Wind Speeds ....... 58 
Table 3.8: 50- and 100-year Downburst Design Wind Speed (mph) at 33 ft ............. 58 
Table 3.9: Assumed ASCE 7-05 Velocity Pressure Variables ................................... 60 
Table 3.10: Design Velocity Pressure (lb/ft2) at 33 ft by Strike Factor ...................... 60 
Table 4.1: TFPV panel producers in the world market (ENF 2010) .......................... 70 
Table 4.2: Properties of Monolithic Annealed, Heat Strengthened and ..................... 78 
Table 4.2: Parameter values used in determining probability of failure using GFPM 88 
Table 4.3: Maximum loads resisted by laminated glass at probability of failures of 
0.008 and 0.5 for each set of dimensions considered using the GFPM .......... 88 
Table 4.4: Mean failure strengths predicted by GFPM model compared to the 
equivalent 3-second load strength from laminated glass test data .................. 88 
Table A.1: Critical Values for Anderson-Darling Test (D'Agostino and Stephens 
1986) ............................................................................................................. 111 
 
 
iv 
 
 
LIST OF FIGURES 
Figure 1.1: BIPV Applications on a Typical Building (adapted from Shaybani, 2010) ..... 6 
Figure 1.2: Fa?ade Failure Due to Hurricane Alicia in 1983, Houston, TX (USACE 2010)
................................................................................................................................. 7 
Figure 2.1: First Downburst Conceptual Model (Fujita 1985) .......................................... 30 
Figure 2.2: Bow Echo Formation and Downburst Locations (Fujita 1978) ...................... 30 
Figure 2.3: Dry and Wet Microburst Diagrams (adapted from Caracena and Flueck 1988)
............................................................................................................................... 30 
Figure 2.4:  Andrews Airforce Base Downburst - August 1, 1983 (Fujita 1985) ............. 31 
Figure 2.5: Doppler Radar Coverage for the Baltimore/Washington NWS (NWS 2010) 31 
Figure 2.6: Coverage Area of Baltimore/Washington NWS by County (NWS 2010) ...... 32 
Figure 2.7: Location of Extensive Downburst Studies in the United States (Wakimoto 
2002) ..................................................................................................................... 32 
Figure 2.8: Downburst Occurrences by State 2001-2003 (Pryor 2009) ............................ 33 
Figure 2.9: GOES Microburst Product overlaid with Radar data (Pryor 2009) ................ 33 
Figure 2.10: 50-year Recurrence Interval for Microburst Peak Wind Speeds (mph) 
(Walter 2007) ........................................................................................................ 34 
Figure 2.11: Baltimore/Washington Storm Event Report - Early Format ......................... 34 
Figure 2.12: Baltimore/Washington Storm Event Report ? Recent Format ...................... 34 
Figure 2.13: National Storm Event Report (same date as Figure 2.11) ............................. 35 
Figure 3.1: Downbursts in the WBMA, 1996-2009 .......................................................... 61 
Figure 3.2: A time history of the July 2nd 2006 Annandale, VA downburst 10 miles from 
the downburst location .......................................................................................... 61 
Figure 3.3: Downbursts by Wind Speed (mph) in WBMA from 1996-2009 (1 mph = 0.45 
m/s) ....................................................................................................................... 62 
Figure 3.4: Type I and III PDFs for Downburst Wind Speeds in the WBMA using method 
of moments (MOM) and maximum likelihood estimation (MLE) ....................... 63 
Figure 3.5: Type I and III EVD CDFs fitted to WBMA Downburst Data from 1996-2009
............................................................................................................................... 63 
 
v 
 
 
Figure 4.1: National Renewable Energy Laboratory (NREL) maximum efficiencies of 
PVs by year ........................................................................................................... 89 
Figure 4.2: Examples of fa?ade BIPVs in (a) curtain wall and (b) rain screen applications 
(Sayigh 2000) ........................................................................................................ 89 
Figure 4.3: a-Si and micromorphous Si typical layers (Kaltschmitt et al. 2007) .............. 90 
Figure 4.4:  (a) AstroEnergy a-Si/?c-Si micromorphous solar panel construction 
(AstroEnergy 2010) .............................................................................................. 90 
Figure 4.5 (a) Schott ASI Thru PV Module, (b) EPV Solar EPV-4X Solar Module (Schott 
2010) ..................................................................................................................... 91 
Figure 4.6: Typical layers of a CI(G)S solar cell (AGC 2010) ......................................... 91 
Figure 4.7: (a) Avancis Powermax CIS module, (b) Nanosolar Utility CIGS Panels 
(Avancis 2010; Nanosolar 2010) .......................................................................... 92 
Figure 4.8: Typical layers of a CdTe solar cell ................................................................. 92 
Figure 4.9: (a) Q-cell CIGS Q.Smart 70-90 PV panel, (b) First Solar CdTe (Q-Cell 2010; 
First Solar 2010). .................................................................................................. 93 
Figure 4.10: Behavior of glass under loading.................................................................... 93 
Figure 4.11: Chart used for determining the stress distribution J from the aspect ratio and 
non-dimensional lateral load (Beason and Morgan 1984) .................................... 94 
Figure 4.12: Fragility Curve (Pf = 0.008) of laminated glass using the GFPM ................. 95 
Figure 4.13: Fragility Curve (Pf = 0.5) of laminated glass using the GFPM ..................... 96 
Figure 4.14: ASTM E1300 design chart for 6 mm laminated glass (ASTM 2009) .......... 97 
Figure 4.15: ASTM E1300 design chart for 6 mm annealed glass (ASTM 2009) ............ 97 
Figure 5.1: Downburst mean velocity profile: comparison with laboratory experiments, 
empirical models and with a typical boundary layer profile (Kim and Hangan 
2007) ................................................................................................................... 101 
Figure A.1: Poisson Pulse Process Diagram Modified to Include a(t) ? Generalized 
Threshold Value as a Function of Time .............................................................. 114 
 
 
1 
 
 
EXECUTIVE SUMMARY 
Extreme wind loading on buildings can be caused by a variety of different weather phenomenon, 
including straight-line wind inducing events known as downbursts. With maximum wind gusts 
up to 168 mph, downbursts have the potential to cause significant damage to modern 
infrastructure, comparable to that of the more commonly-known tornado or hurricane. Among 
the many variables that affect the extent of damage to infrastructure from such events, the 
performance of a building is largely dependent on two factors ? (a) the magnitude of the loads 
induced on a building, and (b) the strength of the building components resisting these loads.  
The first goal of this research is to define a region-specific probabilistic load model of downburst 
winds for Washington D.C.-Baltimore metropolitan area (WBMA). Currently the occurrence of 
downbursts is not systematically documented in the United States due to relatively short history 
of data recording and lack of a sufficiently dense sensor network, making it difficult to 
characterize their behavior to account for them in design wind loads of buildings. The model 
presented, based on data collected from several sources and a modified downburst wind load 
model developed by others, represents the first effort in quantifying the design wind speed for 
downbursts in the WBMA region. 
The second goal of this research is to characterize the behavior of building integrated photovoltic 
(BIPV) panels used in building fa?ades when subjected to wind loading.  Infrastructure today 
continues to progress towards more ?green? construction techniques, promoting the use of 
energy-saving materials that are integrated into the many systems that make up a functioning 
building. One method that has quickly gained attention in recent years is that of BIPVs, in which 
glass-protected thin-film photovoltaic panels are used in the fa?ade and roof of buildings. These 
 
2 
 
 
panels simultaneously act as both a generator of clean energy from the sun?s light, as well as the 
protective building envelope of the structure.  A variety of dimensions, thicknesses of glass, and 
layering techniques are used by manufacturers of BIPVs, thus a survey of manufacturers? data of 
thin-film BIPV models was first conducted to determine the range of properties of these panels 
on the market today. Then, by assuming that the thin-film layer of the BIPV panels can be 
neglected due to its negligible thickness, high stiffness and material strength associated with 
metal oxides, the BIPV panels behavior are modeled using a failure model for similarly 
constructed laminated glass under uniform loading.  
Ultimately the design wind loads determined from the probabilistic downbursts wind model are 
compared with the probabilistic strength distribution of the BIPVs used in building facades. 
Based on these, a failure probability model is derived for BIPVs specific to the Washington 
D.C.-Baltimore metropolitan region.  
New building designs that integrate more costly green technologies should be sustainable and 
resource-efficient, but also still be safe and reliable.  Currently the United States lacks a set of 
standard performance criteria which BIPV panels must pass in order to be widely accepted for 
use in modern infrastructure. Understanding the general behavior and reliability of BIPVs under 
extreme wind loading using this model can help provide guidance as these standards are being 
developed.  
 
 
 
  
 
3 
 
 
CHAPTER 1: INTRODUCTION AND MOTIVATION 
Today, buildings in the United States (U.S.) account for approximately 39% of total energy use 
and 72% of electricity use according to the U.S. Department of Energy (EPA 2009). As energy 
demands in the U.S. continue to remain high, a growing need for more environmentally-
responsible, ?green? design and construction is increasingly apparent.  These new building 
designs should be safe and reliable, and also sustainable and resource-efficient.  In 2007, 26 
states had passed some form of green building legislation, most mandating high-efficiency 
standards for government buildings and encouraging the private building sector through tax 
incentives (King and King 2004; Nitkin 2007).  The State of California became the first state to 
adopt a mandatory, state-wide green building code in January of 2010.   
One of the emerging methods of decreasing energy consumption is through the use of a more 
energy-efficient building envelope, the outer structure of a building, including the facade, 
windows and doors, and roof (Chuwieduk 2003; Arnold 2009). In recent years a new class of 
renewable energy technologies known as BIPVs (Building Integrated Photovoltaics) have been 
developed which function simultaneously as the building envelope and as photovoltaic (PV) 
power generators (Dougherty et al. 2005) (Figure 1.1). For tall buildings in particular with large 
surface areas exposed to sunlight, the potential amount of energy generated from the use of this 
technology in building facades is significant. As an example, one of the largest fa?ade BIPV 
applications is in Ulm, Germany stands 355 ft (102 m) tall and produces 70 megawatt hours each 
year.  
Today many different BIPV technologies are available on the market.  While mono and 
polycrystalline silicon (c-Si) PV technologies currently hold 80-90% of the total market share of 
building-installed PVs in the U.S., thin-film PVs (TFPVs) have several distinct advantages over 
 
4 
 
 
c-Si PVs, and are favored over c-Si for future BIPV applications. For PVs of similar efficiency ? 
approximately a 100 times thinner layer of PV material is needed for TFPV than for c-Si panels. 
Today, a growing number of companies produce TFPV units designed to replace more common 
glass fa?ade panels and have integrated them into the design of buildings throughout the world. 
While in Germany, the world?s largest solar market, BIPV only represents 2% of all installed PV 
products (Runyon 2010), these multi-functional panels, if reliable, can be an integral part of the 
increasing number of environmentally-friendly buildings for a less energy-dependent future. A 
recently published study by GTM Research found that products based on the concept of BIPVs 
are beginning to emerge in the marketplace after more than 20 years of research and 
development (GTM 2010). While markets in Europe are larger and more well-established, the 
U.S. market is developing quickly (GTM 2010). 
 A failure state that threatens the reliability of TFPVs used in building fa?ades is the structural 
failure of the glass elements due to excess loading or stress concentrations. Extreme wind events 
such as downbursts, tornados or hurricanes can cause high winds that can damage these fa?ade 
elements rendering them unusable (Figure 1.2). The focus of this research is centered on TFPVs 
in building facades when subjected to extreme wind loading, specifically concentrating on the 
Washington D.C.- Baltimore Metropolitan area (WBMA). Determining the behavior of this 
system requires two parts. First, the determination of the design wind pressure subjected on the 
thin-film BIPV, and second, a resistance probability distribution of the TFPV panel.  Where the 
TFPVs fail is where the tail of the strength distribution is less than the design wind pressure.  
To determine the design wind pressure on the fa?ade of a building, a case of extreme weather 
know as downbursts is considered. Downbursts cause strong lateral winds for a short period of 
time in a concentrated area on the surface of the earth. Historically these extreme weather events 
 
5 
 
 
have been known to cause severe damage to building facades despite current building code 
requirements because of their high-intensity winds. Downburst events have been recorded in all 
regions of the United States, yet their characteristics, intensity and occurrence rates are not 
systematically recorded today (as compared to tornados or hurricanes), and vary substantially by 
location. When determining wind loads, the American Society of Civil Engineering (ASCE) 7-05 
wind load specifications (2005) indicate different design wind speeds depending on the location 
of construction. To limit the scope of this research to an area of relatively uniform wind loading 
conditions, specifically the Washington D.C.-Baltimore Metropolitan area (WBMA) of the 
United States was chosen to study. The data collected on downbursts in this region is used to 
develop a wind load distribution curve, which is used in a stochastic model to determine the 
design wind speeds.  These wind speeds are converted to design wind pressures using the 
procedure in ASCE 7-05.  
To determine the probabilistic distribution curve of the TFPVs? strength, an understanding of the 
components and their behavior under loading is required. A TFPV panel is considered to have 
failed when the glass has fractured. A review of all TFPV technologies available on the market 
was conducted first to determine the most common materials used and method of construction.  
Similar in structure and construction to laminated glass, a failure model of laminated glass under 
uniform loading is used to construct the resistance distribution curve of TFPVs in BIPV 
applications. By calculating the area where the resistance distribution is less than the design wind 
pressure, the reliability of this system is determined from the corresponding failure probability.  
This thesis is organized into six chapters. Chapter 1 provides the motivation for this research and 
overview of the work done in this study. Chapter 2 offers a general discussion of the weather 
phenomenon of a downburst as well as the different methodologies used in previous studies to 
 
6 
 
 
categorize and characterize these high-wind producing storms. The data available in the WBMA 
on the occurrence of downbursts and the methodology followed in this research to establish this 
dataset are also discussed. Chapter 3 provides details on the conversion process for the 
downburst data into a distribution curve, and a stochastic model used to determine the design 
wind speed and wind pressure. The second part of this thesis focuses on BIPVs. Chapter 4 
discusses a brief history on PVs, the different types of TFPVs and their layers, and introduces 
BIPVs and fa?ade applications. The modeling of TFPVs behavior from a laminated glass model 
is discussed as well as the creation of a distribution curve of resistance.  Chapter 5 compares the 
design wind pressures determined for downbursts in the WBMA and the fa?ade-installed TFPVs 
resistance.  Current building design specifications are discussed.  Chapter 6 summarizes the 
major finding from this research as well as provides a discussion of future work.  
 
Figure 1.1: BIPV Applications on a Typical Building (adapted from Shaybani, 2010) 
Roof 
Facade 
Solar Shading 
 
7 
 
 
 
Figure 1.2: Fa?ade Failure Due to Hurricane Alicia in 1983, Houston, TX (USACE 2010) 
  
 
8 
 
 
CHAPTER 2: DOWNBURSTS  
Extreme wind loading on buildings can be caused by a variety of different weather phenomenon. 
Since 1950 two of these well-know causes of damages, hurricanes and tornados, have been 
reported to the National Weather Service (NWS) approximately 4 to 5 times per year in 
Maryland (MD) and 11 times per year in Virginia (VA) (NCDC 2010). In the coastal regions of 
MD and VA, hurricane events are frequent enough to merit higher design wind speeds in the 
wind load design charts of ASCE-7-05 design load standards in the United States (ASCE 2005). 
In this area, as well as throughout the U.S., a less-known yet equally damaging weather 
phenomenon known as a downburst (wind speed up to 168 mph (75 m/s) ), is estimated to occur 
approximately ten times as frequently as tornados according to the National Weather Service 
(NWS 2010). This extreme weather event, however, is not considered in the ASCE-7-05 design 
standards. Current lack of reliable information on the occurrence of downbursts would make this 
difficult to do so. 
A downburst is created by a column of sinking air that, after hitting the earth?s surface, spreads 
out in all directions, producing damaging straight-line winds, similar, but distinguishable from 
that of tornados.  The reason, in part, why this phenomenon is not well-known or documented 
today is that their relatively small size (less than 10 km across) and short duration (lasting only 2 
to 20 minutes) make them difficult to reliably detect and track. Downburst can occur anywhere, 
and the Washington D.C.-Baltimore Metropolitan area (WBMA) (northern VA, MD, and 
Washington D.C.), similar to other metropolitan areas in the U.S., has experienced downburst 
wind gusts of 130 mph (58 m/s) (Fujita 1985). Maximum horizontal winds of a downburst can be  
168 mph (76 m/s), winds comparable to that of a Category 5 hurricane or an F3 tornado.  
 
9 
 
 
Downbursts, while originally studied because of the significant threat they pose to airborne 
planes during take-off and landing, can also be a threat to structures. In May of 2009 the Dallas 
Cowboys Football practice facility collapsed due to a downburst, injuring 12 people (Gross et al. 
2010). With a roughly estimated occurrence rate higher than hurricanes and tornados and with 
winds comparable to other extreme wind events, it is reasonable to justify this weather 
phenomenon as one that merits further study, particularly for buildings with fa?ade elements 
susceptible to breakage from high wind loading. 
In the WBMA, in order to determine the reliability of Building Integrated Photovoltaics (BIPVs) 
and more specifically thin-film photovoltaic (TFPV) panels installed in building fa?ades, the 
wind loading distribution curve of downburst winds must first be established for the WBMA. To 
determine this distribution this chapter discusses how a downburst is defined and what features 
characterize this extreme-wind causing event. In addition, the methodologies previous studies 
have followed to determine downburst wind speeds for a particular region and the methodologies 
used in this research are discussed. 
2.1 DEFINITION AND ORIGIN OF DOWNBURST CONCEPT 
Downbursts are a weather phenomenon first discovered by Fujita in 1976-1977 after a detailed 
study of two severe-weather related airplane crashes in the U.S. (Fujita 1976;  Fujita and Byers 
1977). Downbursts can be defined as strong, concentrated downdrafts from convective storms 
(i.e. thunderstorms) that can cause damaging divergent winds on or near the ground surface of 
the earth.  The ?jet-like? downdraft that caused two plane crashes in 1975 was first identified in 
1976 and soon after Fujita (1976) and Fujita and Byers (1977) proposed using the term 
?downburst? for this extreme weather event.  
 
10 
 
 
To further study downbursts, the Northern Illinois Meteorological Research on Downbursts 
(NIMROD) field program was established in 1978, which used Doppler radars to successfully 
identify downbursts, verifying their existence. Since then an increasing focus on observational 
and theoretical studies of downbursts has been pursued. 
2.2 DOWNBURST FORMATION 
During a thunderstorm, differences in temperature on the surface of the earth create instability in 
the air, sometimes driving warm, moist air to higher elevations in the atmosphere. This warm, 
dry air rises from the earth?s surface and enters the thunderstorm causing rain to evaporate, 
cooling the dry air rapidly.  The cooled air then becomes too heavy to be supported mid-air and 
subsequently falls back towards the Earth?s surface, creating a downdraft. As the air falls it 
accelerated downwards and finally hits the earth?s surface to create the phenomenon of a 
downburst (Yuh-Lang 2007). 
At touchdown, the downburst is characterized by a shaft of strong downward velocity at its 
center and strong divergence upon impact. Once the air downdraft has hit the ground, it flows 
outward from the center of the downburst in the form of straight-line horizontal winds. Figure 
2.1 is a conceptual drawing of how this process occurs (Fujita 1985).   
2.3 MICROBURSTS AND MACROBURSTS 
Wakimoto (1985) further divided downbursts into two categories based primarily on size ? 
macrobursts and microbursts. Macrobursts are defined as downbursts with diameters of more 
than 2.5 miles (4 km), duration of 5-20 minutes, and wind speeds up to F3 intensity, or 134 mph 
(60 m/s). Microbursts are smaller downbursts of less than 2.5 miles (4 km) in diameter, 2-5 
minutes in duration, with winds as high as 168 mph (75 m/s). A Chicago study yielded 
 
11 
 
 
microburst durations of 3.4 minutes on average, and a second study in Denver (1982) found 
average microburst lasted 2.8 min (Fujita 1985).  
Due to the nature of the dataset used in this study, reported downburst events could not be 
confidently classified as ?microbursts? or ?macrobursts? in many cases, thus the general term 
?downbursts? is used instead. Both microbursts and macrobursts are capable of producing 
extreme winds, and the greatest importance of the downbursts in this research is the wind speed 
produced as a result of these events. 
2.4 WET AND DRY DOWNBURSTS 
Results of an extensive project named JAWS (Joint Airport Weather Studies) completed in 1982 
revealed that there are not just one, but two main conditions in which downbursts can occur. Some 
downbursts are accompanied by heavy rains from thunderstorms, while others are caused by virga 
shafts - observable streaks or shafts of precipitation that fall from a cloud but evaporate before 
reaching the ground (Brown et al. 1982).  This observation further divided downbursts into ?wet? 
and ?dry? categories, which others confirmed in years following (Wolfsen 1983; Caracena et al. 
1983; Wilson et al. 1984; Fujita 1985; Caracena and Flueck 1988). A summary of the characteristics 
of wet and dry downbursts and corresponding diagrams can be found in Table 2.1 and  
Figure 2.3.  
Wet downbursts occur in conditions of simultaneous heavy rains and high winds, relying on the 
downward acceleration of the precipitation to form (Fujita 1983, Doswell 1994).  Much less is 
known, however, about how wet downbursts originate compared to dry downbursts, and thus 
they are correspondingly more difficult to forecast (Doswell 1994). They can be easily identified 
by doppler radar reflectivity data, however, because they have a high reflectivity compared to 
other parts of convective storms (Wilson et al. 1984).  The formation of a bow echo, a specific 
 
12 
 
 
bow-like shaped storm cell as shown in Figure 2.2 has been associated with the occurrence of 
downbursts (Fujita 1978).  
Dry downbursts can also occur in the presence of rain, however little or no rain is most common. 
They are created by evaporation in relatively dry conditions which produces a downward force, 
named negative buoyancy, that propels the air downward (Doswell 1994).  They can be 
identified by a virga, or shaft of precipitation that falls from a cloud then evaporates before 
reaching the ground. For further information readers are referred to Wakimoto (1985).  
Little published information is available on the type of downburst that are most characteristic in 
the WBMA, however an unpublished study by Pryor (2009) suggests that wet downbursts are 
more common than dry downbursts in this region.  Atkins and Wakimoto (1991) suggest that wet 
downbursts are most characteristic of the southeastern U.S.  
It should be noted that because of the different characteristics of downbursts depending on the 
region of study, a discussion on the frequency of occurrence of downbursts in a large area (e.g. 
the Unites States) is too large an area of study. The WBMA was chosen for two reasons. First, 
there has been relatively little research on downburst occurrences in this area of the U.S., thus 
studying this region can lead to a better characterization of the threat of downbursts in this 
region. Second, downburst horizontal straight-line winds produce high-velocity gusts between 
approximately 33-164 ft (10-50 m), similar to the height of high-rise buildings characteristic of 
city centers (Holmes 2002; Kim and Hangan 2007). It is the medium to high rise buildings where 
Building Integrated Photovoltaics (BIPV) in facades are most commonly installed. The WBMA 
contains two major city centers with medium to high rise buildings, one of which is the U.S. 
national?s capitol. 
 
Table 2.1: Dry and Wet Downburst Characteristics (adapted from Rose 2010) 
 
13 
 
 
 
Characteristics Dry Downburst Wet Downburst 
Location of Highest 
Probability (within the U.S.) Dry Climates (Midwest/West) 
Moist or temperate climates 
(Southeast) 
Precipitation Little/None Moderate/Heavy 
Features Below Cloud Base Virga Shafts of strong precipitation reaching the ground 
 
  
 
14 
 
 
 
2.5 WASHINGTON D.C.-BALTIMORE METROPOLITAN AREA 
(WBMA) DOWNBURSTS  
The varying features of a climate of a particular region can cause it to be more or less conducive 
to downburst occurrences of different types.  Downbursts in the eastern U.S. are most likely wet 
downbursts, which originate within convective storms (thunderstorms). In a recent unpublished 
analysis of the downburst occurrences in the eastern part of the U.S., among the areas of the most 
frequent occurrences was the WBMA (Pryor 2009).  The low-level northward flow of warm air 
and moisture from the Atlantic Ocean creates pressure instability, then storm fronts combined 
with the orographic lift (raising the elevation of air mass) caused by the Appalachian Mountains 
produce good conditions for convective thunderstorms, and thus downbursts.   
Further published evidence of downbursts in the WBMA is supported by Fujita (1985). Fujita 
studied a microburst event at Andrews Airforce Base near Washington D.C. that occurred on 
August 1, 1983. The peak wind speeds caused by the downdraft were clocked at approximately 
158 mph (70 m/s), one of the highest speeds recorded in downburst history. A graph of wind 
speed vs. time of this downburst is shown in Figure 2.4.   
2.5.1 WBMA Data Coverage Limits 
The WBMA in this research is defined by the coverage area of the National Weather Service 
(NWS) Forecast Office for the Baltimore/Washington area. This area includes most of Maryland, 
District of Columbia, Northern Virginia, parts of Eastern West Virginia. Figure 2.5 shows the 
coverage area of the main doppler radar for this region and Figure 2.6 shows all the counties 
 
15 
 
 
within the coverage area. This figure also includes the number of Skywarn1 participants, or 
trained citizens who report weather conditions such as downbursts, convective storms, tornados, 
etc. to the NWS. The areas of green are the counties which have the most densely populated 
areas of Skywarn-trained citizens, which also correspond to the cities of Washington D.C. and 
Baltimore.  Doswell (2005) points out severe storm event (such as a downburst) reporting quality 
and methodologies can be inconsistent among different NWS offices, thus in an effort to obtain 
consistent data, only one NWS office was chosen. 
2.5.2 The Convective Season of the WBMA 
Downbursts in the WBMA occur with convective storms.  In this area, and throughout the 
United States, convective storms are more frequent in the warm summer months than in the 
colder months of the year (Kelly et al. 1985). In this research, data on the occurrence of 
downbursts is thus restricted to May-August.  According to Kelly et al. (1985) these four months 
account for 69% of damaging wind storms, 74% of strong wind gusts, and 76% of violent 
windstorms for weather throughout the U.S.   
2.6 METHODOLOGY OF DOWNBURST DATA COLLECTION AND 
ANALYSIS 
Unlike other extreme-wind weather events such as tornados or hurricanes, no comprehensive 
database of occurrences currently exists that officially tracks downbursts, where they occur, their 
strength, wind speed, and other pertinent details.  Thus in order to best predict the probability of 
occurrence of a downburst in the WBMA, a review of previous studies and research was 
                                                 
1 Skywarn website - http://www.weather.gov/skywarn/ 
 
16 
 
 
conducted to assess methods of obtaining data for this research. The chosen methodology is a 
reflection of the culmination of the knowledge obtained from these studies as it is relevant to this 
research.  
2.6.1 A Review of Downburst Studies Conducted in the United States  
The methodologies of studying downburst occurrences throughout the United States can be 
divided into three categories: (1) long-term, sensor network studies, (2) retroactive surface data 
studies, and (3) computer modeling based studies. Of each of these methodologies used to study 
downbursts, the second methodology, retroactive surface data collection, is most relevant for the 
datasets available for the WBMA of the United States. 
2.6.1.1 Long-term, Sensor Network Studies  
One methodology used to study downburst behavior and count occurrences can be classified as a 
long-term sensor network study. These types of studies, while they have been conducted in a 
variety of environments studying different types of downbursts, follow the same general 
methodology of obtaining microburst data on wind speeds, size and counts. This methodology 
requires an extended period of time in which to actively observe the occurrence of downbursts in 
a finite area. In these studies a network of sensors and data recorders such as a Portable 
Automated Mesonet (PAM) is first set up, then the downbursts are identified using the wind 
speed and direction data collected. A PAM consists of a base station connected to a network of 
remote sampling stations. The base station logs all data (wind speed, wind direction, pressure, 
humidity, etc?) obtained from each of the stations (Brock 1977). 
From these studies using a high-density network of sensors much has been learned about the 
characteristics and behaviors of downbursts. Compared to relying on already existing, more 
sporadically placed sensors of varying quality, or relying on trained citizen observers to report 
 
17 
 
 
such events, the generated data from this type of network is much more consistent. 
Understandably, however this methodology also requires a significant amount of resources, time 
and funding to complete. Several detailed studies throughout the United States of this type have 
been conducted in effort to better characterize downburst behaviors. These include the following:  
? Northern Illinois Meteorological Research on Downburst (NIMROD) (1978) 
?  Joint Airport Weather Studies (JAWS) (1982) 
? Classify, Locate and Avoid Wind Shear (CLAWS) (1984) 
? FAA/Lincoln Laboratory Operational Weather Studies (FLOWS) (1984-85) 
? Microburst and Severe Thunderstorm (MIST) (1986) 
? Convection Initiation and Downburst Experiment (CINDE) (1988) 
? Thunderstorm Outflow Project (2002) 
Information and conclusions from these studies can be found in more detail in other papers (e.g. 
McCarthy et al. 1982; Fujita 1985; Rinehart et al. 1987; Wilson 1988; Hjelmfelt 1988; Gast and 
Schroeder 2003). Figure 2.7 indicates the location of several of these studies. The general 
detection methods of each of these studies is similar. The MIST study is explained in more detail 
below.  
During the MIST experiments several observation methods were employed for the detection of 
downbursts. Held in Northern Alabama in 1986, the most important meteorological data to this 
study was surface weather data.  Surface data was provided by a dense PAM network of 41 
observation stations and 30 FLOWS (FAA-Lincoln Laboratory Observational Weather Studies) 
stations (Atkins and Wakimoto 1991). The average spacing between stations was approximately 
1.25 miles (2 km) and data was recorded every 1 minute, with the hopes that all but the smallest 
microburst (< 1.25 miles (2 km)) would be identified by the sensors. PAM data was analyzed 
 
18 
 
 
using a computer algorithm first suggested by Fujita (1985) and Wakimoto (1985) to identify 
downburst occurrences. NCAR (National Center for Atmospheric Research) Doppler radars were 
also spaced out to provide full coverage of the area. The interested reader is referred to these 
studies for specifics on the set of standard conditions required of surface data observations used 
to identify the characteristic wind speeds over short periods of times.   
Studies such as the MIST project have the advantage of a dense network of frequent wind 
observations, a condition that is very advantageous when attempting to observe all downbursts 
events for a given area. For the purposes of this research these studies such as the MIST study 
provide a better understanding of the formation and proper identification of downbursts, but are 
less useful as a methodology to follow.  Unfortunately no such network exists or has been set up 
for the WBMA, meaning that an extensive network of sensors would need to be set up in this 
region and subsequently monitored. Therefore, the data used here must be collected from 
previously recorded sources.  Ideally, a network of sensors set up in a large area for an extended 
period of time would present more reliable data and more information on the downbursts that 
occur beyond an estimated or measured wind speed and is the subject of future work.   
2.6.1.2 Retroactive Surface Data Studies  
More pertinent to this research are the downburst studies that have been conducted retroactively. 
These studies must combine a variety of already-recorded data sources to identify and verify the 
severe weather occurrences such as downbursts. All the data that could be used to identify these 
events already exists, though not typically for the sole purpose of downburst identification. A 
more detailed description of this type of methodology can be found in the closely related field of 
research on tornado occurrences (e.g. Kelly et al. 1978; Brooks et al. 2003; Doswell et al. 2005). 
 
19 
 
 
 In this methodology the initial tally of a specific weather occurrences is gathered from published 
storm reports that are available in two different formats online from the Storm Prediction Center 
(SPC) and National Climatic Data Center (NCDC), both part of the National Oceanic and 
Atmospheric Administration (NOAA)2. This data originates from surface weather observation 
stations on rivers and coast areas, at airports, from human observers, as well as other 
methodologies. And while errors associated with this database are possible due to the variety of 
data sources and methods used to collect the data, since 1972 a concerted effort has been made to 
ensure complete and accurate data (Schaefer and Edwards 1999). 
Within this database, downburst occurrences are typically reported under the category of high 
wind events. For each event the characteristics of the event, including location, date, time, 
duration and estimated or measured wind speed are typically included.  Examples of these data 
entries are found in Figure 2.11,Figure 2.12 and Figure 2.13. 
Several methods of checking the validity of these reports are available. Radar reflectivity 
imagery from Next-Generation Radar (NEXRAD) can be used to verify the existence of a 
convective system (in the shape of bow or crescent shaped echos at the location of the 
downburst) (Fujita 1978).  Additionally, the daily infrared satellite imagery and 850-mb theta-e 
data, where available, can be collectively used to identify if conditions for convective activity are 
favorable (Atkins and Wakimoto 1991).   
Figure 2.8 displays the results from Pryor?s unpublished study (2009) of downburst occurrences 
in the eastern U.S. This data is limited to the timeframe of years 2001-2003, thus more data 
                                                 
2 The NOAA websites for storm report data: http://www.spc.noaa.gov/climo/ (preliminary data), and 
http://www4.ncdc.noaa.gov/cgi-win/wwcgi.dll?wwEvent~Storms (final data) 
 
20 
 
 
beyond these three years would be better for characterization of downburst occurrence in the 
WBMA.    
2.6.1.3 Computer Modeling Based Studies 
Some of the more recent studies of downburst occurrences have utilized computer based data and 
software to collect information on downbursts (e.g. Ellrod et al. 2000; Pryor et al. 2002; Pryor et 
al. 2003).  GOES (Geostationary Operational Environmental Satellite) satellites provide 
continuous monitoring of weather patterns through imager and sounder data. Each GOES 
satellite monitors one area on the surface of the Earth by circling it at speeds matching its 
rotation. GOES Microburst Product is a software program that uses GOES sounder image data to 
generate a map showing areas of elevated likelihood that a microburst could occur (Figure 2.9). 
The closer to red the color, the higher likelihood of occurrence is predicted. 
GOES microburst prediction methods and other computer generated models such as WINDEX 
(McCann 1994) and GUSTEX (Geerts 2001) derived from the GOES data are recent and still 
under evaluation and are thus not used in this research. As an example of their application, in a 
study by Walter (2007) a variety of data, including GOES-derived WINDEX data was collected 
for a 30-year period to determine the number of days in which downbursts were likely to occur 
annually. Figure 2.10 shows the predicted peak wind speeds of downbursts. 
2.7 WBMA DOWNBURST DATA COLLECTION METHODOLOGY 
2.7.1 Datasets Used for Downburst Information Collection 
Compiling a dataset of downburst occurrences for a specific region is ultimately limited to the 
time period in which the downburst weather phenomenon has been recognized as a type of 
severe weather event (i.e., from 1970s to present). Since there has been no formal and consistent 
 
21 
 
 
mechanism for reporting this type of severe weather throughout the years, developing a 
sufficiently large dataset to characterize the downburst activity for the WBMA multiple 
approaches are used.  This section outlines the approach for compiling such a dataset of 
downburst occurrences in the WBMA.  
2.7.1.1 Storm Event Database ? Data Collection Methodology: 
Data was collected using two different approaches. The primary data source originates from the 
NCDC. The searchable and publicly accessible online database of severe weather events called 
the Storm Event Database3 is the same database used in previous studies (e.g. Kelly et al. 1978; 
Brooks et al. 2003; Doswell et al. 2005).  It contains official records of all reported severe 
weather events that have been observed and recorded across the United States from 1950 to the 
present day. The data published in this database is also published on NOAA?s SPC website4 but 
this data is considered preliminary only.  
 In order to merit an entry in the Storm Event Database, the reported weather must be classified 
as ?severe? weather. Severe weather, as defined by the NWS, is weather that produce one or 
more of the following events (NWS 1995):  
? large hail of at least 0.75 in (1.9 cm) diameter 
? damaging winds of at least 58 mph (26 m/s) 
? tornados 
In addition to these, any storm event that causes structural damage is typically considered to be 
severe, even if its intensity was not recorded (Doswell et al. 2005). Since strong wind speeds, 
typically above 58 mph (26 m/s), are one of downbursts? most defining characteristics, it is 
                                                 
3 Storm Event Database - http://www4.ncdc.noaa.gov/cgi-win/wwcgi.dll?wwEvent~Storms 
4 Storm Prediction Center http://www.spc.ncep.noaa.gov/climo/ 
 
22 
 
 
determined in this research to search this database for severe and damaging wind events caused 
by downbursts. Information for this database comes from a variety of human and automated 
sources ? the NWS, Military Services, Federal Aviation Administration, as well as data from 
trained civilian volunteers.  
The Baltimore/Washington section of the NWS publishes monthly summaries of severe weather 
taken from the Storm Event Database, compiling all of the severe weather reports that pertain 
only to the WBMA into one document. Thus to obtain all existing records of downbursts in this 
region, the files containing data for the months in the convective season (May ? August) 
published on the Baltimore/Washington NWS website5 were compiled. This data was then 
searched using the keywords ?downburst,? ?microburst,? and ?macroburst,? for occurrences of 
downburst occurrences within the severe wind events in all available years (1996-2009). Each 
record identified in this compiled set of data was recorded, totaling 78 records identified as 
downburst events.  
The data for each record varied greatly in the amount of detail reported.  Where possible the 
following information was recorded for each downburst occurrence: 
? Date of occurrence 
? Time of downburst 
? City and state where the observation was made 
? Latitude and longitude coordinates 
? Maximum estimated or measured wind speed 
? Any further textual details pertaining to the downburst occurrence given in the storm 
report 
 
                                                 
5 NWS Baltimore/Washington Storm Data - http://www.erh.noaa.gov/lwx/Storms/Strmdata/ 
 
23 
 
 
To ensure that all reported cases within the WBMA were consistent with the NOAA?s 
nationwide Storm Event Database from which the data was extracted, all records were cross 
checked with this database. In all cases records of the same information were found. Additionally 
in many cases where the Baltimore/Washington compiled data had missing information ? 
particularly the latitude/longitude coordinates of the location of the touchdown of the downburst 
? the national database entry contained that piece of information.  
Several examples of the Storm Event Database records from the WBMA are shown in Figure 
2.11, Figure 2.12 and Figure 2.13, illustrating the variance in the detail of reports.  A month?s 
worth of Storm Data includes many of these entries separated by both region (e.g. Central 
Maryland) and by type of report (strong wind, tornado or hail).  The database entries shown from 
the NWS Baltimore/Washington section?s compiled data include the title at the top of each page, 
the region of the particular report of interest and all of the details it gives. Also included are the 
records from the national Storm Event Database of the same events to show the similarities and 
differences between the two types of entries and compare the amount of detail in each.  
Of the records reviewed, all indicated a date, time and location. In many cases the entries also 
listed the estimated or measured wind gust speed, however especially in the earlier entries, the 
wind gust speed was not noted in the weather reports.  45 of the 78 entries reported an estimated 
or measured wind speed of a downburst of 50 knots (58 mph, 26 m/s) or larger. The other 33 had 
either a numerical measurement of the wind speed of the downburst of ?50+ knots? or no wind 
speed estimate at all.  
Due to the small size of downbursts and the lack of densely placed wind speed measuring 
anemometers throughout the WBMA, it is understandable that a wind speed value may not be 
available for downbursts from ground surface data. It is assumed, however, that if the event is 
 
24 
 
 
placed in the Severe Storm Database, then the NWS approved the event?s reporting as fitting the 
criteria of a ?severe? wind event ? that is one over 58 mph (26 m/s). Thus for each record with 
wind data marked as ?50+? or as nothing, an estimated 58 mph (26 m/s) speed was assumed.  
This assumption is consistent with the assumptions used by McDonald and Abbey (1979) who 
applied this idea to tornado datasets. In their dataset, 8.1% of tornados were missing intensity 
estimates, 39.4% lacked adequate information. Thus they assigned a minimum value of intensity 
to any missing data, assuming that if the tornado was reported, it likely existed, but also if it had 
been more noteworthy, more information would have been reported on it (Schaefer et al. 1986).  
In addition to the assumption that all unreported downburst wind speeds are 58 mph (26 m/s),  
the downburst dataset was also considered in which only the downbursts with reported wind 
speeds (estimated or measured) were included for purposes of comparison and analysis. 
In a number of cases the Storm Event Database records indicated the exact wind speed measured 
for the downburst, however in most cases either a range of winds speeds or an estimated wind 
gust is presented. Given the range of values of wind speeds (e.g 60-80 mph), the low (e.g 60 
mph), average (e.g. 70 mph) and high (e.g. 80 mph) estimates were evaluated when creating the 
model of the downburst occurrences in this area in an attempt to encompass the full range of 
possible wind speeds that may have occurred.  
For an understanding of where the estimated wind speeds originate on the Storm Event entries, a 
common way that wind speed is estimated, if exact measurements are not available, is by using 
the Beauford Scale.  The scale uses the end product of a damaging wind event ? damage to 
buildings, trees, etc? to estimate the wind gust speed.  Table 2.2 shows the different severity 
estimates, relating wind speed and resulting damage to each other.  Based on the categories in 
this estimation system, winds above approximately 55 mph (24.6 m/s) can cause significant 
 
25 
 
 
structural damage including snapping of trees and structural damage. In many of the reports 
listed in the Storm Event Database, both building damage and snapped trees were reported.  
Taking the storm reports presented in Figure 2.11 wind speeds were estimated at 61 knots (70 
mph, 31.2 m/s).  At this speed the scale indicates that trees can be broken and structural damage 
can occur. The text in this severe storm report indicates that over 100 trees were knocked down 
and structural damage occurred in a commercial district in Bowie, MD, which seems to align 
with the wind estimates presented on the database entry.  
Many of the estimates that are given in this database come from citizen reporters trained through 
a program known as Skywarn6.  This program sponsored by the NWS teaches citizens about the 
different types of weather and how to report severe weather observed to NWS. For reporting 
estimated wind speeds such as those associated with a downburst, a Skywarn training manual 
suggests observing damage to trees and the environment in an effort to use the Beauford scale to 
measure the estimated wind speed.   
  
                                                 
6  Skywarn - http://www.weather.gov/skywarn/ 
 
26 
 
 
 
 
 
 
 
Table 2.2: The Beaufort Scale - Wind Speed Estimates (adapted from SPC 2010) 
Beaufort 
Number 
Wind Speed in 
mph (m/s) 
Description Effect on Land 
0 < 1 ( 0.4) Calm Calm; smoke rises vertically. 
1 1-3 (0.4-1) Light Air Smoke drift indicates wind direction, vanes still. 
2 4-7 (1-3) Light Breeze Wind felt on face; leaves rustle; vanes begin to move. 
3 8-12 (3-5) Gentle Breeze Leaves, small twigs in constant motion; light flags extended. 
4 13-18 (5-8 Moderate Breeze Dust, leaves and loose paper raised up small branches move. 
5 19-24 (8-11) Fresh Breeze Small trees begin to sway. 
6 25-31 (11-14) Strong Breeze Large branches of trees in motion; whistling heard in wires. 
7 32-38 (14-17) Moderate Gale Whole trees in motion; resistance felt in walking against the wind. 
8 39-46 (17-21)  Fresh Gale Twigs and small branches broken off trees. 
9 47-54 (21-24) Strong Gale Slight structural damages occur; slate/shingles blown off roofs. 
10 55-63 (24-28)  Whole Gale Seldom experienced on land; trees broken; structural damage occurs. 
11 64-72 (28-32) Storm Very rarely experienced on land; usually causes widespread damage. 
12 > 72 (32) Hurricane Force Violent and destructive winds. 
 
  
 
27 
 
 
 
2.7.1.2 News Search via Lexis Nexis ? Data Collection Methodology:  
Considering the gaps and inherent inaccuracies in the data of the Storm Data, a search was also 
performed using the database LexisNexis to search all publications of the major newspapers and 
newsfeeds in the WBMA between the same years of 1996 to 2009. The goal in performing this 
search was to increase the probability of detection of at least the strongest wind events. It is these 
events that are of greatest interested to the structural design of buildings as they have a higher 
potential to damage infrastructure. Through the news database Lexis Nexis the sources searched 
including the major newspapers within the coverage area of the Baltimore/Washington NWS - 
The Washington Post, The Washington Times and The Baltimore Sun among others as well as 
the newsfeeds/newswires for the area ? were searched.  This search resulted in 15 total reports of 
downbursts using the keywords ?downburst?, ?microburst? and ?macroburst?.  Comparing the 
events with the events recorded in the Storm Event Database found that 2 of the 15 records were 
consistent. The two overlapping records from both databases were counted only once, the other 
13 records were added to the first dataset to create a total of 93 records, representing 91 
downburst events.  
It is of some cause for concern that there was not more overlap found between the Storm Event 
Database and the news search performed. Throughout this paper it is recognized that there are 
unavoidable flaws in the dataset, and the data is treated in such a manner.    
2.7.1.3 Accuracy of Data Sources 
Since the diameter of downbursts can be much smaller than the distance between wind data 
collectors or trained civilians in the WBMA, it is highly likely downburst events are not included 
in the dataset, even when multiple data sources are consulted. However the assumption is made 
 
28 
 
 
that if the downburst is strong enough to cause major damage, it is highly likely that it will be 
reported to either a news source or to the NWS. Since the values of the high speed winds (rather 
than the lower speeds) caused by downbursts are of most importance it is reasonable to assume 
that a high percentage of the strong downbursts have been identified in one of the two sources.   
The Storm Event Database entries are only as accurate as the tools and resources used to report 
the event.  In some cases the data is collected from an automated weather station ? here the 
winds are measured with an anemometer and are highly accurate. However in other cases the 
event is reported by a trained civilian with little to no equipment, and thus relies on visual clues 
or the Beauford scale (Table 2.2) for an estimated wind speed.  
2.7.1.4 Methodology Comparison to ASCE 7-05 
In the design of infrastructure in the United States, ASCE (American Society of Civil Engineers) 
7-05 ? ?Minimum Design Loads for Buildings and Other Structures? (2005) is followed in the 
determination of the loads on buildings.  ASCE 7-05 provides a means for determining required 
wind loads which a building must withstand (Peterka and Sohban 1998). While the methodology 
used to create the wind map involved the collection of wind speed data as opposed to just 
downburst data, the similarities to the methodology in this research are listed as follows:  
? The data used in the determination of the design wind speeds for the United States were 
obtained by combining wind data from multiple meteorological stations into one 
?superstation? for a given region (methodology described in Peterka 1992). This is 
similar to the compilation of data from multiple sources across the WBMA obtained for 
the downbursts in this study.   
 
29 
 
 
? These ?superstations? were generally the size of a ?typical state,? particularly in the 
eastern United States because sufficient data existed at multiple stations in the region, 
which is similar to the size of the WBMA region of study. 
? The typical length of record over which correlations were calculated was 15-25 years 
depending on the length of the time period of data available when this map was first 
developed (1998). In this study the length of time is 14 years. 
  
 
 
 
Figure 2.1: First 
Figure 2.2: Bow Echo Formation and 
Figure 2.3: Dry and Wet Microburst Diagrams 
 
 
Downburst Conceptual Model (Fujita 1985) 
Downburst Locations (Fujita 1978)
(adapted from Caracena and Flueck 1988
30 
 
 
 
 
) 
 
31 
 
 
 
Figure 2.4:  Andrews Airforce Base Downburst - August 1, 1983 (Fujita 1985) 
 
 
Figure 2.5: Doppler Radar Coverage for the Baltimore/Washington NWS (NWS 2010)  
 
 
 
 
Figure 2.6: Coverage Area of Baltimore/Washington NWS by County (
Figure 2.7: Location of Extensive Downburst Studies in the United States (Wakimoto 2002)
NWS 2010
32 
 
) 
 
 
 
33 
 
 
 
Figure 2.8: Downburst Occurrences by State 2001-2003 (Pryor 2009) 
 
Figure 2.9: GOES Microburst Product overlaid with Radar data (Pryor 2009) 
 
 
34 
 
 
 
Figure 2.10: 50-year Recurrence Interval for Microburst Peak Wind Speeds (mph) (Walter 2007) 
 
 
 
 
Figure 2.11: Baltimore/Washington Storm Event Report - Early Format 
 
 
 
Figure 2.12: Baltimore/Washington Storm Event Report ? Recent Format  
 
35 
 
 
 
Figure 2.13: National Storm Event Report (same date as Figure 2.11) 
  
 
36 
 
 
CHAPTER 3: DOWNBURST DATA AND DISTRIBUTION 
3.1 SUMMARY OF DATA: 
Over the 14 year period of records from 1996 - 2009 in the WBMA, 93 total downburst events 
were reported during the convective season (May - August) - 78 were reported in the Storm 
Event Database, and 15 were reported in the Lexis Nexis database, with 2 overlapping.  
3.2 ELIMINATION OF EXTRANEOUS DATA ENTRIES: 
Several of the 93 records were eliminated or combined to create a more meaningful and useful 
dataset. Two of the events had records from both the Storm Event and Lexis Nexis databases, 
thus the information from both sources was combined to create one event to represent each 
downburst occurrence.   
Several of downbursts reports occurred on the same day at similar times but in different 
locations, likely meaning one convective storm produced multiple downbursts.  Previous 
climatology studies have chosen to count the number of days, rather than the total number of 
downburst (e.g. Brooks 2003).  The use of ?downburst days? as opposed to the inclusion of all 
downbursts in the dataset is consistent with the approach taken by Brooks et al. (2003).  Brooks 
argues that this approach takes advantage of the best aspects of the limited dataset.  The location 
of the touchdown point and date of occurrence are taken as the most reliable factors and thus it is 
explained that using tornado (or in this case downburst) days gives the estimate of the threat of 
1+ downbursts touching down at any location during a 24 hour period. Using this methodology 
the occurrence of one major convective storm that produced several sightings of downburst does 
not skew the dataset.  One problem that this presents is what downburst, of the several that may 
 
37 
 
 
occur on a given day, should be chosen to represent that day?s downburst occurrences. In this 
research the downburst with the highest wind speed value was chosen to represent the worst case 
scenario. 
Eliminating the extraneous data from the original dataset, a total of 58 downbursts remain. A list 
of each of these events, including date, location and wind speed are listed in Table 3.1. In this 
table the average wind speed values are reported. In the cases where a range of wind speed 
values was listed (rather than a single value), an average of the minimum and maximum 
estimates was taken to determine the average wind speed.  In Figure 3.1 each downburst 
occurrence is represented by a marker. White and green markers indicated unknown or average 
wind speeds from 58 ? 74 mph (26 - 33 m/s), yellow is wind speeds of 75 ? 89 mph (34 - 40 
m/s), and red is 90 mph (40 m/s) and above. 
Of the total 58 downbursts in the final dataset listed in Table 3.1, 46 of these downbursts events 
reported wind speed estimates or measurements and the remaining 12 did not.  This 
supplementary dataset without the assumed ?50+ knot? winds was also analyzed using the 
procedures that follow to fit the distribution curves and find the design wind speed and pressure 
for purposes of comparison with the final dataset.  
 
38 
 
 
 
Table 3.1: Downbursts in the WBMA (in order by date) 1996-2009
Date 
Average Wind 
Speed Approx. Location 
m/s mph 
5/4/1996 31.3 70 New Market, VA 
6/4/1996 26.8 60 Green Hill, MD 
6/14/1996 35.8 80 Mannassas City, VA 
6/24/1996 26.8 60 DC 
7/13/1996 26.8 60 Charlotte Hall, MD 
7/14/1996 40.2 90 Lovettsville, VA 
7/15/1996 26.8 60 Benedict, MD 
7/19/1996 26.8 60 Rosemont, MD 
7/30/1996 26.8 60 Paw Paw, WV 
8/16/1996 31.3 70 Bowie, MD 
8/27/1996 26.8 60 La Plata, MD 
6/18/1997 26.8 60 DC 
6/26/1997 40.2 90 Welltown, VA 
8/17/1997 40.2 90 Deale, MD 
5/31/1998 26.8 60 Flinstone, MD 
6/2/1998 26.8 60 Annapolis MD 
6/13/1998 26.8 60 Scottsville, VA 
6/16/1998 26.8 60 Jefferson, MD 
6/30/1998 25.9 58 Ownings Mills, MA 
7/21/1998 35.8 80 Berkeley 
6/14/1999 31.3 70 Oakton, VA 
7/24/1999 26.8 60 Spotslyvania, VA 
8/14/1999 35.8 80 Leesburg, VA 
5/13/2000 35.8 80 Middleburg, VA 
5/19/2000 26.8 60 Greene, VA 
8/7/2000 29.1 65 DC 
8/8/2000 24.1 54 N VA 
5/27/2001 31.3 70 Sterling, VA 
6/6/2001 29.1 65 Shenandoah, VA 
6/12/2001 29.1 65 NE Carrol, MD 
6/29/2001 26.8 60 Severna Park, MD 
8/10/2001 31.3 70 New Baltimore, VA 
8/11/2001 29.1 65 Edgewood, MD 
5/13/2002 33.5 75 Ellicott City, MD 
5/31/2002 24.6 55 NW Spotsyvania, VA 
6/6/2002 31.3 70 Wilderness Corner, VA 
6/27/2002 36.2 81 Sunderland, MD 
8/23/2002 29.1 65 Urbana, MD 
5/7/2003 31.3 70 Warrenton, VA 
5/9/2003 26.8 60 Albernarle, VA 
6/9/2003 34.0 76 DC 
6/13/2003 24.6 55 SE DC 
6/30/2003 35.8 80 Potomac 
7/4/2003 35.8 80 Glenelg, MD 
7/6/2003 31.3 70 Anne Arundel, MD 
7/9/2003 26.8 60 PG County, MD 
8/22/2003 33.5 75 Culpeper, VA 
8/27/2003 31.3 70 Sterling, VA 
7/14/2004 35.8 80 Solomons Island, MD 
7/23/2005 26.8 60 DC/MD 
7/2/2006 31.3 70 Annandale, VA 
7/19/2006 29.1 65 Roanoke, VA 
7/29/2007 31.3 70 Bel Air, MD 
6/3/2008 35.8 80 Potomac, MD 
6/4/2008 40.2 90 Luray, VA 
7/23/2008 38.0 85 Bethel, VA 
6/9/2009 25.5 57 Annapolis, MD 
7/25/2009 31.3 70 New Market, VA 
7/27/2009 35.8 80 Reistertown, MD 
 
  
 
43 
 
 
3.3 METHOD OF VERIFICATION 
To provide some means of verification of accuracy of this dataset this data is compared to a 
Pryor (2009) who compiled a dataset for downburst activity for the eastern United States for the 
years 2001-2003.  Comparing the two datasets for this time period for the WBMA, the average 
number of events was 7 days or 9 events, which is similar to the 9 average downburst events for 
Virginia, West Virginia and Maryland found by Pryor. Considering the coverage area of the 
WBMA does not encompass all of MD, VA or WV, this is determined to be an acceptable 
number.  
In some cases, recorded surface wind data is used to verify a downburst occurrence. As an 
example, on July 2nd, 2006 a downburst occurrence was reported in Annandale, VA.  The closest 
airport that records wind data is Ronald Reagan Washington National Airport (DCA) 
approximately 9 miles from Annandale.  A time versus wind speed graph is shown in Figure 3.2 
showing the effects of this downburst by the spike in the peak wind speed and rapid change in 
wind direction. The reported wind speeds of this downburst were 70 mph. DCA anemometers 
report a maximum wind speed of 65 mph. This is determined to be reasonable due to the distance 
from the airport to the downburst location. 
3.4 DOWNBURSTS CATEGORIZED BY WIND SPEED 
The downbursts of major concern are those with wind speeds above 90 mph (40 m/s) ? the 
design wind speed for buildings in the WBMA (ASCE 2005). The NWS defines a severe 
weather wind event as one with winds above 50 knots (58 mph, 23 m/s), thus wind reports from 
the Storm Event Database are above this first threshold.  Figure 3.3 shows a histogram of the 
 
44 
 
 
frequency of occurrence of downbursts in the WBMA by wind speed from 1996-2009. In this 
figure the average wind speed values are used if a range of wind speed values is reported.  
3.5 FREQUENCY OF OCCURRENCE OF DOWNBURSTS 
With a dataset of 14 years of downburst occurrences and each of their measured or estimated 
wind speeds, it is now useful to fit a distribution curve to this data in order to model the 
occurrence rate of downbursts. Several methods are used in this study and their results were 
compared to determine the best distribution parameters.   
3.5.1 Downburst Wind Distribution Functions 
3.5.1.1 Extreme Value Distributions  
The family of Extreme Value Distributions (EVD) is good candidates for extreme wind 
frequency analysis (Rohan and Dale 1987). Li (2000) and Oliver et al. (2000) proposed models 
for downburst winds for transmission line design, recommending the use of an EVD to model the 
meteorological data.  Studies have found that the Gumbel distribution (Type I EVD) in particular 
models extreme wind distributions well (Cook 1982; Simiu et al. 2001; Rajabi 2008). Holmes 
and Moriarty (1999) analysis of downburst winds in Australia used the Generalized Extreme 
Value Distribution (GEVD) to conclude the Type I EVD was good fit, but conservative.  The 
Weibull distribution (Type III EVD) and Generalized Pareto distribution have also been used to 
model extreme winds (Simiu and Heckert 1996; Holmes and Moriarty 1999; Pandey et. al 2003).  
The basis of the ASCE 7-05 design wind speed maps for the United States is wind speed data 
fitted with a Type I EVD (Peterka and Shahid 1998).  In an effort to compare the design wind 
speeds of ASCE-7-05 with the values determined from the downburst wind speeds the same 
 
45 
 
 
distribution curve is determined. The Type III EVD is also evaluated for comparison of fit to the 
dataset. The Appendix includes further information on these two distributions.  
Downburst Sample Mean and Standard Deviation  
The sample mean and standard deviation must be computed before the EVD data fitting. To 
determine the sample mean g1876 g3365and standard deviation g2026 g3051 g3365 , the following equations are used, 
where m is the total number of downbursts: 
 g1876 g3365 g3404 1g1865 g3533 g1876g3036
g3040
g3036g2880g2869
 (3.1) 
 
 ?g2934 g3404 g3497 1g1865 g3436 1g4666g1865 g3398 1g4667g3440 g3533g4666g1876g3036 g3398 g1876 g3365 g4667g2870
g3040
g3036g2880g2869
 (3.2) 
 
The sample mean and standard deviation for the minimum, average, and maximum wind speed 
estimates are included in Table 3.2.  
The difference in mean values between the dataset using minimum, average, and maximum wind 
speed respectively is not large because only several of the downburst occurrences were reported 
as a range of wind speed estimates. Most downburst occurrence reports only include one wind 
speed estimate, meaning that this number is used in the computation of the minimum, average 
and maximum wind speed values. 
 
46 
 
 
3.5.2 Determination of Parameters to Fit Data to Distribution 
To fit the downburst wind speed data to Type I and III EVDs, parameter estimation is necessary.  
In the creation of the wind map of the ASCE 7-05, wind data was fitted using the Method of 
Moments (MOM), described by Simiu and Scanlan (1986).  The MOM and Maximum 
Likelihood Estimation (MLE) methods are used in this research. The Appendix includes more 
detailed information on these two methods.  
The MOM calculations were computed by hand, and the MLE parameters were calculated using 
the computer software program EasyFit (MathWave 2010). These values are summarized in 
Table 3.3 and Table 3.4. A graph of the average wind speed PDFs is shown in Figure 3.4. 
3.5.3 Goodness-of-Fit 
The Anderson-Darling (A-D) test (Stephens, 1974) is used to test whether the sample data comes 
from a population of a specific statistical distribution.  It is a modification of the Kolmogorov-
Smirnov (K-S) test, and uses a specific distribution to calculate its critical values.  The K-S test 
is an alternative to the A-D test, but cannot be used if the distribution being tested has parameters 
estimated from the sample data. The Appendix includes more detailed information on the A-D 
and K-S tests. 
The A-D test shows that the Type III EVD provides a better fit for the data because the 
difference between the critical value and the A2 statistic is less for the Type III EVD than the 
Type I EVD. The A-D test also indicated that in both cases the MLE method provided better 
fitting data. Thus the Type III EVD is chosen using the parameters estimated from the MLE 
methodology. 
 
47 
 
 
When the distributions are fit to the data limited only to only the events with measured or 
estimated wind speed values (rather than ?50+ knots?), the distributions provided better fits to 
the data than when the unreported wind speed values are included. The parameters estimated 
using MOM and MLE for Type I and III EVD, however, are not very different between the two 
datasets. Like the dataset that includes the 12 downburst events with wind speeds that were not 
reported, the MLE method for the Type III EVD provides the best fit for the data. 
  
 
48 
 
 
 
Table 3.2: Mean and Standard Deviation of Downburst Wind Speeds  
 Low  Average High  
Mean (g2206 g3365) (mph) 68.05 68.83 69.86 
Standard Deviation (g2763g1824) 9.83 10.02 10.79 
 
 
Table 3.3: Type I Extreme Value Distribution Parameters by Method of Moments 
Wind Speed Estimate g2235 g2246 
Low 0.130 72.474 
Average 0.128 73.337 
High 0.119 74.717 
 
Table 3.4: Type I Extreme Value Distribution Parameters by Maximum Likelihood Estimation 
Wind Speed Estimate g2235 g2246 
Low 0.130 73.63 
Average 0.128 74.33 
High 0.119 75.01 
 
 
Table 3.5: Type III Extreme Value Distribution Parameters by Method of Moments 
Wind Speed Estimate g2235 g2237 
Low 8.23 72.16 
Average 8.17 73.02 
High 7.67 74.34 
 
Table 3.6: Type III Extreme Value Distribution Parameters by Maximum Likelihood Estimation 
Wind Speed Estimate g2235 g2237 
Low 7.13 72.42 
Average 7.12 73.27 
High 6.62 74.59 
 
 
49 
 
 
 
3.5.4 Final Downburst Distribution 
Given the parameters of the Type I and III EVD that are estimated, the following functions 
represents the probability density and cumulative distribution functions (PDF and CDF 
respectively) for the average wind speed values using the values determined by MLE: 
Type I EVD:  
 g1858g4666g1874g4667 g3404  0.128 g1857g2879g4666g3049g2879g2875g2872.g2871g2871g4667g2868.g2869g2870g2876g1857g2879g3032g3127g4666g3299g3127g3123g3120.g3119g3119g4667g3116.g3117g3118g3124  (3.3) 
 
 g1832g4666g1874g4667 g3404  1 g3398 g1857g2879g3032g4666g3297g3127g3123g3119.g3119g3119g3123g4667g3116.g3117g3119 (3.4) 
Type III EVD:  
 g1858g4666g1874g4667 g3404  7.1273.27g2875.g2869g2870  g4666g1874g4667g4666g2875.g2869g2870g2879g2869g4667g1857g2879g4672 g3049g2875g2871.g2870g2875g4673
g3123.g3117g3118
 (3.5) 
 
 g1832g4666g1874g4667 g3404  1 g3398 g1857g2879g4672 g3049g2875g2871.g2870g2875g4673g3123.g3117g3118 (3.6) 
Where f(v) is the PDF of downburst wind speeds and v is the wind speed in mph. F(v) is the CDF 
of downburst wind speeds, which provides the probability that a downburst wind speed is less 
than or equal to a given v.  
These functions are now used in a stochastic model to determine the design wind speed for the 
WBMA due to downburst wind loading for comparison with the ASCE-7-05 50-year design 
wind speed.  
 
50 
 
 
3.6 DOWNBURST DESIGN WIND SPEED MODEL 
A value for the design wind speed for downbursts in the WBMA is needed to compare to the 
wind load capacity of BIPVs in buildings. For the WBMA a 50-year design wind speed of 90 
mph (40 m/s) is specified in ASCE 7-05. The goal is to determine where a downburst 50-year 
design wind speed falls in relation to this value. A stochastic model that includes the CDF of the 
downburst wind speeds is used. 
3.6.1 Assumptions: 
ASCE 7-05 makes several assumptions when creating the map of the design wind speeds for the 
U.S. These same assumptions are assumed in this study: 
? The wind speeds determined for a particular area are for a 50-year, 2 percent probability 
of exceedence. 
? No corrections to the data are made for terrain roughness upwind of the anemometer site 
and wind direction is not accounted for (Peterka and Shahid 1998).  
? The output values for wind speed represent 3-second gust wind speeds in miles per hour 
at 33 feet above ground for Exposure C (terrain with scattered obstructions having 
heights of generally less than 30 ft ? e.g. flat open country, grassland). 
3.6.2 Downburst Model: 
A stochastic model for downburst wind speeds is proposed by Li (2000). This model is derived 
from the Poisson Pulse Process, explained in the Appendix.  The final equation for the 
exceedence probability, g1868g4666g1872g4667, over the threshold wind speed value v, of a set of wind data due to 
downbursts is given as: 
 
51 
 
 
 g1868g4666g1872g4667 g3404 g1833g2914g2912g4666g1874g4667g2019g1872 (3.7) 
Where g2019 is the occurrence rate of downbursts per year, g1872 is time, and v is equal to Vd , the design 
wind speed of the structure.  g1833g3031g3029g4666g1874 g4667 g3404 1 g3398 g1832g4666g1874g4667g2011  , where g1832g3031g3029g4666g1874g4667 is the CDF of downburst 
wind speeds (mph), modified by the probability of an affected area being hit (g2011g4667. Thus 
  g1832g3031g3029g4666g1874g4667 g3404 g1832g4666g1874g4667g2011, where g1832g4666g1874g4667 is the CDF of the downburst wind speed (mph) determined in the 
previous section and g2011 is the modifier, or strike factor.  When a downburst occurs, it may or may 
not necessarily strike a building, thus g2011 is included as a way of taking this into account.  
3.6.3 Discussion on Strike Factor: 
The probability that a downburst strikes a building can be determined  by using geometrical 
probability and introducing a reference area (Li 2000). This can be calculated in several different 
ways; three methods are discusses in this research.  
3.6.3.1 Method 1: Point Probability  
This methodology considers both the size of the downbursts and the total area which the 
downbursts can strike (i.e. the reference area). The original model (Thom 1963) of point 
probability was created to represent the probability of a tornado hitting a certain point within a 
given area using the formula: 
 g1842 g3404 g1878g1191g1827 g2019 (3.8) 
Where g1878g1191 is the mean path area of the tornado in square miles, g2019 is the number of tornados per 
year and A is the reference area and P is the probability of a tornado striking a point in any year. 
Removing the occurrence rate g2019, the strike factor g2011 in this model can be defined as: 
 
52 
 
 
 g2011 g3404 g1878g1191g1827 g3404 mean path areaWBMA area  (3.9) 
or the mean path area over the total area of interest.  
Since 1963, this model has been modified through the incorporation of stochastic models, 
intensity path area relationships, and the effects of structure size. Twisdale and Dunn (1983)  
modified this model to account for the size of a tornado relative to the size of a building it could 
hit.  However, since the diameter of a downburst (as compared to a tornado) is typically 
significantly larger than that of a building, this model is less practical for downbursts. Choosing 
the size of the building to be infinitesimally small compared to the diameter of the downburst (i.e 
the building is represented by a point), the Twisdale and Dunn (1983) model becomes the same 
model of Thom (1963).  
To achieve a similar equation for this research all downburst events and approximating the 
downburst diameters g4666g1849g3031g3029g4667g2919, and path lengths g4666g1838g3031g3029g4667g2919 for the ith downburst are summed, then 
divided by the number of total downbursts n. and the total reference area A. This yields: 
 g2011 g3404 ? g4666g4666g1849g3031g3029g4667g2919 g4666g1838g3031g3029g4667g2919g4667
g3041
g3036g2880g2869
n Ag3415  (3.10) 
 A similar form of the simplified form of the equation above is used in Schaefer et al. (1985) to 
determine tornado risk. The actual length g4666g1849g3031g3029g4667g2919 and width g4666g1838g3031g3029g4667g2919 of the ith downburst path is 
not likely to be known, but if an average value for all downburst lengths and widths is assumed, 
the above equation is changed to: 
 g2011 g3404 g4666g1838g3031g3029g3364g3364g3364g3364g3364g4667g4666g1849g3031g3029g3364g3364g3364g3364g3364g4667g1827  (3.11) 
g1838g3031g3029g3364g3364g3364g3364g3364 and g1849g3031g3029g3364g3364g3364g3364g3364 represent the average values of the downburst path dimensions.  Given a downburst 
duration  g1872g3031g3029g3364g3364g3364g3364 and a speed  g1874g3031g3029g3364g3364g3364g3364g3364 an approximation of the path length can be determined. Thus: 
 
53 
 
 
 g2011 g3404 g4666g1874g3031g3029g3364g3364g3364g3364g3364g1872g3031g3029g3364g3364g3364g3364g4667g4666g1849g3031g3029g3364g3364g3364g3364g3364g4667g1827  (3.12) 
A microburst by definition, last approximately 2-5 minutes (macroburst last longer). The 
translational velocity of a downburst can be about 3 times that of a thunderstorm (Holmes and 
Oliver 2000; Savory et al. 2001).  A study of dry downburst in Colorado observed speeds of 22 ? 
45 mph (10-20 m/s) (Hjelmfelt 1988).  A speed of 45 mph (20 m/s) was assumed by Holmes and 
Oliver (2000).  The reference area A can be determined as the total reference area of the WBMA- 
18896 mi2 (48940 km2) (NWS 2006) following the assumptions made by Li (2000) and Holmes 
(2000). 
Using this method, and assuming the downburst travels at a speed of 45 mph (20 m/s) and lasts 
for 3-minutes (average value from previous studies e.g. Fujita 1985), and also assuming the 
width of the downburst to be 2.5 miles (4 km), g2011 can be calculated as: 
g2011 g3404 g4666g1874g3031g3364g3364g3364g1872g3031g3365 g4667g4666g1849g3031
g3364g3364g3364g3364g4667
g1827 g3404
g468445 mph g1499 3 min
60 minhr
g4685 g46662.5 milesg4667
18896 mig2870 g3404 g2779. g2786g2784 g1824g2778g2777
g2879g2781   
This methodology introduces many variables of which the values are not explicitly known in this 
research. With thorough data that includes information such as the downbursts? translational 
velocities, diameters, and durations this method could be more accurate than other methods. 
However, using the limited dataset generated in this research this number may misrepresent the 
downbursts of the WBMA. This method yields the lowest design wind speeds of the three 
methods proposed.  
 
54 
 
 
3.6.3.2 Method 2: Urban Area Ratio 
A second methodology considers the ratio of the urban areas g1827g3048 in the WBMA to the reference 
area g1827, the total area of the WBMA: 
 g2011 g3404 g1827g3048g1827 g3404 g1847g1870g1854g1853g1866 g1853g1870g1857g1853g1849g1828g1839g1827 g1853g1870g1857g1853 (3.13) 
Urban areas are likely to be the main locations in which medium to high rise buildings are 
constructed. In the event of a downburst, these areas are likely the only locations where building 
fa?ade damage would be of concern. The total urban area g1827g3048 is determined as: 
 g1827g3048 g3404 g3533g4666g1827g3048g4667g3036
g3041
g3036g2880g2869
 (3.14) 
Where g4666g1827g3048g4667g3036 is the area of the ith urban location in which buildings are located, and n is the 
number of urban locations.  
Washington D.C. and Baltimore represent the main urban areas in the WBMA. Taking the areas 
inside the Washington and Baltimore beltways (I-495 and I-695), approximated from using the 
total length of each highway (64 miles and 51 miles respectively) as the approximate 
circumference of a theoretical circle surrounding each urban area (FHA 2010): 
g2011 g3404 g1827g3048g1827 g3404 g1827g3003g3028g3039g3047g3036g3040g3042g3045g3032 g3397 g1827g3024g3028g3046g3035g3036g3041g3034g3047g3042g3041 g3005.g3004.g1827 g3404 207 mi
g2870 g3397 326 mig2870
18896 mig2870 g3404 g2779. g2785g2780g1824g2778g2777
g2879g2779 
When used in the wind speed model equation, this method yields the middle range of wind 
speeds.  It is less conservative than Method 3, but yields higher values than Method 1. 
3.6.3.3 Method 3: Conservative Assumption 
Li (2000) verifies his downburst wind speed model by assuming that if a downburst occurs, it  
will hit a structure.  Design wind codes for transmission lines in Australia assume that this is the 
 
55 
 
 
case. Li (2000) showed that with this assumption, the model produces design wind values 
consistent with the design code. Using this assumption:   
g2011 g3404 1 
This assumption is simple in nature, and is the most conservative. It yields the highest downburst 
occurrence probabilities because the probability of a downburst occurring is not modified from 
the raw data.  
3.7 50-YEAR AND 100-YEAR DOWNBURST DESIGN WIND SPEED 
The three methods discussed above provide a lower, middle and upper bound for the possible 
values of the strike factor.  Using these values the 50- and 100-year design values for buildings 
in the WBMA are computed and summarized in Table 3.8, assuming the values in  
 
 
  
 
56 
 
 
Table 3.7.  
The exceedence probability g1868g4666g1872g4667 over the threshold wind speed value v is: 
 g1868g4666g1872g4667 g3404 g1833g2914g2912g4666g1874g4667g2019g1872 (3.15) 
Substituting g1833g3031g3029g4666g1874g4667 g3404 1 g3398 g1832g4666g1874g4667, modified by g2011, and the design wind speed uni0056g3031 for v, the final 
equation for probability of exceedence becomes: 
 g1868g4666g1872g4667 g3404 g46701 g3398 g1832g3031g3029g4666uni0056g3031g4667g2011g4671g2019g1872 (3.16) 
By definition the return period R is equal to the reciprocal of the annual exceedence probability 
(i.e.  1/ g1868g4666g1872g4667), thus to determine the design wind speeds for a return period of 50 or 100 years, 
and substituting t=1, the follow equation is derived: 
 g1844 g3404
1
g46701 g3398 g1832g4666uni0056g3031g4667g2011g4671g2019 g3404
1
g4680g1857g2879g4672
g2906g3279
g2875g2871.g2870g2875g4673
g3123.g3117g3118
g4681 g2019g2011
 (3.17) 
Solving for Vd, the equation becomes: 
 uni0056g3031 g3404 73.27 g3428g3398ln g3436 1
g2019g2011g1844g3440g3432
g2869 g2875.g2869g2870g3415
 (3.18) 
The values are substituted in to this equation to yield the equation for the 50-year 
exceedence probability are summarized in  
 
 
  
 
57 
 
 
Table 3.7. Using these values the 50- and 100-year design wind speeds are summarized in Table 
3.8.  Assuming a strike factor of 1 the design wind speeds exceed the ASTM-7-05 design wind 
speeds. Using the other two strike factors, both are below the 90 mph design wind speed.    
 
 
 
 
  
 
58 
 
 
Table 3.7: Parameter Values for Determining Downburst Design Wind Speeds 
 
Variable Value 
R 50, 100 years 
g2162g4666g1796g2186g4667 1 g3398 g1857g2879g4672 g2906g3279g2875g2871.g2870g2875g4673g3123.g3117g3118 
g2245 59 g1856g1867g1875g1866g1854g1873g1870g1871g1872 g1856g1853g1877g1871 14 g1877g1870g1871 g3404 4.21 
g2202 1 g1877g1857g1853g1870 
 
 
 
 
 
Table 3.8: 50- and 100-year Downburst Design Wind Speed (mph) at 33 ft 
Strike Factor 
(g2237g4667 
Design Wind Speed (mph) 
50-year 100-years 
0.00297 --* 59.36 
0.0283 79.48 83.23 
1 92.73 95.19 
 
* The design wind speed with this strike factor is too low to be calculated given the CDF used to 
calculate it. 
 
  
 
59 
 
 
3.7.1 Conversion to Wind Pressure  
 To convert the design wind speeds (mph) to pressure (lb/ft2), the equation from the ASCE-7-05 
code (2005) for velocity pressure (Equation 6-15 in the code) is used: 
 g1869g3053 g3404 0.00256g1837g3053g1837g3053g3047g1837g3031g1848g2870g1835 (3.19) 
Where g1837g3031 is the wind directionality factor, g1837g3053 is the velocity pressure exposure coefficient, g1837g3053g3047 
is the topographic factor, and I is the importance factor. An exposure level of C is chosen here to 
be consistent with methodology used in the creation of the ASCE 7-05 design wind speed maps 
(Peterka and Sohban 1998).  
g1878g3034 and g2009 are values that are dependent on this exposure level.  For exposure level C, g2009 g3404 9.5 and  
g1878g3034 g3404 900uni002C based on the values in Table 6-2 of ASCE 7-05 (2005).   
g1837g3053 varies with height z above the ground. To be consistent with the methodology used in the 
creation of the ASCE 7-05 design wind speed maps (Peterka and Sohban 1998) and the 
anemometer height at which the wind speed values were obtained in this study, a height of 33 ft 
(10 m) is used.  
 
 
Table 3.9 defines the assumed values for the variables in Equation 3.19 for the velocity pressure 
calculations. The resulting velocity pressures are calculated (Table 3.10). These values will be 
used later on to calculate failure probability of BIPV panels by comparing to the strength data of 
fa?ade-installed BIPVs. 
 
  
 
60 
 
 
 
 
 
Table 3.9: Assumed ASCE 7-05 Velocity Pressure Variables 
Variable Value ASTM 7-05 Reference 
g1837g3031 .85 Table 6-4 
g1837g3053g3047 1 Section 6.5.7.2 
g1837g3053 2.10g3435g1878 g1878g3034uni2044 g3439g2870 g3080uni2044  Table 6-3 
g1835 1 Section 6.5.5 
 
 
 
 
Table 3.10: Design Velocity Pressure (lb/ft2) at 33 ft by Strike Factor 
Strike Factor 
(g2237g4667 
Design Velocity Pressures (lb/ft2) 
50-year 100-years 
0.00297 -- 8.02 
0.0283 14.39 15.78 
1 19.59 20.64 
 
  
 
 
 
Figure 3
0
10
20
30
40
50
60
70
2100 P.M. 2150 P.M.
W
ind
 Sp
eed
 (m
ph
)
Figure 3.2: A time history of the July 2
.1: Downbursts in the WBMA, 1996-2009 
2200 P.M. 2250 P.M. 2300 P.M.
Wind Speed (mph)
Time
nd 2006 Annandale, VA downburst 
61 
 
 
 
 
62 
 
 
0
5
10
15
20
50 60 70 80 90 100
Do
wn
bu
rst
 C
ou
nt
Wind Speed (mph)   
Figure 3.3: Downbursts by Wind Speed (mph) in WBMA from 1996-2009 (1 mph = 0.45 m/s) 
 
0
0.01
0.02
0.03
0.04
0.05
40 50 60 70 80 90 100 110 120
Type I EVD (MOM)
Type I EVD (MLE)
Type III EVD (MOM)
Type III EVD (MLE)
Pr
ob
ab
ilit
y 
Wind Speed (mph)  
 
63 
 
 
Figure 3.4: Type I and III PDFs for Downburst Wind Speeds in the WBMA using method of 
moments (MOM) and maximum likelihood estimation (MLE) 
0
0.2
0.4
0.6
0.8
1
50 60 70 80 90 100
Ordered Downburst Data
Type I EVD
Type III EVDC
um
ula
tiv
e %
Wind Speed (mph)  
Figure 3.5: Type I and III EVD CDFs fitted to WBMA Downburst Data from 1996-2009 
  
 
64 
 
 
 
CHAPTER 4: BUILDING INTEGRATED PHOTOVOLTAICS 
(BIPV) UNDER UNIFORM LOADING 
Since the first discovery of a photovoltaic (PV) cell capable of converting sunlight to useable 
energy in the late 1800s, PV technology has come a long way in terms of efficiency and wide-
spread use.  Beginning with the development and commercialization of crystalline silicon (c-Si) 
PVs in the 20th century, research focuses in PV technology today have turned more towards thin-
film PVs (TFPVs), which require much less raw materials to produce than c-Si.  It is these 
TFPVs that are being produced and installed in the facades of high-rise buildings to function 
both as a source of power and the building envelope of the building, an application known as 
building integrated photovoltaics (BIPVs). As this technology has grown in popularity, various 
standards have been implemented in different countries to regulate the quality of BIPVs, 
however the potential effect of strong downburst wind-induced loads has not been assessed for 
fa?ade-installed BIPVs.  To understand this potential impact, a review of the most common 
TFPV BIPV technologies, their components and material properties is conducted to determine 
possible ways to assess their behavior under extreme wind loading.  Based on this assessment, a 
model known as the Glass Failure Prediction Model (GFPM) was chosen to use to model the 
behavior of TFPVs under uniform wind loading and then compared with test data to provide 
verification of the model. 
 
65 
 
 
4.1 THE ORIGIN OF PHOTOVOLTAICS 
The idea of photovoltaics first originated in the early 1800s when a French physicist, Alexandre-
Edmond Becquerel, first observed what he called the photovoltaic (PV) effect.  In 1883, Charles 
Fritts, an American inventor, constructed a 1% efficient fully-functioning solar cell (Fritts 
1883).  Many years later in 1954 a crystalline silicon (c-Si) p/n junction was accidentally 
discovered, which generated voltage when exposed to light (Chapin et al. 1954).  This new 
technology was about five times more efficient than Fritts?, however at a high cost of production, 
PVs were not ready for wide-scaled use.  By the 1960s key papers such as those published by 
Prince (1955), Loferski (1956) and Wysocki et al. (1960) among others had developed the basis 
of design of c-Si solar cell operations.  Enough was understood about their design to enable the 
production of PV cells to provide satellites of the 1950s and 60s with sufficient power to sustain 
themselves in space.  
Two pivotal events occurred in 1973 that helped spur increased interest in PV technologies.  
First, the discovery of what is called a ?violet cell? increased the efficiency of even the most 
state-of-the-art c-Si technologies of the time by 30% (Lindmayer 1973).  Second, the U.S. 
Department of Energy was formed under President Nixon?s direction to, in part, support PV and 
other renewable energy technologies.  This renewable energy support was further fueled by the 
world?s first oil embargo later that year.  The decade that followed brought about the maturity of 
the industry of c-Si PVs.  Large solar power companies and manufacturers across the U.S., Japan 
and Europe were established and the start of ?pilot? PV lines to be used for commercial and 
residential applications. Efficiency of c-Si panels reached 20% in 1985 by researchers in 
Australia (Green 1995) and has continued to increase in efficiency slowly over the past 20 years 
 
66 
 
 
(Figure 4.1Error! Reference source not found.Error! Reference source not found.). Today 
most of solar installations are mono- or poly- c-Si PVs, representing an approximately 80-90% 
market share.  Yet as the market continues to grow, the focus of researchers is turning more 
towards the development and improved efficiency of thin-film photovoltaic technologies 
(TFPVs), a potentially lower cost approach for solar power. Further reduction of PV module 
costs is conditional on the availability of raw materials worldwide, meaning TFPVs which 
require lesser quantities of the PV materials are more economically favorable.  C-Si 
technologies, while the cornerstone of the establishment of the photovoltaic industry, has been 
deemed by some more as a technology of today, rather than of the future.   
4.2 DEVELOPMENT OF THIN-FILM PHOTOVOLTAIC (TFPV) 
TECHNOLOGIES  
TFPVs, the second generation PVs after the more common c-Si PVs, have followed a similar 
path of development to that of c-Si PVs, only a few decades later in terms of efficiency and use.  
Research efforts focused on TFPVs began in the 1950s in the U.S. with the publication of results 
of approximately 6% efficiency by Reynolds et al. (1954).  By the 1980s, a-Si (amorphous 
silicon), CIS/CIGS (copper indium gallium selenide), and CdTe/CdS (cadmium telluride) TFPV 
technologies had achieved efficiencies of over 10% during small-scale lab testing, but the 
difficulty of achieving this efficiency for larger areas (larger than approximately 1 cm2) had not 
been overcome. Complications with the scaling up of TFPVs limited their use to powering small 
electronics such as calculators at the time.  In the early 1990s new technologies were developed 
and patented that allowed for larger scale TFPV technologies to be implemented.  As an 
 
67 
 
 
example, BP Solar, a British electric company, acquired patents to electrodeposition of thin-film 
CdTe in 1989, and 10 years later purchased Solarex who focused on a-Si technology.  This 
established BP Solar as one of the leading TFPV solar companies until 2002 when BP Solar 
drew out of all but the c-Si market.  The Japanese PV industry became large contenders 
throughout the 1990s as intensive research on TFPVs led to new device designs, and improved 
material processing (Hegedus and Luque 2002).  
Research and development in TFPVs has continued (Figure 4.1), in part, because it has been 
recognized that the silicon (Si) crystals required to for c-Si technologies are expensive and slow 
to grow compared to thin-film technologies.  Si wafers are nearly 100 times thicker than TFPVs 
(~1-3?m vs 100 ?m).  Today the solar industry continues to develop, but at a faster pace.  Many 
start-up companies have established themselves within the past decade in the U.S. in attempt to 
take advantage of the growing market for PVs, many in c-Si, and others in TFPVs.  While still 
relatively few companies in the United States market TFPVs for commercial applications, the 
thin-film industry continues to take shape through BIPV applications world-wide.  
4.3 BIPV: COMBINING TFPVS AND ARCHITECTURE 
BIPVs, a relatively new area of increased research interest in the field of PVs in recent years, can 
be defined in a broad sense as PV components that replace the look and function of a primary 
building material (Strong 1996). BIPV products, such as TFPVs, act as a weather resistant skin 
of the building, placed often times on the facades or roofs of a building.  The advantage of this 
system over that of more traditional applications of PVs is that no additional land area is needed, 
and the PVs are integrated into the building itself rather installed separately. By simultaneously 
acting as the building envelope material and a power generator, the system provides savings in 
 
68 
 
 
material and electricity costs.  BIPVs do not always come in the form of solar panels, for 
example some companies now produce solar shingles ? or flexible PVs that are shaped like 
traditional shingles. However for fa?ade applications framed solar panels are more common. 
4.3.1 Placement of BIPVs on Buildings: 
BIPVs can function as a building material of several different parts of a building including: 
? Fa?ade  
? Roof  
? Solar shade (for windows or doors) 
? Skylights and windows 
 
In the case of extreme horizontal wind loading, the fa?ade of building will experience these wind 
forces directly, thus fa?ade applications only are considered in this research. For facades, there 
are two distinct applications in which PVs are integrated into a building ? these are (Figure 4.2): 
? Curtain walls 
? Rain screens 
Rain screens are a double layered building envelope which includes an outer layer that sheds 
most precipitation and an inner layer which handles the rest. The water that does penetrate the 
outer layer of this open-joint system is drained through a cavity behind it. In this system the PV 
panels are installed as a part of the outer layer (Figure 4.2a), however this is a less common 
method of utilizing PVs in facades. The focus of this research is on curtain wall applications of 
BIPVs, thus while rain screen application of BIPVs exist, they are not discussed in detail. 
A curtain wall is typically a single layer, aluminum framed wall constructed of in-fills of glass or 
other materials. The BIPV components are usually installed as in-fills between the vertical and 
horizontal aluminum members (Figure 4.2b). A curtain wall does not act as a structural support 
 
69 
 
 
for a building ? thus all its components must only support their own dead load and any external 
forces that are placed on it.   
Understanding how the PV panels are most commonly installed in building facades is important 
in understanding the boundary conditions of the PV panels being subjected to horizontal 
downburst winds.  In traditional curtain wall as well as in BIPV curtain walls most in-fill panels 
are supported by the horizontal and vertical aluminum mullions, creating 4-sided, continuous 
support conditions.   
4.4 TYPES OF TFPV PANELS USED IN BIPV APPLICATIONS AND 
THEIR MATERIAL PROPERTIES 
To better characterize BIPV behavior in fa?ade applications under wind loading, an 
understanding of the construction and material properties of different types of TFPVs used in 
BIPV fa?ade applications is needed. This section discusses the basics of TFPV technologies, 
specifically three TFPVs available on the market for use in BIPV applications: 
(a) Amorphous Silicon (a-Si) 
(b) Copper-Indium-(Gallium)-Selenide (CI(G)S)  
(c) Cadmium Telleride (CdTe)  
A-Si being the most common technology, with CI(G)S and CdTe in second and third 
respectively (Table 4.1Error! Reference source not found.Error! Reference source not 
found.). These three technologies are available most commonly as single layer PV panels.  
Laboratory testing has also shown that increased efficiency can be obtained using tandem and 
triple cells, in which multiple layers of TFPV are applied on top of one another, yet their large 
 
70 
 
 
scale use is still in development.  Similarly other PV materials such as organic cells, 
metamorphic Ga(In)As, GaInP, and other multi-junction TFPVs have had success in research 
laboratories but have not yet been fabricated into large-scale PV panels. Those technologies that 
have succeeded in the PV market thus far can be manufactured with sufficient efficiency in large 
scale PV applications. 
While each PV technology employs slightly different layering structure and thicknesses, the 
basic elements of a TFPV cell are the same.  Sunlight falls on the top surface of the solar cell 
where it penetrates the front glass substrate to electrical contacts (e.g. typically a type of TCO ? 
transparent conducting oxide).  These metal oxides allow light photons to penetrate and be 
absorbed and converted to electrical charges by several layers of semiconductors (n-, (i-) and p-
type).  A back layer of TCO forms the second electrical contact, followed by a layer of polyvinyl 
butyral (PVB), and finally second layer of glass.  These layers within a PV panel are important to 
know and characterize because it is these layers that feel the stresses under high wind loading.  . 
 
Table 4.1: TFPV panel producers in the world market (ENF 2010) 
Solar 
Technology 
Companies Producing 
Solar Panels 
a-Si 131 
CI(G)S 30 
CdTe 4 
 
 
4.4.1 Amorphous Silicon (a-Si) (and Micromorphous Silicon) 
Origin:  With advancements in processing and fabrication methodologies of c-Si PVs, and 
further silicon research, a-Si solar cells were discovered (Chittick 1969) and constructed 
 
71 
 
 
(Carlson 1977), using only about 1% of the silicon needed for typical c-Si cells. Today many of 
the commercial manufacturers and vendors of TFPV panels market a-Si technologies. According 
to ENF, there are 131 current companies producing a-Si products as of the year 2010.  A short 
list of these manufactures is: Astroenergy, Auria, Bankok Solar, UniSolar, Dupont, ENN, EPV 
Solar, FlexCell, Global Solar, Inventux, Kaneka, Mistuishi, Polar PV, Sinonar Solar, Solar Plus, 
Solem, Sunfilm, SunGen, SunStar, TerraSolar, and Tianjin Jinneng Solar. The majority of these 
companies are not U.S.-based.  
Efficiency: When a-Si was first discovered, the efficiency was that of only about 1% (Morel 
1978), however the overall efficiency of a-Si is higher today, but is still significantly less than c-
Si (around 6 - 10%).  Tandem and triple-layer cells called micromorph silicon solar cells in 
which a-Si and very thin layers of crystalline silicon (?c-Si) are deposited on top of one another 
have shown significantly higher efficiencies.   
Structure and Material Properties: Unlike c-Si whose atomic arrangements are regular 
crystals, a-Si has an irregular arrangement of atoms ? allowing approximately 2.5 times more 
light to be absorbed, but less efficiently. Because it is not crystalline in nature and due to its 
extreme thinness, when flexed it will not break ? allowing it to be deposited on to flexible or 
non-flexible substrates. The substrates used are numerous, including ceramics, metals, steel or 
plastics, however for fa?ade applications of TFPVs, a-Si is typically deposited on to glass.  As 
pointed out by Madou (2002), very little is known about the mechanical properties of a-Si itself. 
Its modulus of elasticity is estimated to be 80 +/- 20 GPa (Freund and Suresh 2003). 
To improve efficiency some manufacturers of a-Si products have begun to manufacture 
micromorphous silicon module technologies using multiple layers of silicon such as amorphous 
 
72 
 
 
and a thin layer of crystalline silicon. Several manufacturers? diagrams of their a-Si solar cells 
are shown in Figure 4.3 ? 4.5.  
4.4.2 Copper-Indium-(Gallium)-Diselenide (CI(G)S)  
Origin: The first thin-film CI(G)S PV was fabricated by Kazmerski et al. (1976) around the 
same time as a-Si technologies were being developed.  The first large-scale production of CI(G)S 
PV modules began in 2006 by the Germany company Wurth Solar. Among some of the CI(G)S 
PV panel manufacturers today are: Ascent Solar, Avancis, Dupont, Global Solar, ICP Solar, 
Nanosolar, SoloPower, Solar Frontier, Global Solar and Wurth Solar. Of these many market 
thin-film flexible laminates rather than CIGS solar panels.  
Efficiency: Similar to other TFPV, CI(G)S is far more efficient per unit of material than c-Si ? a 
CI(G)S  film approximately 1 micron thick has the potential to generate a similar power to that 
of a c-Si with a 200 - 300 microns thickness. The maximum efficiency achieved in small area PV 
laboratory testing as of August 2010  is 20.3% (Stuar 2010). The efficiency of larger scale solar 
panels currently average around 13%.  This large discrepancy between lab-scale cells and full-
scale modules is largely due to the difficulties in the development of new and complex 
manufacturing processes that produce CI(G)S cells.  
Structure and Material Properties: Unlike a-Si, CI(G)S TFPVs are crystalline in structure. Of 
the research published concerning CI(G)S under stresses, a study by Chen (1992) discusses that 
substrates of similar thermal expansion coefficients to CI(G)S should be chosen so that when the 
two layers cool from about 500?C, the thin-film will not experience increased tensile or 
compressive stress. Stressed thin-films can exhibit voids and microcracking under tension or 
adhesion failure under compression (Chen 1992). When searching for material properties of 
 
73 
 
 
CI(G)S panels, the most common results for the properties listed below were not for the CI(G)S 
itself, but for the glass specifically designed to be used in the CI(G)S PV panels. As an example, 
AGC Solar, give properties of its glass that is specifically designed for CI(G)S applications 
(AGC 2010). Several manufacturers? diagrams of their CI(G)S panels are shown in Figure 4.6 
and Figure 4.7. 
4.4.3 Cadmium Telleride (CdTe) 
Origin: The first CdTe crystals were synthesized in 1947 by Frerichs (1947), then proposed as a 
possible PV technology in 1956 (Jenny 1954; Kruger 1955; Loferski 1956; de Nobel 1959). Two 
types of CdTe cells emerged, homojunctions (single layer) and heterojunctions (multiple layers), 
however nearly all research focus has concentrated on heterojunction PV cells since the 1960s.  
CdTe has been used in both the p- and n- layers of this TFPV technology, however CdTe as the 
p-layer with stable oxides such as ITO, ZnO, SnO2 and CdS as the n-type layer (e.g. CdTe/ZnO) 
has emerged as most popular. All CdTe cells, CdTe/CdS, which can be manufactured by 
deposition onto either the glass superstrate or substrate, have similar efficiencies. A brief list of 
CdTe TFPV manufacturers are:  Abound Solar, Calyxo, First Solar, MiaSole, Q-cell and 
Primestar. 
Efficiency: Present day small CdTe modules are approximately .5 to 1 m2 in area with efficiency 
being around 10-15%. CdTe cells by MiaSole were recently confirmed to produce efficiencies of 
13.8% in 1 m2 panel, and Solibro achieved an efficiency of 13% (Cheyney 2010). NREL has 
achieved efficiencies of 19.9% in laboratory testing of small CdTe cells (McMahon 2008).  One 
of the main difficulties that manufacturers face with CdTe PV is similar to other TFPV 
technologies, namely, maintaining efficiency on module-sized scales. 
 
74 
 
 
Structure and Material Properties: Flexible CdTe solar cells on polymer substrates have been 
successfully produced at reasonable efficiencies, indicating that flexibility is possible (Romeo et 
al. 2005). Several manufacturers? diagrams of their CdTe solar cells are shown in Figure 4.8 and 
Figure 4.9. 
In summary, each of the three aforementioned TFPV technologies used for BIPV applications 
represents a unique technology discovered and developed for the commercial market to convert 
photons of light from the sun into useable power.  In order for these modules to be available in 
Europe each module must pass a set of standard tests, include one for wind loading (IEC 61646 - 
Thin-film terrestrial photovoltaic (PV) modules ? Design qualification and type approval), 
however in the U.S. solar PV panels, modules, and components are not currently required to 
undergo third-party testing to back up manufacturer claims of durability, quality, or reliability. 
ASTM, however, is working towards a similar standard to that of IEC 61646 in Europe. As a part 
of the qualifications of IEC 61646, TFPV panels must pass a mechanical load test requiring the 
panels to withstand 50.2 psf (2400 Pa) of uniform pressure applied on the front and back for 1 
hour duration each (IEC 2008). However, even with these qualification standards, there are very 
few published studies on the behavior and properties of TFPV panels under uniform loading 
conditions. Instead in several cases the strength and behavior of the glass used on the front and 
back parts of the panels as a way to characterize the TFPV panel strength (Webb 2010). Thus in 
an effort to characterize the behavior of TFPVs, the properties and behavior of glass lites of 
similar structure and dimensions are reviewed and used in this study.  
 
75 
 
 
4.5 MATERIAL LAYERS WITHIN PV PANELS: 
Each of the TFPVs has a similar construction of layers, primarily comprising of two glass lites, 
adhesive (e.g. PVB, EVA), front and back contacts (TCO or metal), and the PV layer(s) (see 
Figure 4.3, Figure 4.6, Figure 4.8). In modeling the behavior of TFPV panels under wind loading 
it is important to review the behavior of these, most importantly the glass superstrate and 
substrate since these are subject to the most compressive and tensile forces under flexural 
loading due to their location on the exterior surfaces of the PV panels. Below is a discussion on 
the types and properties of glass used in TFPV panels, as well as a review of the most common 
types of glass used in PV panels based on data collected from leading BIPV manufacturers? 
published data sheets or product specification.  
4.5.1 Glass 
The role of glass in a TFPV panel is primarily to protect the TFPVs from the elements while still 
allowing light to pass through to allow the TFPV to function properly and efficiently. Thin lites 
of glass can act as both the substrate and superstrate (front and back) of a PV module. According 
to the National Institute of Standards and Technology (NIST) glass has a high theoretical 
strength (approximately 17 GPa), however, the theoretical strength of glass is a property that has 
yet to be taken advantage of due to its highly brittle nature and its susceptibility to tiny surface 
flaws known as Griffith Flaws, that act as stress concentrators in the glass under loading 
conditions (Griffith 1921). Glass cannot deform plastically and thus it can break without early 
warning due to high surface tension.  This also makes glass highly susceptible to localized over-
 
76 
 
 
stressing.  There are a number approaches to strengthen glass to increase its load capacity which 
are discussed in the next section. These methods help avoid its failure due to surface flaws.  
Glass used in TFPV panels is typically soda lime glass formed mainly from three raw materials: 
soda, lime and silica (AMMA 1984).  Heat strengthened (HS) and fully tempered (FT) glass are 
both made from soda lime glass, but have undergone treatment to increase their load capacity. 
The general properties of glass are commonly agreed upon within the building industry, however 
due to its brittle nature and susceptibility to fracture due to surface flaws, the coefficient of 
variation (as high as 25%) of these values is quite high compared to that of other common 
engineering materials (Beason et al. 1998; Pilkington 2010).   
4.5.1.1 Annealed/Float Glass 
Float glass is the most basic form of flat glass used in TFPV panels.  Its name originates from the 
manufacturing process developed by Pilkington Brothers Ltd. in 1959.  In this process the 
ingredients are heated in a 1500?C furnace to form molten glass, which is then poured across a 
surface of molten tin. The liquid glass spreads out and flattens as it cools. The glass is then 
transported horizontally on rollers in one continuous sheet into a long oven to form a uniform 
thickness sheet. A slow and controlled cooling process is necessary to prevent any residual 
stresses in the glass that result from more rapid cooling. Once completed the final glass product 
is typically termed ?annealed? glass, distinguishing it from float glass that has received further 
treatment such as heat strengthening or tempering after the float process.  This glass can be cut, 
machined, drilled, edged and polished after manufacturing. 
Of all flat glass products, annealed glass is typically the most inexpensive to manufacture, but is 
also the weakest.  Because no further heat treatment is applied to the glass, it is highly 
 
77 
 
 
susceptible to failure in tension. When annealed glass fails it breaks into large pieces that can be 
a safety hazard if it falls from a building fa?ade. Table 4.2 shows some of the typical values 
assumed for its behavior. 
4.5.1.2 Heat Strengthened (HS) Glass 
Heat strengthened (HS) glass is formed by applying further heat treatment to float glass.  During 
this process, the glass sheet is heated to its softening point then rapidly cooled with air jets, 
causing the surface and edges of the glass to cool rapidly and contract, while the center is still 
hot and fluid.  As the glass sheet cools, to counteract the state of compression of the surface and 
edges between 3500 and 7500 psi (per ASTM C1048), the center is in a state of tension to 
maintain equilibrium within the material. This is advantageous to the strength of the glass sheet 
because the failure of glass almost always occurs at flaws on the tension surface of the glass.  
HS glass is generally twice as strong as annealed glass of the same thickness and configuration, 
but it has the disadvantage that is cannot be cut or machined in anyway once the heat 
strengthening process has been completed. All drilling of holes, cutting or grinding must take 
place before the HS process is completed. When HS glass breaks, it breaks into large pieces in a 
similar fashion to annealed glass, however it is preferred due to its strength coupled with good 
post-breakage behavior.  Table 4.2 shows some of the typical values assumed for its behavior. 
4.5.1.3 Fully Tempered/Toughened (FT) Glass  
Fully tempered (FT) glass, also known as safety glass, is similar to that of HS glass except that 
the cooling process is faster, creating higher residual compressive stresses on the surface and 
edges of the glass. To be officially considered FT, the surface compression must be over 10000 
psi (per ASTM C1048). Compared to annealed glass, tempered glass can be up to 5 times 
 
78 
 
 
stronger.  It has the same disadvantage as HS glass in that it cannot be modified once tempered.  
When this type of glass breaks it shatters into small pieces. Fully tempered glass also has the 
disadvantages of some loss of flatness due to the increased compressive stress on the glass 
surfaces, which can be difficult to use in laminated glass.  Additionally FT glass has some 
possibility of spontaneous breakage, i.e. failure internally generated by inherent impurities such 
as nickel sulfide within the glass. Table 4.2 shows some of the typical values assumed for its 
behavior. 
 
 
Table 4.2: Properties of Monolithic Annealed, Heat Strengthened and 
Fully Tempered Glass (Button and Pye 1993; GANA 2008; ASTM 2009) 
 
Property Annealed HS  FT  Source 
Modulus of Rupture 
(50% probability of 
breakage) 
5.95 ksi  
(41 MPa) 
 
12.04 ksi 
(83 MPa) 
 
23.93 ksi 
(165 MPa) 
 
Button and 
Pye 1993 
Design Stress 
(0.8% probability of 
breakage) 
2.8 ksi 
(19.3 MPa) 
 
5.60 ksi 
(38.6 MPa) 
 
11.2 ksi 
(77.2 MPa) 
 
GANA 2008 
Residual surface stress Nearly 0 (0 MPa) 3.48 ksi (24 MPa) 10.00 ksi (69 MPa) ASTM 2009 
  
 
79 
 
 
4.5.2 Laminated glass (LG) 
Laminated glass typically consists of two glass lites bonded together with an elastomeric 
interlayer, commonly made of polyvinyl butyral (PVB). Unlike monolithic glass, the behavior of 
this sandwiched glass-PVB-glass laminates is highly dependent on its temperature (Linden et al. 
1984; Reznik and Minor 1986).  The behavior of this glass is believed by some to fall between 
that of a monolithic plate of glass and that of two unbounded plates of glass of equivalent 
thickness (Hooper 1973; Edel 1997; Norville 1998). Because of uncertainties in the exact shear 
response of the interlayer and its role at varying rates and temperatures, the strength of laminated 
glass is usually downgraded as compared to equivalent monolithic glass plates (Foss 1994). 
However, others have found that at room temperature LG of the same geometry and under the 
same load conditions is equal or stronger in strength than monolithic glass of equivalent 
thickness (e.g. Beason et al 1998; Van Duser 1999).  
4.5.3 Laminated glass model for TFPV panels 
The importance of LG in this research is that, in an effort to most accurately represent the 
behavior of TFPV panel under uniform loading conditions, LG offers the closest glass structure 
to that of a TFPV panel.  Because very little detailed information is available on the behavior of 
TFPV panels under loading, the use of LG is determined to be an acceptable alternate due to its 
many similarities in material make-up, size and structure.  Similar to LG, TFPV are typically 
composed of two glass lites with an interlayer of PVB or similar. The difference between the two 
is that in the TFPV panels the interlayer also contains a very thin layer of photovoltaic materials 
and TCOs (metal oxides with much higher modulus and strength than the elastomeric layer).  
Since the photovoltaic material is located near the center of the two glass lites, it experiences 
 
80 
 
 
only very little of the bending stress compared to the exterior glass lites. In addition it has a high 
modulus of elasticity and ability to deform (demonstrated by its used in TFPV flexible laminates 
available on the market today) as compared to glass. 
4.6 TFPV PANEL MANUFACTUERS DATA 
In choosing to make the assumption that LG can be used to model TFPV panels, this implies that 
the behavior of the TFPV panels is dependent on the material properties of the glass lites used, 
thus since there are several different types, dimensions and thicknesses of glass that can be used, 
a limited review of 36 manufacturer?s specifications for their TFPV panels was conducted.  A 
total of 23 a-Si, 9 CI(G)S and 4 CdTe manufacturers? TFPV product specifications were 
reviewed for type of glass used (annealed, HS, FT), glass lite and PVB thickness, and glass size. 
A list of these manufacturers is listed below.  Inconsistencies in the way data was reported were 
observed in terms of how thickness was reported, and in some cases, data was not available.  
TFPV manufacturers surveyed:  
Abound Solar 
AGC Solar 
Ascent Solar 
Astronergy 
Targray 
Auria 
Avancis 
Energyglass 
Bangkok Solar 
Calyxo 
Canadian Solar 
Unisolar 
BP Solar 
Canon  
Canrom 
CNPV 
DelSolar 
Central Electric 
Limited 
Ditecar 
Ditecar 
Bosch Solar 
Eopply 
E-ton Solar 
Evergreen Solar 
Dupont 
ENN 
EPV Solar 
Ertex Solar 
First Solar 
FlexCell 
Global Solar 
ICP Solar 
Inventux 
Kaneka 
MiaSole 
Mistubishi 
Nanosolar 
Polar PV 
PowerFilm Solar 
Primstar Solar 
Q-Cell 
Schott 
Signet Solar 
Sinonar Solar 
Solar Plus 
SolarPower 
SolarFrontier 
Solems 
Sunfilm 
SunGen 
SunStar 
TerraSolar 
Tianjin Jinneng 
Solar 
Unisolar 
Wurth Solar
 
78 
 
 
Assumed Glass Thickness in TFPV Panels 
Of the TFPV panel manufacturers who reported glass thickness, all but one manufacturer used a 
1/8 in (3.0-4.0 mm) front and back glass lites, thus this size is assumed, constituting a total 
thickness of approximately 1/4 inch for all materials. 
4.6.1 Assumed Glass Type in TFPV Panels 
The type of glass used for TFPV panels varied significantly more, however of the manufacturers 
who reported glass type, 67% used annealed glass, 13% used HS glass, and 20% used FT glass 
for the front lite. For the back lite, 50% used annealed glass, 8% used HS, and 42% used FT.  
From these statistics, it appears that the most common type of glass used in LG is annealed glass. 
However, FT comes in a close second, particularly for the back lite of glass.  
For simplicity, annealed glass is chosen as the most representative type of glass used for TFPV 
panels based on the percentages determined. The other advantage of modeling TFPV panels 
using annealed glass is that this is the weakest of the three glass types considered, thus the 
estimate of glass strength determined from laminated glass data using annealed glass provides a 
lower bound estimate of the strength of TFPVs under wind loading. This adopted approach 
should lead to a lower estimate of the strength of the TFPV panels. 
4.6.2 Assumed Geometric Properties 
A wide range of sizes of TFPV panels are available based on the manufacturers data. Many 
manufacturers had multiple sizes of panels available, and others noted they could custom make 
PV panels to fit the needs of the architect or engineer. In general, sizes ranged from 11 to 236 
inches (270 - 6000 mm) per side with varying aspect ratios. The median dimensions fell 
 
79 
 
 
approximately within 39 to 47 inches (1000 - 1200 mm) and the mean were 39 to 59 inches 
(1000 - 1500 mm) depending on which dimension is considered. The aspect ratios ranged from 1 
to 7, however the median and mean aspect ratios were 1.95 and 2.28 respectively. An effort is 
made to match these property values with those of the laminated glass data used to represent 
TFPV behavior.  
4.7 LAMINATED GLASS FAILURE MODEL 
Many studies have been conducted in an effort to better understand the behavior of glass under 
loads. A review of published studies on glass failure was conducted to gather information on the 
failure strengths of glass experimentally determined by others. Monolithic glass failure data is 
reported and analyzed in many studies (e.g. Orr 1957; Hershey and Higgins 1973; Kanabolo and 
Norville 1985; Norville et al 1998).  Laminated glass failure data is also reported and analyzed 
(e.g. Linden et al. 1984; Behr et al 1985; Reznik and Minor 1986; Vallabhan et al 1981; Minor 
and Rezik 1990; King 1996; Norville 1998).   
Analysis of glass failure in the 1960s and early 1970s generated the first design charts used in 
building codes from empirical models by Orr (1957) and Hershey and Higgins (1973).  In the 
late 1970s a new model was developed by PPG Industries (PPG 1979) based on the finite 
element analysis conducted by Tsai and Stewart (1976) and Krall et al. (1981). However due to 
significant controversy over the first glass design charts and the PPG methodology, these design 
charts were not widely adopted into building codes of the time. The current methodology used in 
building codes today was first presented by Beason 1980, Beason and Morgan (1984), Norville 
and Minor (1985), and Beason et al. (1998) in an effort to model the failure probability of glass.  
Termed the Glass Failure Prediction Model (GFPM), this model became the basis of the modern-
 
80 
 
 
day ASTM standard E1300 (Standard Practice for Determining Load Resistance of Glass in 
Buildings) ? that is in use today for the design of glass used in building glass fa?ade elements.  
This method was initially developed for monolithic glass, and later extended to cover the design 
of laminated glass in more recent versions. Due to the complexity and number of factors that can 
affect the strength of a glass under uniform loading, rather than create a new model, the glass 
failure prediction model (GFPM) adopted by ASTM E1300 (2009) is used in this study.   
4.7.1 Glass Failure Prediction Model (GFPM) 
The GFPM is based on a statistical theory of failure for brittle materials advanced by Weibull 
(1939). This method allows the probability of failure of a glass plate to be calculated in terms of: 
? glass plate geometry (dimensions and thickness) 
? load duration 
? elastic properties of glass 
? magnitude of the applied load 
 
The GFPM model used in this research was intended designed to predict the failure of 
monolithic glass, and was later extended to encompass laminated glass (LG) by multiplying its 
output by a constant to obtain the probability of failure value for LG. The original failure 
probability function proposed by Beason (1980) and explained fully in Beason and Morgan 
(1984) is expressed as: 
 g1842g3033 g3404 1 g3398 g1857g2879g3003 (4.1a) 
 
 
 
 
81 
 
 
Where: 
 
g1828 g3404 g1863g4666g1853g1854g4667g3040g2879g2869 g4666g1831g1860g2870g4667g3040 g3436 g1872g303160g3440
g3040
g2869g2874 g1844 g4672g1865uni002C g1869g3548 g1853
g1854g4673 
g3404 g1863g4666g1853g1854g4667g3040g2879g2869 g4666g1831g1860g2870g4667g3040 g3436g1872g303160g3440
g3040
g2869g2874 g1857g3011 
(4.2b) 
 g1836 g3404 ln g4678g1844 g4672g1865uni002C g1869g3548 g1853g1854g4673g4679 (4.3c) 
 
g1869g3548 g3404 g1869g4666g1853g1854g4667
g2870
g1831g1860g2872  
 
(4.4d) 
a, b = dimensions of the glass 
k, m = surface flaw parameters (assumed to be m = 7 and k = 1.365 x 10-29 in2 lb-7 ) (Beason 
1998) 
E = elastic modulus of glass (taken to be 10.4 x 106 psi) (Beason 1998) 
h = thickness of glass lite (for laminated glass this is the total thickness) 
td= load duration  
q = uniform load applied on the glass  
 g1869g3548 = dimensionless load  
J is determined by using a chart that relates the aspect ratio and g1869g3548 with J. (see Figure 4.11). 
To understand why many of these variables are incorporated into this complex equation, each of 
the effects of the variables is discussed in the following sections. 
 
82 
 
 
4.7.2 Effect of Plate Geometry on Glass Strength 
The dimensions of a glass sheet have been found to affect the maximum load that glass can 
withstand.  The area is important because the larger the sheet of glass, the higher the potential for 
surface flaws and thus the more likely the glass may fail.  Resistance to stress corresponds to the 
integrity of the surface, and the presence of microflaws can dramatically lower its strength.   
The thickness of the glass can also determine the stresses in the glass, thus a thicker glass equates 
to a greater load resistance. Because glass typically fails on the tension surface due to its brittle 
nature, the thicker the glass, the smaller the surface stress for a given load.  
The PVB layer thickness has also been shown to affect the total strength of laminated glass 
(Norville 1998; King 1996). PVB typically is available in approximately 0.38 mm (0.02 in) 
increments (e.g. .38, .76, 1.14, 1.52 mm), the most commonly used being 0.38 mm (0.02 in). A 
thicker PVB layer at room temperature was shown in these studies to resist higher loads than 
thinner PVB layers. In the survey conducted by the writer on the commercial TFPV data, most 
BIPV manufacturers did not report the thickness of the PVB layer in their TFPV panels. Thus 
since it has been found that thinner PVB layers are weaker under uniform loading than laminated 
glass with thicker PVB layers, with an intent to be conservative, the lower bound of 0.38 mm 
thickness is assumed for this study, which is consistent with the assumptions made used in 
ASTM E1300 (2009).  
In the GFPM model a and b are the variables that represent the dimensions of the glass panel 
under consideration; these values vary based on the size of the TFPV panel. The thickness h is 
the total thickness of the glass. In the case of LG this would be the sum of the thickness of the 
two glass lites and the PVB layer.  Since the PVB layer thickness is much less than that thickness 
 
83 
 
 
of the glass, it can be neglected, and thus the thickness for the LG is assumed to be 1/4 in. (6 
mm), which includes a front and back glass of 1/8 in. thick based on the commercially available 
TFPV data.   
4.7.3 Glass Strength Dependence on Load Duration  
The behavior of glass under loading is highly time-dependent, thus when specifying a design 
strength of glass, the duration of the load must be specified. Glass can withstand more load for a 
shorter period of time than for a longer time and its load capacity at fracture decreases at 
approximately 1/16th power of load duration under constant load. The current ASTM E1300 
standard (2009) uses a 3-second load duration to parallel the 3-second loads used in the wind 
design charts. Since the downburst wind data is based on 3-second wind gust load data, the GFM 
model used in this research uses a 3-second load duration for consistency among the two data 
sets in this research. If a load is used that is not a 3-second load, this load can be converted to an 
equivalent 3-second load using the following equation from ASTM E1300 (ASTM 2009): 
 g1869g3032g3044 g3404 g1869g3031
g3284 g3428
g1856g3036
3 g3432
g2869uni002Fg2869g2874
 (4.2) 
 
g1869g3032g3044= equivalent 3-second load 
g1856g3036= load duration of the load being converted into an equivalent 3-second load 
g1869g3031g3284= load with duration g1856g3036 being converted to an equivalent 3-second load 
 
84 
 
 
4.7.4 Strength of Weathered Glass 
Over time glass strength degrades due to its exposure to weather, thermal and wind induced 
stresses, among other factors. Thus the strength of newly formed glass is substantially higher 
than that of weathered, in-service glass (Abiassi 1980; Norville and Minor 1985). The design 
parameters m and k used in the above equation are the two parameters that account for this 
degradation in the behavior of the in-service glass. It has been found by experiment that for 
weathered glass these two parameters take the following values: m = 7 and k =  1.365 x 10-29 in12 
lb-7 (Beason and Norville 1989). The use of these surface parameters will result in a lower 
estimation of the strength of glass under loading, thus taking into account the loss it strength due 
to weathering.  
4.7.5 Failure Probability of Annealed Laminated Glass  
4.7.5.1 Conversion of monolithic glass to laminated glass strength 
In order to use the GFPM for LG to model the BIPVs under wind loading, the values for the 
monolithic glass must be converted to the equivalent LG (per the procedure adopted by ASTM 
E1300 (2009)). Several studies consider the behavior of LG as being less than that of its 
monolithic equivalent of equal thickness and dimensions (Hooper 1973; Edel 1997). These 
studies have mostly focused on LG beams rather than plates. Other studies have indicated that 
LG has a strength near to their monolithic equivalent in size and thickness (Minor and Reznik 
1990; Norville 1990). Initially manufacturers of LG chose to use a conservative estimate of LG 
strength by multiplying the monolithic strength by a factor less than one.  The more recent 
version of the ASTM E1300 standard (2009) has been modified to eliminate this strength factor 
and instead adopt similar, but separate methods for monolithic and laminated glass.  Comparing 
 
85 
 
 
the charts used to determine the strength of LG and monolithic glass, however, the current 
ASTM E1300 standard (2009) method is nearly the same (Figure 4.14 and Figure 4.15). This is 
in agreement with the findings of Minor and Reznik (1990) who compared LG failure loads from 
studies by Linden et al (1984) and Reznik and Minor (1986) with that of the failure loads of 
monolithic glass of the same size by Kanabolo and Norville (1985). 
4.7.5.2 BIPV Panel Sizes Chosen for use in GFPM 
Assuming the use of method of GFPM as used in the ASTM E1300 (2009) for predicting the 
failure probability of a LG panel, five cases with different dimensions were of glass were 
considered in this study.  An effort was made to represent the range of gathered data on the 
TFPV solar panels available.  The glass sizes in these five cases were chosen with different 
aspect ratios in an attempt to encompass the majority of the fa?ade-installed TFPV panels in use 
today in the U.S.   
Three of the sizes (38 x 76 inch, 66 x 66 inch and 66 x 96 inch) (965 x 1930 mm, 1676 x 1676 
mm, 1676 x 2438 mm respectively) were chosen to parallel the sizing of the laminated glass 
tested in Linden et al. (1984) and Reznik and Minor (1986) for ease of comparison of the model 
results to actual LG test data. The average and median values of the TFPV panel dimensions 
were found to be approximately 40 inches and the aspect ratios are around 2.1; the 38 x 76 inch 
(965 x 1930 mm) size has similar dimensions and an aspect ratio of 2.11. The two other sizes 
considered have dimensions 39 x 39 inch and 25 x 40.5 inch (1000 x 1000 mm and 629 x 1029 
mm respectively). The 39 x 39 inch panel is the size of the a-Si TFPV panels installed on the 
National Air and Space Museum at Dulles Center in Washington D.C., and was chosen to 
represent a small aspect ratio (1).  The 25 x 40.5 inch dimensions (aspect ratio is 1.62) are from 
 
86 
 
 
the Schott ASI solar panel data sheet (Schott 2010). Schott solar panels are installed in at the 
New York Stillwell Avenue subway station, and were chosen to represent a mid-range aspect 
ratio.  
4.7.5.3 Probability of Failure Calculations 
Using these dimensions, the probability of failure curve for each set of dimensions is computed.  
Table 4.2 shows the variables used in these calculations. The fragility curves are shown in 
Figures 4.12 and 4.13. 
4.7.5.4 Model Verification 
To cross-check that these calculations are consistent with those given by the charts used in the  
ASTM E1300 (2009), using this model the load is evaluated at a probability of failure of 0.008, 
following that of the ASTM E1300 standard. These loads are indicated in Table 4.3. Using the 
ASTM E1300 design chart for 6mm, 4-side simply-supported LG (Figure 4.14) each of these 
sizes are also evaluated.  At a probability of failure of 0.008, all of the values calculated are 
consistent with the design charts commonly used for glass thickness design (ASTM 2009).  
Because the 36 x 76 in, 66 x 66 in and 66 x 96 in panels used to calculate the probability of 
failure in this model are the same sizes as those used in previous studies, by calculating the 
probability of failure at 50%, these values can be compared to the average failure values of the 
laminated glass lites tested in Linden et al. (1984) and Reznik and Minor (1986) to obtain some 
measure of accuracy of the model (Table 4.4). Linden et al. (1984) performed testing on 29 
laminated glass specimens (60 x 96 in) and Reznik and Minor (1986) used 25 and 26 specimens 
(38 x 76 in. and 66 x 66 in. respectively). The reported 60-second loads are converted to 
 
87 
 
 
equivalent 3-second loads in Table 4.4 by multiplying by a factor of 1.2 per Equation X7.1 in 
ASTM E1300 (2009) which converts a load of a given duration to an equivalent 3-second load.  
In all cases laminated glass, the model provides a higher failure strength. This is to be expected 
since the glass properties chosen for use in this model represent the properties of weather, rather 
than new glass (new glass is used in Linden et al (1984) and Reznik and Minor (1986)).  
 
88 
 
 
Table 4.3: Parameter values used in determining probability of failure using GFPM 
 
36 x 76 in. 66 x 66 in. 66 x 96 in. 39 x 39 in. 25 x 40.5 in. 
m 7 7 7 7 7 
k 1.365 x 10-29 1.365 x 10-29 1.365 x 10-29 1.365 x 10-29 1.365 x 10-29 
td 3 sec. 3 sec. 3 sec. 3 sec. 3 sec. 
E 1.04 x 106 psi 1.04 x 106 psi 1.04 x 106 psi 1.04 x 106 psi 1.04 x 106 psi 
h 1/4 in. 1/4 in. 1/4 in. 1/4 in. 1/4 in. 
a 36 in. 66 in. 36 in. 39 in. 25 in. 
b 76 in. 66 in. 76 in. 39 in. 40.5 in. 
Aspect 
ratio 2.11 1 1.45 1 1.62 
 
Table 4.4: Maximum loads resisted by laminated glass at probability of failures of 0.008 and 0.5 for 
each set of dimensions considered using the GFPM 
 
Pf 36 x 76 in. 66 x 66 in. 66 x 96 in. 39 x 39 in. 25 x 40.5 in. 
0.0008 
0.5 
58 
145 
40 
100 
30 
65 
105 
290 
153 
330 
 
 
Table 4.5: Mean failure strengths predicted by GFPM model compared to the equivalent 3-second 
load strength from laminated glass test data  
 
Pf 36 x 76 in. 66 x 66 in. 66 x 96 in. 39 x 39 in. 25 x 40.5 in. 
GFPM (equivalent 3-
second load) (50% failure) 145 100 65 -- -- 
Mean Failure Load (Reznik 
and Minor (1986) and 
Linden et al. (1984)) 
(equivalent 3-second load) 
160 
(134)* 
133 
(111)* 
92.4 
(77)* -- -- 
*Note: The numbers in parenthesis are the 60-second loads before conversion to 3-second equivalent 
loads 
 
  
 
 
 
Figure 4.1: National Renewable Energy Laboratory (NREL) maximum efficiencies of PVs by year
    (a)    
Figure 4.2: Examples of fa?ade BIPVs
 
  
                   (b) 
 in (a) curtain wall and (b) rain screen applications (Sayigh 
2000) 
89 
 
 
 
 
90 
 
 
 
 
Figure 4.3: a-Si and micromorphous Si typical layers (Kaltschmitt et al. 2007) 
 
Figure 4.4:  (a) AstroEnergy a-Si/?c-Si micromorphous solar panel construction (AstroEnergy 
2010) 
 
 
91 
 
 
   
(a)     (b) 
Figure 4.5 (a) Schott ASI Thru PV Module, (b) EPV Solar EPV-4X Solar Module (Schott 2010) 
 
Figure 4.6: Typical layers of a CI(G)S solar cell (AGC 2010) 
 
 
 
92 
 
 
 
   (a)     (b) 
Figure 4.7: (a) Avancis Powermax CIS module, (b) Nanosolar Utility CIGS Panels (Avancis 2010; 
Nanosolar 2010) 
 
 
Figure 4.8: Typical layers of a CdTe solar cell  
 
93 
 
 
  
     (a)           (b) 
Figure 4.9: (a) Q-cell CIGS Q.Smart 70-90 PV panel, (b) First Solar CdTe (Q-Cell 2010; First Solar 
2010).  
 
Figure 4.10: Behavior of glass under loading  
 
94 
 
 
 
Figure 4.11: Chart used for determining the stress distribution J from the aspect ratio and non-
dimensional lateral load (Beason and Morgan 1984) 
 
 
95 
 
 
0
0.002
0.004
0.006
0.008
0.01
0 20 40 60 80 100 120 140 160
36 x 76 in
66 x 66 in 
66 x 96 in
39 x 39 in
20 x 40.5 in
Pf
Wind Load (psf)  
Figure 4.12: Fragility Curve (Pf = 0.008) of laminated glass using the GFPM 
 
 
96 
 
 
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 100 200 300 400
36 x 76 in
66 x 66 in 
66 x 96 in
39 x 39 in
20 x 40.5 in
Pf
Wind Load (psf)  
Figure 4.13: Fragility Curve (Pf = 0.5) of laminated glass using the GFPM 
 
97 
 
 
 
Figure 4.14: ASTM E1300 design chart for 6 mm laminated glass (ASTM 2009) 
 
Figure 4.15: ASTM E1300 design chart for 6 mm annealed glass (ASTM 2009) 
 
98 
 
 
CHAPTER 5: DOWNBURST DESIGN PRESSURE AND TFPV 
LOAD RESISTANCE  
Thus far in Chapter 3 the design velocity pressures associated with the downbursts in the 
WBMA were determined for building fa?ade elements located at 33 ft above the ground, and  for 
three different possible strike factors respectively.  In Chapter 4 the fragility curves were 
determined for five cases with different sized TFPV panels using two 3-mm thick annealed glass 
lites (see Figure 4.12 and 4.13).  It is now possible to compare each of these results to their 
respective existing design standards, as well as compare these two results with each other to 
create a failure prediction model.   
5.1 ASCE 7-05 AND DOWNBURST DESIGN WIND LOADS 
Before comparing the downburst design wind loads with that of the BIPV panel probability of 
failure curves, it is relevant to first compare the current design wind speeds in the WBMA and 
the predicted design wind speeds from downbursts using the model in this research. If the 
predicted wind speed from the downbursts is lower than that of the current design wind speed, it 
seems reasonable to believe there is limited concern of needing additional design considerations 
for downburst winds in the WBMA.  However if the design wind speeds predicted by the 
downburst model are above that of existing design wind speeds, it is reasonable to consider 
addition recommendations for downburst winds be included in future ASCE 7 standards. 
The 50-year design wind speed specified by ASCE 7-05 for the WBMA is 90 mph (40 m/s) 
(ASCE 2005).  Given the three possible methods of calculating the strike factor for downburst 
occurrences in the WBMA, the 50-year design wind speeds at 33 ft  height were calculated to 
 
99 
 
 
range from approximately 80-93 mph (35-40 m/s) (Table 3.8).  This range of numbers is 
approximately the same design wind speed specified for this region.  Using the ASCE 
conversion from wind speed to pressure, this translates to an equivalent uniform wind velocity 
pressure of approximately 14 - 20 psf (689 - 938 Pa). The 100-year design velocity pressure, 
sometimes used for the design of buildings, is only slightly higher, at up to 21 psf (988 Pa). 
Based on the comparison of these two values, it is seen that the ASCE 7-05 design wind values 
may be sufficient for use in the design of buildings for downburst occurrences without 
modification.   
However, an important difference to note when comparing these two design wind speeds is the 
difference in the vertical profile of a downburst and that of the ASCE 7-05 profile. According to 
downburst simulation models created by others (e.g. Kim and Hangan 2007; Woods et al. 2001), 
the vertical wind profile of a downburst peaks at a certain height that varies with the downburst 
diameter then begins to decrease. According to downburst simulation models created by Kim and 
Hangan (2007), the maximum velocity of a downburst wind profile occurs at heights of less than 
5% of the initial downburst diameter. Similar models of downburst vertical wind profiles have 
been created by others based on field and experimental data (e.g. Woods et al. 2001; Ozegura 
and Bowles 1988). Holmes (2002) found that thunderstorm downbursts produce the extreme 
winds at 10 m (33 ft) height.  Following the results of Kim and Hangan (2007), for a downburst 
of 500 m and 1 km diameter, the maximum wind velocities would occur at less than 25 - 50 m 
(82 - 164 ft), then would begin to decrease as height increases.   
The vertical wind profile used in the design of buildings per ASCE 7-05, unlike that of a 
downburst, continually increases with height.  Using this wind speed, the ASTM 7-05 design 
 
100 
 
 
methodology based on factors such as the size of the building, height, etc. can be followed to 
create a vertical wind profile of the design pressures for the building cladding and components 
(see ASCE 7-05 (2005) for a more detailed explanation). This wind profile increases 
significantly in the first 15 ft (5 m) then continues to slowly increase with height at a rate that 
varies with the specified exposure level of the building in question.   
An example of the comparison of these two different profiles is shown in Figure 5.1. In this 
figure, V/Vmax is the normalized wind velocity and Z/Z0.5*Vmax is the height divided by the half-
velocity height. These scales are used so that multiple datasets can be compared on one graph. 
The dotted line represents the atmospheric boundary layer (similar to that of ASCE 7-05) and the 
other data points are results from multiple studies. From this figure, it is clear that the two wind 
profiles are different.  Thus it is possible that above the height of 33 ft, depending on the size of 
the downburst, that the downburst wind speed is greater than that of the wind speed predicted by 
ASCE 7-05. Further analysis and research is needed to confirm this issue.  
 
101 
 
 
 
Figure 5.1: Downburst mean velocity profile: comparison with laboratory experiments, empirical 
models and with a typical boundary layer profile (Kim and Hangan 2007) 
 
5.2 EUROPEAN IEC 61646 MECHANICAL LOAD REQUIREMENTS 
AND GFPM PREDICTED LOAD RESISTANCE 
It is also relevant to compare the predicted load resistance of the TFPV panels from the model 
used in this research with that of the loads used to test TFPV panels.  Currently one standard test 
exists that uses standardized test procedures to test and certify TFPV panels.  This standard, IEC 
61646 ? ?Thin-film terrestrial photovoltaic (PV) modules ? design and qualification and type 
approval? was developed in 1996 to standardize testing of thin-film PVs such as those used in 
BIPV applications. A test standard for the United States specifically has not yet been developed. 
The IEC 61646 standard, recognized throughout much of Europe, tests many different aspects of 
new TFPVs, including tests such as temperature, performance under different solar conditions, 
 
102 
 
 
thermal cycling, damp heat, twisting, and, most important and relevant to this research, 
mechanical load testing. In comparing the predicted performance of TFPVs with the prescribed 
load in IEC 61646 a better understanding of the relevance of the standard load can be 
understood.  
Part of the IEC 61646 test requires that both the front and back of the solar panel withstand two 
cycles of 2400 Pa (50 psf) for 1 hour.  As discussed previously, glass behaves differently based 
on the duration of the load induced on it.  The strength of glass predicted in this research is for a 
3-second load, thus to translate a 1 hour load (3600 seconds) to a 3-second load, for purposes of 
comparison, following ASTM E1300, 2400 Pa is divided by a factor of 0.64 (Table X6.1 in 
ASTM E1300 (2009)), which yields a pressure of 3750 Pa (78.32 psf). For the five sizes used in 
this research, the probability of failure of the TFPV predicted by the GFPM model used in this 
research ranges from less than 8:1000, the design value assumed for ASTM E1300 (39 x 39 and 
20 x 40.5 inch) to a nearly 70% failure probability (66 x 96 in) (see Figure 4.12).  While the 66 x 
66 and 66 x 96 inch sizes are significantly larger than most BIPV panels surveyed, it is worth 
noting their failure probabilities are very large. Table 5.1 lists the failure probabilities for each of 
the five sizes considered.  
 
The percentages in Table 5.1, particularly those of the 36 x 76 inch, 66 x 66 inch and 66 x 96 
inch size BIPVs can be misleading. These sizes are significantly larger than that of the solar 
panels surveyed.  It is likely that if manufacturing solar panels of this size were to be used, that 
thicker glass would be used for the front and back lites, equating to a greater strength and 
resistance to the given load.  
 
103 
 
 
5.3 DOWNBURST DESIGN WIND LOADS AND GFPM PREDICTED 
LOAD RESISTANCE 
Ultimately the design wind loads determined from the probabilistic downbursts wind model are 
now compared with the probabilistic strength distribution of the BIPVs used in building facades. 
Based on these, a failure probability model is derived for BIPVs specific to the Washington 
D.C.-Baltimore metropolitan region.  
From Table 5.2 it is shown that only the largest of the five sizes considered had a large enough 
predicted probability of failure to be recorded.  From this it can be concluded that if only the 33 
ft design wind pressures are considered, then a building?s BIPVs should be able to withstand the 
downburst occurrences in the WBMA.  Further analysis of the vertical profiles of the downburst 
and ASCE 7-05 specified wind load design curves is needed to compare these two in more detail.  
 
  
 
104 
 
 
 
Table 5.1: Failure Probabilities of Five TFPV BIPV Panels at the Equivalent 3-second Uniform 
Pressure Specified by IEC61646 
 
36 x 76 in. 66 x 66 in. 66 x 96 in. 39 x 39 in. 25 x 40.5 in. 
2.76 % 19.1 % 69.9 %* 0.19 % 0.02 %  
  
* This value is very high, however it should be noted that few to none of TFPV are this large in size, regardless if 
they were they likely would use thicker glass to compensate for the large size 
 
 
 
 
Table 5.2: Failure Probabilities of Five TFPV BIPV Panels at the 3-second Uniform Pressure 
Predicted by the Downburst Wind Pressure in the WBMA 
 
Strike Factor Pressure 36 x 76 in. 66 x 66 in. 66 x 96 in. 39 x 39 in. 25 x 40.5 in. 
0.00297 -- -- -- -- -- -- 
0.0283 14.39 Negligible Negligible 0.028 % Negligible Negligible 
1  Negligible 0.028 % 0.17 % Negligible Negligible 
Note: ?Negligible? in this chart indicates a failure probability of less than 0.001% 
 
  
 
105 
 
 
CONCLUSIONS AND FUTURE WORK  
Thin-film building integrated photovoltaics (BIPVs) are technologies that are being implemented 
in buildings today, uniquely producing clean energy from the sun?s light while simultaneously 
acting as the building?s weather-resistant envelope.  Of the many potential causes of failure of 
these primarily glass solar panels, extreme loading can cause failure of the BIPV panel.  
Downbursts have the potential to produce extreme winds capable of causing these panels to fail.  
This research focuses on this issue through a region-specific approach ? utilizing the Washington 
D.C.- Baltimore metropolitan area (WBMA) as the area of analysis. 
In this research first a region-specific probabilistic load model of downburst winds has been 
created for the WBMA.  Downburst data was obtained from both the Storm Event Database, 
available through the National Climactic Data Center and the National Weather Service, and 
through a search conducted of WBMA news papers, news wires, and other publications.  This 
data was then compiled and modeled using Type I and III extreme value distribution functions.  
Using a goodness-of-fit test, the Type III distribution is found to fit the dataset best, although it 
did not provide a satisfactory fit due to lack of sufficient downbursts data in the region.  A 
slightly better fit is found when the dataset is limited only to the downbursts with reported wind 
speeds, using very similar distribution curve parameters. A stochastic model was adopted to 
predict the 50-year wind loads that downbursts occurring in the WBMA would subject to the 
building in this area.  Depending on the strike factor assumed in the stochastic model, the 50-
year design wind speeds (79 ? 93 mph) were both slightly above and below the 50-year design 
wind speeds specified for the WBMA by ASCE 7-05 standards (90 mph). Using the 
 
106 
 
 
methodology of ASCE 7-05, these wind speeds were then converted to velocity pressures for 
purpose of comparison with the strength of the BIPV panels. 
The behavior of BIPVs installed on the facades of buildings subjected to uniform wind loading 
was next accomplished. Assuming that the thin-film layer (TF) and the transparent conducting 
oxide (TCO) can be neglected because of their high stiffness and very small thickness, the TFPV 
panels are modeled with an assumption that their behavior is that of a laminated glass plate. The 
glass failure prediction model (GFPM) is then used to predict the probability of failure of the 
BIPV panels under uniform loading, and with simple supports on four sides.  In comparing the 
strength of the BIPV panels to the load effect of the downburst-induced design wind loads in the 
WBMA, only for large panels (> 66 in) was the probability of failure due to downbursts is 
measurable, although it was still less than the design 8/1000 probability of failure specified by 
the current design standards for glass - ASTM E1300. Caution should be exercised in 
interpreting this since a number of assumptions were made in deriving this. It is also concluded 
that the downburst loads (14-20 psf) are less than the loads at which they are currently required 
by IEC 61646 (60 psf equivalent 3-second load), a standard primarily used in Europe for testing 
of thin-film BIPVs.  
In future work it is proposed that the modeling of TF BIPVs as laminated glass be verified 
through uniform pressure testing.  It is also proposed that the probabilistic model created in this 
research be further expanded to include consideration of the height above the ground on the 
downburst wind speed as it compares to the TF BIPV panels strength.  
  
 
107 
 
 
APPENDIX A 
A.1  EXTREME VALUE DISTRIBUTION TYPES 
A.1.1 Type I ?Extreme Value Distribution  
The Gumbel or Type I extreme value distribution (EVD) has two forms, one based on the 
smallest extreme and the other based on the large extreme. These are called the minimum and 
maximum extremes respectively. For this, as for other wind data cases, the maximum form of the 
Type I EVD is used.  The general formula for the probability density function (PDF) of this 
distribution is: 
g1858g4666g1876g4667 g3404  g2009 g1857g2879g4666g3051g2879g3091g4667g3080g1857g2879g3032g3127g4666g3299g3127g3339g4667g3328 
  The probability distribution function (i.e. cumulative distribution function (CDF)) is: 
g1832g4666g1876g4667 g3404  1 g3398 g1857g2879g3032g4666g3299g3127g3339g4667g3328 
The function is based on two parameters: g2009, the scale parameter, and g2020, the location parameter.  
A.1.2 Type III ?Extreme Value Distribution 
The Weibell or Type III EVD general formula for the PDF of this distribution is: 
g1858g4666g1876g4667 g3404  g2011g2009  g4672g1876 g3398 g2020g2009 g4673
g4666g3082g2879g2869g4667
g1857g2879g4666g4666g3051g2879g3091g4667 g3080uni2044 g4667g3330 
In the case where g2020 g3404 0 (location parameter) then the distribution can be reduced to its simpler 
form called the 2-parameter Type III EVD: 
g1858g4666g1876g4667 g3404  g2009g2011g3080  g4666uni0078g4667g4666g3080g2879g2869g4667g1857g2879g4672
g3051
g3082g4673
g3328
 
  The probability distribution function (i.e. cumulative distribution function, CDF) is: 
 
108 
 
 
g1832g4666g1876g4667 g3404  1 g3398 g1857g2879g4672
g3051
g3082g4673
g3328
 
The function is based on two parameters g2009, the scale parameter, and g2011, the shape parameter.  
A.2 DISTRIBUTION PARAMETER ESTIMATION METHODS 
A.2.1 Method of Moments 
The method of moments is one of the oldest methods to estimate population parameters. In this 
method, it is important to examine two moments: 
Mean ? first moment about the origin 
Variance ? second moment about the mean 
The respective kth moment about the origin for a discrete or continuous random variable is: 
g1839g3038g4593 g3404 g1516 g1876g3038g1858g3051g4666g1876g4667g1856g1876g2998g2879g2998      or     g1839g3038g4593 g3404 ? g1876g3036g3038g1842g3051g4666g1876g3036g4667 
In which x is the random variable, g1858g3051g4666g1876g4667 is its density function, n is the number of elements in 
the underlying sample space of X, and g1842g3051g4666g1876g4667 is the probability density function. The first moment 
about the origin, where k=1 in the above equations yields the mean of X and is denoted as g1876g1191. 
Because the downburst data is discrete rather than continuous, only the discrete equations will be 
shown below.  The generalized function of this is denoted as: 
g1831g4670g1859g4666g1876g4667g4671 g3404 g3533 g1859g4666g1876g3036g4667g1842g3051g4666g1876g3036g4667
g3041
g3036g2880g2869
 
To obtain the parameters for the Type I EVD from sample mean g1876g1191 and standard deviation g2026 , the 
following equations are used, as developed from the more generalized equations above g2011 is the 
Euler-Mascheroni constant g3406 .577 ) in the method of moments: 
 
109 
 
 
g2026 g3404 g3095g3080uni221Ag2874   or  g2009 g3404 g3095g3097uni221Ag2874 
g1876g1191 g3404 g2020 g3397 g3082g3080  or   g2020 g3404 g1876g1191 g3397 .g2873g2875g2875g3080  
To obtain the parameters of the Type III EVD distribution, the coefficient of variation is first 
obtained from the equations of the first and second moments and used to calculate g2009: 
g1829g1848 g3404
g3495uni0393 g46741 g3397 2g2009g4675 g3398 uni0393g2870 g46741 g3397 1g2009g4675
uni0393g2870 g46741 g3397 1g2009g4675
 
Then g2026 is calculated: 
g2026 g3404 g4684 g1876g1191
uni0393 g46741g2009 g3397 1g4675
g4685
g3080
 
 
A.2.2 Maximum Likelihood Estimation 
Maximum likelihood parameter estimation is based on the principle of calculating values of 
parameters that maximize the probability of obtaining the particular sample. The likelihood of 
the sample is the total probability of drawing each item in the sample. The total probability is the 
product of all the individual item probabilities. This product is then differentiated with respect to 
the parameters and the resulting derivatives are set to zero to achieve the maximum. Maximum 
likelihood solutions, however, do not always produce solvable equations.    
Given the Type I EVD as an example, the likelihood function, L, is defined as: 
 
110 
 
 
{ } { }{ }
( ) { }[ ]
??
?
??
? ??+??=
=?????===
?
??
=
==
n
i
ii
n
i
n
i
i
n
i
in
yy
yyyfyyfL
1
11
1
)(exp)(exp
)(expexp)(exp),|(),|,...,(),(
?????
???????????
 
Taking the natural log of both sides: 
{ }[ ]?
=
??+???==
n
i
ii yynLl
1
)(exp)()ln()ln( ?????  
The partial derivative of the equation must be taken for each variable being estimated, thus: 
 
{ }[ ] { }
{ }[ ] { }???
??
===
==
?????+?=???????=??
???=??+??=??
n
i
ii
n
i
i
n
i
iii
n
i
i
n
i
i
yyynyyynl
ynyl
111
11
)(exp)()()(exp)()(
)(exp)(exp
???????????
?????????
 Setting the two partial differentials to zero for their maxima and solving simultaneously for ? 
and ? the MLE can be obtained. The same methodology can be followed for the Type III EVD 
but is not shown here.  
A.3 GOODNESS-OF-FIT TESTS: 
A.3.1 Anderson-Darling (A-D) Test: 
The A-D test is one of several goodness-of-fit techniques, but is only possible to calculate for 
continuous distribution functions. Its purpose is to decide whether to accept or reject Ho, the null 
hypothesis that states that the sample data follows the given distribution.  
It is important to note that the acceptance of Ho for a given level of confidence does not mean 
that the population in fact is that distribution. It means that Ho cannot be rejected. Thus for a 
defined confidence level where Ho is accepted, there is not enough evidence to indicate that the 
data does not follow the given distribution. The A-D distribution is often preferred over the 
 
111 
 
 
Komolgorov-Smirnov (K-S) test because it is more sensitive to deviations in the tails of the 
distribution. 
The A-D is defined by: 
Ho : The data follow the specified distribution 
Ha : The data follow the specified distribution 
For Type I and III distributions the following equations are used to determine A2, the test statistic 
for the A-D test: 
g1827g2870 g3404 g3398g1866 g3398 g34361g1866g3440 g3533g46662g1861 g3398 1g4667g4670lng4666g1875g3036g4667 g3397 ln g46661 g3398 g1875g3041g2879g3036g2878g2869g4667g4671
g3041
g3036g2880g2869
 
Where n is the sample size and w is the cumulative distribution function (CDF) for the 
distribution under consideration. For small samples A2 is modified using the following equation: 
g1827g3040g2870 g3404 g1827g2870 g34361 g3397 0.2uni221Ag1866g3440 
The critical values obtained from the calculation of A2 are the following for different levels of 
significance (?): 
Table A.0.1: Critical Values for Anderson-Darling Test (D'Agostino and Stephens 1986) 
Statistic Values 
g2009 0.1 0.05 0.025 0.01 
A2 0.637 0.757 0.877 1.038 
 
The A-D method is often preferred over the K-S test for the following reasons: 
? The K-S test tends to be more sensitive near the center of the distribution than the tails, 
while A-D gives more weight to the tails. 
 
112 
 
 
? The K-S is not valid if the PDF and CDF parameters are estimated from the data, while 
A-D can be used for this application. 
? The A-D is more sensitive to the lack of fit of the Type III EVD than the K-S test (Evans 
et al. 1989).  
A.3.2 Kolmogorov-Smirnov (K-S) Test 
The K-S test (Kolmogovro 1933) is an alternative to the A-D test, but serves the same purpose of 
determining the acceptance or rejection of Ho, the null hypothesis that states that the sample data 
follows the given distribution.   
The K-S is defined by: 
Ho : The data follow the specified distribution 
Ha : The data does not follow the specified distribution 
Given n ordered data points (smallest to largest), Y1, Y2? Yn, the empirical distribution function 
(ECDF) is defined as: 
g1831g3041 g3404 g1840g4666g1861g4667g1866  
Where N(i) is the number of data points less than Yi.  
The test statistic D, the maximum absolute difference between the values of the cumulative 
probability distribution of the sample size n and the specified cumulative probability distribution 
function, can be determined by the following equation: 
g1830 g3404 mauni0078g2869g3000g3036g3000g3041 g4678g1832g4666g1851g3036g4667 g3398  g1861 g3398 1g1866 uni002C g1861g1866 g3398 g1832g4666g1851g3036g4667g4679 
Where F is the CDF being evaluated. 
 
 
 
A.4 POISSON PULSE PROCES
This process involves the use of three random variables 
(number of downbursts/year), di 
? intensity (mph) of the ?i?th downburst with a probability distribution function (a.k.a 
cumulative probability distribution) of 
square wave represents one downburst with a duration d and an intensity w. The subscript ?i? 
represents the ?i?th downburst over a given period of time. 
 
Figure 
To obtain the probability of exceedence of a given wind speed for a given time peri
downbursts, the probability is first defined by an upcross rate 
speed measured will exceed a threshold or barrier level value 
design life time from t=0 to T. This is explained using th
S AND UPCROSS RATE: 
- ? - occurrence rate of downbursts 
? duration of the ?i?th downburst (with a mean values of ?
Fdb (w). A diagram of this process is shown below. Each 
 
 
A.1:  Poisson Pulse Process (Li 2000) 
? or the rate at which the wind 
a(t) sometime in the structure?s 
e following equation: 
 
113 
d), Wi 
od for 
 
114 
 
 
g1868g4666g1872g4667 g3404  g3505 g1874g4666g2028g4667g1856g2028
g3047
g2868
                 
In this equation p(0) is the probability that wind at a given location exceeds the threshold wind 
speed at the beginning of the evaluation period t=0 ? this value is considered to be zero.  g1874g4666g2028g4667 is 
the upcross rate as described above. The evaluation of this integral is obtained through the 
following process as outlined in Structural Reliability Analysis and Predictions by Melchers 
(1987).  
A process having rectangular pulses of intensity Wi and length di are described as: 
Wg4666uni03BBuni002C uni0064uni002C tuni002C tg2919uni002C g4667 g3404  Wg2919uni002C uni0077hen the uni0064uni006Funi0077nuni0062uni0075rst uni006Funi0063uni0063uni0075rs  
        g3404  0uni002C           at all uni006Fther times      
a(t)
 
Figure A.1: Poisson Pulse Process Diagram Modified to Include a(t) ? Generalized Threshold Value 
as a Function of Time  
The level of the upcross rate g1874g3028g2878g4666g1872g4667 of the pulses above the threshold value can be obtained 
directly from the limit as duration,  g1856 g3639 0.  The probability that W(t) starts out below the 
threshold value a(t) or g4670g1849g4666g1872g4667 g3409 g1853g4666g1872g4667g4671 then by the end of the duration, has crossed it so that 
g1849g4666g1872 g3397 uni2206g1872g4667 is greater than a(t)  or g4670g1849g4666g1872 g3397 uni2206g1872g4667 g3408 g1853g4666g1872g4667g4671 is equal to:  
 
115 
 
 
g1842g4668g4670g1849g4666g1872g4667 g3409 g1853g4666g1872g4667g4671 g1514 g4670g1849g4666g1872 g3397 uni2206g1872g4667 g3408 g1853g4666g1872g4667g4671g4669 
Taking the limit of this expression, multiplied by g2019  or the occurrence rate of these downburst 
wind events, the following expression is created: 
g1874g3028g2878g4666g1872g4667 g3404 limuni2206g3047g1372g2868 g3420 1uni2206g1872 g4670g1842g4666g1873g1868g1855g1870g1867g1871g1871g1861g1866g1859 g1861g1866 uni2206g1872g4667g4671g2019g3424 
                                              g3404 limuni2206g3047g1372g2868 g3420 1uni2206g1872 g4670g1842g4668g4670g1849g4666g1872g4667 g3409 g1853g4666g1872g4667g4671 g1514 g4670g1849g4666g1872 g3397 uni2206g1872g4667 g3408 g1853g4666g1872g4667g4671g4669g2019g3424 
To simplify the following equation into one that can be more easily evaluated the following 
things are considered. First in this case, a(t) represents the threshold value of the wind speed, or 
the design wind speed which is constant throughout time t called Vd. This is substituted in the 
equation: 
 g1874g3028g2878g4666g1872g4667 g3404 limuni2206g3047g1372g2868 g3420 1uni2206g1872 g4670g1842g4668g4670g1849g4666g1872g4667 g3409 g1848g3031g4671 g1514 g4670g1849g4666g1872 g3397 uni2206g1872g4667 g3408 g1848g3031g4671g4669g2011g3424 
Second, the union of two statistically independent events such as those given above is equal to 
the multiplication of the probabilities of each of the events (i.e. g1842g4666g1827 g1514 g1828g4667 g3404 g1842g4666g1827g4667g1842g4666g1828g4667 where A 
represents g4670g1849g4666g1872g4667 g3409 g1853g4666g1872g4667g4671 and B represents g4670g1849g4666g1872 g3397 uni2206g1872g4667 g3408 g1853g4666g1872g4667g4671. Additionally, based on the 
definition of a generalized cumulative distribution function F(x) being the probability of a 
density function g1858g4666g1871g4667 less than a given value x: 
g1832g4666g1876g4667 g3404  g1842g4666g1850 g3409 g1876g4667 g3404  g3505 g1858g4666g1871g4667g1856g1871
g3051
g2879g2998
 
Translated into the terms used in this paper where Vd is the wind speed threshold value: 
g1832g3031g3029g4666g1848g3031g4667 g3404  g1842g4666g1849g3036g4666g1872g4667 g3409 g1848g3031g4667 g3404  g3505 g1874g4666g2028g4667g1856g2028
g3023g3279
g2879g2998
 
 
116 
 
 
The probability that W(t) will be less than a(t) as time approaches zero is equal to the cumulative 
probability function of that density function Fdb(a(t)). Also based on this same definition, the 
probability that g1849g4666g1872 g3397 uni2206g1872g4667 will be larger than a(t) is equal to one minus that cumulative density 
function: 
g1833g4666g1876g4667 g3404  g1842g4666g1850 g3408 g1876g4667 g3404  1 g3398 uni0046g4666g1876g4667 
Translated into the terms used in this paper: 
g1833g3031g3029g4666g1848g3031g4667 g3404  g1842g4666g1849g3036g4666g1872 g3397 uni2206g1872g4667 g3408 g1848g3031g4667 g3404  1 g3398 g1832g3031g3029g4666g1848g3031g4667 
 Substituting these into the original equation, the final equation obtained is: 
g1874g4666g1872g4667 g3404 g1832g3031g3029g4666g1848g3031g4667g1833g3031g3029g4666g1848g3031g4667g2019 
When Vd is large and constant, as is the case in this equation, then g1832g3031g3029g4666g1848g3031g4667 g3406 1 because the wind 
values almost always are below the threshold value except in rare cases, the expression for the 
upcross rate then becomes a much more usable equation: 
g1874g4666g1872g4667 g3404 g1833g3031g3029g4666g1848g3031g4667g2019 
The exceedence probability of a given set of wind data due to downbursts over the threshold 
value Vd is then found by multiplying the upcross rate by time t to get the probability over a give 
time. 
g1868g4666g1872g4667 g3404 g1874g1872 g3404 g1833g3031g3029g4666g1848g3031g4667g2019g1872 
  
 
117 
 
 
REFERENCES 
Abiassi, J. J. (1981). The strength of weathered window glass using surface characteristics. 
Institute for Disaster Research, Texas Tech University, Lubbock, Tex. 
 
AGC Solar. (2010). ?Thin Film CIGS.? < http://www.agc-solar.com/solar-technologies/solar-
photovoltaics/thin-film-cigs.html>. (November 4, 2010).  
 
AIS Glass (AIS). (2010). PVB Laminated Glass. <http://www.aisglass.com/pvb_laminated.asp>. 
(October 29, 2010).  
 
American Architectural Manufacturers Association (AAMA). (1984). Curtain Wall Manual #12 
? Structural Properties of Glass. Schaumburg, IL. 
 
American Society of Civil Engineers (ASCE). (2005). ?Wind Loads.? Minimum Design Loads 
for Buildings and Other Structures. ASCE 7-05.  
 
American Society of Testing and Materials (ASTM). (2004). ?Standard specification for heat-
treated flat glass.? ASTM C1048.  
 
American Society of Testing and Materials (ASTM). (2009). ?Standard practice for determining 
load resistance of glass in buildings.? ASTM E1300. 
 
Arnold, C. (2009). ?Building Envelope Design Guide ? Introduction.? Whole Building Design 
Guide. National Institute of Building Sciences. 
<http://www.wbdg.org/design/env_introduction.php#evol>. (September 17, 2010). 
 
AstroEnergy. (2010). ?Thin-film solar panels?. <http://www.astronergy.com/products.php>. 
(September 23, 2010).  
 
Atkins, N.T., and Wakimoto, R.M. (1991). ?Wet Microburst Activity Over the Southeast United 
States Implications for Forecasting.? Weather and Forecasting, 6 (4), 470-482.  
 
Avancis. (2010). ?Avancis Powermax.? <http://www.avancis.de/en/products/powermaxr/>.  
(September 24, 2010).  
 
B. Sopori, ?Thin-film silicon solar cells.? In: Handbook of Photovoltaic Science and 
Engineering, Wiley-VCH, Weinheim. 320. 
 
Beason, W. L. (1980). "A failure prediction model for window glass," PhD dissertation, Texas 
Tech University, Lubbock, Tex. 
 
 
118 
 
 
Beason, W. L., and Morgan, J. R. (1984). "Glass failure prediction model." J. Struct. Engrg., 
ASCE, 110(2), 197-212. 
 
Beason, W. L., and Norville, S. H. (1989). "Development of a new glass thickness selection 
procedure." Proc., 6th U.S. Nat. Conf. on Wind Engrg., Vol. II, University of Houston, 
Houston, Tex. 
 
Beason, W.L., Kohutek, T.L., and Bracci, J.M. (1998). ?Basis for ASTM E1300 Annealed Glass 
Thickness Selection Charts?. J. Struct. Engrg. 124 (2), 215-221.  
 
Behr, R. A., Minor, J. E., and Linden, M. P. (1986). "Load duration and interlayer thickness 
effects on laminated glass." J. Struct. Engrg., ASCE, 112(6), 1441-1453. 
 
Behr, R. A., Minor, J. E., Linden, M. P., and Vallabhan, C. V. G. (1985). ??Laminated glass units 
under uniform lateral pressure.?? J. Struct. Engrg., 111(5), 1037?1050. 
 
Bennisson, S. J., et al. (2005). ?Strength and Deformation Behavior of Laminated Glass.? Proc. 
Glass Proc. Days Conf., Tampere, Finland.  
 
Bonnet D, Ed, Int. J. Solar Energy, 12, Harwood Academic Publishers, Reading, U.K. (1992). 
 
Brock, F.V., and Govind, P.K. (1977). ?Portable Automated Mesonet in Operation? J. Appl. 
Meteor., 16, 299-310.  
 
Brooks, H., Doswell, C.A. and  Kay, M.P. (2003). ?Climatological estimates of local daily 
tornado probability for the United States.?  Wea. Forecasting, 18, 626?640. 
 
Brown, J.M., Knupp, K.R., and Caracena, F. (1982). ?Destructive winds from shallow high-
based cumulonimbi.? Preprints, Twelfth Conference on Severe Local Storms. American 
Meteorological Society, Boston, 272-275. 
 
Button, D., and Pye, B. (1993). Glass in Buildings. Butterworth Architecture, Reed International 
Books, Oxford, U.K. 
 
Caracena, F., and Flueck, J.A. (1988). ?Classifying and predicting microburst activity in the 
Denver, Colorado, Area.? J. of Aircraft, 25, 525-530. 
 
Caracena, F., McCarthy, J., and Fluck, J. (1983). ?Forcasting the Likelihood of microbursts 
along the front range of Colorado.? Preprints. 13th Conf. on Severe Local Storms, Tulsa, 
Amer. Meteor. Soc., 261-264. 
 
Carlson, D. E. (1977). U.S. Patent No. 4,064,521. 
 
 
119 
 
 
Chapin, D., Fuller, C., and Pearson, G. (1954). ?A New Silicon p-n Junction Photocell for 
Converting Solar Radiation Into Electrical Power.? Journal of Applied Physics, 25, 676-
677. 
 
Chen J et al., Thin Solid Films 219, 183?192 (1992). 
 
Cheyney, T. (2010). ?CIGS PV module conversion efficiency contest heats up, as MiaSole hits 
13.8%, Solibro reaches 13%? PV-tech. <http://www.pvtech.org/news/_a/cigs_ 
module_conversion_efficiency_contest_heats_up_as_miasole_hits_13.8_and/>. 
(September 7, 2010). 
 
Chittick, R. C., Alexander, J. H., and Sterling, H. F.  (1969). J. Electrochem. Soc. 116, 77  
 
Chuwieduk, D. (2003). ?Towards Sustainable-Energy Buildings.? Applied Energy, 76 (1-3), 211-
217. 
 
Contreras M et al. (1999). Prog. Photovolt. 7, 311?316. 
 
Cook, N.J. (1982). ?Towards Better Estimation of Extreme Winds.? Journal of Wind Engr. and 
Indus. Aerodyn., 9(3), 295-323. 
 
D'Agostino, R.B. and Stephens, M.A. (1986). Goodness-of-Fit Techniques, Marcel Dekker, Inc., 
New York,  146. 
 
De Nobel, D. (1959). Philips Res. Rpts 14, 361?399 and 430?492. 
 
Doswell, C. A. (1994). ?Extreme Convective Windstorms: Current Understanding and 
Research.? Proc. U.S.-Spain Workshop on Natr. Haz., Barcelona, Spain, 44-55. 
 
Doswell, C.A., Brooks, H.E., and Kay, M.p. (2005). ?Climatology Estimates of Daily Local 
Nontornadic Severe Thunderstorm Probability for the United States.? Weather and 
Forecasting, 20(4), 577-595.  
 
Dougherty, B. P., Fanney, A. H., and Davis, M. W. (2005). ?Measured Performance of Building 
Integrated Photovoltaic Panels. Round 2.? Journal of Solar Energy Engineering, 127, 314-
323.  
 
Edel, M. T. (1997). ??The effect of temperature on the bending of laminated glass beams,?? MS 
thesis, Department of Civil Engineering, Texas A&M University, College Station, Tex. 
 
Ellrod, G. P., Nelson, J. P.,  Witiw, M. R.,  Bottos, L., and Roeder, W. P. (2000). ?Experimental 
GOES sounder products for the assessment of downburst potential.?  Wea. Forecasting, 15, 
527?542. 
 
120 
 
 
ENF.cn. (2010). Solar Panel PV Modules Company Directory. Surrey, U.K. <http://www.enf.cn/ 
database/panels.html>. (September 7, 2010). 
 
Environmental Protection Agency. (2009). ?Buildings and Their Impacts on the Environment: A 
Statistical Summary.? <http://www.epa.gov/greenbuilding/pubs/gbstats.pdf>. (September 17, 
2010).  
 
EPV Solar. (2010). ?EPV Solar EPV-4X, EPV-5X Solar Module Spec Sheet? 
<http://www.epv.net/productspecs.pdf>. (September 1, 2010). 
 
Evans, J.W., Johnson, R.A., and Green, D.W. (1989) ?Two and three parameter weibull 
goodness-of-fit tests.? Forest Products Laboratory Research Paper FPL-RP-493. United 
States Department of Agriculture (USDA).  
 
Evans, J.W., Johnson, R.A., and Green, D.W. (1989).?Two and three parameter Weibull 
goodness of fit tests.? Research Paper FPL-RP-493, U.S. Department of Agriculture, Forest 
Service, Forest Products Laboratory.  
 
Federal Highway Administration (FHA). (2002). ?FHWA Route Log and Finder List.? 
<http://www.fhwa.dot.gov/reports/routefinder/table2.cfm>. (October 6, 2010). 
First Solar. (2010). ?Cadmium Telluride?. <http://www.firstsolar.com/en/CdTe.php>. 
(September 23, 2010).  
 
Flocker, F.W., and Dharani, L.R. (1996). ?Stress in laminated glass subject to low velocity 
impact.? Engineering Structures. 19(10), 851-856. 
 
Foss, R. V. (1994). ??LAMe hybrid system for glazing design.?? Hybrid neural network and 
expert systems, L. R. Medsker, ed., Kluwer Academic, Norwell, Mass. 
 
Frerichs, R. (1947) Phys. Rev. 72, 594?601. 
 
Freund, L. B., and Suresh, S., (2003). ?Thin film materials.? Cambridge University Press,  96. 
 
Fritts. C. (1883). ?On the Fritts Selenium Cells and Batteries.? Proceedings of the American 
Association for the Advancement of Science. 33, 97. 
 
Fujita, T. T. (1976). "Spearhead Echo and Downburst Near the Approach End of a John F. 
Kermedy Airport Runway.? SMRP Res. Project Research Paper 137. New York City, Dept. 
of Geophysical Sciences, Univ. of Chicago.  
 
Fujita, T. T. (1985). ?The downburst, microburst, and macroburst.? SMRP Res. Paper No. 210, 
NTIS No. PB85-148880, Univ. of Chicago, 122. 
 
 
121 
 
 
Fujita, T. T., and Byers, H. (1977). "Spearhead Echo and Downburst in the Crash of an Airliner." 
Monthly Weather Review, 105, 129-146. 
 
Fujita, T.T. (1978). ?Manual of downburst identification for Project Nimrod.? SMRP Research 
Paper No. 156, University of Chicago. 
 
Fujita, T.T. (1983). ?Microburst wind shear at New Orleans International Airport, Kenner, 
Louisiana, on July 9, 1982.? SMRP Res. Paper No. 199, Univ. of Chicago, 39. 
 
Fujita, T.T. (1985). ?The downburst, microburst, and macroburst.? SMRP Res. Paper No. 210 
[NTIS No. PB85-148880], Univ. of Chicago, 122. 
 
Fujita, T.T. (1985). The Downburst. Report of Projects NIMROD and JAWS, University of 
Chicago. 
 
Gast, K. D., and Schroeder J. L. (2003). ?Supercell Rear-Flank Downdraft as Sampled in the 
2002 Thunderstorm Outflow Experiment,? Proceedings, Eleventh International Conference 
on Wind Engineering, Lubbock, Texas, 2233-2240. 
 
Geerts, B. (2001). ?Estimating downburst-related maximum surface wind speeds by means of 
proximity soundings in New South Wales, Australia.? Wea. Forecasting, 16, 261?269. 
 
Gessert, T. A. ?Review of Photovoltaic Energy Production Using CdTe Thin-Film Modules? 
National Renewable Energy Laboratory, NREL/AB-520-44128.  
 
Glass Association of North America (GANA). (2008). ?GANA Glazing Manual?. 
 
Gleskova, H.;   Wagner, S.;   Suo, Z. (1999). ?Failure Resistance of Amorphous Silicon 
Transistors Under Extreme in-plan strain.? Applied Physics Letters. 75 (19), 3011-3013.  
 
Green, M. A. (1995). Silicon Solar Cells: Advanced Principles and Practice. Bridge Printery, 
Sydney. 
 
Greentech Media Research (GTM). (2010). Building-Integrated Photovoltaics: An Emerging 
Market. <http://www.gtmresearch.com/report/building-integrated-photovoltaics-an-
emerging-market>. (October 24, 2010).  
 
Griffith, A.A. (1921). ?The Phenomenon and Rupture of Flow in Solids? Philosophical 
Transactions of the Royal Society of London, 221, 163-198.  
 
Gross, J. L., Main, J. A., Phan, L.T., Sadek, F.H., Cauffman, S.A., and Jorgensen, D.P. (2010). 
?Final Report on the Collapse of the Dallas Coyboys Indoor Practice Facility, May 2, 2009.? 
NISTIR 7661. National Institute of Standards and Technology (NIST).  
 
122 
 
 
Guardian Industries Corp. (2010). ?Guardian SunGuard Technical Information.? 
<http://www.sunguardglass.com/>. (July 1, 2010). 
 
Hahn, H. et al. (1953). Z. Anorg. Allg. Chem. 271, 153?170. 
 
Hegedus, S., and Luque, A. (2002). ?Status, Trends, Challenges and the Bright Future of Solar 
Electricity from Photovoltaics.? 11-15. 
 
Hershey, R L., and Higgins, T. H. (1973). "Statistical prediction model for glass breakage from 
nominal sonic boom loads." Rep. No. FAARD- 73-79, Booz-Allen Applied Research, Inc., 
Bethesda, MD. 
 
Hess, R. (2004). ?Material Glass.? Structural Engineering International, 14(2), 76-80. 
 
Hjelmfelt, M.R. (1988). ?Structure and life cycle of microburst outflows observed in 
Colorado.? J. Appl. Met. 27, 900?927. 
 
Holmes, J.D. (1999). ?Modeling of extreme thunderstorm winds for wind loading of structures 
and risk assessment.?  Wind Engineering into the 21st Century. Proceedings of Tenth 
International Conference on Wind Engineering Larsen, A., Larose, G.L., Livesey, F.M., eds., 
Denmark. 
 
Holmes, J.D. (2002). ?A re-analysis of recorded extreme wind speeds in region A,? Australian 
Journal of Structural Engineering, 4, 29?40. 
 
Holmes, J.D. (2002). ?A Re-analysis of record extreme wind speeds in region.? A. Aust. J. Struct. 
Eng. 4(1). 
 
Holmes, J.D., and Moriarty, W.W. (1999). ?Application of the generalized Pareto distribution to 
extreme value analysis in wind engineering,? J. Wind Eng. Ind. Aerodyn., 83, 1?10. 
 
Hooper, J. A. (1973). ??On the bending of architectural laminated glass.? Int. J. Mech. Sci., 15, 
309?323. 
 
Hsu, S. (1987). ?Comparison of window glass sample strength estimates: full scale vs concentric 
ring.? M. Thesis, Texas Tech University. 60. 
 
International Electrotechnical Commission (IEC). (2008). ?Thin-film terrestrial photovoltaic 
(PV) modules ? Design qualification and type approval.? IEC 61646.  
 
Jenny, D., and Bube, R. (1954). Phys. Rev. 96, 1190?1191. 
 
Kaltschmitt. M., and Streicher, W. (2007). Renewable Energy: Technology, Economics and 
Environment. Springer, New York, 251. 
 
123 
 
 
Kanabolo, D. C , and Norville, H. S. (1985). "The strength of new window glass plates using 
surface characteristics." NTIS Accession No. PB86-140100, Glass Res.and Testing Lab., 
Texas Tech Univ., Lubbock, Tex., Jul. 
 
Kazmerski, L., White, F., Morgan, G. (1976). Appl. Phys. Lett. 29, 268, 269. 
 
Kelly, D.L., Schaefer, J.T., and Doswell, C.A. (1985). ?Climatology of Nontornadic Severe 
Thunderstorm Events in the United States.? Mon. Wea. Rev, 113, 1997-2014. 
 
Kelly, D.L., Schaefer, J.T., NcNulty, R.P., Doswell, C.A., and Abbey, R.F. (1978). ?An 
Augmented Tornado Climatology.? Mon. Wea. Rev., 106, 1172-1183. 
 
Kim, J., and Hangan, H. (2007). ?Numerical Simulations of Impinging Jets with Applications to 
Downbursts? J. of Wind Engr. and Indus. Aerody, 95 (4), 279-298.  
 
Kim, J.D., and Hangan, H. (2007).  ?Numerical simulations of impinging jets with application to 
downbursts,? Journal of Wind Engineering and Industrial Aerodynamics, 95, 9?298. 
 
King, K.W. (1996). ?The Effect of Interlayer Thickness on Laminated Glass Strength.? M. 
Thesis, Texas Tech University, TX. 
 
King, N. J., and King, B. J. (2004) ?Creating Incentives for Sustainable Buildings: A 
Comparative Law Approach Featuring the United States and the European Union.? Va. Envtl. 
L.J., 23, 397-399. 
 
Kolmogorov, A. (1933). ?Grundbegriffe der Wahrscheinlichkeitsrechnung. Ergbnisse der 
Mathematick un ihrer Grenzgebiete, 2(3). 
 
Krall, W. R, Siskos, W. R, Stewart, R. A., and Spindler, R. G. (1981). "The behavior of float 
glass under uniform wind loading." 4th U.S. Nat. Conf. on Wind Engrg. Res., University of 
Washington, Seattle, Wash. 
 
Kruger, F., de Nobel, D. (1955). J. Electron. 1, 190?202. 
 
Li, C.Q. (2000). ?A Stochastic Model of Severe Thunderstorms for Transmission Line Design.? 
Prob. Engineering Mech., 15(4), 359-364.  
 
Linden, M. P., Minor, J. E., and Vallabhan, C. V. G. (1984). "Evaluation of laterally loaded 
laminated glass units by theory and experiment, supplemental report No. 1." NTIS 
Accession No. PB85-1U532, Glass Res. and Testing Lab., Texas Tech Univ., Lubbock, 
Tex., Jul. 
 
Lindmayer, J., and Allsion, J. ?The violet cell: An improved silicon solar cell.? COMSAT 
Technical Review, 3, 1?22. 
 
124 
 
 
Loferski, J. ?Theoretical considerations governing the choice of the optimum semiconductor for 
photovoltaic solar energy conversion.? Journal of Applied Physics. 27, 777?784. 
 
Madou, M. J. (2002). Fundamentals of microfabrication: the science of miniaturization. CRC 
Press, 298. 
 
McCann, D. W. (1994). ?Windex ? A new index for forecasting microburst potential.? Wea. 
Forecasting, 9, 532-541.  
 
McCarthy, J., Wilson, J.W., and Fujita, T.T. (1982). ?The Joint Airport Weather Studies (JAWS) 
Project.? Bull. Amer. Meteor. Soc., 63, 15?22. 
 
McDonald, J.R, and Abbey, R.F. (1979). ?Comparison of the NSSFC and DAPPLE Tornado 
Data Tapes.? Preprints 11th Conf. on Severe Local Storms, Amer. Meteor. Soc., Boston, MA, 
235-240. 
 
McGuire, G. E. (1988). Semiconductor Materials and Process Technology Handbook. William 
Andrew Publishing, Noyes. 
 
McMahon, S. (2008). ?NREL boots CIGS thin-film solar cell efficiency to record 19.9%.? PV-
Tech. <http://www.pvtech.org/news/_a/nrel_boosts_cigs_thin_film_solar_cell_ 
efficiency_to_ record_199_percent/>. (September 7, 2010). 
 
Melcheres, R.E. (1999). Structural Reliability Analysis and Prediction, 2nd ed. John Wiley & 
Sons, Chichester (reprinted June 2001). 
 
Michael, H., and Seaman, K. (2009). ?The Climate and Weather of Delaware, Maryland and 
Washington D.C.? Weatherwise.  
 
Mimilya-Arroyo, J., Marfaing, Y., Cohen-Solal ,G., and Triboulet, R., (1979). Sol. EnergyMater. 
1, 171. 
 
Minor, J.E., and Reznik, P.L. (1990). ?Failure Strengths of Laminated Glass.? J. Structural 
Engineering. 116(4), 1030-1039. 
 
Mitchell K, Fahrenbruch A, Bube R, J. Appl. Phys. 48, 829?830 (1977). 
 
Morel, D. J., Jhosh, A. K., Feng, F., Stogryn, G. L., Purwin, P. E., Shaw, R. S., and Fishman, C. 
(1978). Appl. Phys. Lett. 32, 495. 
 
Naess, A. (1998). ?Estimation of long return period design values for wind speeds.? Journal of 
Engineering Mechanics,124, 252-9. 
 
Nakazawa T, Takamizawa K, Ito K, Appl. Phys. Lett. 50, 279?280 (1987). 
 
125 
 
 
Nanosolar. (2010). ?Nanosolar Utility Panel Data Sheet.?  <http://www.nanosolar.com/sites/ 
default/files/Nanosolar_Utility_Panel_Data_Sheet_r2.0.pdf>. (August 1, 2010). 
 
National Climactic Data Center (NCDC). (2010). ?Storm Event Database.? National Oceanic and 
Aviation Administration (NOAA). <http://www4.ncdc.noaa.gov/cgi-
win/wwcgi.dll?wwEvent~Storms>. (September 17, 2010). 
 
National Weather Service (NWS). (1995). NWS Operations Manual, Chapter C-40, Silver 
Spring, MD, 61. 
 
National Weather Service (NWS). (2010). ?Downburst.? NWS Forecast Office, Columbia, SC. 
<http://www.erh.noaa.gov/cae/svrwx/downburst.htm>. (September 17, 2010). 
 
National Weather Service (NWS). Weather Forecasting Office Baltimore/Washington. < 
http://www.erh.noaa.gov/er/lwx/>. (September 17, 2010). 
 
National Weather Service. (2010). ?NWS CWFA Areal Coverage Sorted by Area.? 
<http://www.weather.gov/mirs/public/prods/reports/pdf/office_facility/cwfa_sq_mi.pdf>. 
(September 21, 2010). 
 
Nitkin, K. (2007) ?States? Green Building Laws ?Lead By Example.? The Pew, Center on the 
States. <http://www.stateline.org/live/details/story?contentId=257232>. (September 17, 
2010). 
 
Norville, H. S., King, K. W., and Swofford, J. L. (1998). ??Behavior and strength of laminated 
glass.?? J. Engrg. Mech., ASCE, 124(1), 46?53. 
 
Norville, S. H., and Minor, J. E. (1985). "Strength of weathered window glass." Am. Ceramics 
Soc. Bull., 64(11), 1467-1470. 
 
Oliver, S.E, Moriarty, W.W. and Holmes, J.D. (2000). ?A Risk Model for Design of 
Transmission Line Systems Against Thunderstorm Downburst Winds.? Engineering 
Structures, 22, 1173-1179.  
 
Orr, L. (1957). "Engineering properties of glass." Publ. 478, Build. Res. Inst., Nat. Acad. of Sci., 
National Research Council, Washington, D.C. 
 
Oseguera, R.M., and Bowles, R.L. (1988). ?A simple, analytic 3-dimensional downburst model 
based on boundary layer stagnation flow.?  N.A.S.A. Technical Memorandum 100632. 
 
Pandey, M.D., Van Gelder, A.J.M., and Vrijling, J.K. (2003). ?Bootstrap simulations for 
evaluating the uncertainty associated with peaks-over-threshold estimates of extreme wind 
velocity.? Environmetrics, 14, 27?43. 
 
126 
 
 
Peterka, J.A. and Shahid, S. (1998). ?Design Gust Wind Speeds in the United States.? J. of Struc. 
Engrg. , 124(2), 207-214.  
 
Phillips, J., Birkmire, R., and Lasswell, P. (1982). "Stability of Thin-Film Cu2S-Based Solar 
Cells at Voc Under Continuous Illumination," Proc. 16th IEEE Photovoltaic Specialist 
Conf., 719?722. 
Pilkington. (2010). ?Glass Strength? <http://www.pilkington.com/europe/norway/norwegian/ 
building+products/pilkington4architects/about+glass/mechanicalfunctionsofglass/glass+str
ength.htm>. (October 17, 2010).  
 
Ponpon J, Siffert P, Rev. Phys. Appl. 12, 427?431 (1977). 
 
PPG glass thickness recommendations to meet architects' specified 1minute wind load. (1979). 
Tech. ServicesJFlat Glass Div., PPG Industries, Pittsburgh, PA. 
 
Prince, M. (1955). ?Silicon Solar Energy Converters.? Journal of Applied Physics, 26, 534?540. 
 
Pryor (2009). ? Personal communication on ?Downburst Climatology for the Eastern United 
States 2001-2003.? 
 
Pryor, K. L., Ellrod, G. P., and Bailey, A. A. (2002). ?Convective downburst potential using 
GOES sounder derived products.? National Weather Association Electronic Journal of 
Operational Meteorology, EJ1. 
 
Pryor, K. L., Ellrod, G. P., and Bailey, A. A. (2003). ?Recent Improvements to the GOES Wind 
Index.? National Weather Association Electronic Journal of Operational Meteorology, EJ1. 
 
Q-cell. (2010). ?CIGS Technology: For the best performance and design.? <http://www.q-
cells.com/en/solar_modules/module_portfolio/qsmart/ cigs_technology/index.html>. 
(September 23, 2010).  
 
Rajabi, M.R., and Modarres, R. (2008). ?Extreme Value Frequency Analysis of Wind Data from 
Isfahan, Iran.? J. Wind Engr. and Indust. Aerodyn., 96(1), 78-82. 
 
Reynolds, D., Leies, G., Antes, L., and Marburger. R. (1954). ?Photovoltaic effect in cadmium 
sulfide? Physics Review. 96, 533, 534. 
 
Reznik, P. L., and Minor, J. E. (1986). "Failure strengths of laminated glass units." NTIS 
Accession No. PB87-118873, Glass Res. and Testing Lab., Texas Tech Univ., Lubbock, 
Tex., Oct. 
 
Rinehart, R.E., DiStefano, J.T., and Wolfson, M.M. (1987). ?Preliminary Memphis FAA/Lincoln 
Laboratory Operational Weather Studies Results.? Project Report ATC-141, Lincoln 
Laboratory, Massachusetts Institute of Technology. 
 
127 
 
 
Roger Edwards. (2006). "The Online Tornado FAQ". National Weather Service. National 
Oceanic and Atmospheric Administration. <http://www.spc.ncep.noaa.gov/faq/tornado/>. 
(September 9, 2010).  
 
Romeo, A., Khrypunov, G., Kurdesau, F., Batzner, D.L., Zogg, H., and Tiwari, A.N. (2005). 
?High-efficiency flexible CdTe solar cells on polymer substrates.?14th Intl. Photovolt. Sci. 
and Egr. Conf., 90(18-19), 304-3415. 
 
Rose, M. (2010). ?Downbursts.? National Weather Service (NWS). 
<http://www.srh.noaa.gov/ohx/?n=downbursts>. (September 17, 2010). 
 
Runyon, R. (2010). ?BIPV Market Gaining Traction.? Renewable Energy World. 
<http://www.renewableenergyworld.com/rea/news/article/2010/08/bipv-market-gaining-
traction>. (September 16, 2010). 
 
Savory, E., Park, G.A.R., Zeinoddini, M., Toy, N., and Disney, P. (2001). ?Modeling of tornado 
and microburst-induced wind loading and failure of a lattice transmission 
tower.? Engineering Structures, 23, 365?375. 
 
Schaefer, J.T., and Edwards, R. (1999). ?The SPC Tornado/Severe Thunderstorm Database.? 11th 
Conf. Applied Climatology, Dallas, TX.  
 
Schaefer, J.T., Kelly, D.L., and Abbey, R.F. (1986). ?A Minimum Assumption Tornado-Hazard 
Probability Model.? J. Climate Appl. Meteor., 25, 1934-1945.  
 
Schott Solar. (2010). ?BIPV- Building Integrated Photovoltaics with Schott ASI Solar Modules.? 
<http://www.us.schott.com/2009_architecture/english/products/bipv/asiglass.html>. 
(September 1, 2010). 
 
Sedlack, R. (1968). ?Tensile fatigue strength of brittle materials.?  Stanford Research Inst.  
Shay, J., Wernick, J. (1974). Ternary Chalcopyrite Semiconductors: Growth, Electronic 
Properties, and 
 
Shaybani, Reza. (2010). ?Building Integrated Photovoltaics (BIPV).? Scheuten Solar Glass. 
<http://www.sesam-uae.com/greenbuilding/presantation/SOLARDAYS_PPP.pdf> 
(September 10, 2010). 
 
Shockley, W., and Queisser, H. (1961). J. Appl. Phys. 32, 510?519. 
 
Simiu, E., and Heckert, N.A. (1996). ?Extreme Wind Distribution Tails: A Peaks over Threshold 
Approach.? J. Struct. Engrg., 122(5), 539-547.  
 
128 
 
 
Simiu, E., Heckert, N.A., and Filliben, J.J. (1999). ?Comparisons of wind loading estimates 
based on extreme speeds and on square of extreme speeds.? Proceedings, 13th ASCE Eng. 
Mechs. Div. Conference, Baltimore, MD. 
 
Simiu, E., Heckert, N.A., Filliben, J.J., and Johnson, S.K. (2001). ?Extreme wind load estimates 
based on the Gumbel distribution of dynamic pressures: an assessment.? Structural Safety, 
23(3), 221-229.  
 
Stephens, M. A. (1974). ?EDF Statistics for Goodness of Fit and Some Comparisons.? Journal of 
the American Statistical Association, 69, 730-737. 
 
Stephens, M.A. (1974). EDF Statistics for Goodness of Fit and Some Comparisons, Journal of 
the American Statistical Association, 69, 730-737. 
 
Strong, S., (1996). ?World Overview of Building-Integrated Photovoltaics? Proceedings of 25th 
IEEE Photovoltaic Specialist Conference, 1197?1202. 
 
Stuar, B. ?Researchers announce 20.3 percent world record CIGS solar cell efficiency.? PV 
Magazine. <http://www.pv-magazine.com/news/details/beitrag/researchers-announce-new-
203-percent-world-record-cigs-solar-cell-efficiency_100000748/>. (November 4, 2010).  
 
Suri, D., Nagpal K, Chadha G, J. Appl. Cystallogr. 22, 578 (1989) (JCPDS 40-1487). 
 
The National Weather Service (NWS). (2006). ?NWS CWFA areal coverage sorted by area.? 
<http://www.weather.gov/mirs/public/prods/reports/pdf/office_facility/cwfa_sq_mi.pdf>. 
(October 6, 2010).  
 
Thom, H.C.S. (1963). ?Tornado Probabilities.? Mon. Wea. Rev., 91, 730-736. 
 
Topbas, A. (2008). ?An Overview of State-of-the-Art Report for Glass Structures.? Proc. AEI 
Bld. Integ. Solns. Architectural Engineering Institute (AEI), Denver, CO. 
 
Tsai, C. R., and Stewart, R. A. (1976). "Stress analysis of large deflection of glass plates by the 
finite-element method." J. Am. Ceramic Soc., 59(9-10),445-448. 
 
Turner, C., and Frankel, M. (2008). ?Energy Performance of LEED for New Construction 
Buildings: Final Report.? New Buildings Institute. <http://www.usgbc.org/ 
ShowFile.aspx?DocumentID=3930>. (September 17, 2010), 2.  
 
Twisdale, L.A., and Dunn, W.L. (1983). ?Probabilistic analysis of tornado wind risks.? J. Struct. 
Div., 109, 468-488. 
 
United States Army Corp of Engineers (USACE). (2010). ?Buildings and Structures: Causes and 
Characteristics of Hurricane Damage.? USACE National Hurricane Study Program. 
 
129 
 
 
<http://chps.sam.usace.army.mil/ushesdata/Assessments/alicia/buildings_damage.htm>. 
(September 17, 2010).  
 
Vallabhan, C. V. G., and Wang, B. Y.-T. (1981). "Nonlinear analysis of rectangular glass plates 
by finite difference method." Inst. for Disaster Res., Texas Tech University, Lubbock, Tex. 
 
Van Duser, A., Jagota, A, and Bennison, S.J. (1999). ?Analysis of Glass/Polyvinal Butyral 
Laminates Subjected to Uniform Pressure.? J. Engrg. Mech. 125(4), 435-442. 
 
Vink T, Somers M, Daams J, Dirks A, J. Appl. Phys. 70, 4301?4308 (1991). 
 
Wakimoto, R.M. (1985). ?Forecasting dry microburst activity over the high plains.? Monthly 
Weather Rev., 113, 1131-1143. 
 
Walter, J.C. (2007).  ?A new climatology of 25-year, 50-year, and 100-year microburst winds.? 
22nd Conference on Weather Analysis and Forecasting, 5a.6. 
 
Webb, J. (2010). ?Thin specialty glass for thin-film reliability.? European Photovoltaic Solar 
Energy Conference, Valencia. 
 
Weibull, W. (1939). A statistical theory of the strength of materials. 
Ingeniorsvetenskapsakademiens, Handlinger, NR 151, Stockholm, Sweden. 
 
Wesoff, E. ?CIGS: Will technology be revealed before funding dries up?? Seeking Alpha. 
<http://seekingalpha.com/article/139444-cigs-will-technology-be-revealed-before-funding-
dries-up>. (September 7, 2010). 
Wilson, J.W., Moore, J.A., Brant Foote, G., Martner, B., Rodi, A.R., Uttal, T., Wilczak, J.M. 
?Convective Initiation and Downburst Experiment (CINDE).? Bulletin of the American 
Meteorologoical Society, 69(11), 1328-1348. 
 
Wilson, J.W., Roberts, R.D., Kessinger, C., and McCarthy, J. (1984). ?Microburst wind structure 
and evaluation of Doppler radar for airport wind shear detection.? J. Climate 
Appl. Meteor., 23, 898-915. 
 
Wolfson, M. (1983). ?Doppler Radar Observations of an Oklahoma Downburst.? Preprints, 21 
Conf. Radar Meteorology, Edmonton, Amer. Meteor. Soc., 590-595. 
 
Wood, G.S., Kwok, K.C.S.,  Motteram, N.A.  and Fletcher, D.F. (2001) ?Physical and numerical 
modeling of thunderstorm downbursts.?  Journal of Wind Engineering and Industrial 
Aerodynamics 89, 535?552. 
 
Wysocki, J., and Rappaport, P. ?Effect of temperature on photovoltaic solar energy conversion.? 
Journal of Applied Physics, 31, 571?578. 
 
130 
 
 
Xie, J., Uchida, S., Rand, B. P., and Forrest, S. R. (2004). ?4.2% efficient organic photovoltaic 
cell with low series resistances? Appl. Phys. Lett. 84, 3012.  
 
Yamaguchi K, Matsumoto H, Nakayama N, Ikegami S, Jpn. J. Appl. Phys. 16, 1203?1211 
 
Yuh-Lang, L. (2007). Mesoscale Dynamics. The Cambridge University Press, Cambridge U.K., 
275-276.