<%BANNER%>

Measured Hurricane Wind Pressure on Full-Scale Residential Structures: Analysis and Comparison to Wind Tunnel Studies an...


PAGE 1

MEASURED HURRICANE WIND PRESSURE ON FULL-SCALE RESIDENTIAL STRUCTURES: ANALYSIS AND COMPARISON TO WIND TUNNEL STUDIES AND ASCE-7 By LUIS DAVID APONTE-BERMUDEZ A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2006

PAGE 2

Copyright 2006 by Luis D. Aponte-Bermdez

PAGE 3

This document is dedicated to all the people that had been affected by the impact of hurricanes and to all the students and professo rs that had collaborated in the research effort presented in this dissertation.

PAGE 4

ACKNOWLEDGMENTS I would like to extend my gratitude to all the people that have made possible the accomplishment of this work. Especially, I would like to thank my advisor, Dr. Kurtis Gurley, for his wisdom and support for the last couple years. I would also like to thank Dr. Timothy Reinhold for his vision and work in the wind engineering community, who initiated the research work presented in this dissertation. Many thanks go also to the members of my committee and fellow researcher Dr. Forrest Masters for his closeness. In addition I thank Dr. Prevatt and Zhuzhao Liu for their work and contribution in the research work presented. I thank the University of Puerto Rico for its financial support in my doctoral studies and the Florida Department of Community Affairs, National Oceanic and Atmospheric Administration, Florida and South Carolina Sea Grant for funding this research. For their prayers, unconditional and loving support, I thank my wife Cassandra, my parents, my grandparents and my extended family. iv

PAGE 5

TABLE OF CONTENTS page ACKNOWLEDGMENTS .................................................................................................iv LIST OF TABLES ...........................................................................................................viii LIST OF FIGURES ...........................................................................................................ix ABSTRACT .....................................................................................................................xvi CHAPTER 1 INTRODUCTION........................................................................................................1 Background on Hurricanes and Coastal Zone Residential Structures in Florida.........1 Synopsis of the Florida Coastal Monitoring Program..................................................2 FCMP Tower Research.........................................................................................4 FCMP House Research..........................................................................................7 FCMP Post-Hurricane Damage Assessment.......................................................12 Scope of Research.......................................................................................................13 Summary of Original Contributions...........................................................................14 2 QUANTIFIYING WIND LOADS ON STRUCTURES: BACKGROUND..............16 Background on Wind Tunnel Modeling of Pressure Coefficients..............................16 Definition of Pressure Coefficient ................................................................16 pC Wind Tunnel Procedure for Calculation of ..................................................18 pC. Important Considerations & Assumptions in Wind Tunnel vs. Full-scale Analysis............................................................................................................21 Background on ASCE-7 Wind Load Provisions........................................................22 Peaks as Functions of Sampling Frequency...............................................................25 The role of Full-scale and New Wind Tunnel Data to Evaluate Current Assumptions in ASCE-7 based on the Stathopoulos work.....................................25 Previous Full-scale Projects........................................................................................26 Quezon City Experiments....................................................................................27 Aylesbury House.................................................................................................28 Silsoe Structures Building...................................................................................29 Texas Tech Building............................................................................................30 Full-scale Measurements on the Kern P. Pitts Center.........................................31 v

PAGE 6

FCMP Data Set....................................................................................................33 New Wind Tunnel Studies (Clemson) to Accompany the Full-scale Data................34 Closing Remarks.........................................................................................................35 3 FCMP DEPLOYMENT HISTORY, ORGANIZATION AND LOGISTICS FOR THE 2004/2005 SEASONS........................................................................................37 Hurricane Frances (2004)...........................................................................................38 Synoptic History..................................................................................................38 House and Tower Deployment............................................................................38 Hurricane Ivan (2004).................................................................................................40 Synoptic History..................................................................................................40 House and Tower Deployment............................................................................41 Hurricane Jeanne (2004).............................................................................................42 Synoptic History..................................................................................................42 House and Tower Deployment............................................................................43 Hurricane Dennis (2005)............................................................................................45 Synoptic History..................................................................................................45 House and Tower Deployment............................................................................46 Hurricane Wilma (2005).............................................................................................47 Synoptic History..................................................................................................47 House and Tower Deployment............................................................................49 Hurricanes Katrina and Rita (2005)............................................................................51 Closing Remarks.........................................................................................................51 4 ANALYSIS OF FULL-SCALE DATA TO DEFINE PRESSURE COEFFICIENTS.........................................................................................................53 Calculating for Full-scale Data: Methods and Outstanding Issues......................53 pC Applied Equations for ...................................................................................54 pC Uncertainty: Data Sources can change from Storm to Storm and House to House...............................................................................................................63 Example Peak Minimum Calculation with Uncertain Reference Velocity..64 pC Identification of Appropriate Sampling Rate.............................................................66 Study for Identification of Appropriate Sampling Rate.............................................67 Influence of Sampling Rate on Peak value...................................................68 pC Influence of Sampling Rate on Spatial Correlation of Peaks..............................71 Correlation coefficient yx, ..........................................................................71 Coherence function fCyx2, .........................................................................73 Peak-score method....................................................................................74 Peak-score method: application to full-scale data....................................81 Uncertainty of Reference Velocity: Estimating Peak Wind Speed Gust....................83 Wind Speed Reference from House Anemometer: Optimizing Roughness.......84 Wind Speed Reference from House Anemometer: Optimizing Peak Factor......85 Uncertainty on Instantaneous Roof Dynamic Pressure..............................................89 vi

PAGE 7

Uncertainty of Reference Pressure.............................................................................91 Combined Effects of Uncertainties.............................................................................92 Quantification of Uncertainties on : Monte Carlo Approach................................93 pC Number of Monte Carlo Simulations..................................................................95 An Example Monte Carlo Simulation.................................................................96 Presentation of Results from Uncertainty Analysis....................................................98 Closing Remarks.......................................................................................................103 5 COMPARISON OF FULL-SCALE DATA WITH WIND TUNNEL STUDY AND ASCE-7-05......................................................................................................104 Clemson Wind Tunnel Study vs. Full-scale (Same Wind Direction).......................104 Comparison of Mean and RMS Pressure Coefficients......................................106 Comparison of Peak Pressure Coefficients.......................................................111 Interpretation of Results....................................................................................115 Concluding Remarks................................................................................................116 6 PROPOSED GIS TOOLBOX FOR FCMP APPLICATIONS................................118 What is GIS...............................................................................................................118 FCMP Deployment Maps.........................................................................................118 FCMP Aerial Pictures of Tower Sites......................................................................119 GIS Toolbox to Assist Randomized Damage Evaluation Studies............................120 Optimized Statistical Sampling.........................................................................123 Recommendation and Current Deployment Strategies.............................................125 7 CONCLUSION AND RECOMMENDATIONS.....................................................127 pC values Comparisons (Full-scale vs. Wind Tunnel vs. ASCE-7-05)....................127 Future and Ongoing Analysis............................................................................128 Workshop and Benchmark Study with all Major Wind Tunnel Facilities........129 Improvements to House Data Collection System.....................................................130 Maintenance to FCMP Houses.................................................................................131 APPENDIX A AERIAL PICTURES AND MAPS FOR 2004 & 2005 DEPLOYMENTS.............132 B WIND CALCULATION..........................................................................................176 C DYNAMIC PRESSURE COEFFICIENT CALCULATIONS................................178 LIST OF REFERENCES.................................................................................................189 BIOGRAPHICAL SKETCH...........................................................................................192 vii

PAGE 8

LIST OF TABLES Table page 2-1 Suggested values of roughness lengths for various types of terrain....................20 0z 4-1 Sensor list for FCMP house FL-27 hurricane Ivan (2004).......................................59 4-2 Sensor list for FCMP house FL-30 hurricane Ivan (2004).......................................60 4-3 Summary of 12 possible cases to compute full-scale pressure coefficients.............64 4-4 Minimum peak pressure coefficients for FL-27 hurricane Ivan (2004) channel 6..66 4-5 Difference % between 20 Hz and 2 Hz minimum for channel 6 of FL-27.......71 pC 4-6 Difference % between 20 Hz and 2 Hz minimum for channel 23 of FL-27.....71 pC 4-5 Classification of optimum and peak factor for the FL-27 house Ivan (2004).....86 0z 4-6 Classification of optimum and peak factor for the FL-30 house Ivan (2004).....87 0z 4-7 Minimum peak pressure coefficients for FL-30 hurricane Ivan (2004) channel 15, reference wind speed variation...........................................................................89 4-8 Expected minimum peak (20 Hz) for channel 6 for the FL-27 hurricane Ivan (2004).......................................................................................................................91 pC 4-9 Minimum peak pressure coefficients for FL-30 hurricane Ivan (2004) channel 15, reference pressure variation...............................................................................92 4-10 Mean and standard deviation of random variables for FL-27 record #135 hurricane Ivan...........................................................................................................95 5-1 Records for wind tunnel vs. full-scale comparison................................................105 C-1 Sensitivity data of sensor used in the FL-27 house during Ivan (2004).................183 C-2 Sensitivity data of sensor used in the FL-30 house during Ivan (2004).................184 viii

PAGE 9

LIST OF FIGURES Figure page 1-1 Tropical Cyclone activity in the State of Florida since 1950 2005..........................3 1-2 Pictures of FCMP Towers (a) FCMP Tower deployed during Hurricane Charley (2004) (b) FCMP Convoy for Katrina (2005)............................................................6 1-3 FCMP Tower Deployment Activity since 1999 2005 and major hurricane activity for the 2004 05 seasons..............................................................................7 1-4 FCMP Houses locations along: Florida, South Carolina and North Carolina...........9 1-5 Pictures of house work. (a) FCMP personnel prepares cable (b) PVC piping system.......................................................................................................................10 1-6 Pictures of house components. (a) House disconnect box (b) House computer box............................................................................................................................10 2-1 Power Law Wind velocity profiles for various values.........................................19 2-2 FCMP House ID FL-27 model scale 1:50................................................................21 2-3 Aylesbury experimental building with 22.5 pitch roof...........................................28 2-4 Texas Tech University Wind Engineering Research Field Laboratory...................31 2-5 Kern P. Pitts Center located at Southern Shores, NC...............................................32 3-1 FCMP Deployment map of hurricane Frances (2004).............................................39 3-2 FCMP Deployment map of hurricane Ivan (2004)..................................................41 3-3 FCMP Deployment map of hurricane Jeanne (2004)...............................................44 3-4 FCMP Deployment map of hurricane Dennis (2005)..............................................46 3-5 FCMP Deployment map of hurricane Wilma (2005)...............................................50 4-1 Voltage time history of functional data channel 0 for the FL-27 house during hurricane Ivan (2004)...............................................................................................55 ix

PAGE 10

4-2 Voltage time history of a malfunctioning data channel 4 for the FL-27 house during hurricane Ivan (2004)....................................................................................55 4-3 Record #129 of house FL-27 hurricane Ivan (2004), channel 5 malfunctions.........56 4-4 Roof sensor layout configuration for FCMP House FL-27......................................58 4-5 Roof sensor layout configuration for FCMP House FL-30......................................58 4-6 Channel 6 peak minimum vs. resample frequency. (a) For record #131 wind direction is 111 (b) For record #147 wind direction is 180...................................69 pC 4-7 Channel 23 peak minimum vs. resample frequency. (a) For record #131 wind direction is 111 (b) For record #147 wind direction is 180..........................70 pC 4-8 Correlation coefficient for record #131 of FL-27 hurricane Ivan (2004)................73 4-9 Coherence function estimate for record #131 of FL-27 hurricane Ivan (2004).......74 4-10 Random data signals. (a) Original signals x an y, (b) Normalized signals and .............................................................................................................................76 nx ny 4-11 Resultant signal 2nnnyxw ............................................................................77 4-12 Empirical cumulative distributions function for and ....................................78 nx ny 4-13 Empirical cumulative distributions function for ................................................78 nw 4-14 Peak-Score check for uncorrelated data and x y ...............................................79 4-15 Peak-Score check for full-correlated data x and x................................................80 4-16 Peak-Score check for partial correlated data x and w...........................................80 4-17 Peak-Score for channel 6 & 7 in record #131 FL-27 hurricane Ivan (2004)........81 4-18 Peak-Score for channel 6 & 5 in record #131 FL-27 hurricane Ivan (2004)........82 4-19 Peak-Score for channel 6 & 23 in record #132 FL-27 hurricane Ivan (2004)......82 4-20 Aerial picture for the FL-27 house showing optimized 0and peak factor values for hurricane Ivan (2004).........................................................................................86 z 4-21 Aerial picture for the FL-30 house showing optimized 0 and peak factor values for hurricane Ivan (2004).........................................................................................87 z x

PAGE 11

4-22 Coefficient of variation, on 100 trials vs. number of simulations N........................96 4-23 Histogram of minimum (20 Hz duration) from Monte Carlo simulation for channel 6 on record #135 FL-27 hurricane Ivan (2004), N = 1000........................97 pC 4-24 ECDF for minimum (20 Hz duration) from Monte Carlo simulation for channel 6 of record #135 FL-27 hurricane Ivan (2004)...........................................98 pC 4-25 Channel 6 time history of min values for 20 Hz (FL-27 Ivan 2004).................99 pC 4-26 Channel 7 time history of min values for 20 Hz (FL-27 Ivan 2004).................99 pC 4-27 Channel 13 time history of min values for 20 Hz (FL-27 Ivan 2004).............100 pC 4-28 Channel 18 time history of min values for 20 Hz (FL-27 Ivan 2004).............100 pC 4-29 Channel 22 time history of min values for 20 Hz (FL-27 Ivan 2004).............101 pC 4-30 Channel 23 time history of min values for 20 Hz (FL-27 Ivan 2004).............101 pC 5-1 Scale model (1:50 )of the FL-27 house and surrounding structures......................105 5-2 Mean pC full-scale vs. wind tunnel comparison for 110 wind direction.............108 5-3 Mean pC full-scale vs. wind tunnel comparison for 120 wind direction.............108 5-4 Mean pC full-scale vs. wind tunnel comparison for 130 wind direction.............109 5-5 RMS pC full-scale vs. wind tunnel comparison for 110 wind direction..............109 5-6 RMS pC full-scale vs. wind tunnel comparison for 120 wind direction..............110 5-7 RMS pC full-scale vs. wind tunnel comparison for 130 wind direction..............110 5-8 Min pC full-scale vs. wind tunnel comparison for 110 wind direction...............112 5-9 Min pC full-scale vs. wind tunnel comparison for 120 wind direction...............112 5-10 Min full-scale vs. wind tunnel comparison for 130 wind direction...............113 pC 5-11 Max full-scale vs. wind tunnel comparison for 110 wind direction...............113 pC 5-12 Max full-scale vs. wind tunnel comparison for 120 wind direction...............114 pC xi

PAGE 12

5-13 Max full-scale vs. wind tunnel comparison for 130 wind direction...............114 pC 6-1 Deployment map of Hurricane Katrina (2005), generated before and updated during landfall........................................................................................................119 6-2 Aerial picture of FCMP mobile tower T1 deployed in Bella Chasse, LA during hurricane Katrina (2005). See also Figure 6-1.......................................................120 6-3 Wind Swath of Hurricane Ivan 1-minute sustained gust........................................122 6-4 User interface window of the FCMP survey toolbox developed in ArcMap.........124 6-5 Map of Charlotte County showing the selected houses using the FCMP survey toolbox....................................................................................................................125 A-1 FCMP Deployment map of Hurricane Charley (2004)..........................................133 A-2 Aerial imagery of the terrain surrounding tower T0 during Hurricane Charley (2004) at Lakeland, FL...........................................................................................134 A-3 Aerial imagery of the terrain surrounding tower T1 during Hurricane Charley (2004) at Fort Meyers, FL......................................................................................135 A-4 Aerial imagery of the terrain surrounding tower T2 during Hurricane Charley (2004) at Osprey, FL..............................................................................................136 A-5 Aerial imagery of the terrain surrounding tower T3 during Hurricane Charley (2004) at Plant City, FL..........................................................................................137 A-6 FCMP Deployment map of Hurricane Frances (2004)..........................................138 A-7 Aerial imagery of the terrain surrounding tower T0 during Hurricane Frances (2004) at Port Salerno, FL......................................................................................139 A-8 Aerial imagery of the terrain surrounding tower T1 during Hurricane Frances (2004) at Port Salerno, FL......................................................................................140 A-9 Aerial imagery of the terrain surrounding tower T2 during Hurricane Frances (2004) at Vero Beach, FL.......................................................................................141 A-10 Aerial imagery of the terrain surrounding tower T3 during Hurricane Frances (2004) at Fort Pierce, FL........................................................................................142 A-11 FCMP Deployment map of Hurricane Ivan (2004)...............................................143 A-12 Aerial imagery of the terrain surrounding tower T0 during Hurricane Ivan (2004) at Mobile, AL.............................................................................................144 xii

PAGE 13

A-13 Aerial imagery of the terrain surrounding tower T1 during Hurricane Ivan (2004) at Pensacola, FL..........................................................................................145 A-14 Aerial imagery of the terrain surrounding tower T2 during Hurricane Ivan (2004) at Fairhope, AL...........................................................................................146 A-15 Aerial imagery of the terrain surrounding tower T3 during Hurricane Ivan (2004) at Destin, FL...............................................................................................147 A-16 FCMP Deployment map of Hurricane Jeanne (2004)............................................148 A-17 Aerial imagery of the terrain surrounding tower T0 during Hurricane Jeanne (2004) at Orlando, FL.............................................................................................149 A-18 Aerial imagery of the terrain surrounding tower T1 during Hurricane Jeanne (2004) at Sebastian, FL..........................................................................................150 A-19 Aerial imagery of the terrain surrounding tower T2 during Hurricane Jeanne (2004) at Merritt Island, FL....................................................................................151 A-20 Aerial imagery of the terrain surrounding tower T3 during Hurricane Jeanne (2004) at Vero Beach, FL.......................................................................................152 A-21 FCMP Deployment map of Hurricane Dennis (2005)...........................................153 A-22 Aerial imagery of the terrain surrounding tower T0 during Hurricane Dennis (2005) at Navarre, FL.............................................................................................154 A-23 Aerial imagery of the terrain surrounding tower T1 during Hurricane Dennis (2005) at Inlet Beach, FL.......................................................................................155 A-24 Aerial imagery of the terrain surrounding tower T2 during Hurricane Dennis (2005) at Pensacola, FL..........................................................................................156 A-25 Aerial imagery of the terrain surrounding tower T3 during Hurricane Dennis (2005) at Pensacola, FL..........................................................................................157 A-26 Aerial imagery of the terrain surrounding tower T5 during Hurricane Dennis (2005) at Niceville, FL...........................................................................................158 A-27 FCMP Deployment map of Hurricane Katrina (2005)...........................................159 A-28 Aerial imagery of the terrain surrounding tower T0 during Hurricane Katrina (2005) at Bay St. Louis, MS...................................................................................160 A-29 Aerial imagery of the terrain surrounding tower T1 during Hurricane Katrina (2005) at Bella Chasse, LA....................................................................................161 xiii

PAGE 14

A-30 Aerial imagery of the terrain surrounding tower T2 during Hurricane Katrina (2005) at Galliano, LA...........................................................................................162 A-31 Aerial imagery of the terrain surrounding tower T3 during Hurricane Katrina (2005) at Pascagoula, MS.......................................................................................163 A-32 Aerial imagery of the terrain surrounding tower T5 during Hurricane Katrina (2005) at Gulf Port, MS..........................................................................................164 A-33 FCMP Deployment map of Hurricane Rita (2005)................................................165 A-34 Aerial imagery of the terrain surrounding tower T0 during Hurricane Rita (2005) at Port Arthur, TX..................................................................................................166 A-35 Aerial imagery of the terrain surrounding tower T1 during Hurricane Rita (2005) at Houston, TX.......................................................................................................167 A-36 Aerial imagery of the terrain surrounding tower T3 during Hurricane Rita (2005) at Nederland, TX....................................................................................................168 A-37 Aerial imagery of the terrain surrounding tower T5 during Hurricane Rita (2005) at Orange, TX.........................................................................................................169 A-38 FCMP Deployment map of Hurricane Wilma (2005)............................................170 A-39 Aerial imagery of the terrain surrounding tower T0 during Hurricane Wilma (2005) at Everglades City, FL................................................................................171 A-40 Aerial imagery of the terrain surrounding tower T1 during Hurricane Wilma (2005) at Weston, FL.............................................................................................172 A-41 Aerial imagery of the terrain surrounding tower T2 during Hurricane Wilma (2005) at Ochoppi, FL............................................................................................173 A-42 Aerial imagery of the terrain surrounding tower T3 during Hurricane Wilma (2005) at Miami, FL...............................................................................................174 A-43 Aerial imagery of the terrain surrounding tower T5 during Hurricane Wilma (2005) at Naples, FL...............................................................................................175 C-1 FCMP Instrumented houses in the state of Florida................................................179 C-2 FCMP Instrumented houses in the state of South and North Carolina..................180 C-3 Roof sensor layout of the FL-27 House.................................................................181 C-4 Aerial imagery of FCMP House ID FL-27 at Gulf Breeze, FL.............................181 C-5 Roof sensor layout of the FL-30 House.................................................................182 xiv

PAGE 15

C-6 Aerial imagery of FCMP House ID FL-30 at Pensacola Beach, FL......................182 C-7 Mean pressure coefficients for FL-27 during hurricane Ivan (2004).....................185 C-8 RMS pressure coefficients for FL-27 during hurricane Ivan (2004).....................185 C-9 Min peak pressure coefficients 20 Hz for FL-27 during hurricane Ivan (2004)....186 C-10 Max peak pressure coefficients 20 Hz for FL-27 during hurricane Ivan (2004)..186 C-11 Mean pressure coefficients for FL-30 during hurricane Ivan (2004).....................187 C-12 RMS pressure coefficients for FL-30 during hurricane Ivan (2004).....................187 C-13 Min peak pressure coefficients 20 Hz for FL-30 during hurricane Ivan (2004)....188 C-14 Max peak pressure coefficients 20 Hz for FL-30 during hurricane Ivan (2004)...188 xv

PAGE 16

Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy MEASURED HURRICANE WIND PRESSURE ON FULL-SCALE RESIDENTIAL STRUCTURES: ANALYSIS AND COMPARISON TO WIND TUNNEL STUDIES AND ASCE-7 By Luis D. Aponte-Bermdez August 2006 Chair: Kurtis R. Gurley Cochair: Gary R. Consolazio Major Department: Civil and Coastal Engineering Each year hurricanes cause devastating damage along the southeast coastline of the United States, where the State of Florida is one of the most vulnerable. The work presented in this dissertation is the result of full-scale measurements conducted during the last six Atlantic Hurricane Seasons (1999-2005). The primary objective was to quantify over-land near-surface hurricane wind velocity and uplift loads on residential structures using full-scale experiential methods. The research goal is to help reduce hurricane wind damage to residential structures by providing ground-truth data about the intensity of the wind, the resultant loads on residential structures, and the performance of these structures in high winds. The full-scale hurricane data measurement was conducted with two separate data collection systems. The first system consists of portable weather towers deployed in the path of landfalling hurricanes to capture the wind field behavior at a height of 5 and 10 meters, as well as temperature, humidity, rainfall and barometric pressure. The second system uses xvi

PAGE 17

pressure sensors to collect wind pressure data on the roofs of occupied residential structures along the Florida and the Carolinas coastlines. To date, 32 houses along the Florida coastline, 4 along South Carolina, and 2 along the North Carolina coastline have been outfitted with these sensors. The data collected from these houses are compared to wind tunnel model studies on scale models of the subject homes. During the hurricanes of 2004 and 2005 several data sets were collected from homes that experienced sustained hurricane level winds. A total of 16 homes were instrumented during 3 of the 2004 storms, and 6 homes were instrumented over 3 storms in 2005. Data were collected from 9 of these homes in sustained hurricane level winds, a first in experimental wind engineering. Details are provided regarding the deployment of the portable towers and the instrumentation of the coastal homes and analysis of this full-scale data are presented along with comparison between wind tunnel models and the ASCE-7 wind load provisions. Implications regarding the current state of knowledge of extreme wind loading in low-rise structures are provided. Preliminary analysis of the full-scale vs. wind tunnel homes presented in this dissertation suggests that it may potentially impact wind load standards. These comparison studies suggest that the peak negative pressure coefficient obtained from the full-scale data exceeded the ASCE-7 coefficient (component and cladding) for the corresponding roof zones of high suction areas. xvii

PAGE 18

CHAPTER 1 INTRODUCTION Background on Hurricanes and Coastal Zone Residential Structures in Florida Historically hurricanes in the Atlantic basin refer to tropical cyclones that form in the North Atlantic Ocean, Caribbean Sea and the Gulf of Mexico, north of the equator, usually in the Northern Hemisphere, summer or autumn. The Atlantic hurricane season officially runs from June 1 to November 30. The U.S. National Hurricane Center monitors the basin and issues reports, watches and warnings about tropical weather systems for the United States, where other countries around the basin track and issue warnings for tropical weather in their territories. The State of Floridas coastline consist of 580 miles out of the total 2,069 miles in the Atlantic coast portion, and 770 miles out of the 1,631 in Gulf of Mexico coast portion of the U.S.; this represents 28% and 47% respectively (Teachervision.com, 2006). This represent a 36% of the total coastline from the Texas to Maine, consequently Florida is the state with the largest coastline exposure to Tropical Cyclone activity in the Atlantic basin. In terms of coastal population, from the 2003 U.S. Census Bureau and Woods & Poole Economics (W&PE) Inc., the Southeast and the Gulf of Mexico coastal areas are the least populated regions in the United States with a 9% and 13% respectively, where the most populated coastal areas in the U.S. correspond to the Northeast with a 34% followed by the Pacific with a 26% and the Great Lakes with and 18%, but none of these regions (with the exception of Northeast) are in danger of Tropical Cyclones in the 1

PAGE 19

2 Atlantic basin. The Southeast has progressively become a primary destination for retirees and job-seekers. Florida showed the greatest percent coastal population change of any coastal region by percentage between 1980 and 2003. On average in the last decade, new residential construction in Florida in a given year represents 2% of the total residential infrastructure. This growth rate is not projected to slow in the near future, thus protecting this increasing population from the effects of hurricanes is a growing priority. Historically Florida is one of the Southeast states that suffer direct impact of hurricanes. Figures 1-1 shows a map with the Tropical Cyclones that had made direct landfall in the state of Florida, where for the last two years (2004/05) the Tropical Cyclone activity in the Atlantic basin has been above the normal historical average, causing this billion of dollars in the damage. The research presented in this dissertation contributes to understanding and minimizing the effects of hurricane wind loads on residential structures. Synopsis of the Florida Coastal Monitoring Program The Florida Coastal Monitoring Program (FCMP) began as a full-scale research effort in 1998 in response to a need in the wind engineering community to better understand hurricane wind loading on low-rise structures. The program started at Clemson University, with the University of Florida joining the research team in 1999. At present the team also includes Florida International University, Florida Institute of Technology, and the Institute for Business and Home Safety (IBHS) in Tampa, Florida. Funding support has been provided by the Florida Department of Community Affairs (DCA), the National Oceanic and Atmospheric Administration (NOAA), the Federal Emergency Management Agency (FEMA), The Florida Building Commission (within the

PAGE 20

3 DCA), Florida Sea Grant, South Carolina Sea Grant, and the Institute for Business and Home Safety. Figure 1-1 Tropical Cyclone activity in the State of Florida since 1950 2005 The FCMP research activities consist of four distinctive projects. Two consist of separate full-scale data collection systems to collect ground level wind speeds and uplift pressures on the roofs of residential structures. These will be referred to as FCMP Tower and FCMP House. The third consists of post-hurricane damage evaluations to provide data relating winds speed to damage. The fourth project is a destructive testing program on houses, which allows researches to quantify actual construction methods and vulnerability reducing retrofits to their limit states. These four complementary projects provide information needed to develop a more wind resistant infrastructure by quantifying hurricane wind behavior, extreme wind

PAGE 21

4 loading, and the vulnerability of both existing construction methods and proposed cost-effective retrofits and code improvements. The next section will provide details about the first two activities in which the author has been an active participant. FCMP Tower Research A source of great uncertainty in the design of hurricane resistant structures is the actual winds these structures are subjected to, in terms of both sustained winds and the magnitude and frequency of gusts. The commonly used Saffir-Simpson storm intensity rating scale is based upon sustained 1-minute averaged wind speeds over open water. As the wind transitions from open water to land, the influence of terrain and infrastructure alters the behavior of the wind. For example, a Category 3 storm may only produce sustained winds of 100 mph over land, although the SS-scale rates Category 3 as at least 111 mph sustained winds. Conversely, the gustiness of the wind increases over land, which can adversely influence the loads on structures. This disconnect between the public perception of hurricane intensity and the actual loads experienced by structures is an outstanding issue to address as researchers investigate means to mitigate wind damage. The FCMP Tower project is designed to quantify this wind behavior through direct measurements. Instrumented portable towers are placed in the path of oncoming extreme wind events to quantify instantaneous wind speed and direction as the hurricane approaches and impacts land, and moves inland. The FCMP full-scale wind velocity data collection system consists at present of six portable towers designed and built at Clemson University (Poss, 2000). The towers are stored in the off-season at the University of Florida where upgrades and maintenance are performed by the FCMP team members.

PAGE 22

5 During deployment, each tower relays summary data every 15-minutes to a public access website for use by NOAA researchers, emergency managers, and risk modelers. In 2006 this cellular communication-based real-time system is being augmented with satellite transmission modulators to transfer data directly via-satellite to NOAAs Geostationary Operational Environmental Satellite (GOES). The FCMP mobile towers (Figure 1-2) are designed to meet U.S. Department of Transportation (DOT) requirements for transport as a conventional trailer, and withstand peak gust wind speed of 90 m/s (200 mph), which corresponds to a strong Saffir-Simpson Category 5 (Simpson and Riehl, 1981). The FCMP towers mobility and easy assembly, approximately twenty minutes by a three man crew allow deployment in almost in any terrain exposures condition. Tower instruments are located at three levels (3, 5 and 10 m). The data acquisition system measures 3D wind speed and direction at the top two levels and collects temperature, rainfall, barometric pressure, and relative humidity data at the towers base. Two RM Young anemometry systemsa wind monitor and a custom array of three gill propellerscollect data at the 10-m level, which the World Meteorological Organization deems as the standard wind speed observation height. A second array of gill propellers collects wind speed data at the 5-m level to measure winds at the approximate eave height of a single-story home. The tower power system has been upgraded from its original contractor-grade gasoline generator with diesel generators to provide power for a period up to 48 hours. A series of UPC batteries provide an extra 8-10 hours of power when the generator runs out of fuel.

PAGE 23

6 The data acquisition system consists of two separate computer systems. The first consists of a PC that collects the data at a sampling rate of 100 Hz, and stores it digitally onto two separate hard drives. The second system consists of a laptop computer that collects the data at sampling rate of 10 Hz, and stores the data onto a single hard drive. This system was incorporated in the 2003 hurricane season and it is responsible to connect to the Internet via cellular modem and upload statistics summaries through various ftp sites. More details are presented by Masters (2004). (a) (b) Figure 1-2 Pictures of FCMP Towers (a) FCMP Tower deployed during Hurricane Charley (2004) (b) FCMP Convoy for Katrina (2005) The FCMP tower data set contains information from Tropical Cyclones Georges (1998), Dennis (1999), Floyd (1999), Irene (1999), Gordon (2000), Gabrielle (2001), Michelle (2001), Isidore (2002), Lili (2002), Isabel (2003), Bonnie (2004), Charley (2004), Frances (2004), Ivan (2004), Jeanne (2004), Dennis (2005), Katrina (2005), Rita (2005) and Wilma (2005). Facts about the FCMP deployments since 1998 to 2003 are presented by Masters (2004); details about the 2004/05 seasons will be presented in this dissertation. This database to date has provided a crucial source of information when

PAGE 24

7 defining the vulnerability of the infrastructure. Figure 1-3 shows a map of all the tower deployment activity since 1999 2005 and the major hurricane activity of the 2004 05 seasons. Figure 1-3 FCMP Tower Deployment Activity since 1999 2005 and major hurricane activity for the 2004 05 seasons. FCMP House Research Just as the knowledge of wind speeds over land is largely uncertain despite advances in hurricane tracking and intensity forecasting, the wind loads associated with extreme wind speeds are also lacking in first-hand direct quantification. ASCE-7 wind load provisions are based largely on small scale wind tunnel replications conducted on simple structural shapes. There are limits regarding the accuracy of extrapolating the loads measured in these experiments to full-scale, and in particular the interaction of

PAGE 25

8 extreme wind and resultant loading on more typical complex shapes found in the existing housing infrastructure. The FCMP house project was created to bridge the gap between wind tunnel based wind load provisions and the actual loads experienced by full-scale structures. This project consists of measuring wind pressure data directly on the roof, soffit and walls of pre-selected residential houses along the coast of Florida during land falling hurricanes. The homeowners of these occupied structures agreed to collaborate with the FCMP researchers in exchange for retrofits to improve the homes resistance to hurricane winds. Modern high wind rated shingles, wind and impact rated garage doors, and window shutter systems are typical of these incentives. The current FCMP house catalog contains a total of 32 houses in the state of Florida, 4 in the state of South Carolina and 2 in the state of North Carolina. Figure 1-4 shows the distribution of the houses among these three states. The location of these houses had been carefully selected using the historical frequency of land-falling hurricanes in these regions. The homes are spaced at intervals of 16 to 24 km (10 15 mi), and most are within 1.5 km (1 mi) of the coastline. Typically the houses are one or two stories tall have composite shingle roof covering, and the surrounding areas are suburban and relative free of tree cover. Aerial pictures for houses presented in this dissertation are available in Appendix A. The house instrumentation is prepared during the off season. This consists of installing roof brackets to attach the sensors, and exterior wiring to connect the sensors to the computer system. Figure 1-5 (a) shows the wires for the individual sensors, and Figure 1-5 (b) shows the plastic piping containing these wires, installed under the

PAGE 26

9 overhang. All of the wires meet at a disconnect box shown in Figure 1-6 (a), to which the computer data collection system is attached as shown in Figure 1-6 (b). The pressure sensors are mounted on the roof, and the computer system is installed within days of an approaching hurricane and retrieved immediately after the event. Figure 1-4 FCMP Houses locations along: Florida, South Carolina and North Carolina The computer is contained in a 60 Steel Jobsite Box shown in Figure 1-6 (b). The inside is customized to accommodate a PC, CPU batteries (to provide up to 24-36 hours of power), time lapse VCR (to record images of the house during the storm) and miscellaneous tools. The boxs final weight is around 300 lb, heavy enough to resist high winds. The data acquisition system measures data at a sampling rate of 100 Hz. The data are stored digitally into two independent hard drives every 15 minutes. The field instrumentation per house consists of a maximum of twenty eight Microswitch 142 PC-15 absolute pressure transducers. Reference pressure sensors are located inside the house attic and at ground level (yard sensor). In addition, depending to the house configuration,

PAGE 27

10 3-cup-anemometers are installed on the roof of the house, using a 54 inch stem extension from the roof eave. (a) (b) Figure 1-5 Pictures of house work. (a) FCMP personnel prepares cable (b) PVC piping system (a) (b) Figure 1-6 Pictures of house components. (a) House disconnect box (b) House computer box The sensors are sheltered in a 12 in diameter aluminum pan (roof) or square plastic box (wall & soffit), distributed along the roof, soffit, walls and camera mounting base

PAGE 28

11 plate (yard sensor). Detailed technical information is presented by Michot (1999). Some of the most relevant points are voltage signal resolution of pressures to .005 psf, an arbitrary offset voltage that varies from sensor to sensor the transducer calibrations are sensitive to temperature changes. Michot reported a temperature adjustment factor of 0.0144 Volts/F for a sensor circuit of sensitivity of 20.6 psf/Volt. FCMP house sensor calibration tests have been performed in order to find the sensitivity for each sensor. A sealed box allows calibrating twelve sensors at a time. Using an electric air compressor, suction is applied to the box to provide a known pressure. Voltage data from the sensors are collected during a 15 minute interval at a sampling rate of 100 Hz. During the test, pressure inside the box is monitored using a Setra Digital Pressure Gage Model 370. The pressure then is varied inside the box over a range around 810 to 970 mbars. A linear regression is then applied to the sensor voltage data and known pressure measured using the RM Young pressure sensor, thus providing the sensitivity (calibration) factor for each sensor. FCMP analysis of house data collected during hurricane Ivan (2004) will be presented in Chapter 4. Additional data was collected during Frances and Jeanne (2004), and Dennis and Wilma (2005), and is the subject of ongoing analysis. This dissertation will focus on the development of a probabilistic approach to the analysis of the full-scale data, which will be employed for future analysis of these data sets. Companion research at Clemson University focuses on constructing scale models of the homes that collected data in 2004 and 2005, and conducting wind tunnel studies of these models. The loads observed from these tests will be compared with those loads

PAGE 29

12 measured in full-scale in an attempt to identify shortcomings in the model testing procedures used to determine wind load provisions. The analysis of full-scale pressure data collected during actual hurricanes presents numerous challenges in terms of offering accurate results, reasonable uncertainty quantification, and a fair comparison to wind tunnel studies. This is a major topic in this dissertation. FCMP Post-Hurricane Damage Assessment In addition to the quantification and modeling of hurricane winds and structural loading, the assessment and documentation of damage to residential structures after a hurricane event are necessary to identify existing weaknesses and potential solutions. During the spring of 2005, the FCMP teams were joined by Florida A & M University (FAMU) students to conduct an extensive in-field evaluation of homes impacted by the 2004 hurricanes in Florida. The strategy was to randomly sample addresses located within designated peak wind bands for evaluation, thus providing an unbiased survey of damage to typical structures of various ages (Gurley et al., 2006). The random sampling strategy relied upon a database of county-wide residential house information. The randomization process was a rudimentary random order scheme. The outcomes of the study have been very revealing in terms of comparing structural performance as a function of construction age, and peak wind speed. However, since the study was conducted months after the storms occurred, much of the data collection was based upon homeowner interviews rather than first hand observations of damage. Ideally such extensive damage assessments would be conducted in the days immediately after the storm impacts a region in order to capture perishable data that is

PAGE 30

13 best observed first-hand. In order to capitalize on the use of a randomized sampling strategy, a portion of this dissertation deals with the development of a GIS-based tool to employ a series of sampling strategies stratified by wind speed, age, region, construction type, or a combination. This tool contains the database of residential housing from most of the counties in Florida. With such a tool in hand, the next intense hurricane to impact Florida will be immediately followed by a thorough damage assessment that is designed to provide statistically relevant samples of damage to stratifications as determined by researchers in the field. Scope of Research This document presents the FCMP research efforts in further detail, and highlights the original contributions of the author to these efforts. Chapter 2 provides a historical background on the quantification of wind loads, including wind tunnel modeling, ASCE-7, and the role of new full-scale pressure data. Chapter 3 will discuss full-scale data collection methods and present a summary of data collected since 2004. Chapter 4 presents the methods developed for, and results produced from the analysis of full-scale pressure data collected in Florida during hurricane landfall. There are two focuses in this chapter. The first is the identification of a suitable sampling rate for analysis. The second is the identification, modeling and quantification of the influence of uncertainty in the calculated pressure coefficient values to define confidence limits. Analysis of the data collected during Hurricane Ivan (2004) on a low-rise residential structure is presented. Chapter 5 presents a comparison of the full-scale results with those of wind tunnel studies of the FCMP houses as well as ASCE-7. The full-scale pressure coefficients will

PAGE 31

14 be bounded by confidence intervals as determined by Monte Carlo-based methods developed for this research. This comparison of probabilistic full-scale pressure coefficients, from homes in sustained hurricane winds, with results currently used wind tunnel methods and ASCE load provisions, is the first such study available in the open literature. Chapter 6 presents the GIS applications developed by the author to improve deployments strategies and the quality of post-hurricane damage surveys. Finally, Chapter 7 summaries conclusions on the FCMP data analysis of full-scale vs. wind tunnel vs. ASCE7-05 code parameters for low-rise structures, and presents comments and suggestion for future research works. Additional FCMP house analyses are presented in the Appendix C for data sets collected during the events of hurricane Frances (2004), Jeanne (2004), Ivan (2004), Dennis (2005) and Wilma (2005). Summary of Original Contributions The research contributions include: The development of new analysis methods for full-scale pressure data: Identification of uncertainties in the collection and analysis of full-scale data Development of a probabilistic pressure coefficient that incorporates these uncertainties Development of a systematic analysis method to apply to each house data set Identify the ideal down-sampling rate of full-scale data based on observed peak loads at individual sensors, and correlated peaks at multiple spatially separate sensors Analysis of the only full-scale residential pressure data sets available The first comparative study of full-scale, wind tunnel, and ASCE-7 wind loads on structures under sustained hurricane winds:

PAGE 32

15 Comparative study of peak pressure values from full-scale and scaled wind tunnel results Evaluation of the accuracy of ASCE-7 for houses in hurricane prone regions Evaluation of the effects of terrain on pressure coefficients The development of new GIS-based methodologies and functional framework for: Analysis and presentation of post-damage survey data Design and automated selection for damage survey studies in future storms

PAGE 33

CHAPTER 2 QUANTIFIYING WIND LOADS ON STRUCTURES: BACKGROUND This chapter presents background information on quantification and modeling of wind pressure on low-rise structures. The estimation of pressure coefficients from wind tunnel studies is first discussed, followed by the application to wind load provisions. Finally the role of full-scale data and modern wind tunnel methods are presented. Background on Wind Tunnel Modeling of Pressure Coefficients This section provides background information concerning the modeling of wind pressure on low-rise structures. It presents basic concepts and the need for verification and improvement via the full-scale contributions presented later in the document. Some of the early works presented in this section had tremendous influence on the development of wind load provisions for low-rise structures. Definition of Pressure Coefficient pC The pressure coefficient is a non-dimensional value which acts as a means of indicating the local pressure at some point of interest around a body, and which is independent of velocity. It is defined in equation 2-1. pC 221UppCp (2-1) Where: p is the pressure at the point of interest, is the free stream pressure or reference pressure, p is the fluid density and U is the reference or representative velocity near the point of interest. Positive pressure acts toward the surface and negative pressure acts away from the surface. The pressure coefficient can represent an average 16

PAGE 34

17 pressure over a prolonged period, the fluctuating pressure about the mean, or minimum / maximum peak values over a defined duration of constant average pressure. These values commonly evaluated from wind tunnel studies are referred to as mean ( pC pC ), RMS ( pC ), peak minimum () and peak maximum () values, and are calculated using mean, fluctuating, or peak values of pC pC p in Eq. 2-1. In practice, the pressure coefficient is applied by rearranging 2-1 such that the pressure differential in the numerator (the design wind load) can be determined with knowledge of the pressure coefficient and reference wind speed. pC changes from one location to another on the same structure in the same wind field. Factors that influence the value of beyond the parameters in 2-1 include size, roughness and orientation (relative to wind direction) of the surface at which is being determined, the location on that surface (near an edge or in the middle), the turbulence (gustiness) of the approaching wind field, and the terrain surrounding the structure (related to the turbulence). pC pC Typically a model structure is subjected to a turbulent wind field with constant average speed and the pressure coefficients are calculated over a finely spaced grid of different locations over the surface of the structure. This is repeated over a series of wind directions, producing a surface of values for a given wind direction. The worst case pressure coefficient at a given location is then selected from the many wind directions, enveloping the effects of wind from any direction. pC

PAGE 35

18 Wind Tunnel Procedure for Calculation of pC This section summarizes the extensive work conducted by researches to model wind pressure loading in low-rise structures in boundary layer wind tunnels. It also presents some of the details of the FCMP houses modeled at the Clemson University Wind Load Test Facility and some details of the work conducted at the University of Western Ontario (UWO). First a scale model is constructed, where the model size depends on the dimensions and velocity scales of the boundary layer wind tunnel. Normally scale models are constructed of Plexiglas and instrumented with pressure taps that sit flush to the surface in the roof and walls, which allows researchers to heavily instrument scale models in areas of interest like high suction zones (e.g. roof corners and edges). Clemson University researchers gather enough information from the full-scale experiment conducted by the FCMP, such as aerial pictures, topographic maps and detailed measured of the structures, to build a scaled geometric model of the house (shown in Figure 2-2) and other typical neighboring houses constructed of foam. The second step consists of selecting and evaluating the wind speed profile that will be replicated to scale in the wind tunnel. This is also referred to as matching the approach roughness. This roughness is quantified as a roughness length. Table 2-1 presents typical roughness length for common exposures types. As the roughness increases, the mean speed of the wind drops more rapidly from high (well above the structure) to low elevations (near the structures roof). This is often modeled as a log-law or power-law relationship shown in equation 2-2. 0z

PAGE 36

19 2121ggggzzzUzU (2-2) Where the exponent in the equation is dependent upon the roughness and and denotes heights above ground. The level of turbulence increases with lower elevation more rapidly with higher roughness values. Figure 2-1 shows the wind velocity profiles for different values of 1gz 2gz Power Law Wind Velocity Profiles00.20.40.60.810.200.400.600.801.00U(zg1)(zg1/zg2) Coastal Areas = 0.10 Open Terrain = 0.14 Suburban Terrain = 0.22 Center of Large Cities = 0.33 Figure 2-1 Power Law Wind velocity profiles for various values. Since both mean and fluctuating wind speed effects care must be taken to properly represent the roughness of the terrain in the model study. If the scale model in Figure 2-2 is placed in a wind tunnel that replicates smooth open terrain, and the actual full-scale data from that home are from suburban terrain, direct comparisons of data sets are less meaningful. pC

PAGE 37

20 The reference velocity to be used in the denominator of equation 2-1 is then determined by measurements in the wind tunnel. This is typically taken to represent the expected peak value of wind speed for a given duration of time, located at the mean roof height. This is typically quantified in one of two ways: 1) Measure the wind speed at the top of the boundary layer wind tunnel, well above the model. Then by applying the power-law relationship defining wind speeds as a function of elevation, the wind speed at mean roof height is computed. This is the methodology used by Clemson University Wind Tunnel Experiments. 2) Another approach to estimate the reference velocity at mean roof height is presented by Kopp (2005) at the University of Western Ontario. The peak velocity at mean roof height, is defined as VgVVh where V is the mean roof height velocity, V is the root mean square (standard deviation) of the velocity fluctuation, and g is the peak factor, taken as a nominal value of 3.0. V and V are measured in the wind tunnel at the mean roof height without the model in place. Table 2-1 Suggested values of roughness lengths for various types of terrain 0z Type of TerrainCoastala,bOpenbSparsely Built-up SuburbsbTowns, Densely Built-up SuburbsbCenters of Large Citiesbz0(m)0.005 0.010.03 0.100.20 0.400.80 1.202.00 3.00aApplicable to structures directy exposed to winds blowing from open waterbValues of z0 to be used in conjunction with the assumption zd = 0 The model is then placed on a turntable in the wind tunnel and data are collected at different wind angles 0 to 360 by rotating the turntable between tests. This is done for multiple runs for a given wind direction. Finally by applying equation 2-1 the mean, RMS, minimum and maximum peak pressure coefficients are computed using the measured wind tunnel data.

PAGE 38

21 Figure 2-2 FCMP House ID FL-27 model scale 1:50 Important Considerations & Assumptions in Wind Tunnel vs. Full-scale Analysis The evaluation of peak pressure coefficients in wind tunnel and full-scale scenarios is related to the sampling rate duration, where higher (minimum and maximum) peaks are expected for faster sampling rates. This relationship eventually levels off at very fast sampling rates. Another consideration is the correlation of peaks over larger areas to represent the aggregate uplift over large structural components (e.g. sheathing). At faster sampling rates, a given peak pressure observed at one sensor may not be correlated to nearby sensors. in view of the fact that the pressure coefficient in load provisions are intended to represent aggregate loads, for example over a 4x8 piece of sheathing, identifying spurious peaks uncorrelated over small distances is not ultimately useful. The identification of an appropriate sampling rate between the model and full-scale data comparison will be addressed in detail in chapter 4. The model scale and the mean wind speed are also issues when considering the sampling frequency to be used when comparing full-scale and wind tunnel data. The dimensionless proportion in equation 2-2 describes the desired relationship between full-scale and model scale data.

PAGE 39

22 scale-FullModelnBUnBU (2-2) Where U is the roof mean wind speed, n is the sampling rate and B is a physical dimension (for example in a 1:50 scale model B will have a value of 1 in the model and a value of 50 in the full-scale). Typically the full-scale speed and dimension values are not controlled by the researcher, and the model scale is selected for reasons associated with wind tunnel test section size. Thus the full-scale sampling frequency, the model sampling frequency, and the mean wind speed used in the wind tunnel can be controlled to maintain this relationship. Another concern between the model and full-scale data are the issue of stationary data, where stationary refers to mean and fluctuating components of both the wind speed and direction that do not change over significant time duration. This is easily controlled in the wind tunnel as wind direction and speed are within user control, and the turbulence remains at a constant rate determined by the approach roughness in the tunnel. At full-scale the experiment cannot be controlled in this manner. Wind speed, direction and turbulence can fluctuate dramatically over short periods of time. The challenge is to identify segments of data from the larger record that meet the stationary requirement. For example, full-scale data where wind direction does not vary beyond a 5 degree arc and the mean speed remains within 5% of its starting value is very hard to obtain in a 30 minute period but it is possible to obtain in a short period of time. In order to take into account the stationary issues a duration time of 15 minute will be used. Background on ASCE-7 Wind Load Provisions The current Wind Load Provisions of ASCE-7 are based on the wind tunnel study work of Theodore Stathopoulos conducted in the late 1970s at the Boundary Wind

PAGE 40

23 Tunnel Laboratory in the University of Western Ontario (UWO) (Stathopoulos, 1979). Stathopoulos work consist of two terrain models, the first a smooth exposure which produces a profile appropriate to open country conditions whereas the second built-up exposure correspond to suburban conditions. The roughness length in the wind tunnel was estimated to be inches [m] for open country (smooth) and inches [m] for suburban (built-up) exposures. Both of these values are very low compared to the currently used values for open country and sparsely built-up suburban in the range of (0.03 0.10 m) and (0.20 0.40 m) respectively. These values suggest that the wind field velocity profile used during the experiments is best described by todays standards as a coastal terrain exposure, which reflect a change in thinking since the late 1970s (Kopp, 2005). 0z 3104 4100.1 31034 4106.8 The characteristics of the boundary layer flow modeled in the wind tunnel dictate the appropriate length and velocity scales for a rigid pressure model. For the Stathopoulos study these are approximately 1:500 and 1:5 respectively. Thus the time scale is of the order 1:100. The frequency response of the pressure measurements system use in Stathopoulos work was capable of modeling full-scale fluctuations up to about 1 Hz and the sampling rate for each pressure signal exceeds and equivalent full-scale sampling rate of about 10 samples per second (Stathopoulos, 1979). The peak measurements considered thru the study were base on a single extreme value recorded rather than an average peak over several trials, based on the narrow differences for multiple trials with a coefficient of variation not exceeding 10%. Kopp (2005) presents that the corresponding model length scale was relaxed for the experiment, and models with length scales of 1:500, 1:250, and 1:100 were investigated

PAGE 41

24 in the same flow. For example, the model-scale integral scale is effectively halved if a 1:250 model is used in a 1:500 scale flow simulation. Stathopoulos and Surry concluded that the length scale relaxation to 1:250 was the largest possible without significant distortion of the results. The resulting database from the Stathopoulos experiments was, thus, largely based on the test data obtained from 1:250 scale model tests of three roof slopes (1:12, 4:12, and 12:12) and three eave heights (4.9, 7.3, and 9.8 m) in an open country exposure. However, the pressure coefficients from this work have been long used to describe the loading on houses in a wide variety of terrains. ASCE-7 allows a wind speed reduction with built up exposure, but applies the same pressure coefficients determined from these open country wind tunnel experiments. Analysis of both full-scale data and recent wind tunnel experiments suggest that pressure coefficients for components and cladding increase in built-up terrain, perhaps to the point of negating the load reduction allowed due to lower wind speed (Reinhold, 2005) The wind tunnel data was collected using state-of-the-art measurement techniques and equipment of the time at a sampling rate of 1000 Hz at 45 intervals and was then filtered using a low pass cut off frequency of 95 Hz. The sampling time was 30 s for the 1:250 data, which corresponds to a full-scale time of about 25 min and full-scale frequency of 2 Hz. This is a point of debate since preliminary analysis of the FCMP full-scale data show that much higher sampling rates have been found to produce well correlated peaks; this issue will be addressed in more details in chapter 4. There is a need to re-evaluate the current application of the Stathopoulos data set based on recent findings from both full-scale and wind tunnel experiments. It is justified

PAGE 42

25 to evaluate the effects of using faster sampling rates and a range of exposure profiles beyond open country. Peaks as Functions of Sampling Frequency In order to compare Stathopoulos data with the FCMP model/full-scale data it is important to define an equivalent sampling frequency to match the conditions of the wind tunnel experiment. A full-scale sampling frequency of 2 Hz was calculated to be appropriate to compare with the Stathopoulos data presented on the ASCE-7. However, it is also important to present the results of the full-scale analysis in a format that best reflects the capabilities of modern instrumentation. Beyond fair comparisons of this data to the Stathopoulos work, the full-scale data will be down-sampled to the maximum rate capable of capturing correlated peaks across spatially separated sensors. This issue will be address in more details in chapter 4. The role of Full-scale and New Wind Tunnel Data to Evaluate Current Assumptions in ASCE-7 based on the Stathopoulos work Modern methods allow pressure coefficient calculations from data directly measured on full-scale residential structures under extreme winds. The role of full-scale measurements in quantifying wind pressure loadings on low-rise structures during landfalling hurricanes is fundamental to advancing the current state of knowledge of extreme wind loading. Simiu & Scanlan (1996) presents some of the early works of model/full-scale comparisons by Richardson (1991), in which comparisons for pressure on low-rise structures (gable-roof buildings) suggest that the wind tunnel does not accurately model the flow separation on the windward roof, so roof pressures often differ significantly between model and full-scale. Tieleman (1998) states that: because exact understanding of vortex formation and development under separated shear layers is

PAGE 43

26 lacking, the success of any simulation technique must be based on model/full-scale comparison of observed surface pressures and not on some preconceived ideas of flow simulation alone. It has been found in most of the early comparison works that is fairly easy to match mean and rms pressure coefficients between model and full-scale, but the difficulty is found when comparing the extreme minimum (negative peak) pressures observed near roof corners and leading edges. This is attributed to inadequate simulation of the longitudinal and lateral turbulence intensities, their small-scale turbulence content, as well as Reynolds effects. Tieleman (1998) performed a comparative study of the full-scale Texas Tech Building, and concludes that agreement between model and field roof pressure coefficient is only possible when duplicating the two horizontal turbulence intensities and their small-scale turbulence content, and the model turbulence scale exceeds one fifth the magnitude of the scaled-down field scale. Previous Full-scale Projects Previous research efforts in the Wind Engineering community concerning the quantification of wind pressure loading on low-rise building had been carried out extensively by wind tunnel modeling and full-scale instrumentation in the last thirty years. The most relevant full-scale research in this area are: Quezon City experiment in the Philippines (1970s) The structure building work in Aylesbury England (1970s) The structure building work in Silsoe England (1980s) Texas Tech Field Wind Engineering Research Field Laboratory (1980-90s)

PAGE 44

27 To complement these full-scale experiments, wind tunnel modeling had been conducted in order to evaluate and improve wind tunnel simulation techniques. At present ongoing research includes the full-scale measurements on the Kern P. Pitts Center which is being conducted by the Department of Civil and Environmental Engineering, University of Illinois at Urbana-Champaign, and the Department of Civil Engineering, Johns Hopkins University, Baltimore, MD (Porterfield and Jones 2001). The FCMP house program is also ongoing. Analysis of the data sets collected during the Atlantic Hurricane Season of 2004 and 2005 will be presented on this dissertation. Simultaneously wind tunnel models of the FMCP houses are being conducted at Clemson University by Liu (PhD Candidate of the Department and Civil Engineering) under the supervision of Dr. Prevatt. A summary description of each one of the previously mentioned projects is now presented. Quezon City Experiments In the 1970s The National Bureau of Standards (NBS) with the collaboration of the Philippine Atmospheric, Geophysical and Astronomical Services Administration (PAGASA) performed full-scale testing on three single family dwellings in Quezon City, Philippines. The goal of this research was to provide new data set to improve the design criteria used to the design of low-rise structures and validate wind tunnel studies. Wind tunnels studies where conducted at the Virginia Polytechnic Institute and State University. These studies provided key finding between the full-scale data and the wind tunnel model such as strong correlation in the mean and rms pressure coefficient values, which provided a high degree of confidence to the wind tunnel simulations. More details about the findings are presented by Marshall (1976), who reported proper intensity of turbulence is a key factor in generating realistic surface pressure fluctuations on the

PAGE 45

28 wind tunnel model. The positive agreement between model and full-scale dimensionless pressure coefficients, probabilistic distributions, and power spectra suggests that valid wind data can be obtained from wind tunnel test which utilize a relatively large scale and suitable roughness (Marshall 1976). Aylesbury House In the early 1970s the Building Research Establishment (BRE) of the United Kingdom constructed a full-scale two story house (Figure 2-3) to serve as a data collection site for field wind pressure studies on typical low-rise residential buildings. During the years of 1972/74 a large amount of data was collected. Holmes (1982) provide details of the experiment such as: building site location on flat open field, plan dimensions of 7 m x13.3 m, height to the eaves of 5 m, uniquely feature to adjust the roof pitch angle from 5 to 45, pressure transducers positions and characteristics of the full-scale runs. Figure 2-3 Aylesbury experimental building with 22.5 pitch roof

PAGE 46

29 The Aylesbury building is consider a milestone in the wind engineering field, and the initiative for international wind tunnel models conducted around the world included the United Kingdom, the United States, Canada, Australia, and Japan among others. At the time this full-scale date set provided the best database on wind pressure on low-rise building (Holmes, 1982). Silsoe Structures Building The Silsoe Structures Building (SSB) was constructed during 1986/87 at the Silsoe Research Institute (SRI), specifically to undertake full-scale wind pressure measurements. The building was constructed with an optimal eaves geometry offering either traditional sharp eaves or curved eaves of 635 mm radius. The building geometry consists of 24 m long by 12.9 m span by 5.3 m ridge height building with a 10 duo-pitch roof. The surrounding flat terrain is mainly open-country site. Experimental data sets presented by Hoxey (1997) consist of two independent data collection systems. The first conducted by SRI consist of a total of 77 pressure tapping points were installed on the building and measurements were made using a sequence controller which sampled two pressures at one time. A total of 4 records were collected, each one with duration of 4 minutes, which provided 4 mean values of wind pressure data. A complete definition of pressures over the building surface required many hours of recording. The second set of data was conducted by the Building Research Establishment (BRE) which consists of a total of 32 pressure tap points, each mounted locally on the inside of the building. The improvements of this second method allow recording simultaneous measurements with wind velocity sensed by a three-component sonic anemometer. Records of one hour duration were made, and these were partitioned

PAGE 47

30 into six 10 minutes records for the analysis. The typical average reference velocity at the building ridge height measurement was 10 m/s. Wind tunnel studies were conducted at the University of Western Ontario (UWO) and the Building Research Establishment (BRE), described in more detail by Hoxey (1997). Comparisons between the full-scale and wind tunnel models were restricted in the report to comparing mean pressure coefficients, since the mean values were able to be determined with statistical confidence. From the wind tunnel and full-scale analysis Hoxey suggest that wind tunneled may be used to estimate wind loads over the majority of the surface of a building, however regions of separated and conical-vortex flow likely to produce underestimations of surface pressures at model scales. Texas Tech Building In the late 1980s Texas Tech University (TTU) researchers constructed the Wind Engineering Research Field Laboratory (WERFL) in Lubbock, Texas. The facility consists of a permanent experimental building mounted on a turntable. The main goal was to provide full-scale data for comparison with wind tunnel model studies. The building configuration consists of a rectangular flat roof with dimension of 30 ft x 45 ft x 13 ft presented in Figure 2-4. The building location site consists of flat open terrain exposure. A guyed tower is located 150ft west of the building, to provide meteorological data such as wind speed, relative humidity, barometric pressure and temperature. Dearhart (2003) provide a well documented comparison of the extensive wind tunnel model studies that had been carried out to compare with the full-scale data sets. Some of these were conducted by: Colorado State University (CSU), The University of Western Ontario (UWO), the Building Research Institute (BRI) of Japan, and the Wind Load Test Facility (WLTF) at Clemson University.

PAGE 48

31 Cope (1997), stated that several studies have been conducted to compare the full-scale mean, peak, and rms pressure coefficients. Most comparisons match well, but distinct discrepancies have been found between model and full-scale rms and peak minimum pressure coefficients at certain locations, particularly, the areas where high suction occurs due to the shear layer or vortex development exhibited higher peak negative coefficients in full-scale testing than those measured in many wind tunnels. Figure 2-4 Texas Tech University Wind Engineering Research Field Laboratory The discrepancy between the compared data was attributed to the approach flow characteristics (lateral turbulence intensity, smalland large-scale spectral content of approach flow fluctuations), Reynolds number effects, frequency response of the pressure measurement system, and sampling frequency of the acquired data (Bienkiewicz, 1998). Cope (1997) reported positive agreement on mean, rms and peak pressure coefficients by correctly matching the full-scale and wind tunnel time duration. Full-scale Measurements on the Kern P. Pitts Center The full-scale measurements taken on the Kern P. Pitts Center, a low-rise structure owned by the Town of Southern Shores (Figure 2-5) in Southern Shores, North Carolina,

PAGE 49

32 were conducted by the University of Illinois and Johns Hopkins University. The research goals are to collect full-scale wind pressure data to study the wind-induced pressures on this particular low-rise structure. The building instrumentation began in 1997. Since that time some of the most notable data had been obtained from the passage of frontal systems, but data have also been collected in several northeasters and three land-falling hurricanes (Bonnie 1998, Dennis and Floyd 1999), though the structure was not subjected to sustained hurricane force winds. The field site is located in the Town of Southern Shores, on the Outer Banks of North Carolina, about a quarter mile west of the Atlantic Ocean. Figure 2-5 Kern P. Pitts Center located at Southern Shores, NC. Meteorological data are collected at two locations in the vicinity of the structure. Wind speeds are measured at 10 meters above ground level: the first system consist of a three axes ultrasonic anemometer located on a tower 60 feet east of the structure, the second system consist of a propeller-vane anemometer, located on an instrumentation pole above the chimney. Barometric pressure, rainfall, and temperature are also

PAGE 50

33 measured. More details about this research is provided by Porterfield, M. and Jones, N.P. (2001). FCMP Data Set The FCMP wind pressure full-scale data set contains hundreds of hours collected during landfalling tropical cyclones on 22 residential structures. The wind speed ranges fall into the category of hurricane and tropical storm intensities. These data sets will allow computing pressure coefficients from direct full-scale measurements and comparison with wind tunnel models and ASCE-7 wind load provisions. Limitation common to the other experiments above also exist on the FCMP data sets such as uncertainties in the calculation of the pressure coefficients, estimation of the roughness exposure for each house location, and variation of wind direction and wind speed. This will not necessary produce the worst-case loading condition for the evaluated house, but the FCMP house database contains valuable information for the wind engineering community. Since the data was collected at a sampling rate of 100 Hz, this will provide a better understanding of the extreme peak pressures. The 2004 full-scale FCMP data collection effort is the first true sustained hurricane wind pressure data collected on an occupied residential structure. Previous to that the FCMP captured wind pressure data in two houses during tropical storm Isidore (2002). The most current database contains wind pressure data for the following hurricanes: Frances (2004), Ivan (2004), Jeanne (2004), Dennis (2005) and Wilma (2005). Detailed information for the house and tower deployments during the events of the hurricanes of the 2004/05 season are presented in Chapter 3. Details concerning some of the issues previously presented will be discussed in details in chapter 4 along with confidence limits on the pressure coefficient calculations.

PAGE 51

34 New Wind Tunnel Studies (Clemson) to Accompany the Full-scale Data Kopp (2005) states that current wind load provisions for low-rise buildings are based on reductive plots and tables that do not allow designers access to the variation of wind effects with time and space. For most current standards, these plots and tables were determined by obtaining equivalent pressure coefficients that envelope responses calculated from wind tunnel data for a range of assumed structural wind resisting systems. Structural analysis programs are now widely used in the design of structures; these methods can yield accurate structural effects, and can account for both static and dynamic loading. However, the wind loads currently available for low-rise buildings through code provisions do not take advantage of these refined analysis techniques. Preliminary model/full-scale data comparison, conducted at Clemson University, on two FCMP house instrumented during Tropical Storm Isidore (2002) suggest that the ASCE-7 wind load provision underestimate the full-scale negative peak pressure coefficients for components and cladding on houses in suburban terrain (Dearhart 2003). The goal of this dissertation is to provide pressure coefficients from the analyzed data for the vast FCMP house database collected in the last two hurricane seasons of 2004/05 and compare the full-scale pressure coefficients with the ASCE-7 wind load provisions, in order to corroborate the preliminary result presented by Dearhart (2003). Reinhold (2005) suggested from the Dearhart (2003) work that the ASCE-7 peak negative pressure coefficients, instead of enveloping the expected highest negative peak pressure, are likely to underestimate the uplift for critical wind directions. This raises questions about the move towards allowing component and cladding load on low-rise buildings to be calculated for non-open terrain building exposures using a single set of pressure coefficients that were developed from testing models in open terrain conditions.

PAGE 52

35 The need to evaluate the current ASCE-7 design wind loads provision is clearly suggested. The FCMP model/full-scale data analysis is an important step in wind engineering to improve existing wind loading provision in low-rise structures. FCMP wind tunnel simulation can directly include terrain influence by mimicking actual full-scale conditions within the wind tunnel, from turbulence characteristics in the wind field measured at the houses and at the FCMP towers located nearby. The influence of surrounding structures and trees can be accounted for in the wind tunnel. The full-scale measurements only cover a limited range of wind direction for the maximum sustained wind speeds. The wind tunnel simulations (conducted in 5 degree increments over a full 260 degree range) will provide a means to evaluate the worst-case direction for the full-scale data. Chapter 4 will demonstrate that for the mean and RMS pressure coefficient there is a strong match between the full-scale and the wind tunnel models. This is not the case for the minimum and maximum peak pressure coefficients, which strongly suggests, along with the findings of other researchers noted above, that more work needs to be done to re-evaluate the ASCE-7 wind load provisions for homes in non-open terrain. Chapter 4 will also present a method, developed for this dissertation, to directly quantify the effects of uncertainty in the analysis of full-scale data, resulting in a probabilistic format for the pressure coefficients including rationally-based confidence limits on peak coefficients. Thus the combined use of full-scale and wind tunnel studies can produce worst-case full-scale loading with some degree of confidence. Closing Remarks Modern ASCE-7 Wind Load Provision base on the wind tunnel work by Stathopoulos developed in the late 1970s is no longer the state-of-the-art. The equivalent

PAGE 53

36 full-scale sampling rate of 2 Hz is to slow compared with todays instrumentation standards. The modeled exposure condition is considered by todays standard as coastal rather than suburban exposure. Current technology has produced full-scale pressures as well as updated wind tunnel techniques to update the current design methodologies. The work presented in this dissertation focuses largely on the analysis of the full-scale data, incorporating new concepts for the identification of a suitable sampling rate, and the identification, modeling and quantification of the influence of uncertainties on calculated pressure coefficients () values. pC

PAGE 54

CHAPTER 3 FCMP DEPLOYMENT HISTORY, ORGANIZATION AND LOGISTICS FOR THE 2004/2005 SEASONS The last two hurricane seasons were prolific in terms of the number of hurricanes to impact the U.S., the damage caused, and the data collected. The FCMP research effort collected ground level approach wind speed velocity and wind pressure data on low-rise residential structures along the coast of the State of Florida. The FCMP deployment activities are presented in the following section; limited to the events in which house data was collected. Additional deployment information and a portion of the collected data can be accessed at the FCMP web site http://www.ce.ufl.edu/~fcmp The synoptic history and track data for each cyclone was taken from the National Hurricane Center Tropical Cyclone Report archives, available at http://www.nhc.noaa.gov This dissertation focuses in large part on the application of full-scale pressure data to evaluate extreme wind loading on residential structures. This chapter will provide details on the full-scale deployments, and discuss the specific data sets captured in the last two years that are currently undergoing analysis. Aerial picture of the tower deployment sites and house are located in Appendix A. Appendix C presents the sensor roof layout of the instrumented houses along with the sensitivity data of the sensors used. The analysis of every house instrumented in 2004 and 2005 is not presented in this dissertation, as much of that work relies upon wind tunnel testing still being conducted at Clemson University. The content of this chapter serves the purpose of documenting the house instrumentation-related activities of 2004 and 2005. 37

PAGE 55

38 Hurricane Frances (2004) Synoptic History Frances developed of the coast of Africa on 21 August, becoming a hurricane on August 26, and the intensification continued until 31 August, when Frances reached a peak intensity estimated at 125 knots (category 4) as it passed north of the Leeward and Virgin Islands. It was a category 2 hurricane with winds of 85-90 knots over the northwestern Bahamas on 3-4 September. Frances weakened just before Frances made landfall over the southern end of Hutchinson Island, Florida near 0430 UTC 5 September as a Category 2 hurricane. Frances weakened as it moved slowly across the Florida Peninsula, and became a tropical storm just before emerging into the northeastern Gulf of Mexico near New Port Richey early on 6 September. It made a final landfall near the mouth of the Aucilla River in the Florida Big Bend region about 1800 UTC 6 September. House and Tower Deployment In Hurricane Frances FCMP personnel successfully deploy four mobile towers and instrumented five houses, distributed along the Florida Atlantic Coast. Figure 3-1 presents a deployment map indicating the location of the towers and instrumented houses and the track of the hurricane path. The tower labels indicate the highest 3-second gust recorded at that location. The blue triangle icons are FCMP houses that were not instrumented. The purple triangle icons are FCMP houses that were instrumented and collected data. All houses that collected data in Frances were north of the eye on the strong side of the storm, and one was very close to the center of the storm. Table 3-1 provides details of each tower location, the maximum measured 3-second and 1-minute gust at each tower with the corresponding time in UTC.

PAGE 56

39 Figure 3-1 FCMP Deployment map of hurricane Frances (2004) Table 3-1 Tower Data Records in Hurricane Frances (2004) TOWERCITY, STATEGPS COORDINATESMaximum Wind Speed On Site @ 10m ( 1-Min Gust ) Date Time (UTC)Maximum Wind Speed On Site @ 10m ( 3-Sec Gust ) Date Time (UTC)27 08' 53.3" N80 12' 53.9" W28 08' 41.6" N80 35' 49.4" W27 41' 38.1" N80 24' 29.5" W27 26' 50.3" N80 19' 17.9" W9/5/2004 02:40:229/5/2004 15:08:309/5/2004 09:58:519/5/2004 04:01:21T3Fort Pierce, FL9/5/2004 04:03:0281.00 mph @ 17108.32 mph @ 16T2Vero Beach, FL64.02 mph @ 7681.29 mph @ 969/5/2004 09:14:46T1Indian Harbor Beach, FL 72.21 mph @ 25283.02 mph @ 2279/5/2004 16:42:47T0Port Salerno, FL57.33 mph @ 35882.23 mph @ 59/5/2004 02:05:39 Table 3-2 House Data Records in Hurricane Frances (2004) 3-Sec1-Min3-Sec1-MinFL-06Jensen Beach, FL1429/5/2004 14:04:109/7/2004 01:39:047.062915810686FL-04Vero Beach, FL1019/4/2004 21:54:109/5/2005 23:08:23N.A.N.A.N.A.10586FL-03Vero Beach, FL1759/5/2004 14:58:239/7/2004 10:43:37N.A.N.A.N.A.10586FL-02Melbourne Beach, FL2719/3/2004 10:50:409/6/2004 06:44:03N.A.N.A.N.A.9779FL-01Melbourne, FL3179/3/2004 02:14:579/6/2004 09:59:54N.A.N.A.N.A.8771Number of RecordsStart Time (UTC)End Time (UTC)House Anemometer Heigth (m)Local Exposure MeasuredOpen Exposure Estimation*FCMP house database contains 251.5 hrs of data for major storm deployments during 2002 2005*Estimation performed by Applied Research Associates parametric hurricane wind field model (Vickery, et al, 2000) Max Wind Speed (mph) @ Anemometer HeightFCMP House IDCity, StateMax Wind Speed (mph) @ 10m Height Zo = 0.03

PAGE 57

40 Table 3-2 provides details of the house city location, the total number of 15 minutes records collected, anemometer height with maximum 3-second and 1-minute gust (if available) and Applied Research Associates parametric Wind Field Model maximum 3-second and 1-minute gust estimation for open exposure conditions at 10 m height (Vickery et al. 2000). Hurricane Ivan (2004) Synoptic History Ivan developed off the west coast of Africa on 31 August, becoming Tropical Storm Ivan at 0600 UTC 3 September. Ivan reached its first peak intensity of 115 knots at 0000 UTC 6 September, making Ivan the southernmost major hurricane on record. Ivan reached its second peak intensity -140 knots and category 5 strength --12 hours later. As Ivan passed south of Jamaica it weakened to category 4 strength. Ivan rapidly intensified to category 5 strength a second time 11 September, weakened back to a category 4 hurricane on 12 September, and re-strengthened to category 5 for its third and final time when it was about 80 n mi west of Grand Cayman Island. The hurricane brought sustained winds just below category 5 strength to the island. This resulted in widespread wind damage, and a storm surge that completely over swept the island except for the extreme northeastern portions. On 13 September Ivan moved over the northwestern Caribbean Sea. The very warm water in that region helped the hurricane maintain category 5 strength for 30 hours. As Ivan neared the northern U.S. Gulf coast it weakened slowly and made landfall as a 105 knots hurricane (category 3) at approximately 0650 UTC 16 September, just west of Gulf Shores, Alabama. By this time, the eye diameter had increased to 40-50 n mi, which resulted in some of the strongest winds occurring over a narrow area near the

PAGE 58

41 southern Alabama-western Florida panhandle border. After Ivan moved across the barrier islands of Alabama, the hurricane turned north-northeastward across eastern Mobile Bay and weakened into a tropical storm 12 hours later over central Alabama. House and Tower Deployment In Hurricane Ivan FCMP personnel successfully deploy four mobile towers and instrumented six houses, distributed along the Florida Panhandle and Alabama. Figure 3-2 presents a deployment map indicating the location of the towers and instrumented houses and the track of the hurricane path. As-measured peak 3-second wind speeds are indicated on the graph where reliable data was available. All houses are on the strong side of the storm. Figure 3-2 FCMP Deployment map of hurricane Ivan (2004).

PAGE 59

42 Table 3-3 Tower Data Records in Hurricane Ivan (2004) TOWERCITY, STATEGPS COORDINATESMaximum Wind Speed On Site @ 10m ( 1-Min Gust ) Date Time (UTC)Maximum Wind Speed On Site @ 10m ( 3-Sec Gust ) Date Time (UTC)30 38' 39.9" N88 03' 48.1" W30 28' 45.4" N87 11' 12.8" W30 28' 21.0" N87 52' 30.0" W30 23' 44.6" N86 27' 58.9" W9/15/2004 08:24:269/16/2004 06:49:599/16/2004 06:44:58N.A.9/15/2004 08:43:309/16/2004 06:43:189/16/2004 06:44:18N.A.T3Destin, FLN.A.N.A.T2Fairhope, AL67.96 mph @ 6289.23 mph @ 56T1Pensacola, FL 80.29 mph @ 124106.26 mph @ 124T0Mobile, AL49.69 mph @ 32772.66 mph @ 320 Table 3-4 House Data Records in Hurricane Ivan (2004) 3-Sec1-Min3-Sec1-MinFL-30Pensacola, FL2199/14/2004 23:19:289/17/2004 06:20:236.553916511493FL-28Pensacola, FL1859/15/2004 01:56:239/17/2004 00:22:13N.A.N.A.N.A.10585FL-27Gulf Breeze, FL2119/14/2004 20:04:339/17/2004 01:04:036.55382459779FL-26Navarre, FL2479/15/2004 13:22:329/18/2004 03:14:056.09659328973FL-24Destin, FL3149/13/2004 19:23:339/16/2004 18:40:546.09656317863FL-23Destin, FL3039/13/2004 19:37:399/16/2004 23:48:496.09657327561Open Exposure Estimation*Number of RecordsHouse Anemometer Heigth (m)*Estimation performed by Applied Research Associates parametric hurricane wind field model (Vickery, et al, 2000) Max Wind Speed (mph) @ Anemometer HeightFCMP House IDCity, StateMax Wind Speed (mph) @ 10m Height Zo = 0.03FCMP house database contains 154.25 hrs of data for major storm deployments during 2002 2005Local Exposure MeasuredStart Time (UTC)End Time (UTC) Table 3-3 provides details of each tower location, the maximum measured 3-second and 1-minute gust at each tower with the corresponding time in UTC. Table 3-4 provides details of the house city location, the total number of 15 minutes records collected, anemometer height with maximum 3-second and 1-minute gust (if available) and Vickerys maximum 3-second and 1-minute gust estimation for open exposure conditions at 10 m height (Vickery et al. 2000). This Hurricane Ivan data set will be used extensively in later chapters to present the house data analysis techniques developed for this dissertation. In particular, FL-27 and FL-30 will be the subject of detailed analyses. Hurricane Jeanne (2004) Synoptic History Jeanne formed from a wave off of Africa on 7 September, strengthening to a tropical storm on 14 September while it moved over the Leeward Islands. It moved over the southeastern Puerto Rico on 15 September when maximum sustained surface winds

PAGE 60

43 reached 60 knots. Jeanne was a hurricane with 70-knots winds during the Dominican Republic landfall, but then weakened over the rough terrain of Hispaniola. Jeanne's slow forward motion across the Caribbean caused torrential rainfall and flooding along its path, causing thousands to die in Haiti. While Jeanne was dumping rain over the Caribbean countries, Hurricane Ivan moved over the Gulf of Mexico and inland across the southeastern United States. Jeanne moved slowly northward over the southeastern Bahamas as a tropical storm and then moved in a loop about 500 n mi east of the northwestern Bahamas. Jeanne gradually strengthened to a hurricane with 85-knots winds by the time it completed this loop on 23 September. On 24 September, Jeanne moved over its own previous track from a few days earlier and encountered cooler waters caused by upwelling from the hurricane. This is believed to decrease the maximum winds from 85 knots to 70 knots on 24 September. Moving away from the upwelled cooler water, the winds increased to 100 knots (category 3) on 25 September. Jeanne made landfall on the east coast of Florida early on 26 September with the center of its 50-n mi diameter eye crossing the coast at the southern end of Hutchinson Island just east of Stuart at 0400 UTC on 26 September. Maximum winds at landfall are estimated at 105 knots over a very small area north of the center and it is not clear whether these strongest winds reached the coast or remained over water. House and Tower Deployment In Hurricane Jeanne FCMP personnel successfully deploy four mobile towers and instrumented four houses, distributed along the Florida east coast. Figure 3-3, presents a deployment map indicating the location of the towers and instrumented houses and the track of the hurricane path. Significantly, several of the same houses that collected data

PAGE 61

44 during Frances were again employed for Jeanne, providing the opportunity to analyze the peak winds on the same structure during two different events. Table 3-5 provides details of each tower location, the maximum measured 3-second and 1-minute gust at each tower with the corresponding time in UTC. Table 3-6 provides details of the house city location, the total number of 15 minutes records collected, anemometer height with maximum 3-second and 1-minute gust (if available) and Vickerys maximum 3-second and 1-minute gust estimation for open exposure conditions at 10 m height (Vickery et al. 2000). Figure 3-3 FCMP Deployment map of hurricane Jeanne (2004) The lack of anemometer data measured directly at these house locations (due to hip roof configuration) increases the uncertainty in the analysis of pressure coefficients for

PAGE 62

45 these homes. Nearby tower data or the wind field map data from ARA will be used instead. Table 3-5 Tower Data Records in Hurricane Jeanne (2004) TOWERCITY, STATEGPS COORDINATESMaximum Wind Speed On Site @ 10m ( 1-Min Gust ) Date Time (UTC)Maximum Wind Speed On Site @ 10m ( 3-Sec Gust ) Date Time (UTC)28 21' 17.0" N81 26' 10.0" W27 48' 50.7" N80 29' 59.5" W28 38' 25.7" N80 43' 50.0" W27 39' 20.2" N80 24' 49.0" W9/26/2004 04:17:399/26/2004 15:39:179/26/2004 06:47:399/26/2004 11:32:509/26/2004 04:18:039/26/2004 10:59:589/26/2004 06:40:249/26/2004 11:32:11T0Orlando, FLT1Sebastian, FLT2Merritt Island, FLT3Vero Beach, FL68.51 mph @ 32101.81 mph @ 7763.77 mph @ 99106.19 mph @ 3945.34 mph @ 11385.17 mph @ 8547.48 mph @ 9281.47 mph @ 40 Table 3-6 House Data Records in Hurricane Jeanne (2004) 3-Sec1-Min3-Sec1-MinFL-02Melbourne Beach, FL2009/26/2004 17:25:199/28/2004 19:38:55N.A.N.A.N.A.10182FL-01Melbourne, FL1159/25/2004 03:16:579/26/2004 08:07:45N.A.N.A.N.A.9275FL-31Melbourne, FL1779/25/2004 20:34:099/27/2004 16:49:35N.A.N.A.N.A.8771FL-32Merritt Island, FL1619/25/2004 22:51:169/27/2004 15:14:02N.A.N.A.N.A.7964House Anemometer Heigth (m)FCMP house database contains 84.5 hrs of data for major storm deployments during 2002 2005*Estimation performed by Applied Research Associates parametric hurricane wind field model (Vickery, et al, 2000) Number of RecordsStart Time (UTC)End Time (UTC)Max Wind Speed (mph) @ Anemometer HeightFCMP House IDCity, StateMax Wind Speed (mph) @ 10m Height Zo = 0.03Local Exposure MeasuredOpen Exposure Estimation* Hurricane Dennis (2005) Synoptic History Dennis formed from a tropical wave that moved westward from the coast of Africa on 29 June. The system moved through the southern Windward Islands on 4 July and lost organization over the southeastern Caribbean. The system reformed, becoming tropical storm Dennis on 5 July. Dennis reached hurricane strength early on 7 July, then rapidly intensified into a Category 4 hurricane with winds of 120 knots before making landfall in southeastern Cuba on 0245 UTC 8 July. Once offshore the hurricane intensified to Category 4. Maximum sustained winds reached a peak of 130 knots on 8 July, then decreased to 120 knots before Dennis made landfall in Cuba again at 1845 UTC. Dennis traversed a long section of western Cuba before emerging into the Gulf of Mexico 9 July. Dennis regained

PAGE 63

46 strength over the Gulf of Mexico with maximum sustained winds reaching 125 knots on 10 July. The maximum sustained winds decreased to 105 knots before Dennis made landfall on Santa Rosa Island, Florida, between Navarre Beach and Gulf Breeze, about 1930 UTC 10 July. House and Tower Deployment In Hurricane Dennis FCMP personnel successfully deploy five mobile towers and instrumented four houses, distributed along the Florida Panhandle. Figure 3-4, presents a deployment map indicating the location of the towers and instrumented houses and the track of the hurricane path. Figure 3-4 FCMP Deployment map of hurricane Dennis (2005)

PAGE 64

47 Like the case from Jeanne in 2004, Dennis offered the opportunity to instrument houses that had already collected data during a previous storm (Ivan 2004), again providing load data on the same structures for different events. Table 3-7 Tower Data Records in Hurricane Dennis (2005) TOWERCITY, STATEGPS COORDINATESMaximum Wind S p eed On Site @ 10m ( 1-Min Gust ) Date Time (UTC)Maximum Wind S p eed On Site @ 10m ( 3-Sec Gust ) Date Time (UTC)30 24' 01.0" N86 51' 49.0" W30 17' 00.0" N86 01' 52.0" W30 28' 37.3" N87 11' 15.3" W30 23' 41.9" N86 27' 56.6" W30 32' 17.4" N86 29' 24.2" W*Measured at 5m from ground7/10/2005 19:41:417/10/2005 07:15:097/10/2005 18:24:36N.A.7/10/2005 19:30:557/10/2005 07:17:067/10/2005 18:36:51N.A.7/10/2005 19:36:51T0*Navarre, FLT1Inlet Beach, FLT2Pensacola, FLT3Destin, FLT5Niceville, FL99.28 mph @ 280120.71 mph @ 27969.55 mph @ 29580.45 mph @ 289N.A.N.A.63.59 mph @ 7780.76 mph @ 7838.41 mph @ 19961.81 mph @ 2017/10/2005 19:54:31 Table 3-8 House Data Records in Hurricane Dennis (2005) 3-Sec1-Min3-Sec1-MinFL-24Destin, FL1307/10/2005 11:31:137/11/2005 19:57:336.0965928N.A.N.A.FL-26Navarre, FL1497/9/2005 16:24:527/11/2005 05:46:586.0964322N.A.N.A.FL-23Destin, FL1737/9/2005 21:22:537/11/2005 16:47:38N.A.N.A.N.A.N.A.N.A.FCMP house database contains 80.5 hrs of data for major storm deployments during 2002 2005*Estimation performed by Applied Research Associates parametric hurricane wind field model (Vickery, et al, 2000) Max Wind Speed (mph) @ Anemometer HeightFCMP House IDCity, StateMax Wind Speed (mph) @ 10m Height Zo = 0.03Local Exposure MeasuredOpen Exposure Estimation*Number of RecordsStart Time (UTC)End Time (UTC)House Anemometer Heigth (m) Table 3-7 provides details of each tower location, the maximum measured 3-second and 1-minute gust at each tower with the corresponding time in UTC. Table 3-8 provides details of the house city location, the total number of 15 minutes records collected, anemometer height with maximum 3-second and 1-minute gust (if available) and Vickerys maximum 3-second and 1-minute gust estimation for open exposure conditions at 10 m height (Vickery 2000). Hurricane Wilma (2005) Synoptic History Wilma has a complicated story. During the second week of October, an unusually large circulation and a broad area of disturbed weather developed over much of the Caribbean Sea. A more concentrated area of disturbed weather and surface low pressure

PAGE 65

48 formed near Jamaica by 14 October. By October 15 the surface circulation became well-enough defined to designate that a tropical depression had formed, centered about 190 n mi east-southeast of Grand Cayman. The depression moved slowly and erratically westward to west-southwestward for a day or so and then drifted south-southwestward to southward for a day or two. The system is estimated to have become tropical storm Wilma at 0600 UTC 17 October. On 18 October Wilma turned toward the west-northwest and strengthened into a hurricane. Later that day, a remarkable strengthening episode began and continued through early on 19 October. By 0600 UTC 19 October, Wilmas winds had increased to near 150 knots (category 5). In the span of just 24 hours, Wilma had intensified from a 60-knots tropical storm to a 150-knots category 5 hurricane, an unprecedented event for an Atlantic tropical cyclone. Wilma reached its peak sustained wind speed of 160 knots at around 1200 UTC 19 October. During the strengthening episode, Air Force reconnaissance observations indicated that the eye of the hurricane contracted to a diameter of 2 n mi; this is the smallest eye known to National Hurricane Center (NHC) staff. The estimated minimum central pressure at the time of peak intensity is 882 mbar, which is a new record low value for a hurricane in the Atlantic basin. Wilma maintained category 5 status until 20 October, when its winds decreased to 130 knots, and the tiny eye was replaced by one about 40 n mi across. The hurricane would retain this large eye ranging from about 40 to 60 n mi in diameter, for most of the remainder of its lifetime. By 21 October the hurricane turned toward the northwest toward the Yucatan Peninsula. Wilmas maximum winds were still near 130 knots (category 4) when it made landfall on the island of Cozumel around 2145 UTC 21

PAGE 66

49 October, and was probably only slightly weaker (but still category 4 intensity) when it crossed the coast of the Yucatan peninsula about 6 hours later. On 22 October the hurricane moved slowly northward, severely battering the extreme northeastern Yucatan peninsula. Wilma emerged into the southern Gulf of Mexico around 0000 UTC 23 October, with maximum winds of near 85 knots, still a large and powerful hurricane. A vigorous southwesterly steering current accelerated Wilma northeastward toward southern Florida. Wilma strengthened over the southeastern Gulf of Mexico and its winds reached about 110 knots as it approached Florida. Maximum sustained winds were estimated to be near 105 knots (category 3 intensity) when landfall occurred in southwestern Florida near Cape Romano around 1030 UTC 24 October. Moving at a forward speed of 20 to 25 knots, the hurricane crossed the southern Florida peninsula in 4.5 hours, emerging into the Atlantic just southeast of Jupiter around 1500 UTC. Maximum winds had decreased to near 95 knots (category 2) during the crossing of Florida. House and Tower Deployment In Hurricane Wilma FCMP personnel successfully deploy five mobile towers and instrumented two houses, distributed along the Gulf and east coast in the State of Florida. Figure 3-5 presents a deployment map indicating the location of the towers and instrumented houses and the track of the hurricane path. The center of Wilma passed over one of the instrumented houses on Marco Island, and south of the other house near Naples. Table 3-9 provides details of each tower location, the maximum measured 3-second and 1-minute gust at each tower with the corresponding time in UTC.

PAGE 67

50 Figure 3-5 FCMP Deployment map of hurricane Wilma (2005) Table 3-9 Tower Data Records in Hurricane Wilma (2005) TOWERCITY, STATEGPS COORDINATESMaximum Wind Speed On Site @ 10m ( 1-Min Gust ) Date Time (UTC)Maximum Wind Speed On Site @ 10m ( 3-Sec Gust ) Date Time (UTC)25 54' 03.0"N81 18' 41.0"W26 08' 45.0"N80 30' 24.0"W25 52' 05.0"N80 53' 59.0"W25 45' 05.6"N80 22' 25.7"W26 09' 07.4"N81 46' 37.6"W10/24/2005 13:46:55N.A.10/24/2005 12:27:5610/24/2005 12:07:1010/24/2005 13:12:0210/24/2005 12:27:4110/24/2005 11:48:5710/24/2005 13:11:4696.04 mph @ 10481.85 mph @ 279N.A.109.11 mph @ 276N.A.10/24/2005 14:12:02N.A.71.68 mph @ 29393.82 mph @ 29186.85 mph @ 145104.69 mph @ 14169.87 mph @ 92T5Naples, FLT3Miami, FLT2Ochoppi, FLT1Weston, FLT0Everglades City, FL Table 3-10 House Data Records in Hurricane Wilma (2005) 3-Sec1-Min3-Sec1-MinFL-18Marco Island, FL18910/22/2005 18:31:4710/24/2005 17:59:366.0967746N.A.N.A.FL-19Naples, FL17610/22/2005 20:25:2710/24/2005 16:35:57N.A.N.A.N.A.N.A.N.A.House Anemometer Heigth (m)FCMP house database contains 91.25 hrs of data for major storm deployments during 2002 2005*Estimation performed by Applied Research Associates parametric hurricane wind field model (Vickery, et al, 2000) Number of RecordsStart Time (UTC)End Time (UTC)Max Wind Speed (mph) @ Anemometer HeightFCMP House IDCity, StateMax Wind Speed (mph) @ 10m Height Zo = 0.03Local Exposure MeasuredOpen Exposure Estimation*

PAGE 68

51 Table 3-10 provides details of the house city location, the total number of 15 minutes records collected, anemometer height with maximum 3-second and 1-minute gust (if available) and Vickerys maximum 3-second and 1-minute gust estimation for open exposure conditions at 10 m height (Vickery et al. 2000). Hurricanes Katrina and Rita (2005) Florida was not directly impacted by Rita, and only suffered the beginnings of Katrina in south Florida as it passed in to the gulf. Thus no houses collected data from either of these storms. The FCMP did successfully deploy the portable towers to collected wind speed data from these two strong storms. For details and results, interested readers are referred to http://www.ce.ufl.edu/~fcmp. Closing Remarks The full-scale pressure data analysis techniques presented in this dissertation were developed using data collected in the Panhandle during Hurricane Ivan, 2004. An initial review of the complete data set from 2004 and 2005 revealed the site-specific nature of the required analysis (details in a later chapter), and justified a focus on data sets that were among the most complete and reliable. Identifying appropriate values for each of the terms in the pressure coefficient (presented earlier in equation 2-1), involve some judgment and a quantification of uncertainties associated with measurement and conditions. This process became the central focus of the dissertation research as well as the evaluation of the conditions under which a fair comparison can be made among full-scale, wind tunnel, and ASCE-7 loads. Chapter four presents the analysis of the full-scale pressure data with an emphasis on a particular house in the Panhandle the endured sustained hurricane winds during Ivan

PAGE 69

52 in 2004. The assumptions are explained, uncertainties are quantified, and new techniques are developed to provide a probabilistic view of extreme wind loads on residential housing.

PAGE 70

CHAPTER 4 ANALYSIS OF FULL-SCALE DATA TO DEFINE PRESSURE COEFFICIENTS This chapter will present the methods developed and results from the analysis of full-scale pressure data collected in Florida during hurricane landfall. There are two focuses in this chapter. The first is the identification of a suitable sampling rate for analysis. The second is the identification, modeling and quantification of the influence of uncertainty in the calculated values. pC Calculating for Full-scale Data: Methods and Outstanding Issues pC Each FCMP full-scale dataset for a particular instrumented house contains up to 28 dynamic pressure measurements from sensors located along the roof, soffit and walls. It also contains pressure measurements for use as the reference pressure. Recall from Eq. 2-1 / 4-3 that the pressure coefficient is expressed in terms of the difference between the local pressure at the point of interest and the barometric (reference) pressure. The potential barometric pressure reference sensors are located either in the attic of the house, away from the house but on the property (yard sensor), or both. In the case of instrumentation malfunction the reference pressure will be obtained from a nearby FCMP mobile tower. The measurement of the local and reference pressures have uncertainties that will affect the reliability (confidence) of the pressure coefficient estimation. These uncertainties are related to the unique calibration factor of each sensor and the fact that the sensor output voltage is affected by temperature changes presented by Michot (1999) 53

PAGE 71

54 In addition to the local and reference pressure a reference wind speed is also selected to normalize the coefficient. Typically this reference wind speed represents the expected (average) peak speed of the wind as the flow approaches the house. This requires selection of both duration (e.g., 1-minute wind or 3-second wind) and an elevation above local ground level (e.g. mean roof height). To select this speed, most of the FCMP houses have a single anemometer installed ~ 4-6 above the ridgeline, a few houses have two, and some (typically hip roof houses) do not have an anemometer installed. Other sources are available to obtain the reference wind speed, for example a nearby FCMP mobile tower or estimations performed by Applied Research Associates parametric hurricane wind field model (Vickery, et al, 2000). The measurement and calculation of a reference wind speed is another source of uncertainty in the final pressure coefficient value. Applied Equations for pC The data analysis starts by exporting each file into ASCII format, which is then converted into MATLAB format for easy manipulation. The data were collected at a sampling rate of 100 Hz, but for analysis purposes the data were initially down sample to 10 Hz. The following steps were used to analyze the data. First, each of the data channels was validated by visually inspecting the time history of the mean raw voltage measurements to provide a list of the functioning sensors. Table 4-1 and 4-2 presents the result of this process for the FL-27 and FL-30 houses respectively, for the data collected during hurricane Ivan (2004).

PAGE 72

55 0 20 40 60 80 100 120 140 160 2.5 3 3.5 4 4.5 5 5.5 Time History of Channel 0 mean voltage output for FL-27 during Ivan (2004)Record numberVolts Figure 4-1 Voltage time history of functional data channel 0 for the FL-27 house during hurricane Ivan (2004) 0 20 40 60 80 100 120 140 160 -5 0 5 10 Time HIstory of Channel 4 during Ivan (2004)Record #Volts Figure 4-2 Voltage time history of a malfunctioning data channel 4 for the FL-27 house during hurricane Ivan (2004)

PAGE 73

56 Figure 4-1 shows a typical example of a functional channel (yard sensor for reference pressure) and Figure 4-2 shows a channel considered malfunctioning (attic reference sensor) for the FL-27 house. Both examples were collected during Hurricane Ivan, 2004). Typical data error could be generated by electrical fluctuation spikes or other unknown factors, causing data sensors to malfunction for short durations, an example of this situation was observed in record # 129 (15-minutes of data) for the FL-27 during Ivan (2004) from channel 5. The plot in Figure 4-3 shows the time histories of channel 5 and 6, where channel 5 shows an erratic output voltage for a short duration in the 15 minute record, compared with channel 6. Figure 4-4 shows that sensors 5 and 6 are in close proximity, and it is expected that the raw voltage time histories should be similar. 00:00 2:30 05:00 07:30 10:00 12:30 15:00 2 2.2 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8 4 Time (MM:SS)Channel Output (Volts)Record #129 FL-27 Ivan (2004) Channel 5 Channel 6 Figure 4-3 Record #129 of house FL-27 hurricane Ivan (2004), channel 5 malfunctions The offset is not of concern, as this differs from sensor to sensor. The dramatic attenuation and subsequent positive jump of the voltage from sensor is considered a

PAGE 74

57 malfunction. This type of behavior generates erroneous pressure coefficients. Channel five for this particular 15 minutes of data is removed from the dataset. The roof sensor layout configuration and the result for the mean output voltage channel inspection for the FL-27 and FL-30 houses are presented in Figure 4-4, 4-5 and Table 4-1, 4-2 respectively. Refer to Figure 3-2 for the location of these houses relative to the path of Hurricane Ivan. The house instrumentation layout shows the location of each sensor along the roof, walls, and soffit as well the location of the house anemometers, and house orientation. The information presented on the tables shows the sensor type, I.D. of the sensor used for each particular event, channel used to collected data (i), corresponding sensor calibration factor ( ), source of calibration factor and channel status. The calibration source UF 2005 refers to a calibration of the specific sensor that was at that location, conducted by the author at UF. Average value refers to using a mean calibration from the sensors that were calibrated at UF. In this case the specific sensor was not calibrated due to an inability to recover the specific sensor used (lost, no longer functioning, or mislabeled during deployment). Use of the average value is viable due to the close agreement of calibration among the many sensors, but it does introduce additional uncertainty in to the analysis, particularly for the peak value estimates. pC The third step requires a nearby FCMP mobile tower or local ASOS (airport weather) station to access temperature time history. This is necessary in order to apply correction changes to the calibration due to temperature effects on the pressure transducers. For this task the GIS frame work developed by the author is very practical, which allow easy visualization of near ASOS station or mobile towers relative to the

PAGE 75

58 house locations. For the analysis of the two houses FL-27 and FL-30 the temperature reference were obtained from the mobile tower T1 data records (Figure 3-2). Figure 4-4 Roof sensor layout configuration for FCMP House FL-27 Figure 4-5 Roof sensor layout configuration for FCMP House FL-30

PAGE 76

59 Table 4-1 Sensor list for FCMP house FL-27 hurricane Ivan (2004) Plastic BoxCamera034.274Average ValueOKPan Sensor026A134.274Average ValueNGPan Sensor008A234.149UF2005OKPan Sensor027A334.274Average ValueOKPlastic BoxAttic434.274Average ValueNGPan Sensor035A534.274Average ValueOKPan Sensor201634.336UF2005OKPan Sensor92734.154UF2005OKPan Sensor834.274Average ValueNGPan Sensor42934.430UF2005OKPan Sensor1671034.733UF2005OKPan Sensor931134.434UF2005OKPan Sensor811234.203UF2005OKPan Sensor003A1334.588UF2005OKPlastic Box1434.274Average ValueNGPan Sensor721534.420UF2005OKPan Sensor131634.274Average ValueOKPan Sensor841734.147UF2005OKPan Sensor2251834.608UF2005OKPan Sensor341934.147UF2005OKPan Sensor1432034.457UF2005OKPan Sensor1442134.396UF2005OKPan Sensor492234.092UF2005OKPan Sensor1772334.224UF2005OKPan Sensor922434.154UF2005OKPlastic Box2534.274Average ValueNGPan Sensor1222634.380UF2005NGPan Sensor542734.395UF2005NGPan Sensor1062834.230UF2005OKAnemometer 1-29-OKAnemometer 2-30-Not in use Channel iCalibration iCalibration SourceChanStatSensor TypeSensor ID nel us The fourth step requires selecting a source of wind speed for the expected peak 3-seconr es FLd gust value. Various sources are available for this: (a) using the 3-cup anemometemounted on the house, (b) from nearby mobile tower measurement applying the proper adjustment for exposure and height or (c) by using overland wind field model estimationprovided by Peter Vickery from Applied Research Associates (ARA) parametric hurricane wind field model (Vickery, et al, 2000). For the analysis of the two hous27 and FL-30 the wind speed source will be obtained from each house anemometer by

PAGE 77

60 estimating the peak 3-sec gust from the 15-minute mean wind speed value. This will bedone using two different methods. Table 4-2 Sensor list for FCMP hou se FL-30 hurricane Ivan (2004) Plastic BoxCamera034.274Average ValueOKPan Sensor031A134.380UF2005OKPan Sensor70234.348UF2005OKPan Sensor19334.452UF2005OKPlastic Box132434.240UF2005OKPan Sensor15533.867UF2005OKPan Sensor171634.398UF2005OKPan Sensor104734.270UF2005OKPan Sensor238834.409UF2005OKPan Sensor043A934.392UF2005OKPan Sensor044A1034.182UF2005OKPlastic BoxSoffit1134.274Average ValueOKPan Sensor1821234.194UF2005OKPan Sensor045A1334.274Average ValueOKPlastic Box010A1434.389UF2005OKPan Sensor741534.016UF2005OKPan Sensor261634.387UF2005OKPan Sensor050A1734.274Average ValueOKPan Sensor2091834.659UF2005OKPan Sensorguess1934.274Average ValueOKPan Sensor002A2034.274Average ValueOKPan Sensor1932134.190UF2005OKPan Sensor1732234.255UF2005OKPlastic BoxWall2334.274Average ValueOKPan Sensor1162434.345UF2005NGPlastic BoxNot in use2534.274Average ValueNGPlastic BoxSoffit2634.274Average ValueOKPlastic BoxWall2734.274Average ValueNGPlastic BoxAttic2820.600Old SensorOKAnemometer 1-29-OKAnemometer 2-30-Not in useCalibration SourceChanStatSensor TypeSensor IDChannel iCalibration i nel us After all the input parameters are establish for each house scenario, the pressure differe the case where the reference pressure is obtained from a nearby tower. ential needed for calculation of the pressure coefficient can be computed using onof two equations, depending on the dataset conditions: Eq. 4-1 is used for the case where the reference pressure is obtained from the attic or camera (yard) sensors, and Eq. 4-2 for

PAGE 78

61 The evaluation of Eq. 2-1 / 4-1 through 4-3 from the voltages collected by the sensors is not a straightforward operation. The sensitivity of each senso r is described in terms of e difference between atmospheric pressure and the pressure measuen is done s done by including a temperature differential between prea of a linear relationship between pressure and voltage. It is typical that the slopethis sensitivity remains very steady for a given sensor, while the y-intercept is more variable between sensors. The analysis presented here eliminates the need for a y-intercept for the sensorsensitivity by calculating th red during the storm. That is, calculating the absolute dynamic pressure at a givsensor (which requires knowledge of the y-intercept sensitivity), is replaced by calculating the difference between the pressure at that sensor during still winds and that during peak winds (which does not require knowledge of the y-intercept). This by taking the difference in the mean sensor voltage from a still wind period well before the storm and the dynamic voltage measured during the storm. That provides the first term in the numerator in equation 2-1 (first parenthetical term in Eq. 4-1). The second term in the equation (reference pressure) is calculated as a differential in the same way(pre-storm minus during-storm). We do need to account for the change to the slope of the sensor sensitivity causedby temperature dependence. This i nd during-storm conditions. The change in slope due to temperature is a measured, known quantity (Dearhart, 2003). The pressure differential itP in the numerator of Eq. 4-3 can thus be representedby Eq. 4-1. Equation 4-2 is used w hen the source for the reference pressure is not at the house, but from a nearby FCMP tower or other atmospheric data record (airport, etc).

PAGE 79

62 REFREFREFiTemptVtPREF6.200144.00144.00 (4-1) iiiiTempVtVV6.200 REFREFiiiiPtPTempVtVtPi006.200144.0 (4-2) Where, all pressures are in psf: voltage for channel @ time itV i t for: 15min Mean value, Max and Min moving average for durations of {selected pressure duration} voltage for channel @ time (15min Mean) pre-storm data tV i V0 i 0t EF, voltage for reference channel (yard, attic or FCMP tower sensor) @ time R t (15min Mean) EF, voltage for reference channel @ time 0t (15min Mean) pre-storm a R V0dat [F], temperature change from between time 0tand t Temp i [ p, obtained from, atm sf/Volt], sensitivity factor for channel sensor calibration test. i includes the reference sensor REF i ospheric pressure @ time REFtP t from Tower Data (15m in Mean) The mean, root mean square (RMS), maximum and minimum peak pressure coefficients are then calculated using Eq. 4-3, where REF, atmospheric pressure @ time 0t, from Tower Data (15 Min Mean) pre-storm data P0 21 is taken as a constant value of 0.00256 whicass density of air for the standard atmosphere i.e., tempeons h reflects the m rature of 59 F and sea level pressure of 29.92 inches of mercury and dimensi

PAGE 80

63 associated with wind speed in mph. The selection of an appropriate velocity valdenominator is the subject of an upcoming section. For all coefficient calculations, the reference pressure and reference velocity are taken as constant for a given pC sample (over a 15ue for the minute period). The dynamic local pressuile ic local pressure are used for the calculation of r re (first term in Eq. 2-1) is an expected value for calculation of the mean pC, whinstantaneous values for dynamms and peak pressure coefficients. 2321Sec itPtC ipV (4-3)Uncertainty: Data Sources can change from Storm to Storm and House to House ary goal of this dissertation is to identify, model and account for the sources ne such b) Camera (yard) Sensor e Tower or airport an also be estimated from more than one source: ) House Anemometer via optimization of roughness () value for selected wind A prim of uncertainty that affect the final estimates of pC from the full-scale house datasets. O source of uncertainty is the source of data that is used to provide the various parameters in Eqs. 4-1 through 4-3. For a given house, the reference pressure can come from more than one source: a) Attic Sensor c) Near FCMP Mobil The reference wind speed c 0z 1 swath (addressed in a later section) 2) House Anemometer via optimization of Peak-Factor ( g ) for selected win d 4) eld model (Vickery, et al, 2000) swath (addressed in a later section) 3) Near FCMP Tower ARAs parametric hurricane wind fi

PAGE 81

64 These different sources introduce varying degrees of uncertainty in the an alysis, and must be accounted for. The outcome of combining the different pressure sources and wind speed sources are twelve possible cases, which are summarized on Table 4-3. Table 4-3 Summary of 12 possible cases to compute full-scale pressure coefficients 1) House Anemometer Wind speed source Optimum z0Optimum Peak-Factor gCase (1a) Ref. pressure from Case (2a) Ref. pressure from 2) House Anemometer 3) Near FCMP Tower4) ARA's Wind Modela) Attic Sensorattic sensor and wind speed from house anemometer optimization of z0attic sensor and wind speed from house anemometer optimization of Peak-FactorCase (3a) Ref. pressure from attic sensor and wind speed from near FCMP towerCase (4a) Ref. pressure from attic sensor and wind speed from ARA modelb) Yard SensorCase (1b) Ref. pressure from camera sensor and wind speed from house anemometer optimization of z0Case (2b) Ref. pressure from camera sensor and wind speed from house anemometer optimization of Peak-FactorCase (3b) Ref. pressure from camera sensor and wind speed from near FCMP towerCase (4b) Ref. pressure from camera sensor and wind speed from ARA modelc) Near FCMP TowerCase (1c) Ref. pressure from near FCMP tower and wind speed from house anemometer optimization of z0Case (2c) Ref. pressure from near FCMP tower and wind speed from house anemometer optimization of Peak-FactorCase (3c) Ref. pressure from near FCMP tower and wind speed from near FCMP towerCase (4c) Ref. pressure from near FCMP tower and wind speed from ARA modelReference pressure source In some cases, more than one source for reference pressure and reference velocity re available. For example a house dataset that includes an attic sensor, a yard sensor, and a rooftop wind anemometer record covers four of the cases in Table 4-3. None of these analyses would produce precisely the same estimate for any and thus the estimate is its. In the next section analysis results using various cases will be presented. Comparison of the output from each case will demonstrate the uncertainty. Example Peak Minimum Calculation with Uncertain Reference Velocity inarIvan (2004) will be presented. For this dataset the reference pressure source is a yard sensor near the subject home and monitored by the same data collection system used with the house sensors. The reference wind speed (expected 3-second gust at mean roof height) is estimated from the house anemometer (using two different methods), the a pinherently a random variable with quantifiable confidence limp C C In this section prelimy analysis results of the FL-27 house during Hurricane

PAGE 82

65 nearby FCMP tower and the ARA wind field model. Thus the analysis of any givenminute record of data will produce four different estimates of pressure coefficient. Fthis example, the peak minimum pressure coefficient is estimated, which corresponds tothe maximum gust-type suction (uplift) that the sensor experienced in that 15-minute timframe, normalized by the expected peak 3 second wind in the that same time frame. Table 4-4 presents the peak minimum pressure coefficients for channel 6 (see Figure 4-4 for a diagram of the roof and sensor locations) using cases 1b, 2b, 3b and 15-or e 4b (fromo ated me. me frame, using different sourc (all bles of amework. For example, the uncertainty in Table 4-3). The data records 131 to 147 (each record is 15 minutes long) correspond to the FL-27 house during hurricane Ivan (2004) when high winds were impacting the home. The direction of the wind is also noted in Table 4-4 from twsources (FCMP tower and ARA model), as well as the estimated 3-second gust estimfrom four methods. The wind is approaching from the southeast during this time fraReferring to the location of channel 6 in Figure 4-4, this portion of the roof is expected to experience strong suction for this range of wind directions. The last four columns in a given row in Table 4-4 show the spread of peak minimum pC estimates for the same sensor, over the same ti es of wind velocity information in the calculation parameters used in Eq. 4-3other varias identical for each calculation). Note that some estimates exceed (are smaller than) the ASCE-7 value in corner zones of -2.6. This preliminary analysis showthe effects of uncertainty in the pressure coefficient results by accounting for just onethe uncertain parameters in the computation. The full uncertainty analysis later in this chapter will account for a number of contributors to uncertainty in a probabilistic fr

PAGE 83

66 refererate channel 6 nce velocity will be changed from the four discrete values used in the current example to a probabilistic representation over a range of values. The same will be done for the other sources of uncertainty, and Monte Carlo simulation will be used to genethousands of estimates of pC in the place of the four estimates per row in Table 4-4. In this manner the resultant histogram of simulated pC values can be used to determine confidence limits to bound the mean value (the estimate). Comparisons with wind tunnel tests and ASCE-7 wind load provisions will be conducted with knowledge of these confidence limits, and will help determine the appropriate level of concern for discrepancies. Table 4-4 Minimum peak pressure coefficients for FL-27 hurricane Ivan (2004) pC 1) House Zo2) House g3) Tower4) ARA1b2b3b4b1319/16/2004 04:52:4853.8861.4058.2962.02108115-2.99-2.30-2.55-2.25 1329/16/2004 05:07:5655.3161.4867.7262.89106117-2.06-1.67-1.37-1.591339/16/2004 05:23:0458.9264.2370.6363.80108120-2.50-2.10-1.74-2.13 1349/16/2004 05:38:1361.5265.1571.9464.86108122-2.47-2.21-1.81-2.231359/16/2004 05:53:2166.1168.7080.2166.02113125-2.71-2.51-1.84-2.711369/16/2004 06:08:2965.7872.9985.0067.18120128-2.46-1.99-1.47-2.351379/16/2004 06:23:3870.9475.1386.9868.24124132-1.80-1.61-1.20-1.951389/16/2004 06:38:4668.6473.2395.8069.07125136-2.10-1.84-1.08-2.071399/16/2004 06:53:5568.3271.4090.2969.66128140-1.72-1.57-0.98-1.651409/16/2004 07:09:0365.8668.2688.4170.08134145-2.38-2.21-1.32-2.101419/16/2004 07:24:1261.5265.8691.3570.30140149-1.73-1.51-0.79-1.331429/16/2004 07:39:2162.7461.8191.0270.27146154-1.54-1.59-0.73-1.231439/16/2004 07:54:2958.4257.3286.4070.00150159-2.88-2.99-1.32-2.001449/16/2004 08:09:3756.0956.1485.1969.45156163-2.63-2.63-1.14-1.721459/16/2004 08:24:4659.0957.9281.9768.64167168-1.35-1.41-0.70-1.001469/16/2004 08:39:5456.9558.5780.0467.58179172-1.62-1.53-0.82-1.151479/16/2004 08:55:0356.4058.8879.8166.30184176-2.64-2.42-1.32-1.91Record with maximum 3-second gust estimatedRecord where the minimum Cp value of -2.6 from the ASCE 7 is exceede Wind Speed 3-Sec Gust (mph)Wind Direction TowerWind Direction ARAMinimum Peak Cp 10 Hz Record #Date & Tim(UTC) Identification of Appropriate Sampling Rate The original 100 Hz data for the FL-27 house were downsample using segmental average to 10 Hz andd in Table 4-4. Downt l it was used for the preliminary analysis presente sampling the original 100Hz data are necessary to expedite processing and filter ouhigh frequency noise. The peak minimum pressure coefficients are affected by the finadownsample rate. A faster sampling rate will produce larger pressure coefficients than

PAGE 84

67 those obtained for slower sampling rates. The influence of sampling rate can be graphed and a suitable rate identified (see Figure 4-6 and 4-7), beyond which there is not significant increase in peaks. For example, downsampling from 100 Hz to 1 Hz may produce a 1 Hz peak pC of -1.5. Using 10Hz sampling may produce a peak of -1increasing to 20 Hz may produce -1.85. A plateau from plots (Figure 4-6 and 4-7) thapresent the minimum peak pressure as function of resample frequency will be identified for maximum suitable sampling rate. Conversely, (peaks) from faster sampling rates (say 100 Hz) tend to be more localized (less correlated spatially), th .8, while t us reducing their influence on the aggregate loads we sek at h priate sampling rate is identified for the FL-27 Ivan dataset using analysis cle 4-4 will be consi ek to quantify. For example, say the full-scale analysis reveals an extreme peaone sensor. If that peak is not correlated with near neighbor sensors, it is reasonable to say that the aggregate uplift loading observed in the 4x8 ft sheathing panel occupied by those sensors is less severe. Such a peak is of limited use. A correlation study among adjacent sensors will evaluate the sampling rate at which correlation has significance witregard to aggregate loading. Study for Identification of Appropriate Sampling Rate In this section an appro ase 1b from Table 4-3. Records 131 147 presented in Tab dered in this study case since the maximum sustained winds were experienced in this time frame. Figure 4-4 presents the roof sensor layout for the FL-27 house, located inGulf Breeze, FL. This study considers individual sensor and pairs among the sensors (56, 7, 22, 23 and 24).

PAGE 85

68 Influence of Sampling Rate on Peak pC value Plots of the mini mum peak pressure coefficient are generated as function of the amhe original 100 Hz data. The data were filtere 4-e same analywth at ata, s. pressusent resample rate through a series of downsple of t d using the decimate function in MATLAB. The decimation process filters the input data with a lowpass filter and then resamples the resulting smoothed signal at a lower rate (by default, decimate employs an eighth-order lowpass Chebyshev Type I filter). After downsampling the data at various rates a plot is generated (Figure 4-6 and7) of the minimum peak pressure coefficient vs. the sampling frequency. Typical plots are presented using record #131 (mean wind direction of 111) and record #147 (mean wind direction of 180) for channel 6 in Figure 4-6. Th sis is done for channel 23 in Figure 4-7. These figures show a plateau in peak minimum pressure coefficient is reached at 10 ~ 20 Hz before a continued slow grohigher frequencies. It is desired to select a downsample rate of the original 100 Hz dto a sampling rate that is lower than half the original rate, in order to filter out high frequency noise. Thus a downsample rate is set to 20 Hz. Additional analysis related to the downsampling rate based on the peak correlation is presented in the next sectionThe wind tunnel work of Theodore Stathopoulos had an equivalent full-scale sampling rate of 2 Hz, so it is worth comparing the difference obtained in the minimum re coefficient from the 20 Hz vs. 2 Hz full-scale dataset. Table 4-5 and 4-6 prethe comparison for records 131 and 147, for channel 6 and 23 respectively, showing the difference % between the minimum pC values having a maximum difference of 44% anda minimum difference of 16%.

PAGE 86

69 0 10 20 30 40 50 60 70 80 90 100 -3.4 -3.2 -3 -2.8 -2.6 -2.4 -2.2 -2 -1.8Peak Minimum Cp (Case 1b)Resample frequency Record #131r FCMP House FL-27 Ivan 2004 Channel 6 fo (a) 0 10 20 30 40 50 60 70 80 90 100 -3.4 -3.2 -3 -2.8 -2.6 -2.4 -2.2 -2 -1.8 -1.6Peak Minimum Cp (Case 1b)Resample frequencyRecord #147 for FCMP House FL-27 Ivan 2004 Channel 6 (b) Figure 4-6 Channel 6 peak minimum vs. resample frequency. (a) For record #131 wind direction is 111 (b) For record #147 wind direction is 180 pC

PAGE 87

70 0 10 20 30 40 50 60 70 80 90 100 -1.3 -1.2 -1.1 -1 -0.9 -0.8 -0.7 -0.6 -0.5 -0.4 -0.3Record #131CMP House FL-27 Ivan 2004 Channel 23 Peak Minimum Cp (Case 1b)Resample frequency for F (a) 0 10 20 30 40 50 60 70 80 90 100 -2.2 -2 -1.8 -1.6 -1.4 -1.2 -1 -0.8Peak Minimum Cp (Case 1b)Resample frequencyRecord #147 for FCMP House FL-27 Ivan 2004 Channel 23 (b) Figure 4-7 Channel 23 peak minimum vs. resample frequency. (a) For record #131 wind direction is 111 (b) For record #147 wind direction is 180 pC

PAGE 88

71 Giverough 4-6, there t can be of FL-27 n the data provided in Figures 4-6 and 4-7, as well as Tables 4-4 th is the potential that the current ASCE-7 loads for components and cladding, based upon wind tunnel data with an equivalent sampling rate 2 Hz, do not reflect the conservative worst case from extreme winds. This possibility is strengthened if ishown that strong spatial correlation exists among peaks sampled at 20 Hz. Table 4-5 Difference % between 20 Hz and 2 Hz minimum pC for channel 6 13114720-3.13-2.862-2.06-2.0Difference %-41%-34% 3Resample rate (Hz) Record #Minimum Cp Channel 6 Table 4-6 Difference % between 20 Hz and 2 Hz minimum for channel 23 of FL-27 C p 13114720-0.66-1.642-0.42-1.3Difference %-44%-16% 9Resample rate (Hz) Minimum Cp Channel 23Record # Influence of Sampling Rate on Spatial Correlation of Peaks patial correlation of peaksand This section focuses on the influence of sampling rate on s The analysis will be developed using the same house dataset presented in the previous section. Two standard methods of correlation measurement are considered, an additional method is developed to specifically focus on the correlation of peaks. Correlation coefficient yx, The correlation coeie fficnt is a normalized measure of the strength of the linear relatio nship between two random variables x and y (any 2 sensors). The correlation coefficientyx, is defined in equation 4-4 as the ratio of the covariance of two rando m

PAGE 89

72 variables x y to the product of the standard deviations. Uncorrelated data results in correlation coefficient close to 0; strongly correlated data have a correlation coefficient near 1. anda yxyxyxCOV,, (4-4)Where is the covariance defined in Eq. 4-5, yxCOV, yx and represents the value of the random mean variable x and y respectively and E the expected valof the term inside the brackets. Intuitively, covariance is the sure of how much two variables vary together. represent ue mea yxyxyxECOV (4-5)Figure 4-8 shows the correlation coefficient between adjacent sensors channels& 7) ation at of the correlation coefficient is that it presents a simple scalar displa & 7) indicate whether peaks between sensors are correlated. (6 and non-adjacent sensor channels (6 & 23) which are located in opposite roof corners, as a function of resample rate. For adjacent sensors (approximately 2.5 feetapart) it shows a very strong correlation and weak correlation for non-adjacent sensor(approximately 60 feet apart). The correlation drops significantly with distance, as expected. In both plots the correlation also drops as the resample rate increased, indicating both the presence of high frequency noise and the shorter spatial correlhigher frequencies. The limitation y of correlation, containing information across all frequencies in the analyzedsignals (any 2 sensors). Thus the high observed correlation from adjacent sensors (6 may be due mostly to low frequency (non-peak) correlation. That is, it cannot specifically

PAGE 90

73 0 5 10 15 20 25 30 35 40 45 50 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Correlation CoefficientResample frequency (Hz)FCMP House FL-27 Ivan 2004 for record #131 Adjacent Sensors 6 & 7 Non-Adjacent Sensors 6 & 23 Figure 4-8 Correlation coefficient for record #131 of FL-27 hurricane Ivan (2004) Coherence function fCyx2, The coherence function fCyx2, is the ratio of the square of the magnitude of the the product of the autospectral density functions of the two r cross-spectrum density function to andom variables tx and ty e quation 4-6. For all f, the quantity fC2 satisfi102. The advantage of the coherence function is that it shows the correlation yx ,es as a function of frequency. ,fCyx fSfSfSfCyyxxyxyx 2 ,2, (4-6)Figure 4-9 shows the coherence function for adjacent sensors (6 & 7) and non-adjacent sensors (6 & 23). For adjacent sensors it is clearly shown a strong correlation at

PAGE 91

74 lo w frequency, not observed for non-adjacent sensors which are located at opposite roof corners. Such behavior was expected due to the relative location of the sensors. 0 5 10 15 20 25 30 35 40 45 50 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Coherence function estimateFrequency (Hz) FCMP House FL-27 Ivan 2004 for record #131 Adjacent Sensors 6 & 7 Non-Adjacent Sensors 6 & 23 Figure 4-9 Coherence function estimate for record #131 of FL-27 hurricane Ivan (2004) The coherence function still does not isolate peaks however, since even high frequency content is an average over entire signal. That is, the low correlation at higher frequencies does not preclude well correlated peaks. rs share peaks at the same time instant (correlation), while ignoring the low frequency signal content and low magnitude Peak-score method A method that focuses on identifying the correlation among peaks in pairs of random variables was developed for this analysis. The peak-score method is intended to qualitatively identify the degree to which two senso

PAGE 92

75 high frequency content t hat mimic the meaning of correlation coefficient and coherence functits re 4-10 (a) and (b), shows typical signals x and y for close sensors and its norm as shown in Figure 4-11. If the original signals x and y experienced values at the same instant, should have a number of peak values that approach -1.0. This can be qualitatively characterized by comparing the probability contents of and A threshold minimum value for is found such that 5% of the probability of is lower than this value. This is called the threshold value Similarly the 5% threshold for is found, .Figure 4-12 shows graphically the estim on. In the following discussion, the random variables x and y can be thought of as the time histories measured at two sensors, either close or well separated. The analysis starby normalizing each of the two time traces x and y such that the minimum value in each is -1.0 and the mean value is 0. These two normalized signals arenx and ny. Figu alized version respectively. Note that peak negative values can be observed in both signals at the same times. A new signal w is created as the average of x and y, or n n n 2/)(nnnyxwpeak minimum nnx w nx, ny nw x nx TH. ny yTH ation of 1666.0xTH and 1589.0 yTH. An average threshold avgTH is thenxTH and yTH. The probability onw that is below this average threshold of 1628.0avgTH is then found. The peak score is the ratio of throwith the 5% used to established the thresholds for nx and ny. Figure 4-13 shows the established frbability om f is p

PAGE 93

76 value obtained frthe TH, so the peak-score for this example has a value o om f avg 948.0 %0.5%74.4 0 1 2 3 4 5 6 7 8 9x 104 -4 -3 -2 -1 0 1 Signal y 0 1 2 3 4 5 6 7 8 9x 104 -4 -3 -2 -1 0 1 2 Original Signals x and y Signal x (a) 0 1 2 3 4 5 6 7 8 9x 104 -1 -0.5 0 0.5 Signal xnNormalized Signals xn and yn 0 1 2 3 4 5 6 7 8 9x 104 -1 -0.5 0 0.5 Signal yn (b) Figure 4-10 Random data signals. (a) Original signals x an y, (b) Normalized signals and nx ny

PAGE 94

77 0 1 2 3 4 5 6 7 8 9x 104 -2 -1.5 -1 -0.5 0 0.5 1 Signal (xn + yn)Computing new normalized signal wn 0 1 2 3 4 5 6 7 8 9 x 104 -1 -0.5 0 0.5 Signal wn = (xn + yn)/2 Figure 4-11 Resultant signal 2nnnyxw The range of possible peak-score outcomes is from 0 to ~1. If the peaks in the signals and are not correlated, the probability density function (PDF) of will not contain as much probability in the left tail as and and the peak score will not approach 1.0. Conversely, if peaks in x and y regularly occur at the same instant, the peak minimum values in will be of the same magnitude and frequency as those in and and the peak score will approach one. The peak score method is first evaluated using artificial data where the level of correlation is known. Random signals and nx ny nw nx ny nw nx ny x y are digitally created with similar spectral and probability characteristics as those that describe a typical time history of pressure from the full-scale data. These artificial si and gnals x y are uncorrelated. A third signal

PAGE 95

78 w is defined as function of x and y so that partial correlation exists (e.g.,yxyw5.00). x 5., -1 -0.5 0 0.5 0 0.05 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Empirical CDF for xn X: -0.1666Y: 0.05007 x nF(xn) -1 -0.5 0 0.5 0 0. 05 0. 1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Emcal CDF for yn X: -0.1589Y: 0.0501 ynF(yn) piri e 4-12 Empirical cumtive distributions function for and Figurula nx ny -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 wnF(wn)Empirical CDF for wn X: -0.1628Y: 0.04739 Figure 4-13 Empirical cumulative distributions function for nw

PAGE 96

79 The peak-score analysis is then performed using uncorrelated data (x and y), partially correlated data (x and w), and perfectly correlated data (x and x). The result from uncorrelated data and x y is presented in Figure 4-14 for various percentage threshold. A 1% threshold (as suggested above) appears suitable, as uncorrelated data provides a peak score of ~ 0.1 as the resample frequency decrease. Figures 4-15 shows the resulting peak scores for perfectly correlated data (peak score hovers around ~1.0), and Figure 4-16 shows partially correlated data (peak score ~0.8). 0.8 0.5 0.6 0.7 0 5 10 15 20 25 30 35 40 45 50 0 0.1 0.2 0.3 0.4 0.9 1-S)"Peak-Score" check for uncorrelated data x & y "Peakcore" Resample frequency (Hz 1% 2.5% 5% Figure 4-14 Peak-Score check for uncorrelated data and x y The peak score method is qualitative at this time. In the application to full-scale pressure uplift data, we seek to demonstrate that the peaks observed between closely spaced sensors are well correlated.

PAGE 97

80 0 5 10 15 20 25 30 35 40 45 50 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1"Peak-Score" Resample frequency (Hz)"Peak-Score" check for full-correlated data x & x 1% 2.5% 5% Figure 4-15 Peak-Score check for full-correlated datax and x 0 5 10 15 20 25 30 35 40 45 50 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1"Peak-Score" check for partial correlated data x & w "Peak-Score" Resample frequency (Hz) 1% 2.5% 5% Figure 4-16 Peak-Score check for partial correlated data x and w

PAGE 98

81 Peak-score method: application to full-scale data The peak-score is now computed for the full-scale data for a range of resample rates beginning with the original 100 Hz signal and for lower resample rates. This analysis was conducted extensively for the highest wind speed data records for adjacent and non-adjacent sensors, where some of the typical findings are presented in Figures 4-17 thru 4-19, these figures shows the peak-score vs. the resample frequency for sensor pairs of channels (6 & 7), (6 & 5) and (6 & 23) respectively. Figure 4-17 and 4-18 correspond to sensors close to each other (approximately 2.5 feet) and as expected the peak-score suggest that the peak are well correlated and are occurring at the same time. This is not the case for sensors spaced far apart (approximately 60 feet), as shown in Figure 4-19, where the peak-score is significantly lower. 0 10 20 30 40 50 60 70 80 90 100 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1"Peak-Score" for channels 6 & 7Resample frequency (Hz)"Peak-Score" FCMP House FL-27 Ivan 2004 for record #131 Threshold = 1% Figure 4-17 Peak-Score for channel 6 & 7 in record #131 FL-27 h urricane Ivan (2004)

PAGE 99

82 0 10 20 30 40 50 60 70 80 90 100 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1"Porh&Resample frequency (Hz)r record #131 "Peak-Score" FCMP House FL-27 Ivan 2004 fo Threshold = 1% eak-Sce" for cannels 6 5 Figure 4-18 Peak-Score for channel 6 & 5 in record #131 FL-27 hurricane Ivan (2004) 0 10 20 30 40 50 60 70 80 90 100 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1"Peak-Score" for channels 6 & 23Resample frequency (Hz)"Peak-Score" FCMP House FL-27 Ivan 2004 for record #132 Threshold = 1% Figure 4-19 Peak-Score for channel 6 & 23 in record #132 FL-27 hurricane Ivan (2004)

PAGE 100

83 Based on this correlation analysis and the study of peak minimum value as a function of resample rate, the full-scale data will be analyzed using a down-sampled rate of 20 Hz. This represents the rate at which peak minimum values plateau, and also maintains strong correlation of peaks between closely spaced sensors. Uncertainty of Reference Velocity: Estimating Peak Wind Speed Gust It was demonstrated in Table 4-4 and the associated example that the identification of a reference velocity for the denominator in Eq. 4-3 can have a significant impact upon the resultant estimated peak minimum In that example, four methods for obtaining the reference velocity were cited. This section explains these wind speed reference estimation methods. relatively straight forward. The velocity is measured either at the mean roof height of the model, or at the top of the tunnel and the speed extrapolated to the mean roof height. This is not the case for the full-scale datasets, where the wind speed reference value is estimated from either an anemometer on the roof (two methods), a nearby FCMP tower, or from wind field models. For any of these cases the associated uncertainty is larger than that in a wind tunnel, and must be quantified. The best reference to obtain the wind speed velocity is when a house anemometer is present, but unfortunately an operational anemometer on the house is not certain. The two houses used in this section (FL-27 and FL-30 Ivan 2004) had working anemometers during the storm events of hurricane Ivan (2004). The reference wind speed in the denominator of Eq. 4-3 should represent the pC pC pC In wind tunnel experiments the determination of a reference wind speed is expected (average) peak 3-second value. When a house anemometer is available, we

PAGE 101

84 canno The first mcal equat t simply use the peak value observed over the analysis time frame of interest. Thiswould be a single sample peak value, not an expected value. The expected value can be estimated based on the mean and standard deviation of the speed record in two ways. Wind Speed Reference from House Anemometer: Optimizing Roughness ethod identifies a roughness exposure value 0z and uses the classi ion to relate wind duration and exposure as given in Eq. 4-7 (Simiu and Scanlan) 9003ln5.2zzWhere z is the mean roof height, 0211tczUzU(4-7) zU 900seconds), and )(tc an is the mean wind over 15 minutes (900 d are from tables provided in Simiu and Scanlan. is the desire for between the maximum 3-second gust actually measured at the house anemometer and the value estimated from Equation 4-7 was calculated over many of 15-minute records. The mean error was minimized by adjusting inute records need to contain wind coming from similar directions, as is a function of approach terrain and therefore wind d the d zU3 d 3 second gust. The difficulty is that the estimation of an appropriate valueroughness 0z is not well established, and this leaves much ambiguity in the application ofEq. 4-7. This issue was solved by selecting a best 0z roughness value through optimization. Although we cannot directly assign the actual observed 3 second gust as the wind speed reference, we can use this data to help identify an appropriate z value. The error 0 0z These 15-m z 0 direction. In order to complete this task a MATLAB code was developed to finoptimum 0z for data records corresponding to a wind swath of 30 each. The optimize

PAGE 102

85 0z value was then applied in Eq. 4-7 to estimate the expected 3 second gust wind speed reference for any given 15 minute record. Wind Speed Reference from House Anemometer: Optimizing Peak Factor The second method to estimate the 3-second gust from the house anemom eter pt elocity referenced to mean roof height by adding the mean to some factor of the ada s the methodology commonly used in wind tunnel analysis. The studies on genericlow buildings at the University of Western Ontario by Kopp (2005) estimate the peak v hVstandard deviation VgVVh where V is the mean roof height mean velocity, V is the root mean square (standard deviation) of the velocity fluctuations and g is a peak factor. In the wind tunnel, the peak value is taken as a nominal value of 3.0. Following this approach, the mean and standard deviation of the house anemometer are measured for each 15 minute record, and the expected peak value is estimated by mean to some factor of the standard deviation. The value 3.0 cannot be applie adding the d here necessarily, as the assigned value implies a specific terrain (roughness) and test duration used in the wind tunnel. These do not apply to the full-scale data, so an appropriate value of g must be identified via optimization. A MATLAB code was developed to find the appropriate peak factor g that will produce the least mean square error between the actual field measurements and estimated 3-second gust. This task has been done for the same wind direction swaths used to optimize 0z. Aerial pictures had been use to present and compare the optimum 0z and peak factor g values obtained in Figure 4-20 for the FL-27 house and Figure 4-21 for the FL-30 house. For the FL-27 house Table 4-5 present the data record range, the wind swath,

PAGE 103

86 the represent the range of data collected during Table04) optimum 0z and peak factor values. The same information is presented for the FL-30house in Table 4-6. The selected 30 zones Ivan at those locations. 4-5 Classification of optimum 0z and peak factor for the FL-27 house Ivan (20 Wind swathHouse records rangeEstimated z0 Peak Factor60 9042 1201.483.4390 120121 1331.033.33150 180142 1470.593.25180 210148 1600.603.21 120 150134 1410.793.49210 240161 1900.583.16 Wind Swath (60 90)0z= 1.44 & g= 3.43 Wind Swath (90 120) .33 0z= 1.03 & g= 3 Wind Swath 240) Wind Swath (120 150) = 0.79 &= 3.49 (210 0z= 0.58 & g= 3.16 0z g Wind Swath (180 210) Wind Swath (150 180) 0z= 0.60 & g= 3.21 0z= 0.59 & g= 3.25 Figure 4-20 Aerial picture for the FL-27 house showing optimized zand peak factor values for hurricane Ivan (2004) 0

PAGE 104

87 Table 4-6 Classification of optimum 0z and peak factor for the FL-30 house Ivan (2004) Wind swathrangez0 Factor9120 House records Estimated Peak 60 9080 1011.143.440 120102 1210.823.42 150122 1260.273.21210 240138 1770.693.47 150 180127 1300.363.52180 210131 1370.303.13 Wind Swath (60 90) = 1.14 & 0z g Figure 4-21 Aerial picture for the FL-30 house showing optimized 0z and peak factor = 3.44 Wind Swath (90 120) = 0.82 & 0z g = 3.42 Win2= 0.27 & values for hurricane Ivan (2004) The information in Figures 4-21 and 4-22 as well as Tables 4-5 and 4-6 qualitatively follow the behavior that would be expected. The roughness value increases as the approach wind flows over longer stretches of land and dense tree cover, valuable toe 0z and decreases when the fetch between the house and open water shortens. It is also note the relative difference in 0z between the two houses over the sam d Swath (10 150) 0z g = 3.21 Wind Swath (150 180) = 0.36 & 0z g = 3.52 Wind Swath (180 210) = 0.30 & 0z g = 3.13 Wind Swath (210 240) = 0.69 & 0z g = 3.47

PAGE 105

88 direction swaths. For example, consider the 0z value for the direction range 180 210 in Figures 4-20 and 4-21. The aerial view clearly reveals that the FL-30 house has a fetch to the water that is far less dense with trees comared to FL-27 house. This is reflected in their relative values for The estimation of roughness lengthis very important because this value will also be used in a later section to estimate the house wind speed from a near FCMP mobile tower (wind speed source case 3). In cases were the house anemometer is not available the tower data must be converted to reflect the terrain exposure of the house. This involves Eq. 4-7 and requires an estimate of om aerial picture inspThe fourth case is the estimation of the reference wind speed by adjusting the data provided by wind field maps provided by Applied Research Associates (Vickery 2000). The maps consist of contours of sustained and peak wind speeds for a giv(usually 15 minutes). These are by default standardized to a roughness exposure of open terrain.nd height of 10 m, and are converted to the local house terrain and mean roof height for our analysis purposes. p 0z. the 0z 0z fr ection. en time frame 300zm a 0

PAGE 106

89 Table 4-7 Minimum peak pressure coefficien for FL-30 hurricane Ivan (2004) channel ts15, reference wind speed variation 1) House Zo2) House g3) Tower4) ARA1b2b3b4b 1169/16/2004 04:20:4151.6552.4058.3165.90101103-2.11-2.06-1.66-1.31179/16/2004 04:35:5051.4653.4856.8367.92110105-2.13-1.98-1.75-1.21199/16/2004 05:06:0767.811209/16/2004 05:21:1574.10Record #Date & Time (UTC) Wind Speed 3-Sec Gust (mph)Wind Direction TowerWind Direction ARAMinimum Peak Cp 10 Hz 031189/16/2004 04:50:5861.3262.8960.3269.86107108-1.51-1.44-1.56-1.1668.3266.1371.76107111-1.26-1.24-1.32-1.1273.6971.9073.59107114-1.18-1.20-1.26-1.201219/16/2004 05:36:2476.1775.6072.5075.25108117-1.04-1.05-1.15-1.061229/16/2004 05:51:3285.2679.2880.7476.80113121-1.82-2.11-2.03-2.241231241259/16/2004 06:36:59100.0085.3098.0879.47125139-1.37-1.88-1.42-2.161299/16/2004 07:37:3380.0773.4391.9577.90145167-1.57-1.87-1.19-1.666Record with maximum 3-second gust estimated 9/16/2004 06:06:4196.0283.0987.2578.39119126-1.12-1.50-1.36-1.699/16/2004 06:21:5095.8783.4888.4779.37124132-1.65-2.18-1.94-2.411269/16/2004 06:52:07101.4086.3792.7479.38128146-1.99-2.75-2.38-3.261279/16/2004 07:07:1687.6676.4089.9779.16134153-1.53-2.02-1.46-1.881289/16/2004 07:22:2581.7775.2094.0578.68139160-1.60-1.89-1.21-1.731309/16/2004 07:52:4273.1766.8488.1976.87150174-1.61-1.93-1.11-1.4Record where the minimum Cp value of -2.6 from the ASCE 7 is exceed um peak ning anemometer and an FCMP tower in FCMP house datasets do not have this luxury, missing either an anemometer on the house, or no FCMP tower near enough to extrapolate its winds. In these cases the wind speed reference uncertainty still exists, but is not easily defined. For these cases the results Table 4-2 and 4-7 will be used as a guide. Uncertainty on Instantaneous Roof Dynamic Pressure When comparing the full-scale instantaneous roof dynamic pressure vs. the wind tunnel data it is important to clarify that there is no control over full-scale parameters Similar to Table 4-4 which presented the range of minimum peak pressure coefficient using the four reference wind speed cases, Table 4-7 present the minim pressure coefficients for channel 15 (refer to Figure 4-5 for sensor layout of the FL-30 house) computed for the FL-30 using the four estimates of reference wind speed. A large range of values can be observed among the four wind speed estimates and pC values for any given row. The level of uncertainty in the wind speed reference velocity is fairly well definedfor the house datasets that have both a functio the near vicinity, as is the case in Table 4-4 and 4-7. However, many of the existing

PAGE 107

90 such ind re ials of a the same wind speed. The peak observed pressu isms ed e full-scale data. This will require dividing each 1imating the minim as temperature, wind speed and wind direction. This is not the case for the wtunnel experiment where the laboratory environment is controlled and repetitive tests aperformed for constant wind speed and wind direction, providing enough data to produce an expected peak pressure value, rather that a single peak observation. For example, a typical analysis in the Clemson wind tunnel facility consists of 16 independent trgiven house from the same direction and at re coefficient from each trial are then averaged together to produce an expected (mean) peak value. For the full-scale dataset the limiting issue is the time frame in which the winddirection and mean speed can be assumed to be relatively constant. The flow mechanthat produce uplift loads on the roof change with wind direction, thus averaging peak values from an approach angle of 110 with those from 180 is not viable. An expectpeak pressure coefficient will be estimated from th 5 minute segment into 3 segments of 5 minutes each and est um peak pressure coefficient for each segment. The expected peak is obtained from the average of these three single measurements. The uncertainty that remains on this measurement is considerable. The sensor measurements are affected by the temperature changes (quantifiable but uncertain), and in some cases the individual sensor sensitivity i is not available, and a assumed value is assign based on an average of other sensors. Table 4-8 present the analysis performed fthe record range from 131 147 on the FL-27 house to obtain the minimum peak pressure coefficient for a 20 Hz sampling rate, where the average percent difference or

PAGE 108

91 between the expected value from the three 5 minutes segments and the single peak in th15 minutes record ha e s a value of 22%. sis. n ore Tablee Strictly speaking, the pressure coefficients in Table 4-8 are described as the expected peak value of minimum pressure over a five minute duration, sampled at 20 Hz. If the stationarity of the data permitted the use of longer segments of data to estimate asingle peak sample (say 15 minutes rather than 5), the peak coefficients would increase. That conclusion is a straight forward application of the rules of random variable analyThe expected peak value of a stationary signal increases as the duration of the expectatioperiod increases. Simply put, the more data over which one is permitted to look, the mlikely he is to observe a larger peak (on average). 4-8 Expected minimum peak pC (20 Hz) for channel 6 for the FL-27 hurricanIvan (2004) 1) House z01319/16/2004 04:52:4853.88108115-2.41-1.60-2.29-2.10-2.4114%1329/16/2004 05:07:5655.31106117-0-0.93-1.45-1.03-1.4534%1349/16/2004 05:38:1361.52108122-1.74-1.16-2.01-1.64-2.0120%1359/16/2004 05:53:2166.11113125-2.00-1.26-1.07-1.44-2.0032%1379/16/2004 06:23:3870.94124132-1.46-1.86-2.52-1.94-2.521389/16/2004 06:38:4668.64125136-2.28-2.26-1.84-2.13-2.281409/16/2004 07:09:0365.86134145-1.63-2.22-1.93-1.93-2.221419/16/2004 07:24:1261.52140149-2.25-1.74-1.31-1.77-2.251439/16/2004 07:54:2958.42150159-1.75-2.22-2.51-2.16-2.5115%1449/16/2004 08:09:3756.09156163-2.13-2.06-1.60-1.93-2.1310%1469/16/2004 08:39:5456.9517911479/16/2004 08:55:0356.401841 .701339/16/2004 05:23:0458.92108120-1.08-1.04-2.06-1.39-2.0639%1369/16/2004 06:08:2965.78120128-3.37-1.19-1.92-2.16-3.3744%26%7%1399/16/2004 06:53:5568.32128140-1.44-1.89-1.92-1.75-1.929%14%24%1429/16/2004 07:39:2162.74146154-1.67-3.03-2.56-2.42-3.0322%1459/16/2004 08:24:4659.09167168-2.67-1.58-3.14-2.46-3.1424%72-1.21-2.14-2.04-1.80-2.1417%76-2.19-1.96-2.59-2.24-2.5914% A vera g e =22%Single Min 3-Sec Gust (UTC) Direction Direction Difference Segment 1Segment 2Segment 3Expected Record # PeakWind Speed Date & Time Wind TowerWind ARAMinimum Peak Cp for a 20 Hz Duration%Value Record with maximum 3-second gust estimatedRecord where the minimum Cp value of -2.6 from the ASCE 7 is exceed Uncertainty of Reference Pressure inside the house attic, the second is by using the measurement from the sensor located in The reference pressure (second term in numerator in Eq. 2-1) can be established from three different sources. The first is by using the measurement from a sensor located

PAGE 109

92 the camera mounting plate (referred to as a yard sensor) and the third is by using the barometric pressure measurement from a nearby FCMP mobile tower. Pressure coefficient results are presented using the three different reference pressures in order to Each of the three right-most columns uses the house anemometer optimization method for wind speed reference (column 3), and a different source for reference pressure (see Table 4-3). It is clear that the reference pressure from the FCMP tower includes a higher level of uncertainty. The barometric pressure measurement is highly influenced by the location of the instrument relative to the eye of the storm. Thus the barometric pressure at a tower may not be a good representation of barometric pressure at an instrumented house. Table 4-9 Minimum peak pressure coefficients for FL-30 hurricane Ivan (2004) channel 15, reference pressure variation establish which sources are more reliable. Table 4-9 present results of the FL-30 house. 0z 1) House Zo2) House g3) Tower4) ARA1a1b1c116 9/16/2004 04:20:4151.6552.4058.3165.90101103-2.20-2.11-4.69/16/2004 04:35:5051.4653.4856.8367.92110105-2.18-2.13-5.21199/16/2004 05:06:0767.8168.3266.1371.76107111-1.27-1.26-3.001209/16/2004 05:21:1574.1073.6971.9073.59107114-1.20-1.18-2.621229/16/2004 05:51:3285.2679.2880.7476.80113121-1.81-1.82-3.431239/16/2004 06:06:4196.0283.0987.2578.39119126-1.10-1.12-2.471259/16/2004 06:36:59100.0085.3098.0879.47125139-1.35-1.371269/16/2004 06:52:07101.4086.3792.7479.38128146-1.98-1.991271289/16/2004 07:22:2581.7775.2094.0578.68139160-1.61-1.601299/16/2004 07:37:3380.0773.4391.9577.90145167-1.60-1.571309/16/2004 07:52:4273.1766.8488.1976.87150174-1.65-Record with maximum 3-second gust estimatedRecord where the minimum Cp value of -2.6 from the ASCE 7 is exceed 511711189/16/2004 04:50:5861.3262.8960.3269.86107108-1.52-1.51-3.761219/16/2004 05:36:2476.1775.6072.5075.25108117-1.05-1.04-2.421249/16/2004 06:21:5095.8783.4888.4779.37124132-1.63-1.65-3.04-2.71-3.299/16/2004 07:07:1687.6676.4089.9779.16134153-1.53-1.53-3.00-3.18-2.921.61-2.85Wind ion ARAMinimum Peak Cp 10 HzRecord #Date & Time (UTC) DirectWind Speed 3-Sec Gust (mph)Wind Direction Tower Combined Effects of Uncertainties All the identified sources of uncertainty need to be quantified in order to produce full-scale data results with an appropriate level of confidence. A given dataset is

PAGE 110

93 influenced by uncertainties in: dynamic local pressure sensor sensitivity, temperature differential between preand during-storm conditions, appropriate reference velocity, andbarometric pressure sensor. Quantification of Uncertainties on : Monte Carlo Approach combined uncertainties within the estimation of peak pressure coefficients. The method is applied to individual 15-minute lengths of data for all sensors on a given house. A series of values are generated by changing the parameters used in its calculation (Eq. 2-1 & 4-1 through 4-3). Those parameters are the change in temperature ( pC Monte Carlo simulation is employed as a means to directly incorporate the above pC T ), calibration factor for the sensor i ( i ) and estimated 3-second wind gust (). For each one of these parameters a Gaussian probability distribution is assign based on studies of uncertainty on the individual parameters. A Monte Carlo simulation consists of rnning the same peak minimum pressure coefficient analysis for thef data many hundreds of times GustWindSecV3 u same sensor and 15-minute segment o Each analysis uses a set of variable parameters (calibration factori temperature change T and reference wind speed 3-sec wind gustGustWindSecV3) randomly assignvalues based on their assigned distribution. Rather than a single value estimate, C is ed any samples of a random variable.zed to provide the most likely value (mean) and confidence limwithin, for example, 95% of the probability of the random variable). This will p generated as m This rando m variable is then analyits (the range of values for pC pprovide a probabilistic view of the full-scale analysis results to compare with the values C

PAGE 111

94 obtained from the wind tunnel model developed at the Wind Load Test Facility in Clemson University. Lets consider record i. The random variables generated f of this record will have a normal distribution. In the case of the change in temperature, the distribution will have a mean value of the difference be or analysistween the 15 minute average at time and with ard 0t it standard deviation value obtained from the temperature measurement of the recoi. Recall that 0t represents the pre-storm record, see Eq. 4-2. Similarly for the sensitivity variable i of the sensors with missing information a random variable isenerated using the mean value of 32.274 [psf/Volt] and standard deviation of 0.381, based on the statistics from calibrated sensors. This will have a strong impact in the pC comptionsince the sensitivity data for the specific yard and attic sensors (reference pressure sensors) for a given application is not available in most of the house datasets. Finally t reference wind speed will be simulated with a norm distribution with a mean and guta healstandard deviation values obtained from the four cases previously presented to estimate the reference wind speed. An example Monte Carlo simulation will be presented for channel 6 in the FL-27 house for record #135 from hurricane Ivan (2004). Table 4-10 provides the mean and standard deviation used to generate the random variables, where these values are obtained from the corresponding data record. For example the Temp mean and standard deviation values are obtained from the FCMP mo1 temperature measurement corresponding to the time frame of record #135. The mtower was located 11.5 miles NW from the FL-27 house. The mean and sta for the calibration bile tower Tobile ndard deviation values factor 0 assig ned to the yard sensor are obtained from the calibration records performed

PAGE 112

95 on 200 sensors. Finally the mean and standard deviation value corresponding to thereference velocity are obtained the four 3-second gust estimations corresponding to the consider house data record. Table 4-10 Mean and standard deviation of random variables for FL-27 record #135 hurricane Ivan VariableMeanStd Temp-4.7860.44634.2740.381Ref Velocity70.261Record #135 for FL-27 channel 6 06.749 Numit state e.simulation. A suitable number of simulations are identified by generating 100 separate trials, ulations. Each of these 100 trials produces a single mean value of peak minimum measured from the N simulations. If N is a suitable number, then the mean outcomcantly from one trial of N simulations to the next. This is quantified by the coefficient of variation among these 100 trials. Then 100 trials are run for each of a vector of different N values (10, 25, 50, 75, 100, 250, 500, ber of Monte Carlo Simulations It is important to establish an adequate number of Monte Carlo Simulations to generate the pC random variable. This number is dependent upon the limit state equation (Eq. 4-1 through 4-3), the number of random variables on the right side of the limequation, and the distributions assigned to these variables. Typically, the more random variables and larger uncertainties (larger standard deviations), the more simulations are required to generate a statistically meaningful outcom Often the appropriate number is itself determined through Monte Carlo each trial consisting of N sim pCe should not vary signifi

PAGE 113

96 750 and 1000). The coefficient of variati on is computed from the 100 mean values of the gener conservative value of N = 1000 ated peak minimum pC. Figure 4-22 presents this coefficient of variation in the expected mean as the number of simulations N increases. The result suggests that very little additional reliability is gained by using N greater than 250. However a will be used since the computer power is not a significant limiting factor. 0 100 200 300 400 500 600 700 800 900 1000 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 C ofn oefficient variatioNumber of Simulations (N) Figure 4-22 Coefficient of variation, on 100 trials vs. number of simulations N An Example Monte Carlo Simulation Monte Carlo Simulation is now executed for channel 6 on FL-27 from hurricane Ivan, with the parameters previously presented on Table 4-10 for record #135. Figure 4-23 presents the resulting normalized histogram of the peak minimum pressure coefficient using a downsample rate of 20 Hz and 1000 simulations. The empirical cumulative distribution function (ECDF) is then determined to identify the lower and upper 2.5% limits for a 95% confidence interval. Figure 4-24

PAGE 114

97 present the cumulative distribution function. The range of value corresponding to above 2.5% and less than 97.5% are -4.19 and -2.01 respectively. These are the 95% confidence limits for the expected (mean) peak value of -2.89. Based on the current knowledge of uncertainties, the most likely value is -2.89, with 95% confidence that the peak minimum is within [-4.19 & -2.01]. pC pC -6.5 -6 -5.5 -5 -4.5 -4 -3.5 -3 -2.5 -2 -1.5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Min CpFyHistogram of min Cp using Monte Carlo Simulation of 1000Record #135 channel 6 FL-27 Ivan (2004) requenc Figure 4-23 Histogram of minimum (20 Hz duration) from Monte Carlo simulation pC for channel 6 on record #135 FL-27 hurricane Ivan (2004), N = 1000

PAGE 115

98 -6.5 -6 -5.5 -5 -4.5 -4 -3.5 -3 -2.5 -2 -1.5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 CpMinF(CpMin)Empirical CDF of Minimum Cp from Monte Carlo SImulationRecord #135 channel 6 FL-27 Ivan (2004) Figure 4-24 ECDF for minimum pC Hz duration) from Monte Carlo simulation for channel 6 of record #135 FL-27 hurricane Ivan (2004) (20Presentation of Results from Uncertainty Analysis The process presented above to quantify the peak minimum from the full-scale measurement is now applied to study several channels over several hours of extreme wind. From the preliminary analysis of the FL-27 house the critical sensors selected to present the result obtained from the Monte Carlo Simulation are channel 6, 7, 13, 18, 22 and 23 for the records 131 147. This time frame, according to the ARA model, represents strong Tropical Storm and Category 1 winds at the house. Figure 4-25 through 4-30 presents the minimum peak pressure estimation of the using case 1b (Table 4-3), and the expected minimum value obtained from the Monte Carlo Simulation along with the logned value. pC pC pwest and highest confidence limits and the ASCE-7 assi C

PAGE 116

99 Peak minimum Cp value for channel 6, FL-27 Ivan (2004)-7.00-6.00-5.00 -4.00-3.00130132134136138140142144146148Record #Minium Co foHz d -2.000.00mp fr 20 uration -1.00 Mean (Min Cp) Min Cp (Lowest 2.5%) Min Cp (Highest 97.5%) ASCE 7 Min Negative Cp -2.6 Single Min Peak (Case 1b) Figure 4-25 Channel 6 time history of min pC values for 20 Hz (FL-27 Ivan 2004) Peak minimum Cp value for channel 7, FL-27 Ivan (2004)-6.00-5.00-4.00-3.00-2.00-1.000.00130132134136138140142144146148Record #Minimum Cp fo for 20 Hz duration Mean (Min Cp) Min Cp (Lowest 2.5%) Min Cp (Highest 97.5%) ASCE 7 Min Negative Cp -2.6 Single Min Peak (Case 1b) Figure 4-26 Channel 7 time history of min values for 20 Hz (FL-27 Ivan 2004) pC

PAGE 117

100 Peak minimum Cp value for channel 13, FL-27 Ivan (2004)-9.00-8.00-7.00-6.00-5.00-4.00-3.00-2.00-1.000.00130132134136138140142144146148Record #Minimum Cp fo for 20 Hz duration Mean (Min Cp) Min Cp (Lowest 2.5%) Min Cp (Highest 97.5%) ASCE 7 Min Negative Cp -2.6 Single Min Peak (Case 1b) Figure 4-27 Channel 13 time history of minp values for 20 Hz (FL-27 Ivan 2004) C Peak minimum Cp value for channel 18, FL-27 Ivan (2004)-8.00-7.00-6.00-5.00-4.00-3.00-2.00-1.000.00130132134136138140142144146148Record #Minimum Cp fo for 20 Hz duration Mean (Min Cp) Min Cp (Lowest 2.5%) Min Cp (Highest 97.5%) ASCE 7 Min Negative Cp -2.6 Single Min Peak (Case 1b) Figure 4-28 Channel 18 time history of min values for 20 Hz (FL-27 Ivan 2004) pC

PAGE 118

101 Peak minimum Cp value for channel 22, FL-27 Ivan (2004)-7.00-6.00-5.00-4.00-3.00-2.00-1.000.00130132134136138140142144146148Record #Minimum Cp fo for 20 Hz duration Mean (Min Cp) Min Cp (Lowest 2.5%) Min Cp (Highest 97.5%) ASCE 7 Min Negative Cp -2.6 Single Min Peak (Case 1b) Figure 4-29 Channel 22 time history of min values for 20 Hz (FL-27 Ivan 2004) pC Peak minimum Cp value for channel 23, FL-27 Ivan (2004)-3.50-3.00-2.50-2.00-1.50-1.00-0.500.00130132134136138140142144146148Record #Minimum Cp fo for 20 Hz duration Mean (Min Cp) Min Cp (Lowest 2.5%) Min Cp (Highest 97.5%) ASCE 7 Min Negative Cp -2.6 Single Min Peak (Case 1b) Figure 4-30 Channel 23 time history of min values for 20 Hz (FL-27 Ivan 2004) pC

PAGE 119

102 These 95% bounds will change from one house dataset to another as the sources contributing to uncertainty (peak 3-second wind, temperature differential, reference pressure sensitivity, etc.) may be different from one subject house to the next. For example, a house that did not have a functional reference pressure sensor on the premises will be required to use the values reported from the nearest portable tower or other data source. Local atmospheric pressure can be significantly different over a matter of only a few miles depending on the location of the hurricane. This scenario would require a higher uncertainty (larger coefficient of variation) be assigned to the reference pressure value, and result in a larger 95% confidence interval for the calculated minimum peak values. the wind direction for these results, range from winds from the east-south-east (110) to winds out of the south (180). Sensors 6, 7 wind tunnel studies. However, this is not the case. The Stathopoulos durations were on the order of 25 m pC Referring to Figure 4-4 and Table 4-4, and 13 (Figures 4-25 through 4-27) are experiencing strong suction and show a tendency to exceed the -2.6 ASCE value either in the mean or the 95% confidence limit. Once the wind direction goes beyond 140 (~record 141 see Table 4-4) sensors 22, 23 and 18 (Figures 4-28 4-30) are in a stronger separation zone and their expected peak minimumpC values decrease. The -2.6 ASCE-7 value is added to these figures as a point of reference. Directcomparison would require that the duration of time that the peak minimum values are sampled from at full-scale (5 minutes) matches the scaled equivalent duration for the inutes of full-scale data per experiment. If the full-scale data were steady enough to legitimately calculate the expected peak value over a time frame

PAGE 120

103 equivalent to that used in the wind tunnel tests, the full-scale values would be more severe (lower) than they appear in this analysis. This is of concern when comparing the results in Figures 4-25 & 4-26, which currently show a strong likelihood of exceeding tASCE-7 value. he Closing Remarks l that ull-inty certainty in the analysis results. The next chapter continues th-scale pressure data. The results of reces: 1) The focus of the results presented above is not the specific limits that were identified for house subject FL-27, but rather the procedures developed for their identification. Ultimately, the full-scale data are to be used as a tool for evaluating theASCE-7 wind load provisions. This will be conducted using all available full-scale data collected in high winds, which includes many houses and several storms. It is criticasuch comparisons be made with an understanding of the possible variability of the fscale results from one subject to the next. It is possible that the results of the uncertaanalyses now being conducted on all appropriate subject homes will lead to the removal of several subjects (specific house, specific storm) from the useable dataset due to unacceptable un e analysis of the full ent wind tunnel studies of these same subject homes, conducted at Clemson University, will now be incorporated into the analysis to accomplish two objectivevaluate the degree to which full-scale and wind tunnel loads on the same subject agree2) provide a basis to extrapolate the full-scale results to wind directions that were not measured in full-scale.

PAGE 121

CHAPTER 5 CO ng direction. This will verify that model and full-scale results are similar in the mean and fluctuating but not with regard to minimum peak values. The full-scale loading, including the confidence limits from the uncertainty analysis, will then be compared to the ASCE-7 wind load provisions and preliminary conclusions drawn. Direct comparison of wind-tunnel vs. ASCE-7 is being conducted at Clemson University. Clemson Wind Tunnel Study vs. Full-scale (Same Wind Direction) The comparison between the wind tunnel studies conducted at Boundary Layer Wind Tunnel at Clemson University and the full-scale data set will focus on the comparison using the wind direction that cover the range of worst full-scale cases (highest winds) from 110 to 130, for the FL-27 house during hurricane Ivan (2004). The wind tunnel model of the FL-27 is a 1:50 scale plexi-glass model, where twenty-four pressure taps were installed to correspond to the locations of the full-scale pressure sensors, and four hundred and seventy-two additional taps were also installed. Styrofoam models (also correctly scaled) were constructed and placed to simulate the MPARISON OF FULL-SCALE DATA WITH WIND TUNNEL STUDY AND ASCE-7-05 This chapter compares three sources of wind load information, including theASCE-7 loads, loads from the Clemson wind tunnel studies of FCMP houses, and full-scale loads from the FCMP houses. Comparisons are first made between the Clemson wind tunnel study and the full-scale measurements from the same subject structure usithe same incident wind pC 104

PAGE 122

105 effect of the existing neighboring buildings as shown in Figure 5-1. During the experimental runs a total of 16 consecutive data records were collected for each wind direction. 10 increments were used over anpling rate wat mean entire 360 swath, where the sam s 3000 Hz and lowpass filtered to 400 Hz. The wind tunnel wind speed was 5.5 m/s a roof height and 11 m/s at Pitot tube height (top of the tunnel). Figure 5-1 Scale model (1:50 )of the FL-27 house and surrounding structures From the preliminary analysis of the full-scale pressure coefficients in the FL-27 house the strongest wi nds occurred when the wind angle of attack was between 110 ~ 130. Tabnnel mode le 5-1 presents the corresponding full-scale records to match the wind tu l wind directions. Table 5-1 Records for wind tunnel vs. full-scale comparison Full-Scale Record #Direction from T1Direction136 137120 124120 Average Wind Wind Tunnel 134 135108 113110138 139125 128130 The plots in the following sections present results comparing the full-scale analysis to the wind tunnel analysis of the same subject home (FL-27, Hurricane Ivan 2004 data set see Figures 1-4 and 4-22). For any given plot (see 5-2 for example) the assigned wind direction (110, 120 or 130) corresponds to two full-scale records (30 sequential

PAGE 123

106 minutes) of data analyzed for that graph, as defined in Table 5-1. The legend in graphs is specifically defined as the following: FS case 1b: This is the non-probabilistic analysis of full-scale data presented as an reference is a deterministic value estimated from the house anemometer using the deterministic value taken as the mean of the yard sensor over a these example in Chapter 4, referred to as case 1-b in Tables 4-3 and 4-4. The wind speed new roughness optimization scheme. The reference barometric pressure is a given 15-minute record. The graph presents the value of the average of the analysis from two sequential 15-minute records defined in Table 5-1. Wind Tunnel: The average resultant from analysis of multiple sequential wind tunnel records of full-scale equivalent length of ~16 minutes per record. MCS: The average value of the Monte Carlo simulation from two sequential 15-minute records. The result from a given 15-minute analysis is the mean value of random variable of interest (mean, RMS, peak min or peak max ). The analysis is the same procedure described and presented in Figures 4-25 4-32. Low 95%: The lower bound of the 95% confidence interval from the Monte Carlo simulation. nte Carlo e coefficients, showing a positive agreement. Refer to out. For the comparison of the mean pressure coefficient (Fig2. 5-2 5-4), the y-axis scale used is very small to illuminate the specific differences between wind tunnel and ull-scale. However the agreement is quite gooddispla pC pC High 95%: The upper bound of the 95% confidence interval from the Mo simulation. Comparison of Mean and RMS Pressure Coefficients Comparisons between the 21 working roof sensors are presented in Figures 5-2 thru 5-7 for the mean and RMS pressur Figure 4-5 for a display of the sensor locations on the roof lay f when viewed on the larger scale used to y peak minimum pressure coefficients (~ -3.0 through +1.0). The typical error between wind tunnel and full-scale mean are significantly smaller than those reported pC

PAGE 124

107 in previous comparisons using tropical storm data from 2002 (Dearhart, 2003). Importantly, the full-scale and model trends ma tch well. bound the un is preover a 10 degree arc or more in evaluated. measu calibration that wcale sensor values were adjusted with an offset to better match the anthis stage of the work it is consiarkably close in all casesor the 95% confidence limits from the full-scale analysis for every sensor and wind direction, with a few exceptions. With a few exceptions, the wind tunnel data falls within the 95% confidence s of the full-scale analysis. These confidence limits would be expanded further hadcertainty of mean wind direction been included. That is, the wind direction at 120cise for the wind tunnel, but fluctuates for full-scale a 15 minute period. At present the means to quantify this source of error is still being Dearhart (2003) discusses reasons for the offset between the model and full-scalerements, including uncertainties associated with the reference pressure factor, temperatures variations, sensor calibrations and the estimation of the peak gust. Inork the full-s me values. This offset was continued through analysis of peak coefficients. In the current analysis, no such offset adjustment was made. At dered best to offer a direct comparison of as-measured results. The comparisons of RMS pC between model and full-scale are shown in figures 5-5 5-7 for three different wind directions. The comparisons are rem and Figure 5-7 in particular. Recall that the scale more appropriate for fair comparison is on the order of ~ -3.0 through +1.0, which reflects the y-axis scale used fthe peak pC values. As was the case for the mean C comparisons, the wind tunnel value is within p

PAGE 125

108 Full-Scale vs. Wind Tunnel for 110 Wind Directio n -0.5-0.4 -0.3-0.100.20.3Meanp -0.20.10.4Channel C 1234567891011121314151617181920212223242526272829 FS case 1b Wind Tunnel MCS Low 95% High 95% Figure 5-2 Mean C full-scale vs. wind tunnel comparison for 110 wind direction p Full-Scale vs. Wind Tunnel for 120 Wind Directio n 0.30.4 -0.5-0.3-0.200.1Mean Cp -0.4-0.10.21234567891011121314151617181920212223242526272829 Channel FS case 1b Wind Tunnel MCS Low 95% High 95% p Figure 5-3 Mean full-scale vs. wind tunnel comparison for 120 wind direction C

PAGE 126

109 Full-Scale vs. Wind Tunnel for 130 Wind Directio n -0.5-0.4-0.3-0.2-0.100.10.20.30.41234567891011121314151617181920212223242526272829ChannelMean Cp FS case 1b Wind Tunnel MCS Low 95% High 95% Figure 5-4 Mean full-scale vs. wind tunnel comparison for 130 wind direction pC Full-Scale vs. Wind Tunnel for 110 Wind Directio n 00.050.10.150.20.250.30.350.41234567891011121314151617181920212223242526272829ChannelRMS Cp FS case 1b Wind Tunnel MCS Low 95% High 95% Figure 5-5 RMS full-scale vs. wind tunnel comparison for 110 wind direction pC

PAGE 127

110 Full-Scale vs. Wind Tunnel for 120 Wind Directio n 00.050.10.150.20.250.30.350.40.451234567891011121314151617181920212223242526272829ChannelRMS Cp FS case 1b Wind Tunnel MCS Low 95% High 95% Figure 5-6 RMS full-scale vs. wind tunnel comparison for 120 wind direction pC Full-Scale vs. Wind Tunnel for 130 Wind Directio n 00.050.10.150.20.250.30.350.41234567891011121314151617181920212223242526272829ChannelRMS Cp FS case 1b Wind Tunnel MCS Low 95% High 95% Figure 5-7 RMS full-scale vs. wind tunnel comparison for 130 wind direction pC

PAGE 128

111 This agreement is a strong indication that the turbulence intensity used in the wind tunnel is a good representation of the full-scale event, and that the wind tunnel experiment is able to closely replicate the loads in the mean and average fluctuating sense. Comparison of Peak Pressure Coefficients Previous full-scale/wind tunnel model comparisons (Dearhart 2003, Cope 1997, Hoxey 1997) show that it is common to match the mean and RMS pressure coefficient, but not minimum and maximum peaks. Figures 5-8 through 5-13 show the comparisons of the expected minimum and maximum peak values from the wind tunnel and full-scale analysis. The comparison of the minimupeak pressure coefficient are presented in ASCE-7 is included as a reference. As expected, given the approach direction of the maximum wind speed records, the worst cases of negative peak pressures are observed by channels 5, 6, 7 and 13. Each of these sensors is located in strong separation zones for that range of wind directions (see Figure 4-5). In one case (Fig. 5-8) the mean result of the Monte Carlo simulation (MCS) exceeds the ASCE-7 value of -2.6. In several cases in Figs. 5-8 5-10 the 95% confidence limit falls well below this value, and in others the deterministic full-scale analysis falls below this value. In general, and in particular for the sensors in strong separation zones, the comparison of full-scale analysis (MCS or deterministic case 1b) with wind tunnel analysis shows that the full-scale peak minimum values are more severe (lower for peak minimum) than the wind tunnel results (Figs. 5-8 5-10). pCm figures 5-8 through 5-10, where the minimum peak pC values of -2.6 allowed by th e pC

PAGE 129

112 Full-Scale vs. Wind Tunnel for 110 Wind Directio n -4.5-4-3.5-3ChnelMi -2.5-1.5-101234567891011121314151617181920212223242526272829ann -0.5 -2 Cp FS case 1b Wind Tunnel MCS Low 95% High 95% ASCE-7 CpMin Figure 5-8 Min pC full-scale vs. wind tunnel comparison for 110 wind direction Full-Scale vs. Wind Tunnel for 120 Wind Directio n -4-2.5-1ChannelMp -3.5-3-2-1.5-0.50in C 1234567891011121314151617181920212223242526272829 FS case 1b Wind Tunnel MCS Low 95% High 9 5% ASCE-7 CpMin p Figure 5-9 Min full-scale vs. wind tunnel comparison for 120 wind direction C

PAGE 130

113 Full-Scale vs. Wind Tunnel for 130 Wind Directio n -4.5-4-3.5-3-2.5-2-1.5-1-0.501234567891011121314151617181920212223242526272829ChannelMin Cp FS case 1b Wind Tunnel MCS Low 95% High 95% ASCE-7 CpMin Figure 5-10 Minp full-scale vs. wind tunnel comparison for 130 wind direction C Full-Scale vs. Wind Tunnel for 110 Wind Directio n 00.10.20.30.40.50.60.70.80.911234567891011121314151617181920212223242526272829ChannelMax Cp FS case 1b Wind Tunnel MCS Low 95% High 95% ASCE-7 CpMa x Figure 5-11 Maxp full-scale vs. wind tunnel comparison for 110 wind direction C

PAGE 131

114 Full-Scale vs. Wind Tunnel for 120 Wind Directio n 00.20.40.60.811.21234567891011121314151617181920212223242526272829ChannelMax Cp FS case 1b Wind Tunnel MCS Low 95% High 95% ASCE-7 CpMa x Figure 5-12 Max full-scale vs. wind tunnel comparison for 120 wind direction pC Full-Scale vs. Wind Tunnel for 130 Wind Directio n 00.10.20.30.40.50.60.70.80.911234567891011121314151617181920212223242526272829ChannelMax Cp FS case 1b Wind Tunnel MCS Low 95% High 95% ASCE-7 CpMa x Figure 5-13 Max full-scale vs. wind tunnel comparison for 130 wind direction pC

PAGE 132

115 This is in agreement with the findings of Dearhart (2003). The exception appears to be sensors 13, 15 and 16. Figures 5-11 5-13 present the maximum peak pressure coefficient comparison along with the maximum peak pressure coefficient allowed by the ASCE-7 of 0.5. The full-scale and wind tunnel results appear in better agreement than was the case for the minimum peak. Analysis values commonly exceed the 0.5 ASCE value, although this is of relatively small consequence for design against wind loads. Interpretation of Results Figures 5-8 5-10 show that the full-scale data analysis exceeds the ASCE-7 value of -2.6 in multiple cases when the sensor is in a strong separation zone. It also shows a d tunnel experiments. There are three indicators that the values currently assigned for components and cladding in ASCE-7 may require further refinement: 1. The preliminary analyses in this dissertation indicate that the wind tunnel tests are capable of matching full-scale results in the mean and average fluctuating (RMS) sense, but are generally non-conservative for the estimation of peak minimumvalues (corroborating Dearhart 2003). Given that ASCE-7 wind loads are based on wind tunnel studies conducted in open terrain conditions, there is evidence to suggest that peak minimum values for homes in suburban terrain should be factored higher in magnitude. 2. The ASCE-7 value is based on an enveloping procedure, where wind tunnel results are taken over many directions to identify the worst-case loads. The full-scale data analysis in this chapter is only conducted over a range of ~ 40 (dictated by the strongest winds of Ivan). It is likely that there are wind directions that would result in even higher magnitude peak minimum values than were presented in this analysis. Work is ongoing to develop a procedure to extrapolate the results of the wind tunnel study (360 range of analysis) to the full-scale data. That is, differences observed between full-scale and winday at 110, will be applied to pC general trend that the full-scale peak minimum pC values exceed those from the win pC pC pC tunnel for, s determine how to extrapolate the wind tunnel results from 270 to full-scale equivalent peak minimum C. p

PAGE 133

116 3. are currently being applied to the multiple other FCMP houses that collected storm level data during several storms in 2004 and 2005. It is reasonable to expect that further exceedance of ASCE-7 loads will be observed, Concluding Remarks The comparison between the model and full-scale data shows a strong agreement when comparing the mean and RMS pressure coefficients. There is less strong agreement between model and full-scale when comparing peak minimum values. Some of this can be associated with controllable uncertainties produced by the human error, like the improper identification of sensors used at each house location. More reliable wind speed instrumentation to capture the wind field at the house would reduce the uncertainty in the prediction of the 3-second gust reference wind speed and direction. Measurement of temperature at the house site will also reduce uncertainty. Aside from uncertainty in full-scale data, the differences between model and full-scale modescale.full-s that turbulence mismatch between model and full-le for coStrondevela sets collected over many homes with varying terrains and features, from several storms during The analysis techniques developed in this research, as well as wind tunnel studies, hurricane or tropical and that the trend of higher magnitude peak minimum pC values at full-scale vs.wind tunnel will continue. pC peak minimum pressure coefficients reflect a limitation in the ability of a scale l in wind tunnel flow to properly replicate the peak uplift pressures observed in fullThis conclusion is further enforced by the strong agreement between model and cale RMS pC values, indicating scais unlikely to be the cause of differences in peak coefficients. These preliminary results suggest that the peak uplift loads prescribed in ASCE-7 mponents and cladding may be non-conservative for homes in suburban settings. ger evidence is required to back up this suggestion. The full-scale analysis methods oped in this research are currently being applied to the extensive full-scale dat

PAGE 134

117 the 2004 and 2005 seasons. Concurrently, Clemson University is conducting scale model tunnel experiments on these homes. Future work will provide a comprehensive arison of full-scale, wind tunnel, and ASCE-7 loads, and indicate whether a ication is warranted. wind compmodif

PAGE 135

CHAPTER 6 PROPOSED GI S TOOLBOX FOR FCMP APPLICATIONS This chapter presents a contribution to the FCMP research using GIS software to generate deployment maps and aerial pictures of the deployment sites, and to select houses to conduct damage surveys in the aftermath of land falling hurricanes. What is GIS GIS (Geographic Information System) is a database system with software that can analyze and display data using digitized maps and tables for planning and decision-making. A GIS can assemble, store, manipulate, and display geographically referenced data, tying this data to points, lines and areas on a map or in a table. GIS can be used to support decisions that require knowledge about the geographic distribution of people, hospitals, schools, fire stations, roads, weather events, the impact of hazards/disasters, etc. Any location with known latitude and longitude or other geographic grid system can be a part of a GIS (Lauden, 2000). FCMP Deployment Maps In the 2003 season during hurricane Isabel, the FCMP teams started using ArcGIS to present maps with the tower location, weather station, airports and hurricane path (among other information). These maps are made available to the public on the web so that outside data users have an immediate view of the instrument placement. More specifically, the GPS coordinates of the towers accompany a map of their locations relative to the current storm path. As summary packets of wind speed information from individual towers are posted to the web in regular 15-minute intervals, and users know 118

PAGE 136

119 graphically where data points are being collected relative to the storm progress. This layer of context has greatly improved the immediate usefulness of the real-time data. Figure 6-1 present the deployment map generaents of hurricane Katrina (2005), similayments at the re ted in the ev r maps are available to the public for all the FCMP storm deplo search web site http://www.ce.ufl.edu. Figure 6-1 Deployment map of Hurricane Katrina (2005), generated before and updated obile during landfall FCMP Aerial Pictures of Tower Sites To complement the real-time data dissemination, aerial images of the tower deployment sites are posted, providing a sense of the exposure conditions that the mtower is experiencing. This information is crucial to researches and meteorologist that follow the FCMP deployment activities.

PAGE 137

120 The National Hurricane Center uses FCMP real-time data in their wind field models and for calibration of their remote sensing experiments as the storm progresseand makes landfall. These researchers convert the data as-measured to an open exposequivalent, the calculations for which are based in part on the local terrain conditions around the observation point. Thus it is very i s ure mportant to provide details of the deployment site surroundings. Figure 6-2 present the aerial picture of the FCMP mobile tower T1 deployed during hurricane Katrina (2005) in Bella Chasse, LA. Aerial pictures of all FCMP tower deployment are available at the web site immediately after a given tower is deployed. Figur during hurricane Katrina (2005). See also Figure 6-1 GIS Toolbox to Assist Randomized Damage Evaluation Studies tion of damage to areas affected by hurricanes e 6-2 Aerial picture of FCMP mobile tower T1 deployed in Bella Chasse, LAIn addition to the deployment location maps and aerial pictures, a GIS framework is used for statistically relevant documenta

PAGE 138

121 after af s ty. ng more rom the NOAA H*WIND model provides almost immediate representations of the wind swath over the entire region impacted by the storm. These NOAA wind swaths are generated using, among other sources of data, the FCMP real-time tower data. Figure 6-3 presents a map with the wind swath model for Hurricane Ivan (2004). The color contours represent wind swaths in color bands of 10 mph for 1-minute sustained gusts converted to open exposure. Such a map, in combination with the residential structural data, allow the selection of potential structures to perform post-damage survey by year built, roof type, wall type, square footage, and other parameters. Significantly, comparative structural performance can be measured as a function of the peak wind speeds those homes experienced. This tool can ensure that structures of a certain type are selected for evaluation across a series of different wind speed zones in statisticallys random, ensuring that visual bias of dam storm has passed. The GIS platform includes information layers consisting ospecific construction information on the existing Florida houses gleaned from the varioucounty property tax appraiser databases. The Florida property appraiser databases providethe age of construction, location and property identification for each parcel in the counThe attributes (property information) vary from county to county, some containidetailed information than others. This is overlaid with wind swath information (e.g. H*Wind from NOAA or the ARA wind field model) and stratified to select properties for damage inspection. Information f meaningful quantities. The selection process i age is not used to select evaluation subjects.

PAGE 139

122 Figure 6-3 Wind Swath of Hurricane Ivan 1-minute sustained gust Thi s will optimize the efforts of the FCMP damage survey teams, allowing the selectCMP er ys, llection of damage surveys that ive targeting of specific structural types with knowledge of the contours of wind speeds over large portions of the affected region. After hurricane Charley 04, the Fsurvey teams did not have such a tool, and simply wandered into areas known to suffsevere damage. The resulting damage survey data set only represents the hardest hit areas, and does not systematically assess damage to homes of similar age and construction across a variety of known peak wind speeds. The Charley damage survewhile valuable, do not present a complete picture of the relative vulnerability of structures as a function of a range of peak wind speeds. With such a GIS tool, the same teams, in the same time frame, could have generated a co

PAGE 140

123 provide a much more global perspective of the effects of Charley on residential construction. Optimized Statistical Sampling This section presents the GIS toolbox developed to expedite the process of selection house under selected wind speeds, using residential property attributes like effective year of construction. The output will be a preliminary list of potential houses that satisfy the user criteria to perform the survey, which contain the same sample space characteristics of the real sample for this case then entire county been studied. For this study the year build criteria (database field code [EFFYR]) will be used to stratify our sample space, the user will input a range of years to classify the parcel data. A spatial query is performed in order to select houses under the selected wind zone (database field eal sample space chara shape A step by step Pseudo Code is presented, providing detailed information of the user input requirements: code [GRIDCODE]); these houses will then be weighted using the r cteristics. Recent FCMP research effort to conduct post damage survey in the region of Charlotte County, FL have involved intensive user interface use with the selection tools provided by ArcMap software. The process had consisted of selecting the desired house types within a desired wind swath region, while ensuring that correct quantities of thehouses of varying description are selected. This custom application will provide a more time efficient means of randomly pre-selecting the residence on which to document damage. The user output will be a file generated from the selected field of the parcel data layer. From this new layer FCMPmembers will have the necessary information to conduct the study.

PAGE 141

124 1. User Input data layers (Parcel Data and Wind Swath) 2. User Start Application from FCMP Toolbar (house button) 3. User Set Parcel Data and Wind Swath Data Layer ta layers 6. User input desired number of the sample space (default value 200) parcel data (default value 2) range 2 [2001 to 1994]), which represent the year of wind load provision changes 9. User click on Run Parcel Data Statistics button and behind the scene the sample e entire county and the selected wind zone populated for the county and wind zone levels ted wind swath zone base on the specified date range and all valid records select another wind zone to obtain new set of houses 4. User Set Filed Code [EFFYR] criteria to be used in the Query Selection for da5. User set wind swath zone value [GRIDCODE] 7. User input number of construction date range, maximum of four to breakdown the8. User input values for the date range (default values: range 1 [2004 to 2002] andin the State of Florida parcel data is stratified into group base on [EFFYR] provided date ranges for th10. The display Parcel Data Statistics Base on Dates Ranges previous specified is 11. The user have the option to export the portion of the parcel data under the selec12. The user Figure 6-4 User interface wi ndow of the FCMP survey toolbox developed in ArcMap

PAGE 142

125 6-5 Map of Charlotte County showing the selec Figure ted houses using the FCMP survey presented in Figure 6-4, where the summary statistics for the entire county are shown to be similar to the selected house under the selected wind swath zone. Finally Figure 6-5 presents the map of Charlotte County, FL with the houses selected under the Wind Zone 11 (140 150 mph 3-second gusts) for the date range of 1983 to 1970. Recommendation and Current Deployment Strategies The author contribution is to provide the framework to future team members to produce the FCMP deployment maps, aerial pictures, improve logistics in future damage survey studies and deployment strategies. A dataset of the past FCMP deployment activities is available along with the framework to assist in post damage surveys. As technology progresses and GIS systems become more accessible to the public, it toolbox Result from the execution of program user interface window are becomes easier to disseminate deployment information via the internet. An example of

PAGE 143

126 this is the current use of Google Earth Pro since the 2005 season. This commercial software allows an easy means to project the FCMP tower location and simultaneously track the storm progress with the possible storm projected path models. It also provides a real-time access to FCMP team members in the filed (with a Laptop and wireless internet service) to assist in the decision making process and deployment logistics.

PAGE 144

CHAPTER 7 CONCLUSION AND RECOMMENDATIONS pCvalues Comparisons (Full-scale vs. Wind Tunnel vs. ASCE7-05) The pressure coefficient evaluation between the full-scale, wind tunnel models and the current wind load provision of the ASCE-7-05 is intended to provide an accurate view of the actual conditions measured during real hurricane in suburban exposure conditions. The information obtained from these evaluations will be used to assess the wind tunnel modeling techniques and the current wind load provisions ASCE-7-05, which were developed based on wind tunnel studies in the late 1970s in what is today considered coastal (very open) exposure rather that suburban. Preliminary analysis of the FCMP full-scale vs. wind tunnel house data, provided by Dearhart (2003), Reinhold (2005), and from the analysis presented in this dissertation, suggests that the recent availability of extreme wind loading data on full-scale residential structures may potentially impact wind load standards. These comparison studies suggest that the peak negative pressure coefficients obtained from the full-scale data exceeded the ASCE-7 coefficients (components and cladding) for the corresponding roof zones of high suction areas and also the wind tunnel data. The peak load (developed for a fairly coastal to open exposure) in ASCE-7 and the values obtained from the wind tunnel model are un-conservative when comparing to the full-scale peak loads. This dissertation develops and presents new contributions in the processing, analysis and interpretation of full-scale pressure data: 127

PAGE 145

128 1. Probabilistic analysis: The identification, modeling and direct inclusion of uncertainties in collection, processing, and analysis of the full scale data via Monte Carlo simulation offers a new environment with which to evaluate results. Comparisons of wind tunnel and ASCE-7 values to full-scale are made with explicit confidence limits on thees. Further, the degree of confidence (spread of the he usability of a given dataset. Tight limits indicate less uncertainty and loose limits indicate more uncert2. wind speed reference for the normalizing term in has been a source of difficulty speed data to an expected peak value had relied upon standard conversion equations best guess for roughness, combined with a deterministic approach to analysis, ment d n inute records, and the ng locally measured mean and The analysis of the full-scale data is continuing in two directions: 1. The data from other FCMP houses, with different exposures, and experience other for this research. Accompanying wind tunnel studies of these homes is also underway tunnel and ASCE-7 loads will be offered upon completion to the ASCE Wind Load engineering community via a benchmark study, discussed in a forthcoming section. 2. The analysis of the FL-27 Ivan dataset includes only the time frame where winds speeds occurred does not cover of 360o, and it is likely that more extreme -2.6 would be observed with a complete directional dataset. The wind tunnel tests are conducted over a complete 36full-scale data with the wind tunnel result. It is desirable to pursue development of a mum at full-scale and wind tunnel can be extrapolated to directions that are not covered in the full scale dataset. pC pC full-scale valu confidence limits) can be used as an indicator of t ainty in the full-scale estimations. Identification of wind speed reference via optimization: The identification of a in past analyses of full-scale data. Conversion of locally measured mean wind that used a best guess for exposure conditions via the roughness coefficient. This produced a methodology that would produce varying estimate for pC from different experts using the same dataset. This dissertation includes the developof a method to estimate an expected peak wind reference using locally measurewind speed data in a standardized way. The directionally dependent terrain roughness and / or peak factor is identified via minimization of the error betweepredicted and observed 3-second peaks over a series of 15-mresult applied as a standard means of converti pC standard speed data to a rational expected peak wind speed reference. Future and Ongoing Analysis storms is still in progress, following the probabilistic analysis procedures developed at Clemson University. A comprehensive set of results comparing full-scale, wind Committee. The FCMP research group will also seek a consensus among the wind were of tropical storm or hurricane intensity. The wind directions over which these excursions of peak coefficients past the ASCE value of 0o and provide a possible means of extrapolation, from the worst case in the method by which the difference between peak mini pC

PAGE 146

129 T a mons toWorkscale facility. Techniques and assumptions differ among the various boundary layer wind tunne it regardot align proviparticloading of full-scale residentity affiliated wind experts from eate a seaindcol ill mae andforw hat is, developeans to convert the available wind tunnel data from all directi a full-scale equivalent. shop and Benchmark Study with all Major Wind Tunnel Facilities The comparisons between the current dataset of full-scale pressure loads and the model loads only apply to the Clemson University boundary layer wind tunnel l facilities, and the researchers that run them. Thusis possible that conclusions ing the non-conservative nature of peak loads measured in the wind tunnel may nwith the results from other facilities. It is imperative that any recommendations regarding the current wind load sions be the result of consensus within the wind engineering community, ularly from those with experience in experimental methods in wind load evaluation. To address this, preliminary work is now underway to host a UF workshop on wind ial structures. The leading univers the U.S. and Canada (possibly others) will meet in Gainesville to cr benchmark study based on the full-scale FCMP house data collected in 2004-2005 sons. Participants will jointly design this study, which will provide a suite of ependently conducted wind tunnel studies on several of the FCMP houses that lected sustained hurricane winds. All teams will be given identical parameters to work with (home dimensions, terrain information, measured wind field statistics, etc.). UF wintain control of the full-scale data. The results will offer varying degrees of agreement/differences between full-scal model studies of extreme wind. Consensus recommendations will then be put ard to the ASCE Wind Load Committee.

PAGE 147

130 Improvements to House Data Collection System There are several limitations with the current wire-to-wire pressure measurement system: Considerable effort is necessary to prepare the wiring for the houses using PVC pipes, there is an upper limit of 28 sensors per house due to the wire-to-wire system, and the number of houses that can be outfitted to collected data as a storm approaches is limited. To address these restrictions, colleagues at Florida Institute of Technology have developed a first generation wireless version of the pressure transducers (Lapilli, 2004). In this new system each sensor transmits its data to an on-site data collection computer remotely, eliminating the need to connect the individual sensors to the computer by wire. This provides much easier and faster installation of the sensors just prior to a storm. The existing system requires three to four hours of work per house, while the wireless system can be installed in less than one hour. It is feasible to ready up to 12 houses for a given storm rather than the current 4-6 houses, producing a higher volume of valuable data per storm. The wireless system can increase data resolution from a maximum of 28 sensors per house to virtually as many as FCMP can transport. This will dramatically increase the detail of information measured on a given house. There are also several advantages in terms of post-processing the data to produce pressure coefficients, since the data collected using this new sensor does not need to be post processed to make correction due to temperature changes. The prototype wireless system was tested during several storms in 2004 and 2005 with promising results. Funding for additional units with up to date technology is being sought.

PAGE 148

131 Maintenance to FCMP Houses It is important to maintain a quality control of the FCMP full-scale house equipment, first by updating the identification system of the sensors used on the roof, walls, yard and attic. It is a key to the success of the full-scale data analysis to know which sensor was used in order to use the proper sensitivity calibration, rather than asa mean va sign lue, since these incorporate uncertainties in the analysis. Another important upgrade will be to include instrumentation to measure the temperature at the house site, rather than using the measurements from a nearby tower, this will help to reduce the uncertainty associated with the temperatures effects in the sensors. Also it will be useful to install RM Young pressure transducers inside the computer box or in the camera post in order to provide a calibrated source of barometric pressure at the house site. Finally it will be very important to synchronize the time in the computers at the house site using an internet connection, since the timestamp of the tower data correspond to the synchronized time. This will provide a more accurate way to relate the wind speedand temperature measurement from the tower and house site.

PAGE 149

APPENDIX A ents maps for the storm AERIAL PICTURES AND MAPS FOR 2004 & 2005 DEPLOYMENTS This appendix contains aerial pictures of the FCMP mobile towers deployments sites, aerial pictures of the FCMP instrumented houses and deploym activities. 132

PAGE 150

133 Figure A-1 FCMP Deployment map of Hurricane Charley (2004)

PAGE 151

134 Figure A-2 Aerial imagery of the terrain surrounding tower T0 during Hurricane Charley (2004) at Lakeland, FL

PAGE 152

135 Figure A-3 Aerial imagery of the terrain surrounding tower T1 during Hurricane Charley (2004) at Fort Meyers, F L

PAGE 153

136 Figure A-4 Aerial imagery of the terrain surrounding tower T2 during Hurricane Charley (2004) at Osprey, FL

PAGE 154

137 Figure A-5 Aerial imagery of the terrain surrounding tower T3 during Hurricane Charley (2004) at Plant City, F L

PAGE 155

138 Figure A-6 FCMP Deployment map of Hurricane Frances (2004)

PAGE 156

139 Figure A-7 Aerial imagery of the terrain surrounding tower T0 during Hurricane Frances (2004) at Port Salerno, FL

PAGE 157

140 Figure A-8 Aerial imagery of the terrain surrounding tower T1 during Hurricane Frances (2004) at Port Salerno, FL

PAGE 158

141 Figure A-9 Aerial imagery of the terrain surrounding tower T2 during Hurricane Frances (2004) at Vero Beach, FL

PAGE 159

142 Figure A-10 Aerial imagery of the terrain surrounding tower T3 during Hurricane Frances (2004) at Fort Pier ce, FL

PAGE 160

143 Figure A-11 FCMP Deployment map of Hurricane Ivan (2004)

PAGE 161

144 Figure A-12 Aerial imagery of the terrain surrounding tower T0 during Hurricane Ivan (2004) at Mobile, AL

PAGE 162

145 Figure A-13 Aerial imagery of the terrain surrounding tower T1 during Hurricane Ivan (2004) at Pensacola, FL

PAGE 163

146 Figure A-14 Aerial imagery of the terrain surrounding tower T2 during Hurricane Ivan (2004) at Fairhope, AL

PAGE 164

147 Figure A-15 Aerial imagery of the terrain surrounding tower T3 during Hurricane Ivan (2004) at Destin, FL

PAGE 165

148 Figure A-16 FCMP Deployment map of Hurricane Jeanne (2004)

PAGE 166

149 Figure A-17 Aerial imagery of the terrain surrounding tower T0 during Hurricane Jeanne (2004) at Orlando, FL

PAGE 167

150 Figure A-18 Aerial imagery of the terrain surrounding tower T1 during Hurricane Jeanne (2004) at Sebastian, FL

PAGE 168

151 Figure A-19 Aerial imagery of the terrain surrounding tower T2 during Hurricane Jeanne (2004) at Merritt Island, FL

PAGE 169

152 Figure A-20 Aerial imagery of the terrain surrounding tower T3 during Hurricane Jeanne (2004) at Vero Beach, FL

PAGE 170

153 Figure A-21 FCMP Deployment map of Hurricane Dennis (2005)

PAGE 171

154 Figure A-22 Aerial imagery of the terrain surrounding tower T0 during Hurricane Dennis (2005) at Navarre, FL

PAGE 172

155 Figure A-23 Aerial imagery of the terrain surrounding tower T1 during Hurricane Dennis (2005) at Inlet Beach, F L

PAGE 173

156 Figure A-24 Aerial imagery of the terrain surrounding tower T2 during Hurricane Dennis (2005) at Pensacola, FL

PAGE 174

157 Figure A-25 Aerial imagery of the terrain surrounding tower T3 during Hurricane Dennis (2005) at Pensacola, FL

PAGE 175

158 Figure A-26 Aerial imagery of the terrain surrounding tower T5 during Hurricane Dennis (2005) at Niceville, FL

PAGE 176

159 Figure A-27 FCMP Deployment map of Hurricane Katrina (2005)

PAGE 177

160 Figure A-28 Aerial imagery of the terrain surrounding tower T0 during Hurricane Katrina (2005) at Bay St. Louis, MS

PAGE 178

161 Figure A-29 Aerial imagery of the terrain surrounding tower T1 during Hurricane Katrina (2005) at Bella Chasse, LA

PAGE 179

162 Figure A-30 Aerial imagery of the terrain surrounding tower T2 during Hurricane Katrina (2005) at Galliano, LA

PAGE 180

163 Figure A-31 Aerial imagery of the terrain surrounding tower T3 during Hurricane Katrina (2005) at Pascagoula, M S

PAGE 181

164 Figure A-32 Aerial imagery of the terrain surrounding tower T5 during Hurricane Katrina (2005) at Gulf Port, MS

PAGE 182

165 Figure A-33 FCMP Deployment map of Hurricane Rita (2005)

PAGE 183

166 Figure A-34 Aerial imagery of the terrain surrounding tower T0 during Hurricane Rita (2005) at Port Arthur, TX

PAGE 184

167 Figure A-35 Aerial imagery of the terrain surrounding tower T1 during Hurricane Rita (2005) at Houston, TX

PAGE 185

168 Figure A-36 Aerial imagery of the terrain surrounding tower T3 during Hurricane Rita (2005) at Nederland, TX

PAGE 186

169 Figure A-37 Aerial imagery of the terrain surrounding tower T5 during Hurricane Rita (2005) at Orange, TX

PAGE 187

170 Figure A-38 FCMP Deployment map of Hurricane Wilma (2005)

PAGE 188

171 Figure A-39 Aerial imagery of the terrain surrounding tower T0 during Hurricane Wilma (2005) at Everglades City, FL

PAGE 189

172 Figure A-40 Aerial imagery of the terrain surrounding tower T1 during Hurricane Wilma (2005) at Weston, FL

PAGE 190

173 Figure A-41 Aerial imagery of the terrain surrounding tower T2 during Hurricane Wilma (2005) at Ochoppi, FL

PAGE 191

174 Figure A-42 Aerial imagery of the terrain surrounding tower T3 during Hurricane Wilma (2005) at Miami, FL

PAGE 192

175 Figure A-43 Aerial imagery of the terrain surrounding tower T5 during Hurricane Wilma (2005) at Naples, FL

PAGE 193

APPENDIX B WIND CALCULATION This appendix shows representative example of the wind calculation used to convert from different wind exposures, heights and durations. Estimating 3-second gust from measured 15-minute wind speed average from FCMP House data House and Instrument Information FCMP House: FL-27 (Gulf Breeze, FL) Instrument: Anemometer 1 Determine Anemometer Height ftHeightRoof17 ftHeightAnemometer17 ftftftStemAnemometerHeightRoofHeightAnemometer5.215.417 Calculate to be used in Equation From Table 2.3.3 on Wind Effects on Structures tC SecGustU3 [Simiu & Scanlan (1996)] Solve for, using the following proportion: sec11t 31tC 3.22tC sec102t 2 ?sec33tC 32321212tCtCtttCtCtt sec332.2sec3sec10332.2sec1sec10C Similarly, the value for 900 sec is: 85.2sec3C tC 21.0sec900C 176

PAGE 194

177 Obtain Log-Law Design Parameters based on Exposure Category From Table 9.1.1 on Wind Effects on Structures [Simiu & Scanlan (1996)] For Densely Built-up Suburbs mz20.180.00 Lets assume a value of: 10 ftm281.300. From Table 2.3.1 on z Wind Effects on Structures [Simiu & Scan lan (1996)] For Densely Built-up Suburbs 85.4 Lets solve for U z3 ln5.2zz 009003secln5.2zzUUzzSec sec31C 9001C ft5.2185.485.2 ftftftUUftftSec281.35.21ln5.285.421.01281.3ln5.215.215.219003 098.15.21ft 335.25.219003ftSecU U 126.25 .21 5.219003 ftftSecUU

PAGE 195

APPENDIX C PRESSURE COEFFICIENT CALCULATIONS This appendix contains a map of the instrumented houses in the state of Florida, ut per house, e sensor sensitivity data and results in tabular and graphical format of the dynamic ressure coefficients. DYNAMIC South and North Carolina; in addition it includes the sensor distribution layo th p 178

PAGE 196

179 Figure C-1 FCMP Instrumented houses in the state of Florida

PAGE 197

180 Figure C-2 FCMP Instrumented houses in the state of South an d North Carolina

PAGE 198

181 Figure C-3 Roof sensor layout of the FL-27 House Figure C-4 Aerial imagery of FCMP House ID FL-27 at Gulf Breeze, FL

PAGE 199

182 Figure C-5 Roof sensor layout of the FL-30 House Figure C-6 Aerial imagery of FCMP House ID FL-30 at Pensacola Beach, FL

PAGE 200

183 Table C-1 Sensitivity data of sensor used in the FL-27 house during Ivan (2004) Plastic BoxCamera034.274Average ValueOKPan Sensor026A134.274Average ValueNGPan Sensor008A234.149UF2005OKPan Sensor027A334.274Average ValueOKPlastic BoxAttic434.274Average ValueNGPan Sensor035A534.274Average ValueOKPan Sensor201634.336UF2005OKPan Sensor92734.154UF2005OKPan Sensor834.274Average ValueNGPan Sensor42934.430UF2005OKPan Sensor1671034.733UF2005OKPan Sensor931134.434UF2005OKPan Sensor811234.203UF2005OKPan Sensor003A1334.588UF2005OKPlastic Box1434.274Average ValueNGPan Sensor721534.420UF2005OKPan Sensor131634.274Average ValueOKPan Sensor841734.147UF2005OKPan Sensor2251834.608UF2005OK Pan Sensor341934.147UFPan Sensor1432034.457UF 2005OK2005OKPanensor1442134.396UF2005OKPanensor492234.092UF2005OKPan Sensor1772334.224UF2005OKPan Sensor922434.154UF2005OKPlastic Box2534.274Average ValueNGPan Sensor1222634.380UF2005NGPan Sensor542734.395UF2005NGPan Sensor1062834.230UF2005OKAnemometer 1-29-OKAnemometer 2-30-Not in useChannel iCalibration iCalibration SourceChannel StatusSensor TypeSensor ID S S

PAGE 201

184 Table C-2 Sensitivity data of sensor used in the FL-30 house during Ivan (2004) Plastic BoxCamera034.274Average ValueOKPan Sensor031A134.380UF2005OKPan Sensor70234.348UF2005OKPan Sensor19334.452UF2005OKPlastic Box132434.240UF2005OKPan Sensor15533.867UF2005OKPan Sensor171634.398UF2005OKPan Sensor104734.270UF2005OKPan Sensor238834.409UF2005OKPan Sensor043A934.392UF2005OKPan Sensor044A1034.182UF2005OKPlastic BoxSoffit1134.274Average ValueOKPan Sensor1821234.194UF2005OKPan Sensor045A1334.274Average ValueOKPlastic Box010A1434.389UF2005OKPan Sensor741534.016UF2005OKPan Sensor261634.387UF2005OKPan Sensor050A1734.274Average ValueOKPan Sensor2091834.659UF2005OKPan Sensorguess1934.274Average ValueOKPan Sensor002A2034.274Average ValueOKPan Sensor1932134.190UF2005OKPan Sensor1732234.255UF2005OKPlastic BoxWall2334.274Average ValueOKPan Sensor1162434.345UF2005NGPlastic BoxNot in use2534.274Average ValueNGPlastic BoxSoffit2634.274Average ValueOKPlastic BoxWall2734.274Average ValueNGPlastic BoxAttic2820.600Old SensorOKAnemometer 1-29-OKAnemometer 2-30-Not in useCalibration SourceChannel StatusSensor TypeSensor IDChannel iCalibration i

PAGE 202

185 2 3 5 6 7 9 10 11 12 13 15 16 17 18 19 20 21 22 23 24 28 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 Cp Mean for FCMP House FL-27 Case (1b) data records 131 147 Pressure Tap LocationCp Mean Figure C-7 Mean pressure coefficients for FL-27 during hurricane Ivan (2004) 2 3 5 6 7 9 10 11 12 13 15 16 17 18 19 20 21 22 23 24 28 0.05 0.1 0.15 0.2 0.25 0.3 Cp RMS for FCMP House FL-27 Case (1b) data records 131 147Pressure Tap LocationCp RMS Figure C-8 RMS pressure coefficients for FL-27 during hurricane Ivan (2004)

PAGE 203

186 2 3 5 6 7 9 10 11 12 13 15 16 17 18 19 20 21 22 23 24 28 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0 Cp Min for FCMP House FL-27 Case (1b) data records 131 14720 Hz Pressure DurationPressure Tap LocationCp Min Figure C-9 Min peak pressure coefficients 20 Hz for FL-27 during hurricane Iva n (2004) 2 3 5 6 7 9 10 11 12 13 15 16 17 18 19 20 21 22 23 24 28 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 Cp Max for FCMP House FL-27 Case (1b) data records 131 14720 Hz Pressure DurationPressure Tap LocationCp Max Figure C-10 Max peak pressure coefficients 20 Hz for FL-27 during hurricane Ivan (2004)

PAGE 204

187 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 26 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 Cp Mean for (FCMP House FL-30 Case (1a) data records 116 130)Pressure Tap LocationCp Mean Figure C-11 Mean pressure coefficients for FL-30 during hurricane Ivan (2004) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 26 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 Cp RMS for (FCMP House FL-30 Case (1a) data records 116 130)Pressure Tap LocationCp RMS Figure C-12 RMS pressure coefficients for FL-30 during hurricane Ivan (2004)

PAGE 205

188 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 26 -4 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0 Cp Min for (FCMP House FL-30 Case (1a) data records 116 130)20 Hz Pressure DurationPressure Tap LocationCp Min Figure C-13 Min peak pressure coefficients 20 Hz for FL-30 during hurricane Iva n (2004) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 26 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 Cp Max for (FCMP House FL-30 Case (1a) data records 116 130)20 Hz Pressure Duration Pressure Tap Location Cp Max Figure C-14 Max peak pressure coefficients 20 Hz for FL-30 during hurricane Ivan (2004)

PAGE 206

189 LIST OF REFERENCES Aponte, L. (2004). Measurement, validation and dissemination of hu rricane wind data, Masters report, University of Florida, Gainesville, FL. Department of Civil and Coastal Engineering. ASCE-7-05 (2005). Minimum design loads for buildings and other structures. Reston, VA: American Society of Civil Engineers. Bendat, J. S., & Piersol, A. G. (2000) Random data: analysis and measurement procedures (3rd ed.). New York: John Wiley & Sons Inc. Bienkiewicz, B., & Ham, H. (1998). Wind tunnel simulation of TTU flow and building roof pressure, Journal of Wind E ngineering and Industrial Aerodynamics. Volumes 77-78, September, Pages 119-133. Cope, Anne D. (1997). Load duration effects on peak minimum pressure coefficients, Master of Science Thesis, Clemson Univ ersity, Clemson, SC. Civil Engineering Department. Dearhart, Elizabeth A. (2003). Comparis on of field and model wind pressures on residential buildings in tr opical storm winds, Master of Science Thesis, Clemson University, Clemson, SC. Civil Engineering Department. Gurley, K., Davis, R., Ferrera, S-P., Burton, J ., Masters, F., Reinhold, T. & Abdullah, M. (2006). Post 2004 hurricane field survey An evaluation of the relative performance of the standard building code and the Florida building code, Proceedings from American Society of Ci vil Engineers Structures Congress: St. Louis, Missouri, USA, Pages 110. Retrieved from conference CD. Holmes, J. D. (1982). Comparison of model a nd full-scale test of the Aylesbury house, Proceedings from International Works hop on Wind Tunnel Modeling Criteria and Techniques in Civil Engineering Applic ations: Gaithersburg, Maryland, USA, edited by Reinhold, T.A, Pages 605-618. Hoxey, R. P., Richardson, G. M., Robe rtson, A. P., & Short, J. L. (1997). The Silsoe Structures building: Comparisons of pre ssures measured at full-scale and in two wind tunnels, Journal of Wind Engineer ing and Industrial Aerodynamics, Volume 72, November-December, Pages 187-197.

PAGE 207

190 Hoxey, R., & Richards, P. J. ( 1995). Full-scale wind load measurements point the way forward, Journal of Wind Engineering and Industrial Aerodynamics, Volume 57, Issues 2-3, July, Pages 215-224. Kaimal, J. C., & Finnigan, J. J. (1994). Atmo spheric boundary layer fl ows: their structure and measurement. New York: Oxford University Press. Kopp, G.A., Pierre, L.M. St., Surry, D. & H o, T.C.E. (2005). The UWO contribution to the NIST aerodynamic database for wind loads on low buildings: Part 2. Comparison of data with wind load provi sions, Journal of Wind Engineering and Industrial Aerodynamics, Volume 93, Issue 1, January, Pages 31-59. Lapilli, Claudio D. (2004). Wireless pressu re monitoring system, Master of Science Thesis, Florida Institute of Technol ogy, Melbourne, FL. Civil Engineering Department. Laudon, K. C., & Laudon, J. P. (2000). Manageme nt information systems: organization and technology in the networked enterpri se (6th ed.). Upper Saddle River, NJ: Prentice Hall. Lynn, B. A., & Stathopoulos, T. (1985). Wi nd-induced fatigue on low metal buildings, ASCE American Society of Civil Engineer s Journal of Structural Engineering Volume 111, Issue 4, April, Pages 826-829. Masters, F. (2004). Measurement, modeli ng and simulation of ground-level tropical cyclone winds, PhD Dissertation, Univer sity of Florida, Gainesville, FL. Department of Civil and Coastal Engineering. Michot, Brian J. (1999). Full-scale wind pr essure measurements utilizing unobtrusive absolute pressure transducer technology, Master of Science Thesis, Clemson University, Clemson, SC. Civil Engineering Department. Porterfield, M., & Jones, N.P. (2001). T he development of a field measurement instrumentation system for low-rise cons truction, Wind and Structures, Volume 4, Issue 3, Pages 247-260. Poss, David B. (2000). Design and evaluati on of a mobile wind instrumentation tower for hurricane wind measurements, Master of Science Thesis, Clemson University, Clemson, SC. Civil Engineering Department. Reinhold, T. A. (2005). Wind loads on low-ri se buildings: Is One Set of Pressure Coefficients Sufficient for All Types of Terrain?, Conference Proceedings from The Second International Symposium on Wind Effects on Buildings and Urban Environment: Seoul, Korea, Pages 49-57.

PAGE 208

191 Richardson, G. M., & Blackmore, P. A. (1995). The Silsoe structures building: comparison of 1 : 100 model-scale data with full-scale data, Journal of Wind Engineering and Industrial Aerod ynamics, Volume 57, Issues 2-3, July, Pages 191-201. Simiu: fundamentals and applications to design (3rd ed.). New York: John Wiley & Sons Inc. Simp, LA: Stathopoulos, T. (1979). Turbulent wind action on low-rise buildings, Ph.D. thesis, The Stathopoulos, T., & Surry, D. (1983). Scale effects in wind tunnel testing of low buildings, Journal of Wind Engineering and Industrial Aerodynamics, Volume 13, Stathopoulos, T., Surry, D. & Davenport, A.G. (1981). Effective wind loads on flat roofs. Journal of the Structural Division, Volume 107, Issue 2, February, Pages Teachline by States, Source: Department of Commerce, National Oceanic and Atmospheric Administration, National Ocean E., & Scanlan, R. H. (1996). Wind effects on structures son, R. H., & Riehl, H. (1981). The hurricane and its impact. Baton RougeLouisiana State University Press. University of Western Ontario, London, Ontario, Canada. Issues 1-3, December, Pages 313-326. 281-298. ervision.com Table Length of the U.S. Coast Service. Retrieved July 1, 2006, from http://www.teachervision.fen.com/maps/bodies-of-water/725.html

PAGE 209

BIOGRAPHICAL SKETCH Luis D. Aponte-Bermdez was born on October 25, 1977, in Bayamn, Puerto Rico, to Benjamn Aponte-Ayala and Griselle M. Bermdez-Garca. After graduating with hamn, Puerto Rico, yagez campment of Civil Engineering and Surveying, and received his bachelors degree in May 2000. Luis then moved to the United States to 2001 Luis and his wife moved to Gainesville,epartment of Architecture and Interior Design to pursuit her masters degree. Luis then started attending in the University of Florida during spring 2002 to pursue his masters and doctoral degrees. In summer 2002 he started to work under the supervision of Dr. Gurley, in May 2004 he received the title of Master Engineer and continued graduate studies for his doctoral degree. After the completion of the doctoral studies Luis will join the University of Puerto Rico at Mayagez campus, as a faculty member in the Department of Civil Engineering and Surveying. onors from the Disciples of Christ Academy in 1995 at Bay Luis moved to Mayagez, Puerto Rico to attend the University of Puerto Rico, Maus, where he was accepted in the Depart he graduated with the highest honors after completing a 5 year engineering curriculum start working with an engineering consulting firm in Silver Spring, Maryland. In fall Florida, where she was accepted in the D 192


Permanent Link: http://ufdc.ufl.edu/UFE0015714/00001

Material Information

Title: Measured Hurricane Wind Pressure on Full-Scale Residential Structures: Analysis and Comparison to Wind Tunnel Studies and ASCE-7
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0015714:00001

Permanent Link: http://ufdc.ufl.edu/UFE0015714/00001

Material Information

Title: Measured Hurricane Wind Pressure on Full-Scale Residential Structures: Analysis and Comparison to Wind Tunnel Studies and ASCE-7
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0015714:00001


This item has the following downloads:


Full Text












MEASURED HURRICANE WIND PRESSURE ON FULL-SCALE RESIDENTIAL
STRUCTURES: ANALYSIS AND COMPARISON TO WIND TUNNEL STUDIES
AND ASCE-7













By

LUIS DAVID APONTE-BERMUDEZ


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA


2006

































Copyright 2006

by

Luis D. Aponte-Bermudez

































This document is dedicated to all the people that had been affected by the impact of
hurricanes and to all the students and professors that had collaborated in the research
effort presented in this dissertation.















ACKNOWLEDGMENTS

I would like to extend my gratitude to all the people that have made possible the

accomplishment of this work. Especially, I would like to thank my advisor, Dr. Kurtis

Gurley, for his wisdom and support for the last couple years. I would also like to thank

Dr. Timothy Reinhold for his vision and work in the wind engineering community, who

initiated the research work presented in this dissertation. Many thanks go also to the

members of my committee and fellow researcher Dr. Forrest Masters for his closeness. In

addition I thank Dr. Prevatt and Zhuzhao Liu for their work and contribution in the

research work presented. I thank the University of Puerto Rico for its financial support in

my doctoral studies and the Florida Department of Community Affairs, National Oceanic

and Atmospheric Administration, Florida and South Carolina Sea Grant for funding this

research. For their prayers, unconditional and loving support, I thank my wife Cassandra,

my parents, my grandparents and my extended family.
















TABLE OF CONTENTS



A C K N O W L E D G M E N T S ................................................................................................. iv

LIST OF TABLES .................................................... ............ ............. .. viii

LIST OF FIGURES ......... ........................................... ............ ix

ABSTRACT ........ .............. ............. ...... ...................... xvi

CHAPTER

1 IN TR OD U CTION ............................................... .. ......................... ..

Background on Hurricanes and Coastal Zone Residential Structures in Florida .........1
Synopsis of the Florida Coastal Monitoring Program ...............................................2
FCM P Tow er R research ............................................................................. 4
FCMP House Research................................................. 7
FCM P Post-Hurricane Damage Assessment.....................................................12
Scope of R esearch............ .................. .................... .. .... ............... .. .... .. 13
Sum m ary of Original Contributions .................................. ............... ............... 14

2 QUANTIFIYING WIND LOADS ON STRUCTURES: BACKGROUND .............. 16

Background on Wind Tunnel Modeling of Pressure Coefficients..............................16
Definition of Pressure Coefficient C, .................................................. 16
Wind Tunnel Procedure for Calculation of C ........................ ............... 18
Important Considerations & Assumptions in Wind Tunnel vs. Full-scale
Analysis....................................................... .. ...... .... ........ 21
Background on ASCE-7 Wind Load Provisions ................................................22
Peaks as Functions of Sam pling Frequency .............................. .... ............... .... 25
The role of Full-scale and New Wind Tunnel Data to Evaluate Current
Assumptions in ASCE-7 based on the Stathopoulos work.............................. 25
Previous Full-scale Projects........ ................. ................................... ............... 26
Q uezon City Experim ents......................................................... ............... 27
A ylesbury H ou se .............................. .... ...................... .. ...... .... ..... ...... 28
Silsoe Structures Building ............................................................................. 29
T exas T ech B uilding................... .. .................................... ...... .. ............. 30
Full-scale Measurements on the Kern P. Pitts Center.......................................31









F C M P D ata Set.......................................................................33
New Wind Tunnel Studies (Clemson) to Accompany the Full-scale Data ...............34
C lo sing R em ark s .................................................................................... 3 5

3 FCMP DEPLOYMENT HISTORY, ORGANIZATION AND LOGISTICS FOR
THE 2004/2005 SEASONS ........... .. .................................. 37

H hurricane F rances (2004) ........................................ .............................................38
Synoptic History .................. ......... ............................ ............... 38
H house and Tow er D eploym ent............................................ .......................... 38
H u rrican e Iv an (2 004)......................................................................... .................. 4 0
Synoptic History ............ .... ........................ ... ...............40
House and Tower Deployment...... ....................... ...............41
H hurricane Jeanne (2004)....................... ....... .................................... ............... 42
Synoptic History ............ .... ........................ ... ...............42
House and Tower Deployment...... ....................... ...............43
H hurricane D ennis (2005) ................................................. ............................... 45
Synoptic History ............ .... ........................ ... ...............45
H house and Tow er D eploym ent............................................ .......................... 46
H hurricane W ilm a (2005).................................................. ............................... 47
Synoptic History ............ .... ........................ ... ...............47
H house and Tow er D eploym ent...................................................................... 49
H hurricanes K atrina and Rita (2005)....................................... ......................... 51
Closing Remarks .............. .... ......... ..................51

4 ANALYSIS OF FULL-SCALE DATA TO DEFINE PRESSURE
COEFFICIENTS ......... ... .......... .......................................... 53

Calculating C, for Full-scale Data: Methods and Outstanding Issues...................53
A applied Equations for C ................... ............ ............................54
Uncertainty: Data Sources can change from Storm to Storm and House to
H o u se ......................... ..... .... ... .......... .........................6 3
Example Peak Minimum C, Calculation with Uncertain Reference Velocity..64
Identification of Appropriate Sampling Rate .............. ..... ....................66
Study for Identification of Appropriate Sampling Rate ...................................67
Influence of Sampling Rate on Peak C, value.............................. ............... 68
Influence of Sampling Rate on Spatial Correlation of Peaks............. ..............71
Correlation coefficient Px,y .......................... ... ............ ............. 71
Coherence function C ) .............. ........................................... 73
"Peak-score" m ethod................... ......................................................... 74
"Peak-score" method: application to full-scale data ...................................81
Uncertainty of Reference Velocity: Estimating Peak Wind Speed Gust ..................83
Wind Speed Reference from House Anemometer: Optimizing Roughness .......84
Wind Speed Reference from House Anemometer: Optimizing Peak Factor......85
Uncertainty on Instantaneous Roof Dynamic Pressure ..............................................89









U uncertainty of R reference Pressure ............................... ........................................... 91
Combined Effects of Uncertainties......................... ................................... ....... 92
Quantification of Uncertainties on Cp: Monte Carlo Approach..............................93
Number of M onte Carlo Simulations .......................................................... 95
An Example M onte Carlo Simulation................................. ............. ........... 96
Presentation of Results from Uncertainty Analysis..............................................98
Closing R em arks .................. .................................... .............. ... 103

5 COMPARISON OF FULL-SCALE DATA WITH WIND TUNNEL STUDY
A N D A S C E -7 -0 5 .................................................................. .......... ................. ... 10 4

Clemson Wind Tunnel Study vs. Full-scale (Same Wind Direction).....................104
Comparison of Mean and RMS Pressure Coefficients.................................... 106
Comparison of Peak Pressure Coefficients ................................................ 111
Interpretation of R results ........................................................ ............. ..115
Concluding Remarks .................. ........................................ .... ......... 116

6 PROPOSED GIS TOOLBOX FOR FCMP APPLICATIONS ............................118

W h at is G IS ............. ... ...... ............. ...................................................... 1 18
FCMP Deployment Maps ........... .. ......... ........................ 118
FCM P Aerial Pictures of Tower Sites ........................................ ......... ............... 1 19
GIS Toolbox to Assist Randomized Damage Evaluation Studies..........................120
O ptim ized Statistical Sam pling ......... ........ ................ ......... ..................... 123
Recommendation and Current Deployment Strategies..............................125

7 CONCLUSION AND RECOMMENDATIONS ............................................. 127

C, values Comparisons (Full-scale vs. Wind Tunnel vs. ASCE-7-05)....................127
Future and Ongoing Analysis........................... ... .. ................128
Workshop and Benchmark Study with all Major Wind Tunnel Facilities ........129
Improvements to House Data Collection System...............................................130
M maintenance to FCM P H ouses ...................................................... .... ........... 131

APPENDIX

A AERIAL PICTURES AND MAPS FOR 2004 & 2005 DEPLOYMENTS ............132

B W IN D C A L C U L A T IO N .......................................................................................... 176

C DYNAMIC PRESSURE COEFFICIENT CALCULATIONS.............................178

L IST O F R E F E R E N C E S ...................................................................... ..................... 189

BIOGRAPHICAL SKETCH ............................................................. ............... 192
















LIST OF TABLES


Table page

2-1 Suggested values of roughness lengths z0 for various types of terrain ....................20

4-1 Sensor list for FCMP house FL-27 hurricane Ivan (2004) .................................59

4-2 Sensor list for FCMP house FL-30 hurricane Ivan (2004) .................................60

4-3 Summary of 12 possible cases to compute full-scale pressure coefficients............64

4-4 Minimum peak pressure coefficients for FL-27 hurricane Ivan (2004) channel 6 ..66

4-5 Difference % between 20 Hz and 2 Hz minimum C for channel 6 of FL-27 .......71
P

4-6 Difference % between 20 Hz and 2 Hz minimum Cp for channel 23 of FL-27 .....71

4-5 Classification of optimum z0 and peak factor for the FL-27 house Ivan (2004).....86

4-6 Classification of optimum z0 and peak factor for the FL-30 house Ivan (2004).....87

4-7 Minimum peak pressure coefficients for FL-30 hurricane Ivan (2004) channel
15, reference w ind speed variation...................................... ......................... 89

4-8 Expected minimum peak Cp (20 Hz) for channel 6 for the FL-27 hurricane Ivan
(2 0 0 4 ) ............................................................................ 9 1

4-9 Minimum peak pressure coefficients for FL-30 hurricane Ivan (2004) channel
15, reference pressure variation ........................................ .......................... 92

4-10 Mean and standard deviation of random variables for FL-27 record #135
hurricane Ivan ................................................... ................. 95

5-1 Records for wind tunnel vs. full-scale comparison...........................................105

C-1 Sensitivity data of sensor used in the FL-27 house during Ivan (2004) ...............83

C-2 Sensitivity data of sensor used in the FL-30 house during Ivan (2004)...............84















LIST OF FIGURES


Figure p

1-1 Tropical Cyclone activity in the State of Florida since 1950 2005..........................3

1-2 Pictures of FCMP Towers (a) FCMP Tower deployed during Hurricane Charley
(2004) (b) FCMP Convoy for Katrina (2005)................... ...................6

1-3 FCMP Tower Deployment Activity since 1999 2005 and major hurricane
activity for the 2004 05 seasons. ........................................ ........................ 7

1-4 FCMP Houses locations along: Florida, South Carolina and North Carolina ..........9

1-5 Pictures of house work. (a) FCMP personnel prepares cable (b) PVC piping
sy ste m ...................................... ................................................... 1 0

1-6 Pictures of house components. (a) House disconnect box (b) House computer
b o x ................... ......................................................... ................ 1 0

2-1 Power Law Wind velocity profiles for various a values............... ....................19

2-2 FCM P House ID FL-27 model scale 1:50.................................... ............... 21

2-3 Aylesbury experimental building with 22.50 pitch roof ...................................28

2-4 Texas Tech University Wind Engineering Research Field Laboratory ...................31

2-5 Kern P. Pitts Center located at Southern Shores, NC........................................32

3-1 FCMP Deployment map of hurricane Frances (2004)...........................................39

3-2 FCMP Deployment map of hurricane Ivan (2004). .......................................41

3-3 FCMP Deployment map of hurricane Jeanne (2004) .....................................44

3-4 FCMP Deployment map of hurricane Dennis (2005) ............................................46

3-5 FCMP Deployment map of hurricane Wilma (2005).............................................50

4-1 Voltage time history of functional data channel 0 for the FL-27 house during
hurricane Ivan (2004) ............ ... ..... ...... ........... .... .. .. ...... ...... 55









4-2 Voltage time history of a malfunctioning data channel 4 for the FL-27 house
during hurricane Ivan (2004)......................................................... ............... 55

4-3 Record #129 of house FL-27 hurricane Ivan (2004), channel 5 malfunctions.........56

4-4 Roof sensor layout configuration for FCMP House FL-27............... .......... 58

4-5 Roof sensor layout configuration for FCMP House FL-30...................................58

4-6 Channel 6 peak minimum C vs. resample frequency. (a) For record #131 wind
direction is 1110 (b) For record #147 wind direction is 1800................................69

4-7 Channel 23 peak minimum C vs. resample frequency. (a) For record #131
wind direction is 1110 (b) For record #147 wind direction is 1800........................70

4-8 Correlation coefficient for record #131 of FL-27 hurricane Ivan (2004) ................73

4-9 Coherence function estimate for record #131 of FL-27 hurricane Ivan (2004).......74

4-10 Random data signals. (a) Original signals x any, (b) Normalized signals x, and
y ..................................................................................... 7 6

4-11 Resultant signal w, = (x + y )/2 .................................... .......... ............... 77

4-12 Empirical cumulative distributions function for x, and y, .................................78

4-13 Empirical cumulative distributions function for w, ...........................................78

4-14 "Peak-Score" check for uncorrelated data x and y .............................................79

4-15 "Peak-Score" check for full-correlated data x and x .............................................80

4-16 "Peak-Score" check for partial correlated data x and w .......................................80

4-17 "Peak-Score" for channel 6 & 7 in record #131 FL-27 hurricane Ivan (2004)........81

4-18 "Peak-Score" for channel 6 & 5 in record #131 FL-27 hurricane Ivan (2004)........82

4-19 "Peak-Score" for channel 6 & 23 in record #132 FL-27 hurricane Ivan (2004)......82

4-20 Aerial picture for the FL-27 house showing optimized z0 and peak factor values
for hurricane Iv an (2004) .............................................................. .....................86

4-21 Aerial picture for the FL-30 house showing optimized z0 and peak factor values
for hurricane Iv an (2004) .............................................................. .....................87









4-22 Coefficient of variation, on 100 trials vs. number of simulations N........................96

4-23 Histogram of minimum C, (20 Hz duration) from Monte Carlo simulation for
channel 6 on record #135 FL-27 hurricane Ivan (2004), N = 1000 ......................97

4-24 ECDF for minimum C, (20 Hz duration) from Monte Carlo simulation for
channel 6 of record #135 FL-27 hurricane Ivan (2004)...........................................98

4-25 Channel 6 time history of min C values for 20 Hz (FL-27 Ivan 2004) ...............99

4-26 Channel 7 time history of min C values for 20 Hz (FL-27 Ivan 2004) ...............99

4-27 Channel 13 time history of min C values for 20 Hz (FL-27 Ivan 2004)............100

4-28 Channel 18 time history of min C values for 20 Hz (FL-27 Ivan 2004)............100

4-29 Channel 22 time history of min C values for 20 Hz (FL-27 Ivan 2004)............101

4-30 Channel 23 time history of min C values for 20 Hz (FL-27 Ivan 2004)............101

5-1 Scale model (1:50 )of the FL-27 house and surrounding structures ......................105

5-2 Mean C full-scale vs. wind tunnel comparison for 1100 wind direction............108

5-3 Mean C full-scale vs. wind tunnel comparison for 1200 wind direction............108

5-4 Mean C full-scale vs. wind tunnel comparison for 1300 wind direction............109

5-5 RMS C full-scale vs. wind tunnel comparison for 1100 wind direction............109

5-6 RMS C full-scale vs. wind tunnel comparison for 1200 wind direction............110

5-7 RMS C full-scale vs. wind tunnel comparison for 1300 wind direction............110

5-8 Min C full-scale vs. wind tunnel comparison for 1100 wind direction .............112

5-9 Min C full-scale vs. wind tunnel comparison for 1200 wind direction .............112

5-10 Min C full-scale vs. wind tunnel comparison for 1300 wind direction .............113

5-11 Max C full-scale vs. wind tunnel comparison for 1100 wind direction.............113

5-12 Max C full-scale vs. wind tunnel comparison for 1200 wind direction.............114









5-13 Max C full-scale vs. wind tunnel comparison for 1300 wind direction...............114

6-1 Deployment map of Hurricane Katrina (2005), generated before and updated
during landfall .................................... ............................... ..........119

6-2 Aerial picture of FCMP mobile tower T1 deployed in Bella Chasse, LA during
hurricane Katrina (2005). See also Figure 6-1 .................................... ........120

6-3 Wind Swath of Hurricane Ivan 1-minute sustained gust....................................122

6-4 User interface window of the FCMP survey toolbox developed in ArcMap.........124

6-5 Map of Charlotte County showing the selected houses using the FCMP survey
to o lb o x ................................. ......................................................... ............... 12 5

A-i FCMP Deployment map of Hurricane Charley (2004)............ ...............133

A-2 Aerial imagery of the terrain surrounding tower TO during Hurricane Charley
(2004) at L akeland, FL ................................................ ............................... 134

A-3 Aerial imagery of the terrain surrounding tower T1 during Hurricane Charley
(2004) at Fort M eyers, FL .................. ............ ....................................135

A-4 Aerial imagery of the terrain surrounding tower T2 during Hurricane Charley
(2004) at O sprey F L ...................... .... .................................................... 136

A-5 Aerial imagery of the terrain surrounding tower T3 during Hurricane Charley
(2004) at Plant City, FL ............................................ .. .......... .... ............. 137

A-6 FCMP Deployment map of Hurricane Frances (2004) .................................... 138

A-7 Aerial imagery of the terrain surrounding tower TO during Hurricane Frances
(2004) at Port Salerno, FL ............................................. ............................. 139

A-8 Aerial imagery of the terrain surrounding tower T1 during Hurricane Frances
(2004) at Port Salerno, FL ............................................. ............................. 140

A-9 Aerial imagery of the terrain surrounding tower T2 during Hurricane Frances
(2004) at V ero B each, FL .............................................. ............................. 141

A-10 Aerial imagery of the terrain surrounding tower T3 during Hurricane Frances
(2004) at Fort Pierce, FL ........... .................. ........ ............................... 142

A-11 FCMP Deployment map of Hurricane Ivan (2004) ............................................143

A-12 Aerial imagery of the terrain surrounding tower TO during Hurricane Ivan
(2004) at M obile, A L ......................... ...... ................ ............... .... .......... 144









A-13 Aerial imagery of the terrain surrounding tower T during Hurricane Ivan
(2004) at P ensacola, FL ................................................ .............................. 145

A-14 Aerial imagery of the terrain surrounding tower T2 during Hurricane Ivan
(2004) at Fairhope, AL .......................................................... ............... 146

A-15 Aerial imagery of the terrain surrounding tower T3 during Hurricane Ivan
(2004) at D estin, FL ............. ....................... ........ ...... ... ............ 147

A-16 FCMP Deployment map of Hurricane Jeanne (2004)........................................ 148

A-17 Aerial imagery of the terrain surrounding tower TO during Hurricane Jeanne
(2004) at Orlando, FL............... .... ........................ .......149

A-18 Aerial imagery of the terrain surrounding tower T during Hurricane Jeanne
(2004) at Sebastian, FL ...................... ................ ............. ..............150

A-19 Aerial imagery of the terrain surrounding tower T2 during Hurricane Jeanne
(2004) at M erritt Island, FL ......................................................... ............... 151

A-20 Aerial imagery of the terrain surrounding tower T3 during Hurricane Jeanne
(2004) at V ero B each, FL .............................................. ............................. 152

A-21 FCMP Deployment map of Hurricane Dennis (2005) .......................................153

A-22 Aerial imagery of the terrain surrounding tower TO during Hurricane Dennis
(2005) at Navarre, FL ..... ........................................................... .... 154

A-23 Aerial imagery of the terrain surrounding tower T1 during Hurricane Dennis
(2005) at Inlet B each, FL .............................................. ............................. 155

A-24 Aerial imagery of the terrain surrounding tower T2 during Hurricane Dennis
(2005) at P ensacola, FL ................................................ .............................. 156

A-25 Aerial imagery of the terrain surrounding tower T3 during Hurricane Dennis
(2005) at P ensacola, FL ................................................ .............................. 157

A-26 Aerial imagery of the terrain surrounding tower T5 during Hurricane Dennis
(2005) at N iceville, FL .................. ............................. .. .. .... .. ........ .. 158

A-27 FCMP Deployment map of Hurricane Katrina (2005)................. .................... 159

A-28 Aerial imagery of the terrain surrounding tower TO during Hurricane Katrina
(2005) at B ay St. L ouis, M S.......................... .......................... ............... 160

A-29 Aerial imagery of the terrain surrounding tower T1 during Hurricane Katrina
(2005) at Bella Chasse, LA ............ ............. ................... 161









A-30 Aerial imagery of the terrain surrounding tower T2 during Hurricane Katrina
(2005) at G alliano, L A ........................................................... .. .................... 162

A-31 Aerial imagery of the terrain surrounding tower T3 during Hurricane Katrina
(2005) at P ascagoula, M S............................................... ............. ............... 163

A-32 Aerial imagery of the terrain surrounding tower T5 during Hurricane Katrina
(2005) at G ulf Port, M S........................... .................. .................. ............... 164

A-33 FCMP Deployment map of Hurricane Rita (2005)............................. .............165

A-34 Aerial imagery of the terrain surrounding tower TO during Hurricane Rita (2005)
at P ort A rthu r, T X ......................................................................... ................... 16 6

A-35 Aerial imagery of the terrain surrounding tower T during Hurricane Rita (2005)
at Houston, TX .................................... ............................... ....... 167

A-36 Aerial imagery of the terrain surrounding tower T3 during Hurricane Rita (2005)
at N ederland, TX .................................................... ..... .. ........ .... 168

A-37 Aerial imagery of the terrain surrounding tower T5 during Hurricane Rita (2005)
at O ran g e, T X ................................................... ................ 16 9

A-38 FCMP Deployment map of Hurricane Wilma (2005) ................. ... ..................170

A-39 Aerial imagery of the terrain surrounding tower TO during Hurricane Wilma
(2005) at Everglades City, FL ...................................................... .............. 171

A-40 Aerial imagery of the terrain surrounding tower T1 during Hurricane Wilma
(2005) at W eston, F L ....................... .. .... ................ .............................172

A-41 Aerial imagery of the terrain surrounding tower T2 during Hurricane Wilma
(2005) at O choppi, FL .................. ......................................... ........ 173

A-42 Aerial imagery of the terrain surrounding tower T3 during Hurricane Wilma
(2005) at M iam i, FL ................................................ .. ...... .. ............ 174

A-43 Aerial imagery of the terrain surrounding tower T5 during Hurricane Wilma
(2005) at Naples, FL....... ......................................... .... .............. 175

C-1 FCM P Instrumented houses in the state of Florida................................................179

C-2 FCMP Instrumented houses in the state of South and North Carolina ................180

C-3 Roof sensor layout of the FL-27 House ................. ....... ...... ............... 181

C-4 Aerial imagery of FCMP House ID FL-27 at Gulf Breeze, FL ...........................181

C-5 Roof sensor layout of the FL-30 House ............. ............. ................ ............ 182









C-6 Aerial imagery of FCMP House ID FL-30 at Pensacola Beach, FL................182

C-7 Mean pressure coefficients for FL-27 during hurricane Ivan (2004) .....................185

C-8 RMS pressure coefficients for FL-27 during hurricane Ivan (2004) ...................185

C-9 Min peak pressure coefficients 20 Hz for FL-27 during hurricane Ivan (2004).... 186

C-10 Max peak pressure coefficients 20 Hz for FL-27 during hurricane Ivan (2004) ..186

C-11 Mean pressure coefficients for FL-30 during hurricane Ivan (2004).....................187

C-12 RMS pressure coefficients for FL-30 during hurricane Ivan (2004) ...................187

C-13 Min peak pressure coefficients 20 Hz for FL-30 during hurricane Ivan (2004).... 188

C-14 Max peak pressure coefficients 20 Hz for FL-30 during hurricane Ivan (2004) ... 188













Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy

MEASURED HURRICANE WIND PRESSURE ON FULL-SCALE RESIDENTIAL
STRUCTURES: ANALYSIS AND COMPARISON TO WIND TUNNEL STUDIES
AND ASCE-7

By

Luis D. Aponte-Bermudez

August 2006

Chair: Kurtis R. Gurley
Cochair: Gary R. Consolazio
Major Department: Civil and Coastal Engineering

Each year hurricanes cause devastating damage along the southeast coastline of the

United States, where the State of Florida is one of the most vulnerable. The work

presented in this dissertation is the result of full-scale measurements conducted during the

last six Atlantic Hurricane Seasons (1999-2005). The primary objective was to quantify

over-land near-surface hurricane wind velocity and uplift loads on residential structures

using full-scale experiential methods.

The research goal is to help reduce hurricane wind damage to residential structures

by providing "ground-truth" data about the intensity of the wind, the resultant loads on

residential structures, and the performance of these structures in high winds. The full-

scale hurricane data measurement was conducted with two separate data collection

systems. The first system consists of portable weather towers deployed in the path of

landfalling hurricanes to capture the wind field behavior at a height of 5 and 10 meters, as

well as temperature, humidity, rainfall and barometric pressure. The second system uses









pressure sensors to collect wind pressure data on the roofs of occupied residential

structures along the Florida and the Carolinas coastlines. To date, 32 houses along the

Florida coastline, 4 along South Carolina, and 2 along the North Carolina coastline have

been outfitted with these sensors. The data collected from these houses are compared to

wind tunnel model studies on scale models of the subject homes.

During the hurricanes of 2004 and 2005 several data sets were collected from

homes that experienced sustained hurricane level winds. A total of 16 homes were

instrumented during 3 of the 2004 storms, and 6 homes were instrumented over 3 storms

in 2005. Data were collected from 9 of these homes in sustained hurricane level winds, a

first in experimental wind engineering. Details are provided regarding the deployment of

the portable towers and the instrumentation of the coastal homes and analysis of this full-

scale data are presented along with comparison between wind tunnel models and the

ASCE-7 wind load provisions. Implications regarding the current state of knowledge of

extreme wind loading in low-rise structures are provided. Preliminary analysis of the full-

scale vs. wind tunnel homes presented in this dissertation suggests that it may potentially

impact wind load standards. These comparison studies suggest that the peak negative

pressure coefficient obtained from the full-scale data exceeded the ASCE-7 coefficient

(component and cladding) for the corresponding roof zones of high suction areas.














CHAPTER 1
INTRODUCTION

Background on Hurricanes and Coastal Zone Residential Structures in Florida

Historically hurricanes in the Atlantic basin refer to tropical cyclones that form in

the North Atlantic Ocean, Caribbean Sea and the Gulf of Mexico, north of the equator,

usually in the Northern Hemisphere, summer or autumn. The Atlantic hurricane season

officially runs from June 1 to November 30. The U.S. National Hurricane Center

monitors the basin and issues reports, watches and warnings about tropical weather

systems for the United States, where other countries around the basin track and issue

warnings for tropical weather in their territories.

The State of Florida's coastline consist of 580 miles out of the total 2,069 miles in

the Atlantic coast portion, and 770 miles out of the 1,631 in Gulf of Mexico coast portion

of the U.S.; this represents 28% and 47% respectively (Teachervision.com, 2006). This

represent a 36% of the total coastline from the Texas to Maine, consequently Florida is

the state with the largest coastline exposure to Tropical Cyclone activity in the Atlantic

basin.

In terms of coastal population, from the 2003 U.S. Census Bureau and Woods &

Poole Economics (W&PE) Inc., the Southeast and the Gulf of Mexico coastal areas are

the least populated regions in the United States with a 9% and 13% respectively, where

the most populated coastal areas in the U.S. correspond to the Northeast with a 34%

followed by the Pacific with a 26% and the Great Lakes with and 18%, but none of these

regions (with the exception of Northeast) are in danger of Tropical Cyclones in the









Atlantic basin. The Southeast has progressively become a primary destination for retirees

and job-seekers. Florida showed the greatest percent coastal population change of any

coastal region by percentage between 1980 and 2003. On average in the last decade, new

residential construction in Florida in a given year represents 2% of the total residential

infrastructure. This growth rate is not projected to slow in the near future, thus protecting

this increasing population from the effects of hurricanes is a growing priority.

Historically Florida is one of the Southeast states that suffer direct impact of

hurricanes. Figures 1-1 shows a map with the Tropical Cyclones that had made direct

landfall in the state of Florida, where for the last two years (2004/05) the Tropical

Cyclone activity in the Atlantic basin has been above the normal historical average,

causing this billion of dollars in the damage.

The research presented in this dissertation contributes to understanding and

minimizing the effects of hurricane wind loads on residential structures.

Synopsis of the Florida Coastal Monitoring Program

The Florida Coastal Monitoring Program (FCMP) began as a full-scale research

effort in 1998 in response to a need in the wind engineering community to better

understand hurricane wind loading on low-rise structures. The program started at

Clemson University, with the University of Florida joining the research team in 1999. At

present the team also includes Florida International University, Florida Institute of

Technology, and the Institute for Business and Home Safety (IBHS) in Tampa, Florida.

Funding support has been provided by the Florida Department of Community Affairs

(DCA), the National Oceanic and Atmospheric Administration (NOAA), the Federal

Emergency Management Agency (FEMA), The Florida Building Commission (within the










DCA), Florida Sea Grant, South Carolina Sea Grant, and the Institute for Business and

Home Safety.


Alabama
Georgia






Florda





Hurricanes over Florida 1950-2005
CAT







Hurricane Landfalls in Florida (1950 2005)



Figure 1-1 Tropical Cyclone activity in the State of Florida since 1950 2005

The FCMP research activities consist of four distinctive projects. Two consist of

separate full-scale data collection systems to collect ground level wind speeds and uplift

pressures on the roofs of residential structures. These will be referred to as FCMP Tower

and FCMP House. The third consists of post-hurricane damage evaluations to provide

data relating winds speed to damage. The fourth project is a destructive testing program

on houses, which allows researches to quantify actual construction methods and

vulnerability reducing retrofits to their limit states.

These four complementary projects provide information needed to develop a more

wind resistant infrastructure by quantifying hurricane wind behavior, extreme wind









loading, and the vulnerability of both existing construction methods and proposed cost-

effective retrofits and code improvements. The next section will provide details about the

first two activities in which the author has been an active participant.

FCMP Tower Research

A source of great uncertainty in the design of hurricane resistant structures is the

actual winds these structures are subjected to, in terms of both sustained winds and the

magnitude and frequency of gusts. The commonly used Saffir-Simpson storm intensity

rating scale is based upon sustained 1-minute averaged wind speeds over open water. As

the wind transitions from open water to land, the influence of terrain and infrastructure

alters the behavior of the wind. For example, a Category 3 storm may only produce

sustained winds of 100 mph over land, although the SS-scale rates Category 3 as at least

111 mph sustained winds. Conversely, the gustiness of the wind increases over land,

which can adversely influence the loads on structures. This disconnect between the public

perception of hurricane intensity and the actual loads experienced by structures is an

outstanding issue to address as researchers investigate means to mitigate wind damage.

The FCMP Tower project is designed to quantify this wind behavior through direct

measurements. Instrumented portable towers are placed in the path of oncoming extreme

wind events to quantify instantaneous wind speed and direction as the hurricane

approaches and impacts land, and moves inland.

The FCMP full-scale wind velocity data collection system consists at present of six

portable towers designed and built at Clemson University (Poss, 2000). The towers are

stored in the off-season at the University of Florida where upgrades and maintenance are

performed by the FCMP team members.









During deployment, each tower relays summary data every 15-minutes to a public

access website for use by NOAA researchers, emergency managers, and risk modelers. In

2006 this cellular communication-based real-time system is being augmented with

satellite transmission modulators to transfer data directly via-satellite to NOAA's

Geostationary Operational Environmental Satellite (GOES).

The FCMP mobile towers (Figure 1-2) are designed to meet U.S. Department of

Transportation (DOT) requirements for transport as a conventional trailer, and withstand

peak gust wind speed of 90 m/s (200 mph), which corresponds to a strong Saffir-Simpson

Category 5 (Simpson and Riehl, 1981). The FCMP towers' mobility and easy assembly,

approximately twenty minutes by a three man crew allow deployment in almost in any

terrain exposures condition.

Tower instruments are located at three levels (3, 5 and 10 m). The data acquisition

system measures 3D wind speed and direction at the top two levels and collects

temperature, rainfall, barometric pressure, and relative humidity data at the tower's base.

Two RM Young anemometry systems-a wind monitor and a custom array of three gill

propellers-collect data at the 10-m level, which the World Meteorological Organization

deems as the standard wind speed observation height. A second array of gill propellers

collects wind speed data at the 5-m level to measure winds at the approximate eave

height of a single-story home. The tower power system has been upgraded from its

original contractor-grade gasoline generator with diesel generators to provide power for a

period up to 48 hours. A series of UPC batteries provide an extra 8-10 hours of power

when the generator runs out of fuel.









The data acquisition system consists of two separate computer systems. The first

consists of a PC that collects the data at a sampling rate of 100 Hz, and stores it digitally

onto two separate hard drives. The second system consists of a laptop computer that

collects the data at sampling rate of 10 Hz, and stores the data onto a single hard drive.

This system was incorporated in the 2003 hurricane season and it is responsible to

connect to the Internet via cellular modem and upload statistics summaries through

various ftp sites. More details are presented by Masters (2004).

.. .... ... .. ... ... ..
iii..... .ii .iiiiiiiiiiiiiii .i















(a) (b)
Figure 1-2 Pictures of FCMP Towers (a) FCMP Tower deployed during Hurricane
Charley (2004) (b) FCMP Convoy for Katrina (2005)

The FCMP tower data set contains information from Tropical Cyclones Georges

(1998), Dennis (1999), Floyd (1999), Irene (1999), Gordon (2000), Gabrielle (2001),

Michelle (2001), Isidore (2002), Lili (2002), Isabel (2003), Bonnie (2004), Charley

(2004), Frances (2004), Ivan (2004), Jeanne (2004), Dennis (2005), Katrina (2005), Rita

(2005) and Wilma (2005). Facts about the FCMP deployments since 1998 to 2003 are

presented by Masters (2004); details about the 2004/05 seasons will be presented in this

dissertation. This database to date has provided a crucial source of information when










defining the vulnerability of the infrastructure. Figure 1-3 shows a map of all the tower

deployment activity since 1999 2005 and the major hurricane activity of the 2004 05

seasons.



FCMP Towers by year, storm name


""- -" .. 7T ...... "-7






; ::: 1W :. .













FCMP Deployments for Atlantic Hurricane Seasons of 1999 2005


Figure 1-3 FCMP Tower Deployment Activity since 1999 2005 and major hurricane
activity for the 2004 05 seasons.

FCMP House Research

Just as the knowledge of wind speeds over land is largely uncertain despite

advances in hurricane tracking and intensity forecasting, the wind loads associated with

extreme wind speeds are also lacking in first-hand direct quantification. ASCE-7 wind

load provisions are based largely on small scale wind tunnel replications conducted on

simple structural shapes. There are limits regarding the accuracy of extrapolating the

loads measured in these experiments to full-scale, and in particular the interaction of
exrm win "pesaeas2'cigi is-ad ietqatfcain SE7wn

lodpoisosaebae aglyo ml saewn tne eliain onutdo
simplestrucural sapes.There re liits reardin the acu acyo xrpltn h

loads~ ~ ~ ~ ~ ~ ~~~~~~~~7 -esrdi'-s xeietst ulsae n nprtclrteitrcino









extreme wind and resultant loading on more typical complex shapes found in the existing

housing infrastructure.

The FCMP house project was created to bridge the gap between wind tunnel based

wind load provisions and the actual loads experienced by full-scale structures. This

project consists of measuring wind pressure data directly on the roof, soffit and walls of

pre-selected residential houses along the coast of Florida during land falling hurricanes.

The homeowners of these occupied structures agreed to collaborate with the FCMP

researchers in exchange for retrofits to improve the home's resistance to hurricane winds.

Modem high wind rated shingles, wind and impact rated garage doors, and window

shutter systems are typical of these incentives.

The current FCMP house catalog contains a total of 32 houses in the state of

Florida, 4 in the state of South Carolina and 2 in the state of North Carolina. Figure 1-4

shows the distribution of the houses among these three states. The location of these

houses had been carefully selected using the historical frequency of land-falling

hurricanes in these regions. The homes are spaced at intervals of 16 to 24 km (10 15

mi), and most are within 1.5 km (1 mi) of the coastline. Typically the houses are one or

two stories tall have composite shingle roof covering, and the surrounding areas are

suburban and relative free of tree cover. Aerial pictures for houses presented in this

dissertation are available in Appendix A.

The house instrumentation is prepared during the off season. This consists of

installing roof brackets to attach the sensors, and exterior wiring to connect the sensors to

the computer system. Figure 1-5 (a) shows the wires for the individual sensors, and

Figure 1-5 (b) shows the plastic piping containing these wires, installed under the










overhang. All of the wires meet at a disconnect box shown in Figure 1-6 (a), to which the

computer data collection system is attached as shown in Figure 1-6 (b). The pressure

sensors are mounted on the roof, and the computer system is installed within days of an

approaching hurricane and retrieved immediately after the event.


J. ,, ,-.














FCMP Houses in Florida FCMP Houses in North and South Carolina

Figure 1-4 FCMP Houses locations along: Florida, South Carolina and North Carolina

The computer is contained in a 60" Steel Jobsite Box shown in Figure 1-6 (b). The

inside is customized to accommodate a PC, CPU batteries (to provide up to 24-36 hours

of power), time lapse VCR (to record images of the house during the storm) and

miscellaneous tools. The box's final weight is around 300 lb, heavy enough to resist high

winds.

The data acquisition system measures data at a sampling rate of 100 Hz. The data

are stored digitally into two independent hard drives every 15 minutes. The field

instrumentation per house consists of a maximum of twenty eight Microswitch 142 PC-

15 absolute pressure transducers. Reference pressure sensors are located inside the house

attic and at ground level (yard sensor). In addition, depending to the house configuration,






10


3-cup-anemometers are installed on the roof of the house, using a 54 inch stem extension

from the roof eave.


I '- 7
(a) (b)
Figure 1-5 Pictures of house work. (a) FCMP personnel prepares cable (b) PVC piping
system


I.I...W __


(a) (b)
Figure 1-6 Pictures of house components. (a) House disconnect box (b) House computer
box

The sensors are sheltered in a 12 in diameter aluminum pan (roof) or square plastic

box (wall & soffit), distributed along the roof, soffit, walls and camera mounting base









plate (yard sensor). Detailed technical information is presented by Michot (1999). Some

of the most relevant points are voltage signal resolution of pressures to +0.005 psf, an

arbitrary offset voltage that varies from sensor to sensor the transducer calibrations are

sensitive to temperature changes. Michot reported a temperature adjustment factor of

0.0144 Volts/F for a sensor circuit of sensitivity of 20.6 psf/Volt.

FCMP house sensor calibration tests have been performed in order to find the

sensitivity for each sensor. A sealed box allows calibrating twelve sensors at a time.

Using an electric air compressor, suction is applied to the box to provide a known

pressure. Voltage data from the sensors are collected during a 15 minute interval at a

sampling rate of 100 Hz. During the test, pressure inside the box is monitored using a

Setra Digital Pressure Gage Model 370. The pressure then is varied inside the box over a

range around 810 to 970 mbars. A linear regression is then applied to the sensor voltage

data and known pressure measured using the RM Young pressure sensor, thus providing

the sensitivity (calibration) factor for each sensor.

FCMP analysis of house data collected during hurricane Ivan (2004) will be

presented in Chapter 4. Additional data was collected during Frances and Jeanne (2004),

and Dennis and Wilma (2005), and is the subject of ongoing analysis. This dissertation

will focus on the development of a probabilistic approach to the analysis of the full-scale

data, which will be employed for future analysis of these data sets.

Companion research at Clemson University focuses on constructing scale models

of the homes that collected data in 2004 and 2005, and conducting wind tunnel studies of

these models. The loads observed from these tests will be compared with those loads









measured in full-scale in an attempt to identify shortcomings in the model testing

procedures used to determine wind load provisions.

The analysis of full-scale pressure data collected during actual hurricanes presents

numerous challenges in terms of offering accurate results, reasonable uncertainty

quantification, and a fair comparison to wind tunnel studies. This is a major topic in this

dissertation.

FCMP Post-Hurricane Damage Assessment

In addition to the quantification and modeling of hurricane winds and structural

loading, the assessment and documentation of damage to residential structures after a

hurricane event are necessary to identify existing weaknesses and potential solutions.

During the spring of 2005, the FCMP teams were joined by Florida A & M

University (FAMU) students to conduct an extensive in-field evaluation of homes

impacted by the 2004 hurricanes in Florida. The strategy was to randomly sample

addresses located within designated peak wind bands for evaluation, thus providing an

unbiased survey of damage to typical structures of various ages (Gurley et al., 2006). The

random sampling strategy relied upon a database of county-wide residential house

information. The randomization process was a rudimentary random order scheme.

The outcomes of the study have been very revealing in terms of comparing

structural performance as a function of construction age, and peak wind speed. However,

since the study was conducted months after the storms occurred, much of the data

collection was based upon homeowner interviews rather than first hand observations of

damage.

Ideally such extensive damage assessments would be conducted in the days

immediately after the storm impacts a region in order to capture perishable data that is









best observed first-hand. In order to capitalize on the use of a randomized sampling

strategy, a portion of this dissertation deals with the development of a GIS-based tool to

employ a series of sampling strategies stratified by wind speed, age, region, construction

type, or a combination. This tool contains the database of residential housing from most

of the counties in Florida. With such a tool in hand, the next intense hurricane to impact

Florida will be immediately followed by a thorough damage assessment that is designed

to provide statistically relevant samples of damage to stratifications as determined by

researchers in the field.

Scope of Research

This document presents the FCMP research efforts in further detail, and highlights

the original contributions of the author to these efforts.

Chapter 2 provides a historical background on the quantification of wind loads,

including wind tunnel modeling, ASCE-7, and the role of new full-scale pressure data.

Chapter 3 will discuss full-scale data collection methods and present a summary of

data collected since 2004.

Chapter 4 presents the methods developed for, and results produced from the

analysis of full-scale pressure data collected in Florida during hurricane landfall. There

are two focuses in this chapter. The first is the identification of a suitable sampling rate

for analysis. The second is the identification, modeling and quantification of the influence

of uncertainty in the calculated pressure coefficient values to define confidence limits.

Analysis of the data collected during Hurricane Ivan (2004) on a low-rise residential

structure is presented.

Chapter 5 presents a comparison of the full-scale results with those of wind tunnel

studies of the FCMP houses as well as ASCE-7. The full-scale pressure coefficients will









be bounded by confidence intervals as determined by Monte Carlo-based methods

developed for this research. This comparison of probabilistic full-scale pressure

coefficients, from homes in sustained hurricane winds, with results currently used wind

tunnel methods and ASCE load provisions, is the first such study available in the open

literature.

Chapter 6 presents the GIS applications developed by the author to improve

deployments strategies and the quality of post-hurricane damage surveys.

Finally, Chapter 7 summaries conclusions on the FCMP data analysis of full-scale

vs. wind tunnel vs. ASCE7-05 code parameters for low-rise structures, and presents

comments and suggestion for future research works.

Additional FCMP house analyses are presented in the Appendix C for data sets

collected during the events of hurricane Frances (2004), Jeanne (2004), Ivan (2004),

Dennis (2005) and Wilma (2005).

Summary of Original Contributions

The research contributions include:


The development of new analysis methods for full-scale pressure data:

Identification of uncertainties in the collection and analysis of full-scale data
Development of a probabilistic pressure coefficient that incorporates these
uncertainties
Development of a systematic analysis method to apply to each house data set
Identify the ideal down-sampling rate of full-scale data based on observed
peak loads at individual sensors, and correlated peaks at multiple spatially
separate sensors
Analysis of the only full-scale residential pressure data sets available


The first comparative study of full-scale, wind tunnel, and ASCE-7 wind loads on
structures under sustained hurricane winds:






15


* Comparative study of peak pressure values from full-scale and scaled wind
tunnel results
* Evaluation of the accuracy of ASCE-7 for houses in hurricane prone regions
* Evaluation of the effects of terrain on pressure coefficients
The development of new GIS-based methodologies and functional framework for:

* Analysis and presentation of post-damage survey data
* Design and automated selection for damage survey studies in future storms















CHAPTER 2
QUANTIFIYING WIND LOADS ON STRUCTURES: BACKGROUND

This chapter presents background information on quantification and modeling of

wind pressure on low-rise structures. The estimation of pressure coefficients from wind

tunnel studies is first discussed, followed by the application to wind load provisions.

Finally the role of full-scale data and modern wind tunnel methods are presented.

Background on Wind Tunnel Modeling of Pressure Coefficients

This section provides background information concerning the modeling of wind

pressure on low-rise structures. It presents basic concepts and the need for verification

and improvement via the full-scale contributions presented later in the document. Some

of the early works presented in this section had tremendous influence on the development

of wind load provisions for low-rise structures.

Definition of Pressure Coefficient C
P

The pressure coefficient C, is a non-dimensional value which acts as a means of

indicating the local pressure at some point of interest around a body, and which is

independent of velocity. It is defined in equation 2-1.

C -P-P.
pU2 (2-1)
2 pU2
Where: p is the pressure at the point of interest, p, is the free stream pressure or

reference pressure, p is the fluid density and U is the reference or representative

velocity near the point of interest. Positive pressure acts toward the surface and negative

pressure acts away from the surface. The pressure coefficient can represent an average









pressure over a prolonged period, the fluctuating pressure about the mean, or minimum /

maximum peak values over a defined duration of constant average pressure. These

C values commonly evaluated from wind tunnel studies are referred to as mean (C-),

RMS (C-), peak minimum (C.) and peak maximum (C.) values, and are calculated


using mean, fluctuating, or peak values ofp in Eq. 2-1.

In practice, the pressure coefficient is applied by rearranging 2-1 such that the

pressure differential in the numerator (the design wind load) can be determined with

knowledge of the pressure coefficient and reference wind speed.

Cp changes from one location to another on the same structure in the same wind

field. Factors that influence the value of Cp beyond the parameters in 2-1 include size,

roughness and orientation (relative to wind direction) of the surface at which Cp is being

determined, the location on that surface (near an edge or in the middle), the turbulence

gustinesss) of the approaching wind field, and the terrain surrounding the structure

(related to the turbulence).

Typically a model structure is subjected to a turbulent wind field with constant

average speed and the pressure coefficients are calculated over a finely spaced grid of

different locations over the surface of the structure. This is repeated over a series of wind

directions, producing a surface of Cp values for a given wind direction. The worst case

pressure coefficient at a given location is then selected from the many wind directions,

enveloping the effects of wind from any direction.









Wind Tunnel Procedure for Calculation of C

This section summarizes the extensive work conducted by researches to model

wind pressure loading in low-rise structures in boundary layer wind tunnels. It also

presents some of the details of the FCMP houses modeled at the Clemson University

Wind Load Test Facility and some details of the work conducted at the University of

Western Ontario (UWO).

First a scale model is constructed, where the model size depends on the dimensions

and velocity scales of the boundary layer wind tunnel. Normally scale models are

constructed of Plexiglas and instrumented with pressure taps that sit flush to the surface

in the roof and walls, which allows researchers to heavily instrument scale models in

areas of interest like high suction zones (e.g. roof covers and edges). Clemson University

researchers gather enough information from the full-scale experiment conducted by the

FCMP, such as aerial pictures, topographic maps and detailed measured of the structures,

to build a scaled geometric model of the house (shown in Figure 2-2) and other typical

neighboring houses constructed of foam.

The second step consists of selecting and evaluating the wind speed profile that will

be replicated to scale in the wind tunnel. This is also referred to as matching the approach

roughness. This roughness is quantified as a roughness length zo. Table 2-1 presents

typical roughness length for common exposures types. As the roughness increases, the

mean speed of the wind drops more rapidly from high (well above the structure) to low

elevations (near the structure's roof). This is often modeled as a log-law or power-law

relationship shown in equation 2-2.












U()= U( zg2
zg2)


(2-2)


Where the exponent a in the equation is dependent upon the roughness and zg


and zg2 denotes heights above ground. The level of turbulence increases with lower


elevation more rapidly with higher roughness values. Figure 2-1 shows the wind velocity

profiles for different values of a .


PoFer Law- ind Velocity Prfiles


0


0.20


-- Coastal Areas a = 0.10

-- Open Terrain a- =0.14

-A-- Slurhan Terrain a 0.22

-- Center of Large Cities a 0.33


0.60

Uzgl)


Figure 2-1 Power Law Wind velocity profiles for various a values.

Since both mean and fluctuating wind speed effects Cp, care must be taken to


properly represent the roughness of the terrain in the model study. If the scale model in

Figure 2-2 is placed in a wind tunnel that replicates smooth open terrain, and the actual

full-scale data from that home are from suburban terrain, direct comparisons of data sets

are less meaningful.










The reference velocity to be used in the denominator of equation 2-1 is then

determined by measurements in the wind tunnel. This is typically taken to represent the

expected peak value of wind speed for a given duration of time, located at the mean roof

height. This is typically quantified in one of two ways: 1) Measure the wind speed at the

top of the boundary layer wind tunnel, well above the model. Then by applying the

power-law relationship defining wind speeds as a function of elevation, the wind speed at

mean roof height is computed. This is the methodology used by Clemson University

Wind Tunnel Experiments. 2) Another approach to estimate the reference velocity at

mean roof height is presented by Kopp (2005) at the University of Western Ontario. The

peak velocity at mean roof height, is defined as Vh = V + gV, where V is the mean roof


height velocity, V is the root mean square (standard deviation) of the velocity fluctuation,

and g is the peak factor, taken as a nominal value of 3.0. V and V are measured in the

wind tunnel at the mean roof height without the model in place.

Table 2-1 Suggested values of roughness lengths z0 for various types of terrain
Towns,
Sparsely Centers of
Type of ,b Densely
p Coastalab Openb Built-up D Large
Terrain b Built-up b
Suburbs Suburbsb Cities
Suburbs
zo(m) 0.005 -0.01 0.03 -0.10 0.20 -0.40 0.80 1.20 2.00 3.00
SApplicable to structures direct exposed to winds blowing from open water
b Values of z0 to be used in conjunction with the assumption z d = 0


The model is then placed on a turntable in the wind tunnel and data are collected at

different wind angles 0 to 3600 by rotating the turntable between tests. This is done for

multiple runs for a given wind direction. Finally by applying equation 2-1 the mean,

RMS, minimum and maximum peak pressure coefficients are computed using the

measured wind tunnel data.





















Figure 2-2 FCMP House ID FL-27 model scale 1:50

Important Considerations & Assumptions in Wind Tunnel vs. Full-scale Analysis

The evaluation of peak pressure coefficients in wind tunnel and full-scale scenarios

is related to the sampling rate duration, where higher (minimum and maximum) peaks are

expected for faster sampling rates. This relationship eventually levels off at very fast

sampling rates.

Another consideration is the correlation of peaks over larger areas to represent the

aggregate uplift over large structural components (e.g. sheathing). At faster sampling

rates, a given peak pressure observed at one sensor may not be correlated to nearby

sensors. in view of the fact that the pressure coefficient in load provisions are intended to

represent aggregate loads, for example over a 4x8 piece of sheathing, identifying spurious

peaks uncorrelated over small distances is not ultimately useful. The identification of an

appropriate sampling rate between the model and full-scale data comparison will be

addressed in detail in chapter 4.

The model scale and the mean wind speed are also issues when considering the

sampling frequency to be used when comparing full-scale and wind tunnel data. The

dimensionless proportion in equation 2-2 describes the desired relationship between full-

scale and model scale data.









UfM fU (2-2)
BI Model nBI Full-scale

Where Uis the roof mean wind speed, n is the sampling rate and B is a physical

dimension (for example in a 1:50 scale model B will have a value of 1 in the model and a

value of 50 in the full-scale). Typically the full-scale speed and dimension values are not

controlled by the researcher, and the model scale is selected for reasons associated with

wind tunnel test section size. Thus the full-scale sampling frequency, the model sampling

frequency, and the mean wind speed used in the wind tunnel can be controlled to

maintain this relationship.

Another concern between the model and full-scale data are the issue of stationary

data, where stationary refers to mean and fluctuating components of both the wind speed

and direction that do not change over significant time duration. This is easily controlled

in the wind tunnel as wind direction and speed are within user control, and the turbulence

remains at a constant rate determined by the approach roughness in the tunnel. At full-

scale the experiment cannot be controlled in this manner. Wind speed, direction and

turbulence can fluctuate dramatically over short periods of time. The challenge is to

identify segments of data from the larger record that meet the stationary requirement. For

example, full-scale data where wind direction does not vary beyond a 5 degree arc and

the mean speed remains within 5% of its starting value is very hard to obtain in a 30

minute period but it is possible to obtain in a short period of time. In order to take into

account the stationary issues a duration time of 15 minute will be used.

Background on ASCE-7 Wind Load Provisions

The current Wind Load Provisions of ASCE-7 are based on the wind tunnel study

work of Theodore Stathopoulos conducted in the late 1970s at the Boundary Wind









Tunnel Laboratory in the University of Western Ontario (UWO) (Stathopoulos, 1979).

Stathopoulos' work consist of two terrain models, the first a "smooth" exposure which

produces a profile appropriate to open country conditions whereas the second "built-up"

exposure correspond to suburban conditions. The roughness length z0 in the wind tunnel

was estimated to be 4 x103 inches [ 1.0 x 104 m] for open country (smooth) and

34 x 10 3 inches [ 8.6 x 104 m] for suburban (built-up) exposures. Both of these values

are very low compared to the currently used values for open country and sparsely built-up

suburban in the range of (0.03 0.10 m) and (0.20 0.40 m) respectively. These values

suggest that the wind field velocity profile used during the experiments is best described

by today's standards as a coastal terrain exposure, which reflect a change in thinking

since the late 1970s (Kopp, 2005).

The characteristics of the boundary layer flow modeled in the wind tunnel dictate

the appropriate length and velocity scales for a rigid pressure model. For the Stathopoulos

study these are approximately 1:500 and 1:5 respectively. Thus the time scale is of the

order 1:100. The frequency response of the pressure measurements system use in

Stathopoulos' work was capable of modeling full-scale fluctuations up to about 1 Hz and

the sampling rate for each pressure signal exceeds and equivalent full-scale sampling rate

of about 10 samples per second (Stathopoulos, 1979).

The peak measurements considered thru the study were base on a single extreme

value recorded rather than an average peak over several trials, based on the narrow

differences for multiple trials with a coefficient of variation not exceeding 10%. Kopp

(2005) presents that the corresponding model length scale was relaxed for the

experiment, and models with length scales of 1:500, 1:250, and 1:100 were investigated









in the same flow. For example, the model-scale integral scale is effectively halved if a

1:250 model is used in a 1:500 scale flow simulation. Stathopoulos and Surry concluded

that the length scale relaxation to 1:250 was the largest possible without significant

distortion of the results. The resulting database from the Stathopoulos experiments was,

thus, largely based on the test data obtained from 1:250 scale model tests of three roof

slopes (1:12, 4:12, and 12:12) and three eave heights (4.9, 7.3, and 9.8 m) in an open

country exposure. However, the pressure coefficients from this work have been long used

to describe the loading on houses in a wide variety of terrains. ASCE-7 allows a wind

speed reduction with built up exposure, but applies the same pressure coefficients

determined from these open country wind tunnel experiments. Analysis of both full-scale

data and recent wind tunnel experiments suggest that pressure coefficients for

components and cladding increase in built-up terrain, perhaps to the point of negating the

load reduction allowed due to lower wind speed (Reinhold, 2005)

The wind tunnel data was collected using state-of-the-art measurement techniques

and equipment of the time at a sampling rate of 1000 Hz at 450 intervals and was then

filtered using a low pass cut off frequency of 95 Hz. The sampling time was 30 s for the

1:250 data, which corresponds to a full-scale time of about 25 min and full-scale

frequency of 2 Hz. This is a point of debate since preliminary analysis of the FCMP full-

scale data show that much higher sampling rates have been found to produce well

correlated peaks; this issue will be addressed in more details in chapter 4.

There is a need to re-evaluate the current application of the Stathopoulos data set

based on recent findings from both full-scale and wind tunnel experiments. It is justified









to evaluate the effects of using faster sampling rates and a range of exposure profiles

beyond open country.

Peaks as Functions of Sampling Frequency

In order to compare Stathopoulos data with the FCMP model/full-scale data it is

important to define an equivalent sampling frequency to match the conditions of the wind

tunnel experiment. A full-scale sampling frequency of 2 Hz was calculated to be

appropriate to compare with the Stathopoulos data presented on the ASCE-7. However, it

is also important to present the results of the full-scale analysis in a format that best

reflects the capabilities of modern instrumentation. Beyond fair comparisons of this data

to the Stathopoulos work, the full-scale data will be down-sampled to the maximum rate

capable of capturing correlated peaks across spatially separated sensors. This issue will

be address in more details in chapter 4.

The role of Full-scale and New Wind Tunnel Data to Evaluate Current Assumptions
in ASCE-7 based on the Stathopoulos work

Modem methods allow pressure coefficient calculations from data directly

measured on full-scale residential structures under extreme winds. The role of full-scale

measurements in quantifying wind pressure loadings on low-rise structures during

landfalling hurricanes is fundamental to advancing the current state of knowledge of

extreme wind loading. Simiu & Scanlan (1996) presents some of the early works of

model/full-scale comparisons by Richardson (1991), in which comparisons for pressure

on low-rise structures (gable-roof buildings) suggest that the wind tunnel does not

accurately model the flow separation on the windward roof, so roof pressures often differ

significantly between model and full-scale. Tieleman (1998) states that: because exact

understanding of vortex formation and development under separated shear layers is









lacking, the success of any simulation technique must be based on model/full-scale

comparison of observed surface pressures and not on some preconceived ideas of flow

simulation alone.

It has been found in most of the early comparison works that is fairly easy to match

mean and rms pressure coefficients between model and full-scale, but the difficulty is

found when comparing the extreme minimum (negative peak) pressures observed near

roof corners and leading edges. This is attributed to inadequate simulation of the

longitudinal and lateral turbulence intensities, their small-scale turbulence content, as

well as Reynolds effects. Tieleman (1998) performed a comparative study of the full-

scale Texas Tech Building, and concludes that agreement between model and field roof

pressure coefficient is only possible when duplicating the two horizontal turbulence

intensities and their small-scale turbulence content, and the model turbulence scale

exceeds one fifth the magnitude of the scaled-down field scale.

Previous Full-scale Projects

Previous research efforts in the Wind Engineering community concerning the

quantification of wind pressure loading on low-rise building had been carried out

extensively by wind tunnel modeling and full-scale instrumentation in the last thirty

years. The most relevant full-scale research in this area are:

Quezon City experiment in the Philippines (1970's)

The structure building work in Aylesbury England (1970's)

The structure building work in Silsoe England (1980's)

Texas Tech Field Wind Engineering Research Field Laboratory (1980-90's)









To complement these full-scale experiments, wind tunnel modeling had been

conducted in order to evaluate and improve wind tunnel simulation techniques. At

present ongoing research includes the full-scale measurements on the Kern P. Pitts Center

which is being conducted by the Department of Civil and Environmental Engineering,

University of Illinois at Urbana-Champaign, and the Department of Civil Engineering,

Johns Hopkins University, Baltimore, MD (Porterfield and Jones 2001). The FCMP

house program is also ongoing. Analysis of the data sets collected during the Atlantic

Hurricane Season of 2004 and 2005 will be presented on this dissertation. Simultaneously

wind tunnel models of the FMCP houses are being conducted at Clemson University by

Liu (PhD Candidate of the Department and Civil Engineering) under the supervision of

Dr. Prevatt. A summary description of each one of the previously mentioned projects is

now presented.

Quezon City Experiments

In the 1970's The National Bureau of Standards (NBS) with the collaboration of

the Philippine Atmospheric, Geophysical and Astronomical Services Administration

(PAGASA) performed full-scale testing on three single family dwellings in Quezon City,

Philippines. The goal of this research was to provide new data set to improve the design

criteria used to the design of low-rise structures and validate wind tunnel studies.

Wind tunnels studies where conducted at the Virginia Polytechnic Institute and

State University. These studies provided key finding between the full-scale data and the

wind tunnel model such as strong correlation in the mean and rms pressure coefficient

values, which provided a high degree of confidence to the wind tunnel simulations. More

details about the findings are presented by Marshall (1976), who reported proper intensity

of turbulence is a key factor in generating realistic surface pressure fluctuations on the









wind tunnel model. The positive agreement between model and full-scale dimensionless

pressure coefficients, probabilistic distributions, and power spectra suggests that valid

wind data can be obtained from wind tunnel test which utilize a relatively large scale and

suitable roughness (Marshall 1976).

Aylesbury House

In the early 1970's the Building Research Establishment (BRE) of the United

Kingdom constructed a full-scale two story house (Figure 2-3) to serve as a data

collection site for field wind pressure studies on typical low-rise residential buildings.

During the years of 1972/74 a large amount of data was collected. Holmes (1982) provide

details of the experiment such as: building site location on flat open field, plan

dimensions of 7 m x13.3 m, height to the eaves of 5 m, uniquely feature to adjust the roof

pitch angle from 5 to 450, pressure transducers positions and characteristics of the full-

scale runs.




















Figure 2-3 Aylesbury experimental building with 22.50 pitch roof
Figure~~~~~~. ..... .................idi g it 2 .5 p tc r o









The Aylesbury building is consider a milestone in the wind engineering field, and

the initiative for international wind tunnel models conducted around the world included

the United Kingdom, the United States, Canada, Australia, and Japan among others. At

the time this full-scale date set provided the best database on wind pressure on low-rise

building (Holmes, 1982).

Silsoe Structures Building

The Silsoe Structures Building (SSB) was constructed during 1986/87 at the Silsoe

Research Institute (SRI), specifically to undertake full-scale wind pressure

measurements. The building was constructed with an optimal eaves geometry offering

either traditional sharp eaves or curved eaves of 635 mm radius. The building geometry

consists of 24 m long by 12.9 m span by 5.3 m ridge height building with a 100 duo-pitch

roof. The surrounding flat terrain is mainly open-country site.

Experimental data sets presented by Hoxey (1997) consist of two independent

data collection systems. The first conducted by SRI consist of a total of 77 pressure

tapping points were installed on the building and measurements were made using a

sequence controller which sampled two pressures at one time. A total of 4 records were

collected, each one with duration of 4 minutes, which provided 4 mean values of wind

pressure data. A complete definition of pressures over the building surface required many

hours of recording. The second set of data was conducted by the Building Research

Establishment (BRE) which consists of a total of 32 pressure tap points, each mounted

locally on the inside of the building. The improvements of this second method allow

recording simultaneous measurements with wind velocity sensed by a three-component

sonic anemometer. Records of one hour duration were made, and these were partitioned









into six 10 minutes records for the analysis. The typical average reference velocity at the

building ridge height measurement was 10 m/s.

Wind tunnel studies were conducted at the University of Western Ontario (UWO)

and the Building Research Establishment (BRE), described in more detail by Hoxey

(1997). Comparisons between the full-scale and wind tunnel models were restricted in the

report to comparing mean pressure coefficients, since the mean values were able to be

determined with statistical confidence. From the wind tunnel and full-scale analysis

Hoxey suggest that wind tunneled may be used to estimate wind loads over the majority

of the surface of a building, however regions of separated and conical-vortex flow likely

to produce underestimations of surface pressures at model scales.

Texas Tech Building

In the late 1980's Texas Tech University (TTU) researchers constructed the Wind

Engineering Research Field Laboratory (WERFL) in Lubbock, Texas. The facility

consists of a permanent experimental building mounted on a turntable. The main goal

was to provide full-scale data for comparison with wind tunnel model studies. The

building configuration consists of a rectangular flat roof with dimension of 30 ft x 45 ft x

13 ft presented in Figure 2-4. The building location site consists of flat open terrain

exposure. A guyed tower is located 150ft west of the building, to provide meteorological

data such as wind speed, relative humidity, barometric pressure and temperature.

Dearhart (2003) provide a well documented comparison of the extensive wind

tunnel model studies that had been carried out to compare with the full-scale data sets.

Some of these were conducted by: Colorado State University (CSU), The University of

Western Ontario (UWO), the Building Research Institute (BRI) of Japan, and the Wind

Load Test Facility (WLTF) at Clemson University.









Cope (1997), stated that several studies have been conducted to compare the full-

scale mean, peak, and rms pressure coefficients. Most comparisons match well, but

distinct discrepancies have been found between model and full-scale rms and peak

minimum pressure coefficients at certain locations, particularly, the areas where high

suction occurs due to the shear layer or vortex development exhibited higher peak

negative coefficients in full-scale testing than those measured in many wind tunnels.

















Figure 2-4 Texas Tech University Wind Engineering Research Field Laboratory

The discrepancy between the compared data was attributed to the approach flow

characteristics (lateral turbulence intensity, small- and large-scale spectral content of

approach flow fluctuations), Reynolds number effects, frequency response of the pressure

measurement system, and sampling frequency of the acquired data (Bienkiewicz, 1998).

Cope (1997) reported positive agreement on mean, rms and peak pressure coefficients by

correctly matching the full-scale and wind tunnel time duration.

Full-scale Measurements on the Kern P. Pitts Center

The full-scale measurements taken on the Kern P. Pitts Center, a low-rise structure

owned by the Town of Southern Shores (Figure 2-5) in Southern Shores, North Carolina,









were conducted by the University of Illinois and Johns Hopkins University. The research

goals are to collect full-scale wind pressure data to study the wind-induced pressures on

this particular low-rise structure.

The building instrumentation began in 1997. Since that time some of the most

notable data had been obtained from the passage of frontal systems, but data have also

been collected in several northeasters and three land-falling hurricanes (Bonnie 1998,

Dennis and Floyd 1999), though the structure was not subjected to sustained hurricane

force winds. The field site is located in the Town of Southern Shores, on the Outer Banks

of North Carolina, about a quarter mile west of the Atlantic Ocean.

















Figure 2-5 Kern P. Pitts Center located at Southern Shores, NC.

Meteorological data are collected at two locations in the vicinity of the structure.

Wind speeds are measured at 10 meters above ground level: the first system consist of a

three axes ultrasonic anemometer located on a tower 60 feet east of the structure, the

second system consist of a propeller-vane anemometer, located on an instrumentation

pole above the chimney. Barometric pressure, rainfall, and temperature are also









measured. More details about this research is provided by Porterfield, M. and Jones, N.P.

(2001).

FCMP Data Set

The FCMP wind pressure full-scale data set contains hundreds of hours collected

during landfalling tropical cyclones on 22 residential structures. The wind speed ranges

fall into the category of hurricane and tropical storm intensities. These data sets will

allow computing pressure coefficients from direct full-scale measurements and

comparison with wind tunnel models and ASCE-7 wind load provisions. Limitation

common to the other experiments above also exist on the FCMP data sets such as

uncertainties in the calculation of the pressure coefficients, estimation of the roughness

exposure for each house location, and variation of wind direction and wind speed. This

will not necessary produce the worst-case loading condition for the evaluated house, but

the FCMP house database contains valuable information for the wind engineering

community. Since the data was collected at a sampling rate of 100 Hz, this will provide a

better understanding of the extreme peak pressures.

The 2004 full-scale FCMP data collection effort is the first true sustained hurricane

wind pressure data collected on an occupied residential structure. Previous to that the

FCMP captured wind pressure data in two houses during tropical storm Isidore (2002).

The most current database contains wind pressure data for the following hurricanes:

Frances (2004), Ivan (2004), Jeanne (2004), Dennis (2005) and Wilma (2005). Detailed

information for the house and tower deployments during the events of the hurricanes of

the 2004/05 season are presented in Chapter 3. Details concerning some of the issues

previously presented will be discussed in details in chapter 4 along with confidence limits

on the pressure coefficient calculations.









New Wind Tunnel Studies (Clemson) to Accompany the Full-scale Data

Kopp (2005) states that current wind load provisions for low-rise buildings are

based on "reductive plots and tables" that do not allow designers access to the variation

of wind effects with time and space. For most current standards, these plots and tables

were determined by obtaining equivalent pressure coefficients that envelope responses

calculated from wind tunnel data for a range of assumed structural wind resisting

systems. Structural analysis programs are now widely used in the design of structures;

these methods can yield accurate structural effects, and can account for both static and

dynamic loading. However, the wind loads currently available for low-rise buildings

through code provisions do not take advantage of these refined analysis techniques.

Preliminary model/full-scale data comparison, conducted at Clemson University,

on two FCMP house instrumented during Tropical Storm Isidore (2002) suggest that the

ASCE-7 wind load provision underestimate the full-scale negative peak pressure

coefficients for components and cladding on houses in suburban terrain (Dearhart 2003).

The goal of this dissertation is to provide pressure coefficients from the analyzed data for

the vast FCMP house database collected in the last two hurricane seasons of 2004/05 and

compare the full-scale pressure coefficients with the ASCE-7 wind load provisions, in

order to corroborate the preliminary result presented by Dearhart (2003).

Reinhold (2005) suggested from the Dearhart (2003) work that the ASCE-7 peak

negative pressure coefficients, instead of enveloping the expected highest negative peak

pressure, are likely to underestimate the uplift for critical wind directions. This raises

questions about the move towards allowing component and cladding load on low-rise

buildings to be calculated for non-open terrain building exposures using a single set of

pressure coefficients that were developed from testing models in open terrain conditions.









The need to evaluate the current ASCE-7 design wind loads provision is clearly

suggested. The FCMP model/full-scale data analysis is an important step in wind

engineering to improve existing wind loading provision in low-rise structures.

FCMP wind tunnel simulation can directly include terrain influence by mimicking

actual full-scale conditions within the wind tunnel, from turbulence characteristics in the

wind field measured at the houses and at the FCMP towers located nearby. The influence

of surrounding structures and trees can be accounted for in the wind tunnel. The full-scale

measurements only cover a limited range of wind direction for the maximum sustained

wind speeds. The wind tunnel simulations (conducted in 5 degree increments over a full

260 degree range) will provide a means to evaluate the worst-case direction for the full-

scale data.

Chapter 4 will demonstrate that for the mean and RMS pressure coefficient there is

a strong match between the full-scale and the wind tunnel models. This is not the case for

the minimum and maximum peak pressure coefficients, which strongly suggests, along

with the findings of other researchers noted above, that more work needs to be done to re-

evaluate the ASCE-7 wind load provisions for homes in non-open terrain.

Chapter 4 will also present a method, developed for this dissertation, to directly

quantify the effects of uncertainty in the analysis of full-scale data, resulting in a

probabilistic format for the pressure coefficients including rationally-based confidence

limits on peak coefficients. Thus the combined use of full-scale and wind tunnel studies

can produce worst-case full-scale loading with some degree of confidence.

Closing Remarks

Modem ASCE-7 Wind Load Provision base on the wind tunnel work by

Stathopoulos developed in the late 1970s is no longer the state-of-the-art. The equivalent









full-scale sampling rate of 2 Hz is to slow compared with today's instrumentation

standards. The modeled exposure condition is considered by today's standard as coastal

rather than suburban exposure. Current technology has produced full-scale pressures as

well as updated wind tunnel techniques to update the current design methodologies.

The work presented in this dissertation focuses largely on the analysis of the full-

scale data, incorporating new concepts for the identification of a suitable sampling rate,

and the identification, modeling and quantification of the influence of uncertainties on

calculated pressure coefficients (C,) values.














CHAPTER 3
FCMP DEPLOYMENT HISTORY, ORGANIZATION AND LOGISTICS FOR THE
2004/2005 SEASONS

The last two hurricane seasons were prolific in terms of the number of hurricanes to

impact the U.S., the damage caused, and the data collected. The FCMP research effort

collected ground level approach wind speed velocity and wind pressure data on low-rise

residential structures along the coast of the State of Florida. The FCMP deployment

activities are presented in the following section; limited to the events in which house data

was collected. Additional deployment information and a portion of the collected data can

be accessed at the FCMP web site http://www.ce.ufl.edu/-fcmp. The synoptic history and

track data for each cyclone was taken from the National Hurricane Center Tropical

Cyclone Report archives, available at http://www.nhc.noaa.gov.

This dissertation focuses in large part on the application of full-scale pressure data

to evaluate extreme wind loading on residential structures. This chapter will provide

details on the full-scale deployments, and discuss the specific data sets captured in the

last two years that are currently undergoing analysis. Aerial picture of the tower

deployment sites and house are located in Appendix A. Appendix C presents the sensor

roof layout of the instrumented houses along with the sensitivity data of the sensors used.

The analysis of every house instrumented in 2004 and 2005 is not presented in this

dissertation, as much of that work relies upon wind tunnel testing still being conducted at

Clemson University. The content of this chapter serves the purpose of documenting the

house instrumentation-related activities of 2004 and 2005.









Hurricane Frances (2004)

Synoptic History

Frances developed of the coast of Africa on 21 August, becoming a hurricane on

August 26, and the intensification continued until 31 August, when Frances reached a

peak intensity estimated at 125 knots (category 4) as it passed north of the Leeward and

Virgin Islands. It was a category 2 hurricane with winds of 85-90 knots over the

northwestern Bahamas on 3-4 September.

Frances weakened just before Frances made landfall over the southern end of

Hutchinson Island, Florida near 0430 UTC 5 September as a Category 2 hurricane.

Frances weakened as it moved slowly across the Florida Peninsula, and became a tropical

storm just before emerging into the northeastern Gulf of Mexico near New Port Richey

early on 6 September. It made a final landfall near the mouth of the Aucilla River in the

Florida Big Bend region about 1800 UTC 6 September.

House and Tower Deployment

In Hurricane Frances FCMP personnel successfully deploy four mobile towers and

instrumented five houses, distributed along the Florida Atlantic Coast. Figure 3-1

presents a deployment map indicating the location of the towers and instrumented houses

and the track of the hurricane path. The tower labels indicate the highest 3-second gust

recorded at that location. The blue triangle icons are FCMP houses that were not

instrumented. The purple triangle icons are FCMP houses that were instrumented and

collected data. All houses that collected data in Frances were north of the eye on the

strong side of the storm, and one was very close to the center of the storm. Table 3-1

provides details of each tower location, the maximum measured 3-second and 1-minute

gust at each tower with the corresponding time in UTC.

















































Florida Coastal Monitoring Program *FCMP Towers SASOS
Deployment for Hurricane Frances Hurricane Frances *BUOY
September 4-6, 2004 AFCMP Houses C-MAN
For more information visit our web page at http://www.ce.ufl.edu/-fcmp AInstrumented Houses WSR-88D


Figure 3-1 FCMP Deployment map of hurricane Frances (2004)




Table 3-1 Tower Data Records in Hurricane Frances (2004)
TOWER CITY, STATE GPS COORDINATES Max mum 1ind Spee Date Time (UTC) Maximum nd Speed Date Time (UTC)
On Site (i, 10m (1-Min Gust) On Site (i 10m 3-Sec Gust)
27" 08' 53 3" N
TO Port Salemo, FL 0 1'53" 57 33 mph @ 358 9/5/2004 02 05 39 82 23 mph @ 5 9/5/2004 02 40 22
Indian Harbor 28" 08' 41 6" N
T1 Ind Harbor FL 80 08' 41 72 21 mph@ 252 9/5/2004 16 42 47 83 02 mph@ 227 9/5/2004 15 08 30
T Beach, FL 80 35' 49 4" W
T2 Vero Beach, FL 24' 5" 64 02 mph @ 76 9/5/2004 09 14 46 81 29 mph @ 96 9/5/2004 09 58 51
27 26' 50 3" N
T3 Fort Pierce, FL 81 00 mph@ 17 9/5/2004 04 03 02 108 32 mph @ 16 9/5/2004 04 01 21



Table 3-2 House Data Records in Hurricane Frances (2004)
Local Exposure Open Exposure
Measured Estimation*

FCMP Number of House Max Wind Speed Max Wind Speed
House ID City, State Records Start Time (UTC) End Time (UTC) Anemometer (mph) @ (mph) @ 10m
Heigth (m) Anemometer Height
Height Zo = 0.03
3-Sec 1-Min 3-Sec 1-Min
FL-06 Jensen Beach, FL 142 9/5/200414:04:10 9/7/200401:39:04 7.062 91 58 106 86
FL-04 Vero Beach, FL 101 9/4/200421:54:10 9/5/2005 23:08:23 N.A. N.A. N.A. 105 86
FL-03 Vero Beach, FL 175 9/5/2004 14:58:23 9/7/2004 10:43:37 N.A. N.A. N.A. 105 86
FL-02 Melbourne Beach, FL 271 9/3/2004 10:50:40 9/6/2004 06:44:03 N.A. N.A. N.A. 97 79
FL-01 Melbourne, FL 317 9/3/2004 02:14:57 9/6/2004 09:59:54 N.A. N.A. N.A. 87 71
FCMP house database contains 251.5 hrs of data for major storm deployments during 2002 2005
*Estimation performed byAppliedResearch Associates parametric hurricane wind field model (Vickery, et al, 2000)









Table 3-2 provides details of the house city location, the total number of 15 minutes

records collected, anemometer height with maximum 3-second and 1-minute gust (if

available) and Applied Research Associates' parametric Wind Field Model maximum 3-

second and 1-minute gust estimation for open exposure conditions at 10 m height

(Vickery et al. 2000).

Hurricane Ivan (2004)

Synoptic History

Ivan developed off the west coast of Africa on 31 August, becoming Tropical

Storm Ivan at 0600 UTC 3 September. Ivan reached its first peak intensity of 115 knots at

0000 UTC 6 September, making Ivan the southernmost major hurricane on record. Ivan

reached its second peak intensity -- 140 knots and category 5 strength --12 hours later. As

Ivan passed south of Jamaica it weakened to category 4 strength.

Ivan rapidly intensified to category 5 strength a second time 11 September,

weakened back to a category 4 hurricane on 12 September, and re-strengthened to

category 5 for its third and final time when it was about 80 n mi west of Grand Cayman

Island. The hurricane brought sustained winds just below category 5 strength to the

island. This resulted in widespread wind damage, and a storm surge that completely over

swept the island except for the extreme northeastern portions.

On 13 September Ivan moved over the northwestern Caribbean Sea. The very

warm water in that region helped the hurricane maintain category 5 strength for 30 hours.

As Ivan neared the northern U.S. Gulf coast it weakened slowly and made landfall

as a 105 knots hurricane (category 3) at approximately 0650 UTC 16 September, just

west of Gulf Shores, Alabama. By this time, the eye diameter had increased to 40-50 n

mi, which resulted in some of the strongest winds occurring over a narrow area near the







41


southern Alabama-western Florida panhandle border. After Ivan moved across the barrier

islands of Alabama, the hurricane turned north-northeastward across eastern Mobile Bay

and weakened into a tropical storm 12 hours later over central Alabama.

House and Tower Deployment

In Hurricane Ivan FCMP personnel successfully deploy four mobile towers and

instrumented six houses, distributed along the Florida Panhandle and Alabama. Figure 3-

2 presents a deployment map indicating the location of the towers and instrumented

houses and the track of the hurricane path. As-measured peak 3-second wind speeds are

indicated on the graph where reliable data was available. All houses are on the strong side

of the storm.




IGso'lOoz

Alabama
Florida
Mississippi 1.- ,,: _















rLl IL.


Deployment for Hurricane Ivan L .. E
Seplemn er 13-16. 200d H ,""h "


Figure 3-2 FCMP Deployment map of hurricane Ivan (2004).








42



Table 3-3 Tower Data Records in Hurricane Ivan (2004)
TOWER CITY,STATE GPS COORDINATES MaAm Mi Se Date Time (UTC) Oaft dS Date Time (UTC)
On Site (~i 10m (1-Min Gust) OnSite10m 3-Sec Gus)
30 38' 39 9" N
TO Mobile, AL 88 08 1" 49 69 mph @ 327 9/15/2004 08 24 26 72 66 mph @ 320 9/15/2004 08 43 30
880 03 48 1 "W
30 28' 45 4" N
T Pensacola, FL 28 80 29 mph@ 1240 9/16/2004 06 49 59 106 26 mph@ 1240 9/16/2004 06 43 18
870 11 12 8" W
30 28'210"N
T2 Farhope, AL 2821 0 67 96 mph @ 62 9/16/2004 06 44 58 89 23 mph @ 56 9/16/2004 06 44 18
870 52' 30 0" W
30 23' 446" N
T3 Destn, FL NA NA NA NA
86 27' 58 9" W


Table 3-4 House Data Records in Hurricane Ivan (2004
Local Exposure Open Exposure
Measured Estimation*
FCMP Number of House Max Wind Speed Max Wind Speed
House ID City, State Records Start Time (UTC) End Time (UTC) Anemometer (mph) @ (mph) @ 10m
Heigth (m) Anemometer Height
Height Zo= 0.03
3-Sec 1-Min 3-Sec 1-Min
FL-30 Pensacola, FL 219 9/14/2004 23:19:28 9/17/200406:20:23 6.553 91 65 114 93
FL-28 Pensacola, FL 185 9/15/2004 01:56:23 9/17/200400:22:13 N.A. N.A. N.A. 105 85
FL-27 Gulf Breeze, FL 211 9/14/2004 20:04:33 9/17/200401:04:03 6.553 82 45 97 79
FL-26 Navarre, FL 247 9/15/2004 13:22:32 9/18/2004 03:14:05 6.096 59 32 89 73
FL-24 Destin, FL 314 9/13/2004 19:23:33 9/16/2004 18:40:54 6.096 56 31 78 63
FL-23 Destin, FL 303 9/13/2004 19:37:39 9/16/200423:48:49 6.096 57 32 75 61
FCMP house database contains 154.25 hrs of data for major storm deployments during 2002 2005
*Estimation performed byAppliedResearch Associates parametric hurricane wind field model (Vickery, et al, 2000)


Table 3-3 provides details of each tower location, the maximum measured 3-second


and 1-minute gust at each tower with the corresponding time in UTC. Table 3-4 provides


details of the house city location, the total number of 15 minutes records collected,


anemometer height with maximum 3-second and 1-minute gust (if available) and


Vickery's maximum 3-second and 1-minute gust estimation for open exposure conditions


at 10 m height (Vickery et al. 2000).


This Hurricane Ivan data set will be used extensively in later chapters to present the


house data analysis techniques developed for this dissertation. In particular, FL-27 and


FL-30 will be the subject of detailed analyses.


Hurricane Jeanne (2004)


Synoptic History


Jeanne formed from a wave off of Africa on 7 September, strengthening to a


tropical storm on 14 September while it moved over the Leeward Islands. It moved over


the southeastern Puerto Rico on 15 September when maximum sustained surface winds









reached 60 knots. Jeanne was a hurricane with 70-knots winds during the Dominican

Republic landfall, but then weakened over the rough terrain of Hispaniola. Jeanne's slow

forward motion across the Caribbean caused torrential rainfall and flooding along its

path, causing thousands to die in Haiti. While Jeanne was dumping rain over the

Caribbean countries, Hurricane Ivan moved over the Gulf of Mexico and inland across

the southeastern United States.

Jeanne moved slowly northward over the southeastern Bahamas as a tropical storm

and then moved in a loop about 500 n mi east of the northwestern Bahamas. Jeanne

gradually strengthened to a hurricane with 85-knots winds by the time it completed this

loop on 23 September. On 24 September, Jeanne moved over its own previous track from

a few days earlier and encountered cooler waters caused by upwelling from the hurricane.

This is believed to decrease the maximum winds from 85 knots to 70 knots on 24

September. Moving away from the upwelled cooler water, the winds increased to 100

knots (category 3) on 25 September. Jeanne made landfall on the east coast of Florida

early on 26 September with the center of its 50-n mi diameter eye crossing the coast at

the southern end of Hutchinson Island just east of Stuart at 0400 UTC on 26 September.

Maximum winds at landfall are estimated at 105 knots over a very small area north of the

center and it is not clear whether these strongest winds reached the coast or remained

over water.

House and Tower Deployment

In Hurricane Jeanne FCMP personnel successfully deploy four mobile towers and

instrumented four houses, distributed along the Florida east coast. Figure 3-3, presents a

deployment map indicating the location of the towers and instrumented houses and the

track of the hurricane path. Significantly, several of the same houses that collected data











during Frances were again employed for Jeanne, providing the opportunity to analyze the

peak winds on the same structure during two different events. Table 3-5 provides details

of each tower location, the maximum measured 3-second and 1-minute gust at each tower

with the corresponding time in UTC. Table 3-6 provides details of the house city

location, the total number of 15 minutes records collected, anemometer height with

maximum 3-second and 1-minute gust (if available) and Vickery's maximum 3-second

and 1-minute gust estimation for open exposure conditions at 10 m height (Vickery et al.

2000).


26i1 (OZ HL .
FL I

FL-

6I .15C.OZ Florida

2,11 SOOZ 2E. 1 (lOZ I pr,



2i,0900Z 16,70.0Z


26,0500Z 2660300Z



Florida Coastal Monitoring Program *"'* :::
Deployment for Hurricane Jeanne r .
Septc-mber 25-26. 2004 r- H...


Figure 3-3 FCMP Deployment map of hurricane Jeanne (2004)



The lack of anemometer data measured directly at these house locations (due to hip


roof configuration) increases the uncertainty in the analysis of pressure coefficients for












these homes. Nearby tower data or the wind field map data from ARA will be used


instead.


Table 3-5 Tower Data Records in Hurricane Jeanne (2004)
TOWER CITY,STATE GPS COORDINATES MaDueidSped Date Time (UTC) Oant 1dSs Date Time (UTC)
On Site (~i 10m (1-Min Gust) OnSite10m 3-Sec Gus)
28 21' 170"N
TO Orlando, FL 8 2'10 0" 45 34 mph@ 113 9/26/2004 15 39 17 68 51 mph@ 32 9/26/2004 10 59 58
810 26 10 0" W
27 48' 50 7" N
Tl Sebastian, FL 50 7 85 17 mph@ 85 9/26/2004 06 47 39 101 81 mph@ 77 9/26/2004 06 40 24
800 29' 59 5" W
28" 38' 25 7" N
T2 Meritt Island, FL 28 47 48 mph @ 92 9/26/2004 113250 63 77mph@ 99 9/26/2004113211
800 43' 50 0" W
27 39' 20 2"N
T3 Vero Beach FL 82' 9 81 47mph@ 40 9/26/2004041803 106 19 mph@ 39 9/26/2004041739


Table 3-6 House Data Records in Hurricane Jeanne (2004)
Local Exposure Open Exposure
Measured Estimation*
FCMP Number of House Max Wind Speed Max Wind Speed
House ID City, State Records Start Time (UTC) End Time (UTC) Anemometer (mph) @ (mph) @ 10m
Heigth (m) Anemometer Height
Height Zo= 0.03
3-Sec 1-Min 3-Sec 1-Min
FL-02 Melbourne Beach, FL 200 9/26/2004 17:25:19 9/28/2004 19:38:55 N.A. N.A. N.A. 101 82
FL-01 Melbourne, FL 115 9/25/2004 03:16:57 9/26/200408:07:45 N.A. N.A. N.A. 92 75
FL-31 Melbourne, FL 177 9/25/2004 20:34:09 9/27/2004 16:49:35 N.A. N.A. N.A. 87 71
FL-32 Merritt Island, FL 161 9/25/2004 22:51:16 9/27/200415:14:02 N.A. N.A. N.A. 79 64
FCMP house database contains 84.5 hrs of data for major storm deployments during 2002 2005
*Estimation performed byAppliedResearch Associates parametric hurricane wind field model (Vickery, et al, 2000)


Hurricane Dennis (2005)


Synoptic History


Dennis formed from a tropical wave that moved westward from the coast of Africa


on 29 June. The system moved through the southern Windward Islands on 4 July and lost


organization over the southeastern Caribbean. The system reformed, becoming tropical


storm Dennis on 5 July.


Dennis reached hurricane strength early on 7 July, then rapidly intensified into a


Category 4 hurricane with winds of 120 knots before making landfall in southeastern


Cuba on 0245 UTC 8 July. Once offshore the hurricane intensified to Category 4.


Maximum sustained winds reached a peak of 130 knots on 8 July, then decreased to 120


knots before Dennis made landfall in Cuba again at 1845 UTC. Dennis traversed a long


section of western Cuba before emerging into the Gulf of Mexico 9 July. Dennis regained








46



strength over the Gulf of Mexico with maximum sustained winds reaching 125 knots on


10 July. The maximum sustained winds decreased to 105 knots before Dennis made


landfall on Santa Rosa Island, Florida, between Navarre Beach and Gulf Breeze, about


1930 UTC 10 July.


House and Tower Deployment


In Hurricane Dennis FCMP personnel successfully deploy five mobile towers and


instrumented four houses, distributed along the Florida Panhandle. Figure 3-4, presents a


deployment map indicating the location of the towers and instrumented houses and the


track of the hurricane path.




70 2IOZ


Florida


Alabama
FL.26
FL FL '.. ,, .
-', il .1 "U ,',/:n
FL .2




Alabama Georgia

Mississippi


Florida


Florida Coastal Monitoring Program
Deployment for Hurricane Dennis
July 8-10. 2005


* .:,r I r M,,, -, I '- I ,
,;I Ii I H ,,- ,_- rl II J
I F


Figure 3-4 FCMP Deployment map of hurricane Dennis (2005)


1 1 I I 1








47



Like the case from Jeanne in 2004, Dennis offered the opportunity to instrument


houses that had already collected data during a previous storm (Ivan 2004), again


providing load data on the same structures for different events.


Table 3-7 Tower Data Records in Hurricane Dennis (2005)
TOWER CITY,STATE GPS COORDINATES MaO m) Date Time (UTC) On t d S Date Time (UTC)
On Site (a 10m (1-Min Gust)On Site 10m Sec
30 24' 01 0" N
TO* Navarre, FL 3 2' 41 0" N 99 28 mph@ 2800 7/10/2005 07 17 06 120 71 mph@ 2790 7/10/2005 07 15 09
86o 51 49 0" W
Tl Inlet Beach FL 30 000" N 69 55 mph@ 2950 7/10/2005 18 36 51 80 45 mph@ 2890 7/10/2005 18 24 36
30 86' 370152 0"
T2 Pensacola, FL 3028137" N NA NA NA NA
878 11' 15 3" W
T3 Destm, FL 30 2'41 9" N 63 59 mph@ 77 7/10/2005 19 36 51 80 76 mph@ 78 7/10/2005 19 30 55
______ _____________ 86 27 566W ________________ ____
T5 Nceville, FL 30 3'174 38 41 mph@ 1990 7/10/2005 19 54 31 6181 mph 201 7/10/2005 19 41 41
*Measured at 5m from ground


Table 3-8 House Data Records in Hurricane Dennis (2005)
Local Exposure Open Exposure
Measured Estimation*
FCMP Number of House Max Wind Speed Max Wind Speed
House ID City, State Records Start Time (UTC) End Time (UTC) Anemometer (mph) @ (mph) @ 10m
Heigth (m) Anemometer Height
Height Zo 0.03
3-Sec 1-Min 3-Sec 1-Min
FL-24 Destin, FL 130 7/10/2005 11:31:13 7/11/2005 19:57:33 6.096 59 28 N.A. N.A.
FL-26 Navarre, FL 149 7/9/2005 16:24:52 7/11/2005 05:46:58 6.096 43 22 N.A. N.A.
FL-23 Destin, FL 173 7/9/200521:22:53 7/11/2005 16:47:38 N.A. N.A. N.A. N.A. N.A.
FCMP house database contains 80.5 hrs of data for major storm deployments during 2002 2005
*Estimation performed byAppliedResearch Associates parametric hurricane wind field model (Vickery, et al, 2000)


Table 3-7 provides details of each tower location, the maximum measured 3-second


and 1-minute gust at each tower with the corresponding time in UTC. Table 3-8 provides


details of the house city location, the total number of 15 minutes records collected,


anemometer height with maximum 3-second and 1-minute gust (if available) and


Vickery's maximum 3-second and 1-minute gust estimation for open exposure conditions


at 10 m height (Vickery 2000).


Hurricane Wilma (2005)


Synoptic History


Wilma has a complicated story. During the second week of October, an unusually


large circulation and a broad area of disturbed weather developed over much of the


Caribbean Sea. A more concentrated area of disturbed weather and surface low pressure









formed near Jamaica by 14 October. By October 15 the surface circulation became well-

enough defined to designate that a tropical depression had formed, centered about 190 n

mi east-southeast of Grand Cayman.

The depression moved slowly and erratically westward to west-southwestward for

a day or so and then drifted south-southwestward to southward for a day or two. The

system is estimated to have become tropical storm Wilma at 0600 UTC 17 October.

On 18 October Wilma turned toward the west-northwest and strengthened into a

hurricane. Later that day, a remarkable strengthening episode began and continued

through early on 19 October. By 0600 UTC 19 October, Wilma's winds had increased to

near 150 knots (category 5). In the span of just 24 hours, Wilma had intensified from a

60-knots tropical storm to a 150-knots category 5 hurricane, an unprecedented event for

an Atlantic tropical cyclone. Wilma reached its peak sustained wind speed of 160 knots at

around 1200 UTC 19 October. During the strengthening episode, Air Force

reconnaissance observations indicated that the eye of the hurricane contracted to a

diameter of 2 n mi; this is the smallest eye known to National Hurricane Center (NHC)

staff. The estimated minimum central pressure at the time of peak intensity is 882 mbar,

which is a new record low value for a hurricane in the Atlantic basin.

Wilma maintained category 5 status until 20 October, when its winds decreased to

130 knots, and the tiny eye was replaced by one about 40 n mi across. The hurricane

would retain this large eye ranging from about 40 to 60 n mi in diameter, for most of the

remainder of its lifetime. By 21 October the hurricane turned toward the northwest

toward the Yucatan Peninsula. Wilma's maximum winds were still near 130 knots

(category 4) when it made landfall on the island of Cozumel around 2145 UTC 21









October, and was probably only slightly weaker (but still category 4 intensity) when it

crossed the coast of the Yucatan peninsula about 6 hours later. On 22 October the

hurricane moved slowly northward, severely battering the extreme northeastern Yucatan

peninsula. Wilma emerged into the southern Gulf of Mexico around 0000 UTC 23

October, with maximum winds of near 85 knots, still a large and powerful hurricane.

A vigorous southwesterly steering current accelerated Wilma northeastward toward

southern Florida. Wilma strengthened over the southeastern Gulf of Mexico and its winds

reached about 110 knots as it approached Florida. Maximum sustained winds were

estimated to be near 105 knots (category 3 intensity) when landfall occurred in

southwestern Florida near Cape Romano around 1030 UTC 24 October. Moving at a

forward speed of 20 to 25 knots, the hurricane crossed the southern Florida peninsula in

4.5 hours, emerging into the Atlantic just southeast of Jupiter around 1500 UTC.

Maximum winds had decreased to near 95 knots (category 2) during the crossing of

Florida.

House and Tower Deployment

In Hurricane Wilma FCMP personnel successfully deploy five mobile towers and

instrumented two houses, distributed along the Gulf and east coast in the State of Florida.

Figure 3-5 presents a deployment map indicating the location of the towers and

instrumented houses and the track of the hurricane path. The center of Wilma passed over

one of the instrumented houses on Marco Island, and south of the other house near

Naples.

Table 3-9 provides details of each tower location, the maximum measured 3-second

and 1-minute gust at each tower with the corresponding time in UTC.



















































Florida Coastal Monitoring Program FCMP Towers A FCMP Houses SASOS
Deployment for Hurricane Wilma I Hurricane Wilma A Instrumented Houses BUOY
Monday, October 24, 2005 C-MAN
For more information visit our web page at http://www.ce.ufl.edu/-fcmp WSR-88D


Figure 3-5 FCMP Deployment map of hurricane Wilma (2005)




Table 3-9 Tower Data Records in Hurricane Wilma (2005)
TOWER CITY, STATE GPS COORDINATES Ma -iu 1MdiSped Date Time (UTC) OnMa Sit mm Sped) Date Time (UTC)
On Site (i, 10m (1-Min Gust)On Site 10m
TO Everglades City, 25 54 03 "N 71 68 mph@ 293 10/24/2005 12 27 56 93 82 mph@ 291 10/24/2005 12 27 41
FL 81 18'41 0"W
Tl Weston, FL 26 08 450N 86 85 mph@ 145 10/24/2005 12 07 10 104 69 mph@ 141 10/24/2005 11 48 57
80_ 30'_24 O"W
25" 52' 05 0"N
T2 Ochoppl, FL 80 55 00W 81 85 mph@ 279 10/24/2005 13 12 02 109 11 mph@ 276 10/24/2005 13 11 46
80 53' 59 0"W
25" 45' 05 6"N
T3 Miami, FL' 05 69 87 mph@ 92 10/24/2005 14 12 02 96 04mph@ 104 10/24/2005 13 46 55
80 22' 25 7"W
26 09' 07 4"N
T5 Naples, FL 81 46' 37 6"-W NA NA NA



Table 3-10 House Data Records in Hurricane Wilma (2005)
Local Exposure Open Exposure
Measured Estimation*

FCMP Number of House Max Wind Speed Max Wind Speed
House ID City, State Records Start Time (UTC) End Time (UTC) Anemometer (mph) @ (mph) @ 10m
Heigth (m) Anemometer Height
Height Zo= 0.03
3-Sec 1-Min 3-Sec 1-Min
FL-18 Marco Island, FL 189 10/22/2005 18:31:47 10/24/2005 17:59:36 6.096 77 46 N.A. N.A.
FL-19 Naples, FL 176 10/22/2005 20:25:27 10/24/2005 16:35:57 N.A. N.A.. A. I N.A. N.A.
FCMP house database contains 91.25 hrs of data for major storm deployments during 2002 2005
*Estimation performed byAppliedResearch Associates parametric hurricane wind field model (Vickery, et al, 2000)


BM W M230W 0VW 8130'0DW 810DW O300W BO0W W300W









Table 3-10 provides details of the house city location, the total number of 15

minutes records collected, anemometer height with maximum 3-second and 1-minute

gust (if available) and Vickery's maximum 3-second and 1-minute gust estimation for

open exposure conditions at 10 m height (Vickery et al. 2000).

Hurricanes Katrina and Rita (2005)

Florida was not directly impacted by Rita, and only suffered the beginnings of

Katrina in south Florida as it passed in to the gulf. Thus no houses collected data from

either of these storms.

The FCMP did successfully deploy the portable towers to collected wind speed

data from these two strong storms. For details and results, interested readers are referred

to http://www.ce.ufl.edu/-fcmp.

Closing Remarks

The full-scale pressure data analysis techniques presented in this dissertation were

developed using data collected in the Panhandle during Hurricane Ivan, 2004. An initial

review of the complete data set from 2004 and 2005 revealed the site-specific nature of

the required analysis (details in a later chapter), and justified a focus on data sets that

were among the most complete and reliable. Identifying appropriate values for each of

the terms in the pressure coefficient (presented earlier in equation 2-1), involve some

judgment and a quantification of uncertainties associated with measurement and

conditions. This process became the central focus of the dissertation research as well as

the evaluation of the conditions under which a fair comparison can be made among full-

scale, wind tunnel, and ASCE-7 loads.

Chapter four presents the analysis of the full-scale pressure data with an emphasis

on a particular house in the Panhandle the endured sustained hurricane winds during Ivan






52


in 2004. The assumptions are explained, uncertainties are quantified, and new techniques

are developed to provide a probabilistic view of extreme wind loads on residential

housing.














CHAPTER 4
ANALYSIS OF FULL-SCALE DATA TO DEFINE PRESSURE COEFFICIENTS

This chapter will present the methods developed and results from the analysis of

full-scale pressure data collected in Florida during hurricane landfall. There are two

focuses in this chapter. The first is the identification of a suitable sampling rate for

analysis. The second is the identification, modeling and quantification of the influence of

uncertainty in the calculated Cp values.

Calculating Cp for Full-scale Data: Methods and Outstanding Issues

Each FCMP full-scale dataset for a particular instrumented house contains up to 28

dynamic pressure measurements from sensors located along the roof, soffit and walls. It

also contains pressure measurements for use as the reference pressure. Recall from Eq. 2-

1 / 4-3 that the pressure coefficient is expressed in terms of the difference between the

local pressure at the point of interest and the barometric (reference) pressure. The

potential barometric pressure reference sensors are located either in the attic of the house,

away from the house but on the property (yard sensor), or both. In the case of

instrumentation malfunction the reference pressure will be obtained from a nearby FCMP

mobile tower. The measurement of the local and reference pressures have uncertainties

that will affect the reliability (confidence) of the pressure coefficient estimation. These

uncertainties are related to the unique calibration factor of each sensor and the fact that

the sensor output voltage is affected by temperature changes presented by Michot (1999)









In addition to the local and reference pressure a reference wind speed is also

selected to normalize the coefficient. Typically this reference wind speed represents the

expected (average) peak speed of the wind as the flow approaches the house. This

requires selection of both duration (e.g., 1-minute wind or 3-second wind) and an

elevation above local ground level (e.g. mean roof height). To select this speed, most of

the FCMP houses have a single anemometer installed 4-6' above the ridgeline, a few

houses have two, and some (typically hip roof houses) do not have an anemometer

installed. Other sources are available to obtain the reference wind speed, for example a

nearby FCMP mobile tower or estimations performed by Applied Research Associates

parametric hurricane wind field model (Vickery, et al, 2000). The measurement and

calculation of a reference wind speed is another source of uncertainty in the final pressure

coefficient value.

Applied Equations for C,

The data analysis starts by exporting each file into ASCII format, which is then

converted into MATLAB format for easy manipulation. The data were collected at a

sampling rate of 100 Hz, but for analysis purposes the data were initially down sample to

10 Hz. The following steps were used to analyze the data.

First, each of the data channels was validated by visually inspecting the time

history of the mean raw voltage measurements to provide a list of the functioning

sensors. Table 4-1 and 4-2 presents the result of this process for the FL-27 and FL-30

houses respectively, for the data collected during hurricane Ivan (2004).




















































Figure 4-1 Voltage time history of functional data channel 0 for the FL-27 house d

hurricane Ivan (2004)


Figure 4-2 Voltage time history of a malfunctioning data channel

during hurricane Ivan (2004)


4 for the FL-27 house


Time Hstoy of Channel 0 mean voltagee output for FL-27 during Ikan (2004)
5.5




5



4.5 --




4I-
4.5- -- -- -rF------rF-----r7---- --7r----- T----- T------








o
3.5 -----
3o
0
o o

3 2------------------------------------ -----------




2.5
0 20 40 60 80 100 120 140 160
FRcord number


during


Time Hstoy of Channel 4 during kan (2004)
10










5 0
I

00




0





6 o
O--





0 20 40 60 8 100 120 140 160
Record #











Figure 4-1 shows a typical example of a functional channel (yard sensor for

reference pressure) and Figure 4-2 shows a channel considered malfunctioning (attic

reference sensor) for the FL-27 house. Both examples were collected during Hurricane

Ivan, 2004). Typical data error could be generated by electrical fluctuation spikes or other

unknown factors, causing data sensors to malfunction for short durations, an example of

this situation was observed in record # 129 (15-minutes of data) for the FL-27 during

Ivan (2004) from channel 5. The plot in Figure 4-3 shows the time histories of channel 5

and 6, where channel 5 shows an erratic output voltage for a short duration in the 15

minute record, compared with channel 6. Figure 4-4 shows that sensors 5 and 6 are in

close proximity, and it is expected that the raw voltage time histories should be similar.


FRcord #129 FL-27 Ian (2004)
4





24
3.8- -- -




32.4- -








3.2 Channel 6


000 2:30 05:00 0730 10:00 12:30 15:00
Time (VIVI:SS)

Figure 4-3 Record #129 of house FL-27 hurricane Ivan (2004), channel 5 malfunctions

The offset is not of concern, as this differs from sensor to sensor. The dramatic

attenuation and subsequent positive jump of the voltage from sensor is considered a









malfunction. This type of behavior generates erroneous pressure coefficients. Channel

five for this particular 15 minutes of data is removed from the dataset.

The roof sensor layout configuration and the result for the mean output voltage

channel inspection for the FL-27 and FL-30 houses are presented in Figure 4-4, 4-5 and

Table 4-1, 4-2 respectively. Refer to Figure 3-2 for the location of these houses relative to

the path of Hurricane Ivan. The house instrumentation layout shows the location of each

sensor along the roof, walls, and soffit as well the location of the house anemometers, and

house orientation. The information presented on the tables shows the sensor type, I.D. of

the sensor used for each particular event, channel used to collected data (i),

corresponding sensor calibration factor (a ), source of calibration factor and channel

status. The calibration source "UF 2005" refers to a calibration of the specific sensor that

was at that location, conducted by the author at UF. "Average value" refers to using a

mean calibration from the sensors that were calibrated at UF. In this case the specific

sensor was not calibrated due to an inability to recover the specific sensor used (lost, no

longer functioning, or mislabeled during deployment). Use of the average value is viable

due to the close agreement of calibration among the many sensors, but it does introduce

additional uncertainty in to the analysis, particularly for the peak Cp value estimates.

The third step requires a nearby FCMP mobile tower or local ASOS (airport

weather) station to access temperature time history. This is necessary in order to apply

correction changes to the calibration due to temperature effects on the pressure

transducers. For this task the GIS frame work developed by the author is very practical,

which allow easy visualization of near ASOS station or mobile towers relative to the







58


house locations. For the analysis of the two houses FL-27 and FL-30 the temperature

reference were obtained from the mobile tower T1 data records (Figure 3-2).


SPFCMP House ID: FL-27
Gulf Breeze, FL 3256 I

Figure 4-4 Roof sensor layout configuration for FCMP House FL-27


S2 3 4
I I I I


26 24 3 23 21 20


5 G
I I


CMP House ID FL-30
ensacola Beach, FL 32507


Figure 4-5 Roof sensor layout configuration for FCMP House FL-30










Table 4-1 Sensor list for FCMP house FL-27 hurricane Ivan (2004)
Channel Calibration Calibration Channel
Sensor Type Sensor ID
i ai Source Status
Plastic Box Camera 0 34.274 Average Value OK
Pan Sensor 026A 1 34.274 Average Value NG
Pan Sensor 008A 2 34.149 UF2005 OK
Pan Sensor 027A 3 34.274 Average Value OK
Plastic Box Attic 4 34.274 Average Value NG
Pan Sensor 035A 5 34.274 Average Value OK
Pan Sensor 201 6 34.336 UF2005 OK
Pan Sensor 92 7 34.154 UF2005 OK
Pan Sensor 8 34.274 Average Value NG
Pan Sensor 42 9 34.430 UF2005 OK
Pan Sensor 167 10 34.733 UF2005 OK
Pan Sensor 93 11 34.434 UF2005 OK
Pan Sensor 81 12 34.203 UF2005 OK
Pan Sensor 003A 13 34.588 UF2005 OK
Plastic Box 14 34.274 Average Value NG
Pan Sensor 72 15 34.420 UF2005 OK
Pan Sensor 13 16 34.274 Average Value OK
Pan Sensor 84 17 34.147 UF2005 OK
Pan Sensor 225 18 34.608 UF2005 OK
Pan Sensor 34 19 34.147 UF2005 OK
Pan Sensor 143 20 34.457 UF2005 OK
Pan Sensor 144 21 34.396 UF2005 OK
Pan Sensor 49 22 34.092 UF2005 OK
Pan Sensor 177 23 34.224 UF2005 OK
Pan Sensor 92 24 34.154 UF2005 OK
Plastic Box 25 34.274 Average Value NG
Pan Sensor 122 26 34.380 UF2005 NG
Pan Sensor 54 27 34.395 UF2005 NG
Pan Sensor 106 28 34.230 UF2005 OK
Anemometer 1 29 OK
Anemometer 2 30 Not in use

The fourth step requires selecting a source of wind speed for the expected peak 3-

second gust value. Various sources are available for this: (a) using the 3-cup anemometer

mounted on the house, (b) from nearby mobile tower measurement applying the proper

adjustment for exposure and height or (c) by using overland wind field model estimation,

provided by Peter Vickery from Applied Research Associates (ARA) parametric

hurricane wind field model (Vickery, et al, 2000). For the analysis of the two houses FL-

27 and FL-30 the wind speed source will be obtained from each house anemometer by










estimating the peak 3-sec gust from the 15-minute mean wind speed value. This will be

done using two different methods.

Table 4-2 Sensor list for FCMP house FL-30 hurricane Ivan (004)
Channel Calibration Calibration Channel
Sensor Type Sensor ID
i ai Source Status
Plastic Box Camera 0 34.274 Average Value OK
Pan Sensor 031A 1 34.380 UF2005 OK
Pan Sensor 70 2 34.348 UF2005 OK
Pan Sensor 19 3 34.452 UF2005 OK
Plastic Box 132 4 34.240 UF2005 OK
Pan Sensor 15 5 33.867 UF2005 OK
Pan Sensor 171 6 34.398 UF2005 OK
Pan Sensor 104 7 34.270 UF2005 OK
Pan Sensor 238 8 34.409 UF2005 OK
Pan Sensor 043A 9 34.392 UF2005 OK
Pan Sensor 044A 10 34.182 UF2005 OK
Plastic Box Soffit 11 34.274 Average Value OK
Pan Sensor 182 12 34.194 UF2005 OK
Pan Sensor 045A 13 34.274 Average Value OK
Plastic Box 010A 14 34.389 UF2005 OK
Pan Sensor 74 15 34.016 UF2005 OK
Pan Sensor 26 16 34.387 UF2005 OK
Pan Sensor 050A 17 34.274 Average Value OK
Pan Sensor 209 18 34.659 UF2005 OK
Pan Sensor guess 19 34.274 Average Value OK
Pan Sensor 002A 20 34.274 Average Value OK
Pan Sensor 193 21 34.190 UF2005 OK
Pan Sensor 173 22 34.255 UF2005 OK
Plastic Box Wall 23 34.274 Average Value OK
Pan Sensor 116 24 34.345 UF2005 NG
Plastic Box Not in use 25 34.274 Average Value NG
Plastic Box Soffit 26 34.274 Average Value OK
Plastic Box Wall 27 34.274 Average Value NG
Plastic Box Attic 28 20.600 Old Sensor OK
Anemometer 1 29 OK
Anemometer 2 30 Not in use

After all the input parameters are establish for each house scenario, the pressure

differential needed for calculation of the pressure coefficient can be computed using one

of two equations, depending on the dataset conditions: Eq. 4-1 is used for the case where

the reference pressure is obtained from the attic or camera (yard) sensors, and Eq. 4-2 for


the case where the reference pressure is obtained from a nearby tower.









The evaluation of Eq. 2-1 / 4-1 through 4-3 from the voltages collected by the

sensors is not a straightforward operation. The sensitivity of each sensor is described in

terms of a linear relationship between pressure and voltage. It is typical that the slope of

this sensitivity remains very steady for a given sensor, while the y-intercept is more

variable between sensors.

The analysis presented here eliminates the need for a y-intercept for the sensor

sensitivity by calculating the difference between atmospheric pressure and the pressure

measured during the storm. That is, calculating the absolute dynamic pressure at a given

sensor (which requires knowledge of the y-intercept sensitivity), is replaced by

calculating the difference between the pressure at that sensor during still winds and that

during peak winds (which does not require knowledge of the y-intercept). This is done

by taking the difference in the mean sensor voltage from a still wind period well before

the storm and the dynamic voltage measured during the storm. That provides the first

term in the numerator in equation 2-1 (first parenthetical term in Eq. 4-1). The second

term in the equation (reference pressure) is calculated as a differential in the same way

(pre-storm minus during-storm).

We do need to account for the change to the slope of the sensor sensitivity caused

by temperature dependence. This is done by including a temperature differential between

pre- and during-storm conditions. The change in slope due to temperature is a measured,

known quantity (Dearhart, 2003).

The pressure differential AP)(t in the numerator of Eq. 4-3 can thus be represented

by Eq. 4-1. Equation 4-2 is used when the source for the reference pressure is not at the

house, but from a nearby FCMP tower or other atmospheric data record (airport, etc).









0.0144
( V(t), -VO + ATempx 0. a, x a,
S0.=0144 ) (4-1)
SAVe(tp)20.6 X aREF
S2T0.144

V(t), V + ATemp x -014 x a, x a-
AP(t), 20.6 a, (4-2)
(P(t)REF ~OREF

Where, all pressures are in psf:

V(t),, voltage for channel i @ time t for: 15min Mean value, Max and Min
moving average for durations of {selected pressure duration}

Vo voltage for channel i @ time to (15min Mean) pre-storm data

V(t)REF, voltage for reference channel (yard, attic or FCMP tower sensor)
@ time t (15min Mean)

VJo voltage for reference channel @ time to (15min Mean) pre-storm
data

ATemp [F], temperature change from between time to and t

a, [psf/Volt], sensitivity factor for channel i, obtained from sensor
calibration test. i includes the reference sensor REF

P(t)REF, atmospheric pressure @ time t, from Tower Data (15min Mean)

POREF atmospheric pressure @ time to, from Tower Data (15 Min Mean)
pre-storm data

The mean, root mean square (RMS), maximum and minimum peak pressure

coefficients are then calculated using Eq. 4-3, where / p is taken as a constant value of

0.00256 which reflects the mass density of air for the standard atmosphere i.e.,

temperature of 59 F and sea level pressure of 29.92 inches of mercury and dimensions









associated with wind speed in mph. The selection of an appropriate velocity value for the

denominator is the subject of an upcoming section.

For all coefficient calculations, the reference pressure and reference velocity are

taken as constant for a given Cp sample (over a 15-minute period). The dynamic local

pressure (first term in Eq. 2-1) is an expected value for calculation of the mean Cp, while

instantaneous values for dynamic local pressure are used for the calculation of rms and

peak pressure coefficients.

AP(t),
c 1 p 2 (4-3)
2 pV3 Sec

Uncertainty: Data Sources can change from Storm to Storm and House to House

A primary goal of this dissertation is to identify, model and account for the sources

of uncertainty that affect the final estimates of Cp from the full-scale house datasets. One

such source of uncertainty is the source of data that is used to provide the various

parameters in Eqs. 4-1 through 4-3. For a given house, the reference pressure can come

from more than one source:

a) Attic Sensor
b) Camera (yard) Sensor
c) Near FCMP Mobile Tower or airport


The reference wind speed can also be estimated from more than one source:

1) House Anemometer via optimization of roughness (z0) value for selected wind
swath (addressed in a later section)
2) House Anemometer via optimization of Peak-Factor (g) for selected wind
swath (addressed in a later section)
3) Near FCMP Tower
4) ARA's parametric hurricane wind field model (Vickery, et al, 2000)







64


These different sources introduce varying degrees of uncertainty in the analysis,


and must be accounted for. The outcome of combining the different pressure sources and


wind speed sources are twelve possible cases, which are summarized on Table 4-3.


Table 4-3 Summary of 12 possible cases to compute full-scale pressure coefficients
Wind speed source
1) House Anemometer 2) House Anemometerar FCMP Tower 4) 's WindModel
Opt m o O m P o g 3) Near FCMP Tower 4) ARA's Wind Model
Optimum z Optimum Peak-Factor g
Case (la) Ref pressure from Case (2a) Ref pressure from
Case (la) pressure from. pressure from Case (3a) Ref. pressure from Case (4a) Ref. pressure from
a a) Attic attic sensor and wind speed attic sensor and wind speed a p a
Sr attic sensor and wind speed attic sensor and wind speed
Sensor from house anemometer from house anemometer
f trom near F CMP tower from ARA model
o optimization of z optimization of Peak-Factor from near FCMP tower from A model
Case (lb) Ref pressure from Case (2b) Ref pressure from
SCase e fm Case (3b) Ref pressure from Case (4b) Ref pressure from
Z b) Yard camera sensor and wind speed camera sensor and wind speed
o camera sensor and wind speed camera sensor and wind speed
S Sensor from house anemometer from house anemometer from near FCMP tower from ARA model
. from near F CMP tower from ARA model
Soptimization of z optimization of Peak-Factor
Case (ic) Ref pressure from Case (2c) Ref pressure from
Sc) Near near FCMP tower and wind near FCMP tower and wind Case (3c) Ref pressure from Case (4c) Ref pressure from
P FCMP near FCMP tower and wind near FCMP tower and wind
speed from house anemometer speed from house anemometer r near FCP tower sed fm AA mod
Tower . speed from near FCMP tower speed from ARA model
Tower optimization of 0 optimization of Peak-Factor


In some cases, more than one source for reference pressure and reference velocity


are available. For example a house dataset that includes an attic sensor, a yard sensor, and


a rooftop wind anemometer record covers four of the cases in Table 4-3. None of these


analyses would produce precisely the same estimate for any Cp, and thus the estimate is


inherently a random variable with quantifiable confidence limits. In the next section


analysis results using various cases will be presented. Comparison of the output from


each case will demonstrate the uncertainty.


Example Peak Minimum C Calculation with Uncertain Reference Velocity


In this section preliminary analysis results of the FL-27 house during Hurricane


Ivan (2004) will be presented. For this dataset the reference pressure source is a yard


sensor near the subject home and monitored by the same data collection system used with


the house sensors. The reference wind speed (expected 3-second gust at mean roof


height) is estimated from the house anemometer (using two different methods), the









nearby FCMP tower and the ARA wind field model. Thus the analysis of any given 15-

minute record of data will produce four different estimates of pressure coefficient. For

this example, the peak minimum pressure coefficient is estimated, which corresponds to

the maximum gust-type suction (uplift) that the sensor experienced in that 15-minute time

frame, normalized by the expected peak 3 second wind in the that same time frame.

Table 4-4 presents the peak minimum pressure coefficients for channel 6 (see

Figure 4-4 for a diagram of the roof and sensor locations) using cases lb, 2b, 3b and 4b

(from Table 4-3). The data records 131 to 147 (each record is 15 minutes long)

correspond to the FL-27 house during hurricane Ivan (2004) when high winds were

impacting the home. The direction of the wind is also noted in Table 4-4 from two

sources (FCMP tower and ARA model), as well as the estimated 3-second gust estimated

from four methods. The wind is approaching from the southeast during this time frame.

Referring to the location of channel 6 in Figure 4-4, this portion of the roof is expected to

experience strong suction for this range of wind directions.

The last four columns in a given row in Table 4-4 show the spread of peak

minimum C estimates for the same sensor, over the same time frame, using different

sources of wind velocity information in the calculation parameters used in Eq. 4-3 (all

other variables identical for each calculation). Note that some estimates exceed (are

smaller than) the ASCE-7 value in corner zones of -2.6. This preliminary analysis shows

the effects of uncertainty in the pressure coefficient results by accounting for just one of

the uncertain parameters in the computation.

The full uncertainty analysis later in this chapter will account for a number of

contributors to uncertainty in a probabilistic framework. For example, the uncertainty in











reference velocity will be changed from the four discrete values used in the current


example to a probabilistic representation over a range of values. The same will be done


for the other sources of uncertainty, and Monte Carlo simulation will be used to generate


thousands of estimates of Cp in the place of the four estimates per row in Table 4-4. In


this manner the resultant histogram of simulated Cp values can be used to determine


confidence limits to bound the mean value (the Cp estimate). Comparisons with wind


tunnel tests and ASCE-7 wind load provisions will be conducted with knowledge of these


confidence limits, and will help determine the appropriate level of concern for


discrepancies.


Table 4-4 Minimum peak pressure coefficients for FL-27 hurricane Ivan (2004) channel 6

i f .,.....,. .il
il. .l .,.. ... j IT I I 'r rr r
131 9/16/2004 04 52 48 53 88 61 40 5829 6202 108 115 -2 30 -255 -225
132 9/16/2004 05 0756 5531 61 48 6772 6289 106 117 -206 -1 67 -1 37 -1 59
133 9/16/2004 05 2304 5892 6423 7063 63 80 108 120 -250 -210 -1 74 -2 13
134 9/16/2004 05 38 13 61 52 65 15 71 94 6486 108 122 -2 47 -221 -1 81 -2 23
135 9/16/2004 05 53 21 6611 6870 8021 6602 113 125 -2 51 -1 84
136 9/16/2004 06 08 29 6578 7299 8500 6718 120 128 -246 -1 99 -1 47 -2 35
137 9/16/2004 06 23 38 70.94 75.13 8698 6824 124 132 -1 80 -1 61 -1 20 -1 95
138 9/16/2004 06 38 46 6864 73 23 95.80 6907 125 136 -2 10 -1 84 -1 08 -207
139 9/16/2004 06 53 55 6832 71 40 9029 6966 128 140 -1 72 -1 57 -098 -1 65
140 9/16/2004 07 09 03 65 86 68 26 88 41 7008 134 145 -238 -221 -1 32 -210
141 9/16/2004 07 24 12 61 52 65 86 91 35 70.30 140 149 -1 73 -1 51 -079 -1 33
142 9/16/2004 07 39 21 6274 61 81 91 02 7027 146 154 -1 54 -1 59 -073 -1 23
143 9/16/2004075429 5842 5732 8640 7000 150 159 -1 32 -200
144 9/16/2004 08 0937 5609 56 14 85 19 6945 156 163 -1 14 -1 72
145 9/16/2004 08 2446 5909 5792 81 97 6864 167 168 -1 35 -1 41 -0 70 -1 00
146 9/16/2004 08 3954 5695 58 57 8004 6758 179 172 -1 62 -1 53 -0 82 -1 15
147 9/16/2004 08 5503 5640 58 88 7981 6630 184 176 -2 42 -1 32 -1 91

SRecord with maximum 3-second gust estimated
-Record where the minimum Cp value of-2. from the ASCE 7 is exceed


Identification of Appropriate Sampling Rate

The original 100 Hz data for the FL-27 house were downsample using segmental


average to 10 Hz and it was used for the preliminary analysis presented in Table 4-4.


Downsampling the original 100Hz data are necessary to expedite processing and filter out


high frequency noise. The peak minimum pressure coefficients are affected by the final


downsample rate. A faster sampling rate will produce larger pressure coefficients than









those obtained for slower sampling rates. The influence of sampling rate can be graphed

and a suitable rate identified (see Figure 4-6 and 4-7), beyond which there is not

significant increase in peaks. For example, downsampling from 100 Hz to 1 Hz may

produce a 1 Hz peak Cp of-1.5. Using 10Hz sampling may produce a peak of-1.8, while

increasing to 20 Hz may produce -1.85. A plateau from plots (Figure 4-6 and 4-7) that

present the minimum peak pressure as function of resample frequency will be identified

for maximum suitable sampling rate.

Conversely, (peaks) from faster sampling rates (say 100 Hz) tend to be more

localized (less correlated spatially), thus reducing their influence on the aggregate loads

we seek to quantify. For example, say the full-scale analysis reveals an extreme peak at

one sensor. If that peak is not correlated with near neighbor sensors, it is reasonable to

say that the aggregate uplift loading observed in the 4x8 ft sheathing panel occupied by

those sensors is less severe. Such a peak is of limited use. A correlation study among

adjacent sensors will evaluate the sampling rate at which correlation has significance with

regard to aggregate loading.

Study for Identification of Appropriate Sampling Rate

In this section an appropriate sampling rate is identified for the FL-27 Ivan dataset

using analysis case lb from Table 4-3. Records 131 147 presented in Table 4-4 will be

considered in this study case since the maximum sustained winds were experienced in

this time frame. Figure 4-4 presents the roof sensor layout for the FL-27 house, located in

Gulf Breeze, FL. This study considers individual sensor and pairs among the sensors (5,

6, 7, 22, 23 and 24).









Influence of Sampling Rate on Peak C value

Plots of the minimum peak pressure coefficient are generated as function of the

resample rate through a series of downsample of the original 100 Hz data. The data were

filtered using the decimate function in MATLAB. The decimation process filters the

input data with a lowpass filter and then resample's the resulting smoothed signal at a

lower rate (by default, decimate employs an eighth-order lowpass Chebyshev Type I

filter). After downsampling the data at various rates a plot is generated (Figure 4-6 and 4-

7) of the minimum peak pressure coefficient vs. the sampling frequency.

Typical plots are presented using record #131 (mean wind direction of 1110) and

record #147 (mean wind direction of 1800) for channel 6 in Figure 4-6. The same

analysis is done for channel 23 in Figure 4-7. These figures show a plateau in peak

minimum pressure coefficient is reached at 10 20 Hz before a continued slow growth at

higher frequencies. It is desired to select a downsample rate of the original 100 Hz data,

to a sampling rate that is lower than half the original rate, in order to filter out high

frequency noise. Thus a downsample rate is set to 20 Hz. Additional analysis related to

the downsampling rate based on the peak correlation is presented in the next sections.

The wind tunnel work of Theodore Stathopoulos had an equivalent full-scale

sampling rate of 2 Hz, so it is worth comparing the difference obtained in the minimum

pressure coefficient from the 20 Hz vs. 2 Hz full-scale dataset. Table 4-5 and 4-6 present

the comparison for records 131 and 147, for channel 6 and 23 respectively, showing the

difference % between the minimum C, values having a maximum difference of 44% and

a minimum difference of 16%.









69




FRcord #131 for FCMP House FL-27 Ian 2004 Channel 6
-18 -



-2 -- -- -- --I
-2



-2.2














-3.4








0 10 2D 33 43 50 60 70 a 90 100
FRsamrple frequency



(a)



Fbcord #147 for FCIP Hbuse FL-27 Ken 2004 Channel 6
-1.6 -






2 -
i 2 .2 \ - - - - -












-2. -


-.----2.









-3. 4
c^---.
32

























0 10 20 30 40 50 60 70 80 90 100
Ftesample frequency



(ab)
direction is I I (b) Ford #147 record #1 47 wind direction is 180















a




I a2



-3-
3.4----------------------i~~~r





















0 10 20 30 40 50 60 70 80 90 100
RFsample frequency



(b)

Figure 4-6 Channel 6 peak minimum C vs. resample frequency. (a) For record #131 wind

direction is 1110 (b) For record #147 wind direction is 1800









70




Fecord #131 for FCMP House FL-27 en 2004 Channel 23
-04-.3


-0.4- -- ---


-05




-0.7 ------- ---
-06"""


-0.6 -\- - - -- -- ---- - -
6.

2 -0.8 -- - - - I- -

4 .9 - - I v - ^ -


-1.1 ------- --



-1.2 -



-1.3
0 10 20 30 40 50 60 70 a 90 100
FRsample frequency


(a)



Fecord #147 for FCMP House FL-27 en 2004 Channel 23
408



-1 -



-1.2 - -
--- -I- --- -I--
























(b)

Figure 4-7 Channel 23 peak minimum C vs. resample frequency. (a) For record #131 wind
-1direction is (b) For record #147 wind direction is 180
14


:' -1 .6 - e n - r - - n -
S16,


-1.8C J L I








-2.2
0 10 20 33 43 50 60 70 m 9 i0m
Ftsamle freqency



(b)
Figure 4-7 Channel 23 peak minimum C vs. resample frequency. (a) For record #131 wind

direction is 1110 (b) For record #147 wind direction is 1800










Given the data provided in Figures 4-6 and 4-7, as well as Tables 4-4 through 4-6,

there is the potential that the current ASCE-7 loads for components and cladding, based

upon wind tunnel data with an equivalent sampling rate 2 Hz, do not reflect the

conservative worst case from extreme winds. This possibility is strengthened if it can be

shown that strong spatial correlation exists among peaks sampled at 20 Hz.

Table 4-5 Difference % between 20 Hz and 2 Hz minimum C, for channel 6 ofFL-27
Minimum Cp Channel 6
Record #
Resample rate
(Hz) 131 147
20 -3.13 -2.86
2 -2.06 -2.03
Difference % -41% -34%

Table 4-6 Difference % between 20 Hz and 2 Hz minimum C for channel 23 of FL-27
P
Minimum Cp Channel 23
Record #
Resample rate R d
(Hz) 131 147
20 -0.66 -1.64
2 -0.42 -1.39
Difference % -44% -16%

Influence of Sampling Rate on Spatial Correlation of Peaks

This section focuses on the influence of sampling rate on spatial correlation of

peaks. The analysis will be developed using the same house dataset presented in the

previous section. Two standard methods of correlation measurement are considered, and

an additional method is developed to specifically focus on the correlation of peaks.

Correlation coefficient Px,

The correlation coefficient is a normalized measure of the strength of the linear

relationship between two random variables x and y (any 2 sensors). The correlation

coefficientpxy is defined in equation 4-4 as the ratio of the covariance of two random









variables x andy to the product of the standard deviations. Uncorrelated data results in a

correlation coefficient close to 0; strongly correlated data have a correlation coefficient

near 1.

COV
Px,y y (4-4)


Where COVx,, is the covariance defined in Eq. 4-5, px and py represents the mean

value of the random variable x and y respectively and E[ ] represent the expected value

of the term inside the brackets. Intuitively, covariance is the measure of how much two

variables vary together.

COVxy = E (x- p )(y- py (4-5)

Figure 4-8 shows the correlation coefficient between adjacent sensors channels (6

& 7) and non-adjacent sensor channels (6 & 23) which are located in opposite roof

corners, as a function of resample rate. For adjacent sensors (approximately 2.5 feet

apart) it shows a very strong correlation and weak correlation for non-adjacent sensor

(approximately 60 feet apart). The correlation drops significantly with distance, as

expected. In both plots the correlation also drops as the resample rate increased,

indicating both the presence of high frequency noise and the shorter spatial correlation at

higher frequencies.

The limitation of the correlation coefficient is that it presents a simple scalar

display of correlation, containing information across all frequencies in the analyzed

signals (any 2 sensors). Thus the high observed correlation from adjacent sensors (6 & 7)

may be due mostly to low frequency (non-peak) correlation. That is, it cannot specifically

indicate whether peaks between sensors are correlated.







73



FCMP House FL-27 Ivan 2004 for record #131
I1 - - - ---- -- - - -- -


0.9 8

0.8
a7- -- ----- ------ ----------- t----- ------t -----
0.7




0. I -
0.--------- ------------------ ------ ------
B 0.5



0. --







--E Non-Adjacent Sensors 6 & 23
0 5 10 15 2 25 30 35 40 45 50
Resample frequency (-t)


Figure 4-8 Correlation coefficient for record #131 of FL-27 hurricane Ivan (2004)

Coherence function C2


The coherence function C2(f) is the ratio of the square of the magnitude of the


cross-spectrum density function to the product of the autospectral density functions of the


two random variables x(t) and y(t) equation 4-6. For allf the quantity C2 (f) satisfies


0 < C (f ) <1. The advantage of the coherence function is that it shows the correlation


as a function of frequency.
0 2 - - I- I- - - - - - -

































Cx,y S ()S,,(f)


Figure 4-9 shows the coherence function for adjacent sensors (6 & 7) and non-


adjacent sensors (6 & 23). For adjacent sensors it is clearly shown a strong correlation at
adjacent sensors (6 & 23). For adjacent sensors it is clearly shown a strong correlation at











low frequency, not observed for non-adjacent sensors which are located at opposite roof

corners. Such behavior was expected due to the relative location of the sensors.


FCMP House FL-27 Ivan 2004 for record #131

S- Acjacent Sensors 6 & 7
S-----. Non-Adjacent Sensors 6 & 23


0.5

0.4
(-



0.3

0.2


5 10 15 20 25
Frequency (Fz)


30 35 40 45 50


Figure 4-9 Coherence function estimate for record #131 of FL-27 hurricane Ivan (2004)

The coherence function still does not isolate peaks however, since even high

frequency content is an average over entire signal. That is, the low correlation at higher

frequencies does not preclude well correlated peaks.

"Peak-score" method

A method that focuses on identifying the correlation among peaks in pairs of

random variables was developed for this analysis. The "peak-score" method is intended

to qualitatively identify the degree to which two sensors share peaks at the same time

instant (correlation), while ignoring the low frequency signal content and low magnitude









high frequency content that mimic the meaning of correlation coefficient and coherence

function.

In the following discussion, the random variables x and y can be thought of as the

time histories measured at two sensors, either close or well separated. The analysis starts

by normalizing each of the two time traces x and y such that the minimum value in each

is -1.0 and the mean value is 0. These two normalized signals are x, and y,.

Figure 4-10 (a) and (b), shows typical signals x andy for close sensors and its

normalized version respectively. Note that peak negative values can be observed in both

signals at the same times. A new signal w, is created as the average of x, and y,, or

w, = (x + y) / 2 as shown in Figure 4-11. If the original signals x and y experienced

peak minimum values at the same instant, w, should have a number of peak values that

approach -1.0.

This can be qualitatively characterized by comparing the probability contents of

x y,, and w,. A threshold minimum value for x, is found such that 5% of the

probability of x, is lower than this value. This is called the threshold value THx.

Similarly the 5% threshold for y, is found, TH .Figure 4-12 shows graphically the

estimation of THx = -0.1666 and THy = -0.1589 An average threshold THag is then

established from THx and TH The probability of w, that is below this average

threshold of THavg = -0.1628 is then found. The peak score is the ratio of this probability

with the 5% used to established the thresholds for x, and y,. Figure 4-13 shows the








76



value obtained from the TH ,I so the "peak-score" for this example has a value of



4.74%/5.0% = 0.948.


(b)
Figure 4-10 Random data signals. (a) Original signals x any, (b) Normalized signals x,

and y,


Cnginal Signals x and y


Norrnmazed Silgals xn and n
05



0)

-05 -



0 1 2 3 4 5 6 7 8 9
x 10

05










x 10











Cornpcting new normalized signal w

0.5
0 --------------------
+ 0-
x -.5 -





x 104

0.5









0 1 2 3 4 5 6 7 8 9
xl4






Figure 4-11 Resultant signal w, = (x + y)/2


The range of possible "peak-score" outcomes is from 0 to ~1. If the peaks in the


signals x, and y, are not correlated, the probability density function (PDF) of w, will


not contain as much probability in the left tail as x, and y,, and the peak score will not


approach 1.0. Conversely, if peaks in x and y regularly occur at the same instant, the peak


minimum values in w, will be of the same magnitude and frequency as those in x, and


y,, and the peak score will approach one.


The peak score method is first evaluated using artificial data where the level of

correlation is known. Random signals x and y are digitally created with similar spectral


and probability characteristics as those that describe a typical time history of pressure

from the full-scale data. These artificial signals x and y are uncorrelated. A third signal









78




w is defined as function of x and y so that partial correlation exists



(e.g., w(x,y)= 0.5x + 0.5y).


Errpirical CF for xn


Errpical CCF for y


X -0 1666
Y 005007 --
- ---5 05

-05 0 05


X -0 1589
Y 0-- -Y0501


-05 0 05


Figure 4-12 Empirical cumulative distributions function for x, and y,


Errpirical CDF for w


0.9


0.8


0.7-


0.6


S0.5-


0.4-


0.3-


0.2


0.1-


X -0 1628
- -Y 0 04739
U,,-


-1 -0.8 -0.6 -0.4
w


-0.2 0 0.2


Figure 4-13 Empirical cumulative distributions function for w,












The "peak-score" analysis is then performed using uncorrelated data (x and y),


partially correlated data (x and w), and perfectly correlated data (x and x). The result from


uncorrelated data x and y is presented in Figure 4-14 for various percentage threshold.


A 1% threshold (as suggested above) appears suitable, as uncorrelated data provides a


peak score of 0.1 as the resample frequency decrease. Figures 4-15 shows the resulting


peak scores for perfectly correlated data (peak score hovers around -1.0), and Figure 4-


16 shows partially correlated data (peak score -0.8).



"Peak-Score" check for uncorrelated data x & y
1-


09--- 25%
5%

08

07

06


05-

-04-

03
0.6 -- -- -- I- --
0.57-- -- -- -- -- -- -- -- -- -- -- --- -- -- -- -- -- -- -- ---


0.4 - -- - - --- -- - --- -- -- -- ---






02---- -

01---- -

0
0 5 10 15 2 25 3J 35 40 45 5
Resarrle frequency (1-)


Figure 4-14 "Peak-Score" check for uncorrelated data x and y


The peak score method is qualitative at this time. In the application to full-scale


pressure uplift data, we seek to demonstrate that the peaks observed between closely


spaced sensors are well correlated.









80




"Peak-Score" check for full-correlated data x & x
1 - - -


0.9 -


0.8--


0.7--


0.6 -





0.4 ---------- -- --- -


0.3 -


0.2 -


01 -E- 25%

5%
0
0 5 10 15 23 25 30 35 40 45 50
Fesarrple frequency (Ft)


Figure 4-15 "Peak-Score" check for full-correlated data x and x





"Peak-Score check for partial correlated data x & w
1 -





0.8


0.7 -


0.6 I


0.5--- ---- --- -- --


0.4 ----- ------ ----


0.3 ----- ------


0.2 -






0 5 10 15 23 25 30 35 40 45 50
Fesarrple frequency ()



Figure 4-16 "Peak-Score" check for partial correlated data x and w











"Peak-score" method: application to full-scale data

The "peak-score" is now computed for the full-scale data for a range of resample


rates beginning with the original 100 Hz signal and for lower resample rates. This


analysis was conducted extensively for the highest wind speed data records for adjacent


and non-adjacent sensors, where some of the typical findings are presented in Figures 4-


17 thru 4-19, these figures shows the "peak-score" vs. the resample frequency for sensor


pairs of channels (6 & 7), (6 & 5) and (6 & 23) respectively.


Figure 4-17 and 4-18 correspond to sensors close to each other (approximately 2.5


feet) and as expected the peak-score suggest that the peak are well correlated and are


occurring at the same time. This is not the case for sensors spaced far apart


(approximately 60 feet), as shown in Figure 4-19, where the peak-score is significantly


lower.


"Peak-Score" FCMP -buse FL-27 ian 2004 for record #131
1L
-e-- Threshod 1%
0.9---





I 0.6----- ---- - - - - - -
0.8 -.,

0.7
S06







0 0.3 -

0.2 -

0 1 - i i ---- - i - -- i -

0
0 10 20 30 40 50 60 70 80 90 10D
Sample frequency (Ft)


Figure 4-17 "Peak-Score" for channel 6 & 7 in record #131 FL-27 hurricane Ivan (2004)









82





"Peak-Score" FCMP -buse FL-27 ian 2004 for record #131
1
--0- Threshold =1%
0.9 -


0.8


O 0.7
to
06
-0.6 ---- I---- -



S0.5 ----


0.4 - -- ----


=0.3 -- ---


0.2 I


0.1 L I L


0
0 10 20 30 40 50 60 70 80 90 100
FRsarple frequency (Hz)



Figure 4-18 "Peak-Score" for channel 6 & 5 in record #131 FL-27 hurricane Ivan (2004)


Figure 4-19 "Peak-Score" for channel 6 & 23 in record #132 FL-27 hurricane Ivan (2004)


"Peak-Score" FCMP 1-buse FL-27 ian 2004 for record #132
1 -
---- Threshold =1%
0.9 -


0.8


P 0.7--
of











0.3
S0.2 ---- ---- -iF


0.5 -1-----


0.1 0- -- - ---- --- -- -- -----_






0
0 10 2D 30 40 50 60 70 80 90 10D
Fsarrple frequency (-z)









Based on this correlation analysis and the study of peak minimum C value as a

function of resample rate, the full-scale data will be analyzed using a down-sampled rate

of 20 Hz. This represents the rate at which peak minimum Cp values plateau, and also

maintains strong correlation of peaks between closely spaced sensors.

Uncertainty of Reference Velocity: Estimating Peak Wind Speed Gust

It was demonstrated in Table 4-4 and the associated example that the identification

of a reference velocity for the denominator in Eq. 4-3 can have a significant impact upon

the resultant estimated peak minimum C In that example, four methods for obtaining

the reference velocity were cited. This section explains these wind speed reference

estimation methods.

In wind tunnel experiments the determination of a reference wind speed is

relatively straight forward. The velocity is measured either at the mean roof height of the

model, or at the top of the tunnel and the speed extrapolated to the mean roof height. This

is not the case for the full-scale datasets, where the wind speed reference value is

estimated from either an anemometer on the roof (two methods), a nearby FCMP tower,

or from wind field models. For any of these cases the associated uncertainty is larger than

that in a wind tunnel, and must be quantified.

The best reference to obtain the wind speed velocity is when a house anemometer is

present, but unfortunately an operational anemometer on the house is not certain. The two

houses used in this section (FL-27 and FL-30 Ivan 2004) had working anemometers

during the storm events of hurricane Ivan (2004).

The reference wind speed in the denominator of Eq. 4-3 should represent the

expected (average) peak 3-second value. When a house anemometer is available, we