|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 MODELING HURRICANECAUSED TREE DEBRIS IN HOUSTON, TEXAS By BENJAMIN K. THOMPSON A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2009
2 2009 Benjamin K. Thompson
3 To hurricaneaffected communities and urban forest managers of the Southeastern United States
4 ACKNOWLEDGMENTS There are many whom I wish to thank for their encouragement and support throughout all phases and aspects of my graduate career. I first wish to acknowledge the USDA Forest Service for providing the funding required of this project. I am also grateful to my professors and advisors at the University of Florida who provided opportunities for me to learn more than I ever thought possible in only two years. In particular, I thank my co chairs Dr. Francisco Escobedo and Dr. Christina Staudhammer, and committee members Dr. Corene Matyas, and Dr. Youliang Qiu. Deserving of a special thank you are two people in particular. First is Elizabeth Binford for all she has done to help me continue to push forward and maintain my sense of focus as I struggle with daily c hallenges. Second is my brother, Eric Thompson, for his hard work, sacrifice, and continued commitment to managing my home back in Washington State. My academic achievements would not have been possible without him. I also thank the friends Ive made i n Gainesville who have shown me what it means to be a graduate student and why its great to be a Florida Gator: Avinash Deshpande, Aaron King, Rachelle Yankelevitz, Cynnamon Dobbs, Lucretia Masaya, Darren Pecora, and Alicia Lawr ence In addition, I thank all my friends and family who provided me with continued support and encouragement over the two year duration of my graduate career
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS .................................................................................................. 4 LIST OF TABLES ............................................................................................................ 7 LIST OF FIGURES .......................................................................................................... 8 ABSTRACT ..................................................................................................................... 9 CHAPTER 1 INTRODUCTION .................................................................................................... 11 Literature Review .................................................................................................... 11 Objectives ............................................................................................................... 19 Hypotheses ............................................................................................................. 19 2 METHODS .............................................................................................................. 20 Data Collection ....................................................................................................... 21 Dataset 1: 2001 Houston UFORE data. ........................................................... 21 Dataset 2: Post Ike 2008 Houston UFORE and debris data. ............................ 23 Dataset 3: NOAA NHC H*Wind data. ............................................................... 24 Dataset 4: USGS NLCD 2001 Land Cover data ............................................... 26 Synthesis of Datasets ............................................................................................. 27 Statistical Procedures ............................................................................................. 32 Verification and Comparison of Results .................................................................. 34 3 RESULTS ............................................................................................................... 43 Debris Models ......................................................................................................... 43 Full Model ......................................................................................................... 43 Final Model ....................................................................................................... 43 Alternate Model ................................................................................................ 44 Development of Debris Estimates ........................................................................... 44 Results of Hypothesis Testing (Hypothesis 1) ........................................................ 46 Results of Hypothesis Testing (Hypothesis 2) ........................................................ 47 4 DISCUSSION ......................................................................................................... 60 Review of Statistical Models and Comparison of Results ....................................... 60 Review of Hypothesis Testing ................................................................................. 64 Appl icability of Findings .......................................................................................... 65 Limitations ............................................................................................................... 68
6 5 CONCLUSIONS ..................................................................................................... 71 APPENDIX: A FRAMEWORK FOR DEVELOPING A GISBASED DEBRIS ESTIMATION TOOL ............................................................................................... 77 Introduction ............................................................................................................. 77 Methods .................................................................................................................. 78 Results .................................................................................................................... 84 Discussion: ............................................................................................................. 87 Conclusion .............................................................................................................. 90 LIST OF REFERENCES ............................................................................................... 91 BIOGRAPHICAL SKETCH ............................................................................................ 95
7 LIST OF TABLES Table page 2 1 Descriptions of sampled USGS land cover classifications .................................. 41 2 2 Descriptive statistics of variables selected for analysis of post st orm effects from hurricane Ike on Houston Texas urban forests. ......................................... 42 3 1 Explanatory variables tested............................................................................... 55 3 2 Analysis of variance (ANOVA) table for final model ............................................ 56 3 3 Analysis of variance (ANOVA) table for final model ............................................ 56 3 4 Analysis of variance (ANOVA) table for alternate model .................................... 56 3 5 Modeled debris estimates based on alternate model predictions and comparison with other debris model estimates and official debris volumes as recorded on post Ike project worksheet data from the city of Houston. .............. 57 3 6 ANOVA table for biomass debris model ............................................................. 58 3 7 Differences between prestorm biomass and post storm debris within Houston city limits using urban forest effects (UFORE) plots ............................. 58 3 8 ANOVA table for land cover debris model .......................................................... 59 3 9 Averaged estimates of prestorm biomass and modeled post storm debris by land cover classification ...................................................................................... 59
8 LIST OF FIGURES Figure page 2 1 Research area and 2008 remeasurement UFORE plots ................................... 37 2 2 H*Wind model gridded points data for the greater Houston area ....................... 38 2 3 Sampled land covers in the greater Houston area .............................................. 39 2 4 Scatter matrix of the dependent variable and a sample of explanatory variables ............................................................................................................. 40 3 1 Plot of predicted versus residual values for full model ........................................ 48 3 2 Plot of predicted versus residual values for final model ...................................... 49 3 3 Plot of predicted versus residual values for alternate model .............................. 50 3 4 Interpolated raster of post storm tree debris estimates based on alternate model predictions (n= 348) ................................................................................. 51 3 5 Interpolated raster of post storm tree debris estimates within Houston city limits based on alternate model predictions (n= 348) ......................................... 52 3 6 Interpolation raster of prestorm, standing tree biomass in cubic yards/acre (n=34) ................................................................................................................. 53 3 7 Interpolation raster of measured post storm, downed tree debris in cubic yards per acre (n=34) ......................................................................................... 54
9 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science MODELING HURRICANECAUSED TREE DEBRIS IN HOUSTON, TEXAS By Benjamin K. Thompson December 2009 Chair: Francisco Escobedo Major: Forest Resources and Conservation Local governments estimations of post storm vegetation debris in hurricane affected areas of the Southeastern United States are limited by a lack of science based hurricane planning and response tools. A stratified subset of permanent Urban Forest Effects (UFORE) research plots within the city of Houston, Texas originally established in 2001 were selected for reinventory and measurement of debris following Hurricane Ike which struck the Houston region on September 13th 2008. Three statistical models for estimating post hurricane tree debris were developed. Model input parameters includ ing v ariables characterizing urban forest structure, wind behavior and land cover we re derived from the 2001 Houston UFORE dataset, the National Hurricane Centers H*Wind dataset, and the United States Geological Surveys National Land Cover Database, respectiv ely. The statistical models estimate tree debris based on alignment of preand post storm data in the region. Results of statistical modeling suggest that nonstorm variables have greater influence over variation in debris estimates than stormrelated v ariables. Land c over wa s tested as a proxy variable for pre storm urban forest biomass and post storm tree debris so that debris estimates stratified by land cover may be applied in cities without UFORE data. Estimates of prestorm, urban
10 forest biomass we re established to test the statistical relationship between pre storm biomass and post storm debris. Testing of both land cover as a proxy variable and the biomass debris relationship we re performed so that volumetric estimates of debris produced through statistical modeling may be simplified, spatially modeled to determine debris locations, and applied to other hurricaneaffected cities in the Southeastern United States.
11 CHAPTER 1 INTRODUCTION Literature Review Hurricane impact s to forested landscapes consist of physical damages, socio economi c burdens, and ecological changes (Everham and Brokaw, 1996; Pielke and Landsea, 1998) Physical damages to forested landscapes include tree defoliation, breakage, and uprooting of individual trees (Foster and Boose, 1992; Francis and Gillespie, 1993; Duryea et al 2007a; Durye a et al., 2007b) Damage to many trees across landscapes can result in standlevel or regional impacts (Foster and Boose, 1992; Stanturf et al 2007; Oswalt and Oswalt, 2008) which result in bot h short and long term changes in ecological health, productivity, and structure of forested ecosystems (Foster and Boose, 1992; Everham and Brokaw, 1996; Stanturf et al 2007) Landscape scale disturbances obligate fore st managers and communities with financial costs of damage mitigation and implement ation of storm recovery effort s (Pielke and Landsea, 1998; Stanturf et al., 2007; Staudhammer et al., 2009) Many studies of hurricaneforest interactions are based on a single event and are typically reviews of storm observations, post storm assessments, evaluations of storm recovery processes, or a combination of these research approache s (Foster and Boose, 1992; Francis and Gillespie, 1993; Zimmerman et al 1994; Everham and Brokaw, 1996; Pielke and Landsea, 1998; Batista and Platt, 2003; Tanner and Bellingham, 2006; Kupfer et al 2008; Oswalt and Oswalt, 2008) Additionally many studies focus on standlevel hurricane effects (Everham and Brokaw, 1996; Stanturf et al 2007) and report impacts on forest ecosystem components (Francis and Gillespie, 1993; Everham and Brokaw, 1996; Kupfer et al. 2008; Oswalt and Oswalt, 2008) T he extent and
12 severity of hurricane dam ages to forest ecosystem s are affected by hurricane characteristics including size, speed, direction, maximum wind speeds and wind distribution patterns (Margoles and Umpierre, 2005; Stanturf et al 2007; Kupfer et al 2008; Escobedo et al 2009) Complexities inherent to hurricanegenerated winds result in variable patterns and scales of damages across landscapes (Tanner et al 1991; Foster and Boose, 1992; Stanturf et al 2007) The literature offering insights into hurricane effects on forested lands capes may also be helpful for determining how urban trees respond to catastrophic winds (Duryea et al 2007a; Duryea et al 2007b) Few studies have investigated landscapescale hurricane impacts to vegetation in urban landscapes (Everham and Brokaw, 1996; Duryea et al., 2007a; Duryea et al., 2007b; Burley et al., 2008; Escobedo et al., 2009; Staudhammer et al., 2009) R esearch has documented hurricane effects on specific trees and individual species within urban areas (Francis and Gillespie, 1993; Duryea et al. 2007a; Dur yea et al. 2007b) yet it remain s difficult to characterize urban landscapes with such data because variation in individual tree or species specific hurricane response s at large spati al scales is unknown. T here is a growing scientific interest in the ecological functioning of urban ecos ystems (Cadenasso et al 2007) and that urban landsc apes may be studied using similar methods to forested landscapes (Zipperer et al 2000) Urban forest managers (e.g. planners or urban foresters) and researchers of urban forests have been successful in adapting forestry practices to the study and management of trees in urban areas (Nowak, 1993; Nowak et al 1996; Nowak, 2002; Xiao et al 2004; Myeong et al 2006) A n example of adapting forestry practices to urban landscapes comes from the
13 USDA Forest Services Urban Forest Effects (UFORE) model (Nowak, 19 93, 2006) The UFORE models design and methods are based on those used by the USDA Forest Services Forest Inventory Analysis (FIA) Program which has been used to make landscapescale assessments of forested ecosystems (Nowak, 1993; Kupfer et al 2008; Oswalt and Oswalt, 2008) In addition, l ands capescale assessments of forested areas are often facilitated by Geographic Information Systems (GIS) (Nowak et al 1996; Xiao et al 2004; Myeong et al 2006; Sales et al 2007; Blackard et al 2008; Hu and Wang, 2008; Oswalt and Oswalt, 2008) Methodologies such as extrapolations of groundbased data, statistical modeling, spatial interpolations, and modeling through combinations of related spatial datasets have been used by researchers to estimate biomass at large spatial scales (Sales et al 2007; Zeng et al 2007; Blackard et al 2008; Hu and Wang, 2008) Emergency managers in Broward County, Florida assert that spatial relationships between land cover and tree biomass can lend insights to locating areas of greater tree debris in the event of a hurricane (Margoles and Umpierre, 2005) Studies sugges t that classification of land cover is an effective way to represent the complexity and variability of urban landscapes (Anderson, 1977; Ridd, 1995; Cadenasso et al 2007) and that degrees of vegetation cover in the landscape are important to development of land cover classifications that are ecological ly relevant (Buyantuyev et al 2007) Temporal and spatial distributions of trees and tree cover are influenced by landuse and land cover change throughout urban systems (Ridd, 1995; N owak et al 1996; Hu and Wang, 2008) S patial distributions of trees as well as their
14 quantity, species composition, and dimensions are cited as principally important in describing urban forest structure (X iao et al 2004) Urban forest structure is influenced by three main factors: complexities of urban landscapes nonconsecutive and variable land cover types; specific land uses within those cover types; and the natural environmental characteristics within and surrounding urban areas (Nowak et al 1996; Xiao et al 2004) Despit e that m any land cover and landuse classifications do not adequately convey the true ecological functioning of urban systems ( Cadenasso et al ., 2007), it remains important to quantify the complexities and heterogeneity of urban landscapes to accurately as sess urban ecological functions (Cadenasso et al 2007) Ecological integrity within and among land cover classes is of primary importance to the accuracy of ecological models applied to the landscape (Buyantuyev et al 2007) Positive correlations between tree dam age and wind speed imply that hurricane damage t o urban forests can be modeled (Duryea et al 2007b) Kupfer et al (2008) suggest that modeling of hurricane damage is possible in spite of an absence of finescale wind data. Others contend that any debris estimation method cannot easily, accurately, and consistently predict realistic results due to complexity and variation of damagedependent variables (Stanturf et al 2007) H owever several models have been developed to estimate hurricanecaused tree debris (COES, 2005; Margoles and Umpierre, 2005; FEMA, 2006; Escobedo et al 2009) The Hazards U.S. Multiple Hazards (HAZUS MH) m odel is disaster management and loss estimation software developed by the Federal Emergency Management Agency ( FEMA) The hurricane module of the HAZUS model calculates estimates of
15 debris based on tree density and tree height data, and also incorporates spatial hurricane wind data from the National Hurricane Center s (NHC) H*wind Model. Based on a single study of windthrow among Ponderosa Pine trees, the model assumes that all trees respond to wind uniformly (FEMA, 2006) and that only trees greater than 30 feet in height will generate debris (FEMA, 2006) In Virginia and North Carolina, this model over estimated tree debris by roughly 90% and 41% respectively (FEMA, 2006; Escobedo et al 2009) The United States Army Corps of Engineers developed a Hurricane Debris E stimating Model based on data collected from hurricanes Frederic1, Hugo2, and Andrew3. M odel input values include the number of households within a given area, extents of vegetation cover, density of commercial dev elopment, precipitat ion, storm category according to the Saffir Simpson scale, and wind speed (COES, 2005) Geographic Information Systems ( GIS ) are used to estimate wind speeds derived from meteorological data analyzed in 5mile wide b ands across hurricaneaffected areas. This model has a reported accuracy of within +/ 30% (COES, 2005; FEMA, 2007) The HurDET model was developed by emergency management staf f in Broward County, FL in 2005 as a GIS based decision support tool for managing hurricane debris. It estimates debris in three categories: building and construction, vegetation, and sediment from storm surge effects. The model incorporates spatial information from 1 Hurricane Frederic made landfall near Mobile, Alabama on 12 September 1979 as a strong Category 3 hurricane on the Saffir Simpson Scale and affected coastal areas of Alabama and Mississippi. 2 Hurricane Hugo made landfall along the South Carolina coast on 21 September 1989 as a Category 4 storm on the Saffir Simpson Scale. 3 Hurricane Andrew made landfall on 24 August 1992 as a Category 5 Hurricane on the Saffir Simpson Scale and caused widespread damage across Southern portions of Miami Dade County, FL, before crossing over the Florida Peninsula to regain strength over the Gulf of Mexico and make a second landfall on the Louisiana coast on 26 August 1992 as a Category 3 hurricane on the Saffir Simpson Scale.
16 seven different land cover classifications and organizes data at the level of county traffic analysis zones. This model categorizes debris according to assumed debris volumes per tree and assumed proportions of tree cover in hardwood trees ve rsus palms. Results of this model vary based on assumed rates of impact on tree cover according to Saffir Simpson storm categories (Margoles and Umpie rre, 2005) The Storm Damage Assessment Protocol Florida Adaption (SDAP FL) is a new extension of the Storm Damage Assessment Protocol (SDAP) included with the USDA Forest Services tree inventory and management software package known as i Tree Storms4. The Florida Adaptation is a Microsoft Excel based tool completed in 2008 (Escobedo et al 2009) It is based on a statistical analysis of tree debris data from project worksheets submitted to FEMA during the 20042005 Florida hurricane season (Escobedo et al 2009) Due to its recent development this tool has not yet been tested with actual hurricane events. Existing debris models with the exception of the SDAP FL, are empirical in nature and have not all been tested as predictive tools (COES, 2005; Margoles and Umpierre, 2005; FEMA, 2006; Escobedo et al 2009) Additional scientific research at the landscape level is needed to guide management of trees and forests for hurricane effects at varying spat ial scales (Stanturf et al 2007) All existing models rely on spatial data (e.g. land cover or spatially modeled wind data) to generate volumetric debris estimates, however none produce spatial estimat e s of debris (Margoles and Umpierre, 2005; FEMA, 2006; Escobedo et al 2009) 4 I Tree is a software package developed by the USDA Forest Service which contains research based tools used to analyze urban forests structure and function and calculate monetary values for associated benefits.
17 Modeling of post hurricane tree debris can improve efficiency of forest managers response to hurricanes (Margoles and Umpierre, 2005; Stanturf et al 2007) Improvements to storm response through debris prediction may include identification of equipment staging sites, location of debris collection centers, procurement of contractors, enlistment of additional human and labor resources, and estimation of costs associated with cleanup and disposal of hurricane debris (Margoles and Umpierre, 2005; Stanturf et al 2007) Resulting damages and associated costs of hurricanes can easily exceed US $1 million per communit y, per storm (Burban and Andresen, 1994; Pielke and Landsea, 1998) This includes damages to both private property and municipal infrastructure. Furthermore, clean up, mitigation, and recovery efforts require large investments in human resources, equipment rental, and contract work (Margoles and Umpierre, 2005) Estimated cost s associated with Hurricane Katrina5 totaled US $81 billion, making this event the most destructive natural disaster in the history of the United States (Oswalt and Oswalt, 2008) Hurricanes generate hundreds of tons of vegetation debris that account for significant portions of total hurricane recovery costs (Margoles and Umpierre, 2005) Research on hurricane affected Florida cities from the record breaking 20042005 hurricane seasons showed that cleanup and disposal of vegetation debris cost communities between US $3,000 and $4 million. On average, the Florida communities 5 Hurricane Katrina made landfall on 29 August 2005 as a Category 3 hurricane on the Saffir Simpson Scale and affected coastal regions of the states of Texas, Louisiana, and Mississippi, only after skirting the Southeastern tip of Florida on 25 August 2005 as a weak Category 1 hurricane.
18 sampled as part of a study in 2008 spent US $704,045 per c ommunity, per storm for vegetation debris disposal (Staudhammer et al 2009) L ocal governments in the U.S. are responsible for assessing quantity of hurricane debris in units of cubic yards and debris disposal costs to meet requirements for reimbursement of Federal monies (FEMA, 2007) Communities must document debris and costs using standardized f orms known as project worksheets to receive federal financial assistance (FEMA, 2007) Project worksheet records are also a data source for researchers investigating debris generation in urban areas affected by hurricanes (Escobedo et al 2009; Staudhammer et al 2009) Urbanized regions of the Southeastern U.S. are annually threatened by hurricanes and tropical storms. I n spite of hurricane fr equency in this region, there is still a great need among local governments for hurricane planning and response tools that can be used to manage hurricane events (Burban and Andresen, 1994; Margoles and Umpierre, 2005) This need was reali zed in the aftermath of Hurricane Andrew in 1992, as plans to deal with hurricane damaged trees were lacking among municipal governments in Florida (Burban and Andresen, 1994) Development of hurricane response tools is further supported by research that suggests hurricane events will likely increase in frequency and intensity throughout the coming decades (Pielke and Landsea, 1998; Balsillie, 2002) In addi tion, the Southeastern U.S. is one of the fastest growing regions of the United States (Campbell, 1997) Such rapid growth suggests that inherent risks to life and property from hurricane events will also increase as urbanizing areas of the Southeastern U.S. continue to develop (Pielke and Landsea, 1998)
19 Objectives The objective of this thesis is to develop a model to estimate landscapescale tree debris volumes in units of cubic yards/acre within urban, hurricaneaff ected ecosystems of the S outheastern U.S. This model and analysis can provide a framework and methods for developing a GIS based tool that communities can use to generate volumetric and spatial es timates of post storm tree debris (See Appendix A) In addition to post storm debris, estimates of intere st include prestorm urban forest biomass. This study will develop a statistical debris model utilizing spatial data as model inputs. This research is unique in that i t will investigate estimations of tree debris as a percentage of total prehurricane biomass to improve spatial estimates To more accurately assess the distribution of prestorm tree debris in cities urban forests, relationships between tree biomass and landcover will be examined. Hypotheses Two principal hypotheses guide this study : Hypothesis 1: Pre storm tree biomass is positively and significantly correlated with post storm tree debris. Hypothesis 2: Land cover data is a statistically significant proxy variable that characterizes prestorm tree biomass and post storm tree debris.
20 CHAPTER 2 METHODS The first steps in this study were to simplify research goals through identification of data sources and opportunities for analysis. This study used urban forest data from hurricaneaffected areas of the Southeastern U.S. to characterize urban forest variability Urban Forest Effects (UFORE) data were available for the cities of Gainesville, Miami, Pensacola and Tampa, Florida (FL), and Houston, Texas (TX). P ost storm measurements of hurricanecaus ed tree debris were collected in Houston, TX, after t he landfall of Hurricane Ike i n 2008. Thus, alignment of preand post storm data focused the investigation on the c ity of Houston, Texas Analysis for this project was performed using English units rather than Metric units because the dependent variable ( e.g. debris) i s reported to the Federal Emergency Management Agency (FEMA) in units of cubic yards (FEMA, 2006, 2007) Therefore, r eporting estimates of debris usin g English units facilitates user friendliness of proposed models and may increase models appeal among potential U.S. users All spa tial analysis procedures were performed with the software package ArcGIS 9.3 and the ArcGIS 9 .3 Spatial Analyst extension. All statistical a nalyses were performed using the statistical analysis software SAS (version 9.1) and the statistical package R (version 2.9.1). In particular, the SAS procedure PROC GLM (General Linear Model) was selected due to the capacity of this procedure to incorporate and test categorical variables such as land cover Methods of analysis are based on those used by Oswalt and Oswalt (2008) in their investigation of forest damages in Mississippi caused by Hurricane Katrina in 2005. In this study, r esearchers visited permanent FIA plots to quantify the scope and extent of
21 damage to trees and forests. Data were then compiled at the plot level and interpolated spatially using GIS to generate maps of damage. Their r esults were assessed by comparing plots prestorm conditions with post storm damage. Research by Kupfer et al. (2008) utilized methods similar to Oswalt and Oswalt (2008) but also introduced hurricane data, such as a variable for each plots distance to the hurricane track and variables for wind speed, wind direction, wind steadiness, and wind duration, from the spatial dataset produced from the NHCs H*wind model. Methods associated with incorporation of USGS land cover data to estimate biomass (and thereby estimate debris according to Hypothesis 1 of this research) at large spatial scal es are based on citations from Blackard et al. (2008). Data Collection Four spatial d atasets contain variables used in this investigation. Dataset 1: 2001 Houston UFORE data These data represent prestorm variability of the Houston areas regional urban forests. The following treelevel and plot level variables were collected: Plot location data: geographic coordinates of plot centroid, street address, and distance & direction measurements to permanent reference objects such as fire hydrants, streetlights, or corners of buildings for plot relocation purposes Land use category (as defined by data collectors) Distribution of plot surface covers : % cement, % asphalt, % pervious rock, % maintained grass, % unmaintained gras s, % shrub, % tree, % duff/mulch, % herbaceous, % water Tree measurements : species, # of stems, number of trees, diameter at breast height (DBH; 4.5 feet) height, crown dimensions, % missing, % dieback, crown light exposure, distance & direction from plot center
22 The research area was defined as the eight Texas Counties surrounding the Houston metropolitan area ( Figure 2 1) which were in nonattainment of the Environmental Protection Agencys (EPA) minimum thresholds for air quality established by the Clean Air Act of 1970 (Merritt, 2009, personal communication) Results of the study were intended to demonstrate importance of urban trees to air quality and classify urban trees as options for reducing air pollution in State Implementation Plans as outlined in section 110 of the Clean Air Act (Merritt, 2009, personal communication. A UFORE study typically consists of 200 randomly located 0.1 acre circular plots within research areas most often defined by political boundaries (Nowak, 2002) F urthermore, woody shrubs and trees with DBH greater than 1 are considered trees and measured for inclusion in the data (Nowak, 2002) The Houston study however did not utilize typical UFORE design elements (Nowak, 2005) Under a modified UFORE protocol, t he Houston study implemented 332 systematicallylocated 0.6 acre circular plots throughout the study area (Nowak, 2005) The study area was divided into a systematic grid and four discrete points were randomly located within each square of the grid (Merritt, 2009, personal communicati on) In urban areas, three of four points /grid square were r andomly selected as UFORE plot centroids ; in suburban areas, two of four points/grid square were randomly selected as UFORE plot centroids ; and in rangeland/agricultural areas, one of four plots/ grid square was randomly selected as a UFORE plot centroid (Merritt, 2009, personal communication) In addition, since the plot s were increased in size from 0.1 to 0.6 acres only trees with DBH greater than 5 were measured to increase the speed and effi ciency of data collection (Nowak, 2005;
23 Merritt, 2009). To capture variability of small trees (<=1 diameter to >5 diameter), a 0. 0 75 acre microplot was randomly located within each UFORE plot (Nowak, 2005) The UFORE model outputs from the 2001 study consist of various measured and estimated urban forest structure and function variables. For each tree on each plot, UFORE outputs include: species, diameter (cm), height (m), condition (e.g. good, fair, poor), leaf area (m2), leaf biomass (kg), carbon sequestration ( kg/year) carbon storage (kg), tree value (U.S.D), notation if tree was located along a street and notation if tree w as a native species (Nowak, 2002) UFORE data were used in this analysis based on its documented common use in other studies of urban forests (Nowak, 2002, 2006) and because of its similarity to data, objectives, and methods used by the Oswalt and Oswalt (2008) study. Dataset 2: Post Ike 2008 Houston UFORE and debris data. This dataset included volumetric measurements of post storm6 tree debris in addition to the following treeand plot level variables, collected according to methods used in the 2001 Houston UFORE study: Plot location data: geographic coordinates of plot centroid, street address, distance & direction to permanent referenc e objects Treelevel va riables: species, # of stems, # of trees, diameter h eight crown dimensions, crown light exposure, % of crown damage, % of defoliation, notation of wind damage and distance & direction from plot center Plot levelvariables: % Tree Cover and %Palm Cover P lot level measurements of tree debris volumes 6 Hurricane Ike made landfall along the Texas coastline on 13 September 2008 as a strong Category 2 Hurricane on the Saffir Simpson Scale and affected local communities throughout the greater Houston metropolitan area.
24 These data were collected in October and November 2008, approximately one month after Hurricane Ike made landfall. A stratified random subset of treed UFORE plots (roughly 10% of the original plots measured) w ere selected for reinventory and measurement of onthe ground tree debris The process of plot selection for reinventory was twotiered. Plots with at least one tree greater than 20 in height and within 15m of a building were selected in the first tie r. This selection of plots was mainly located west of the storm t rack and clustered within Houston city limits A second tier was selected to better represent the geographical distribution of plots in relation to the path of the storm. Criteria f or the second tier of plots dictated that selected plots must have at least one tree greater than 20 but also be distributed east of the stor m track or outside the Houston city limits Plots problematic to revisit due to access or safety issues p lots severely damaged by hurricane storm surge effects or plots having undergone major landuse change between 2001 and 2008 were eliminated from the selection of remeasurement plots according to expertise and consultation provided by the Texas Forest Service (Merritt, 2009). As a result of this process, t hirty four plots were re inventor ied and analy zed (Figure 2 1). Dataset 3: NOAA NHC H*Wind data These data we re produced by the National Hur rican e Center (NHC) at the National Oceanographic and Atmospheric Administration (NOAA) to model behavior of hurricane winds a t varying spatial scales The H*W ind model estimates hurricane wind behavior and assigns estimates to discrete spatial locations across landscapes (Powell et al 1998; Powell et al 2004) E stimate s of winds behavior are based on analyses of data from aircraft flyovers, ships, groundbased towers, and research buoys located across spatial and temporal extents of A tlantic tropical cyclones (Powell and Houston, 1996;
25 Powell et al 2004) D atasets produced by the H*Wind model are developed in conjunction with NHC forecast periods, which take place every six hours while storms are over open ocean and every three hours once a storm makes landfall (Powell and Houston, 1996; Powell et al 1998; Powell et al 2004) H*Wind variables include: Maximum sustained surface wind speed (in m/s, knots, and mph). Wind direction (in degrees) Wind steadiness (indexed value, 0 1 ) Wind duration (in minutes) M aximum sustained surface wind speed refers to winds blowing at a consistent rate of speed for a minimum of one minute, at a distance of 10 meters from the ground surface (Powell et al 2004) The threshold of one minute is a standard measure used by the NHC to designate wind speeds of tropical stor ms and hurricanes. Wind speed rate s are reported in three units: meters/second, knots, and miles/hour. The direction variable report s direction of the maximum sustained winds at the time they were recorded (Powell et al 2004) Steadiness values measure consistency of the maximum sustained winds direction for all recorded wind speeds This variable is more specifically defined as the variation in directional change from mean wind directions over the course of the storm, where low values indicate increased variation in wind direction compared to high values which indicate relative wind steadiness (Powell et al 2004; Kupfer et al 2008) Duration refers to the amount of time maximum winds were blowing at the time they were recorded, however duration values are only assigned to H*Wind data points for modeled wind speeds greater than the minimum wind speed threshold for designation as a category 1 hurricane on the Saffir Simpson scale ( e.g. 74 mph)
26 The H*Wind data are spatially expressed as a grid of points. Modeled values for each of the four variables listed above are attributed to each point on the grid. A combined dataset is produced at the conclusion of the storm that contains maximum values for each point of the grid recorded throughout the storms duration (Powell et al 2004) This output of H*Wind data was used to characterize storm winds for analysis of tree debris caused by Hurricane Ike ( Figure 2 2). A lternative ly, the user could determine which forecast period dataset would be best for use in the debris estimation model ; however, the latter option was deemed more appropriate for this project. The H*Wind data were specifically included in this study based on methods of Kupfer et al. (2008). Use of this dataset is further supported by its incorporation with the HAZUSMH hurricane module. Also, in general, H*Wind data w ere selected to characterize hurricane wind behavior according to the assumption that wind behavior is partly responsible for damage to urban vegetation (Duryea et al 2007a) Dataset 4: USGS NLCD 2001 Land Cover data T hese data are produced by the United States Geological Surveys (USGS) National Land Cover Dataset (NLCD) (Homer et al 2007) This dataset classifies land cover across the geographical extent of the United States, including Alaska and Hawaii Land cover classifications are represented by 30m2 rasterized pixels and were developed according to alignment among several datasets including mul ti seasonal Landsat 5 and Landsat 7 satellite imagery, Forest Inventory & Analysis (FIA) data from the USDA Forest Service, crop land data from the National Agricultural Statistics Se rvice, and ancillary data including maps, orthoimagery, and the 1992 USGS Nat ional Land Cover Dataset (Homer et al 2007)
27 Five of sixteen land cover classifications within the eight count y research area were sampled for this study (Table 21) The five sampled land covers were : Developed High Intensity (DH), Developed Medium Intensity (DM), Deve loped Low Intensity (DL), Developed Open Space (DO) and Woody Wetlands (WW) These land covers were classified based on alignment with locations of post storm re measurement UFORE plots and represent a clear majority of the land area within Houston City Limits (Figure 2 3). USGS land cover data were incorporated in the analysis to represent variation across urban landscapes (Anderson, 1977; Ridd, 1995; Cadenasso et al 2007) and to test the hypothesis that land cover can be used as a proxy variable to characterize prestorm biomass and post storm debris in urban forests ( H ypothesis 2) based on alignment with 2001 Houston UFORE data. This land cover dataset is appropriate because it is standardized across broad regions, readily available, and free to download from the USGS NLCD Multi zone Download Site ( www.mrlc.gov/nlcd_multizone_map.php). Synthesis of Datasets Spatial data for County and City boundaries were downloaded from the Texas General Land Office ( www. glo .state.tx .us ), which is the State of Texas clearinghouse for GIS data. Counties of interest included Brazoria, Chambers, Fort Bend, Galveston, Harris, Liberty, Montgomery, and Waller (Nowak, 2005) City limits of interest were those for the City of Houston. County data were used to spatially identify the study area and the Houston city limit s layer was used to facilitate comparisons of model results with project worksheet data. Data were displayed in Albers projection using North
28 American Datum 1927; this datum and projection is used by the Texas General Land Office to represent data layers of Texas c ounties and citie s. The USGS land cover R egion 10 raster dataset was downloaded and extracted u sing the research area data layer as a mask to isola te land covers of interest Land covers were then assigned to re measurement UFORE plots by e xtracting land covers raster ce ll values to points representing the plots. This procedure appends land cover attributes to UFORE plot data. The most common land covers included: Woody Wetlands (WW); Developed O pen Space (DO); Developed Low Intensity (DL); and Developed High Intensity (DH). Four plots were found to have land covers different from the four previously listed. In these cases, one of the common land covers (WW, DO, DM, DL) found to be directly adjacent was reassig ned to remeasurement UFORE plots based on proximity. The intent of this procedure was to limit the number of land covers included in the analysis. Although reducing the number of categories can reduce model precision, a categorical varia ble such as land cover reduces the experimental error degree s of freedom by one for each category, thus minimizing categories can have a mitigating effect of increasing the sensitivity of statistical tests. Moreover, with four plots having unique land cov ers, a perfect correlation would be made between the dependent variable (debris) and the explanatory variable of land cover. This is neither realistic nor practical for developing a sound statistical model. Hurricane wind attributes were also assigned bas ed on proximity to re measured UFORE plots. The grid of H*W ind data points was overlain with the re measurement UFORE plot layer. A join operation was performed to append H*Wind attributes to
29 UFORE data based on the shortest distance between remeasurement UFORE plots and the nearest H*wind data point The distance to track variable was derived by first downloading a layer of points representing the official forecast track of Hurricane Ike made available by the NHC through their website (Kupfer et al., 2008) A smoothed line connecting each of the points representing the hurricane track was generated; this created the best estimate of the storm s track. Once overlain with UFORE points, the A rcGIS Near tool was use d to determine proximities of each plot to the storm track. Plot to track d istance measur ements were then appended to UF ORE data Results of GIS procedures listed above include an attribute table for 2 008 UFORE plots containing data on debris volume, land cover and wind c haracteristics. These data were concatenated with prestorm data summarized at the plot level from the 2001 Houston UFORE study. Three steps were taken to summarize 2001 dataset at the plot level First, all trees with DBH less than 5 were eliminated from the dataset since these small trees were recorded only within 0.075 acre microplots during the 2001 survey. Although this posed a problem for summarizing plot variables because not all trees could be accounted for, elimination of these data is supported by results of Dury ea et al (2007a) which indicate that small trees are not responsible for significant hurricane damage in urban forests (Duryea et al 2007a) Second, all but 34 plots remeasur ed in 2008 were eliminated from the 2001 UFORE dataset to alig n preand post storm data. R esult s of these procedures reduced 2001 UFORE data to trees with DBH >= 5 within 34 UFORE plots representing 2008 post storm d ebris measurements. Thirdly, 2001 UFORE variables were summarized by calculating the total number of trees; average
30 DBH; average height; and total carbon storage per plot Carbon storage was then mathematically converted to fresh weight biomass. Carbon storage as estimated by the UFORE model can be used to determine biomass. Tree biomass is not esti mated by the UFORE model however the UFORE model calculates biomass as part of the process of calculating carbon storage (Nowak, 2002) Carbon storage is determined by multiplying dry weight biomass values by 0.5 (Nowak, 1993; McPherson, 1994; Hu and Wang, 2008) Thus, UFORE outputs of carbon storage were converted to dry weight biomass values by multiplying carbon storage outputs by two. Hurricanecaused tree debris however, will be fresh weight rather than dry weight (FEMA, 2007) To obtain values for fresh weight biomass, the dry weight biomass values require an additional multiplication of two (Nowak, 1993; McPherson, 1994) To general ize relat ionships between carbon storage, dry weight, and fr esh weight, stored carbon is roughly 25% of trees total weight and dry weight reflects roughly 50% of trees total weight. A master dataset was constructed from plot level tree, land cover, wind behavior, and distanceto track variables. This dataset included fields for: plot identification #, debris (in cubic yards), av erage tree diameter (in inches), average tree height (in feet), total number of trees, total tree biomass (in pounds /plot), percent tre e cover, land cover classification, maximum surface wind speed (in mph), wind steadiness (indexed value), wind duration (in minutes), and distance to track (in miles). Debris volumes collected from the 2008 reinventory were used as the dependent variable in statistical analyses; plot level variables as described above were used as explanatory variables.
31 Average diameter and average height variables were included for statistical analysis based on literature that shows strong correlations with tree biomass (Jenkins et al 2003) especially at large spatial scales (Sales et al 2007; Blackard et al 2008) Tree biomass was included for analysis based on the assertion that c arbon storage is highly correlated with tree and forest inventory measurements such as tree height and diameter (Jenkins et al 2003) and is influenced by natural disturbances such as h urricanes (Jenkins et al 2003) The number of trees per plot and percentage of tree cover were included for analysis according to findings that measures of density help to explain urban forest structure (Xiao et al 2004) and m ight be statistically significant predictors of hurricanecaused tree damages and associated debris (Stanturf et al 2007; Escobedo et al 2009) Other UFORE variables such as tree crown dimensions, tree condition, percentage of missing foliage, and type and percentage of surface covers were excluded from analysis due to a lack of evidence in the literature demonstrating landscapescale relationships between these variables and either biomass or damage and debris estimation, especially in urban landscapes (Francis and Gillespie, 1993; Everham and Brokaw, 1996; Jenkins et al 2003; Sales et al 2007; Duryea et al 2007a; Duryea et al 2007b; Blackard et al 2008; Kupfer et al 2008; Oswalt and Oswalt, 2008) Also excluded from analys is was the wind direction variable from H*Wind data. S ignificance associated with wind direction is specifically related to the unique angle and direction of Hurricane Ikes track in relation to the City of Houstons orientation to the Gulf coastline. The statistical transferability of this variable to other cities is therefore inappropriate. The wind duration variable from the H*Wind dataset was also
32 dropped from the analysis because duration values were only assigned to H*wind data points with modeled wind speeds greater than the Saffir Simpson category 1 threshold of 74 mph (Powell et al 1998; Powell et al 2004) Nine of 34 remeasurement UFORE plots were assigned to H*Wind data points with wind speeds estimated to be less than 74 mph and therefore had no assigned value for duration. Any tree damages experienced in conditions of nonhurricane force winds cannot use duration as an explanatory variable for variation of debris estimates. The distance to track variable was incorporated based on Kupfer et al. (2008) in their use of this variable to model hurricane damage in Mississippis DeSoto National Forest after Hurricane Katrina in 2005. Landuse data was excluded from analysis in favor of land cover data because classifications of l and use are socially and politically applied, whereas land cover classifications are applied based on observations (Ridd, 1995; Cadenasso et al 2007) Statistical Procedures A scatter matrix was generated in SAS for noncategorical variables used in this study to look for relationships between and among the data. A subset of these graphs is shown in Figure 2 4 Exploratory data analysis included descriptive statistics ( Table 22 .) and diagnostic tests (including those for Cooks D, DFFits, DFBetas, and Studentized Residuals performed in R, version 2.9.1) which highlighted outlying observations. In a comparison of strictly numerical variables, the debris value of one obs ervation ( observation # 34) appeared to be an outlier in the dataset. However considering it in context of other explanatory variables, this observation was within an acceptable statistical range for its Woody Wetlands land cover class Woody Wetlands was the land cover class
33 characterized as having the highest average debris volume, and the plot itself was located east of the hurricane track and just outside of Ikes radius of maximum winds (National Hurricane Center, 2008) in the far northwest corner of Liberty County. The plot experienced 8090 mph winds on average according to the H*Wind dataset Examination of the scatter matrix revealed no obvious relationships between any of the noncategorical variables and debris. This could imply that more post storm data should be collected for future analyses or that other variables need to be sought out and tested. This could also imply that a complex multivariate rel ationship exists between groups of two or more explanatory variables which cannot be captured in a twodimensional graph. A n algorithm was employed with the statistical analysis s oftware R (v ersion 2.9.1) to identify variables most desirable for inclusion as model inputs based on four statistical criteria: R2 values, A djusted R2 values, Mallows CP criteria, and Akaikes Information Criteria (AIC). This analysis cannot accommodate categorical variables and land cover was therefore not included. Comparis ons of variables selected through tests of these criteria reveal that number of trees, average diameter, average height, tree biomass, % tree cover, distance to the hurricane track, wind steadiness, and maximum windspeed are significant variables for use as model inputs. Results reported below represent this subset of quantitative variables: CRITERIA R2: R2value = 0.4873 CRITERIA ADJUSTED R2: Adjusted R2value = 0.3512 CRITERIA Mallow`s Cp: Cp value = 8.02 CRITERIA AIC: AICvalue = 62.47
34 To further identify specific variables for inclusion in a statistical model, a stepwise regression analysis was used to systematically eliminate variables according to their contributions to explanations of debris estimates variability. After performing statistical analyses using all selected variables in a Full model, this model is sequentially rerun after eliminating the variable with the lowest subsequent contri bution to variability of estimates This process is repeated until measures of model perfor mance (e.g. R2 and Root Mean Square Error values ) stabilize Verification and Comparison of Results Completion of a statistical model provides the opportunity to create plot specific, volumetric estimates of debris by applying the model to 2001 Houston UFORE plots. Spatial estimation of debris across the Houston city landscape is possible by spatially interpolating measured plot specific volumetric debris estimates. In addition to estimates of debris, measurements of debris recorded during the 2008 post storm re measurement of Houston UFORE plots and biomass estimates derived from 2001 Houston UFORE plots can also be spatially interpolated so that visual comparisons can be made between: Pre storm (2001) biomass and measured post storm debris Pre storm (2 001) biomass and statistically modeled post storm debris Measured post storm debris and statistically modeled post storm debris All spatial interpolations were first conducted using pre and post storm data from the 34 plots remeasured in 2008. Interpol ations performed using these plots could then be more accurately compared. Post storm debris values per plot (both measured and modeled) were subtracted from prestorm biomass values to obtain a mathematical
35 estimate of the mean differences between prest orm biomass and post storm debris in units of cubic yards. Additional interpolations of biomass and modeled debris were performed using all 348 original (2001) UFORE plots. The proposed model is based upon five distinct land cover classifications withi n the Houston city limits. Debris estimates for unsampled land covers were assigned in two ways: 1.) unsampled land covers having measured UFORE plot values for average tree DBH and height were assigned the mean value of sampled land cover parameter values; or 2.) unsampled land covers without measured UFORE plot values for average DBH and height (e.g. UFORE plots without trees) were assigned a debris value of zero. Performing interpolations using all plots provided more accurate visual estimates of debri s. These interpolations also more accurately characterized debris within the Houston city limits so that comparisons could be made with other derived debris values such as a mathematical extrapolation based on mean plot debris values, project worksheet records of total debris removed, and results of other debris models applied to the study area. A comparison with actual project worksheet records from the C ity of Houston w as used to verify the proposed model results, as project worksheet data submitted to FEMA represent the official record of debris removed by the City of Houston. Other model results for comparison were those from the U.S. Army Corps of Engineers Hurricane Debris Estimati ng Model and the Storm Damage Assessment Protocol FL Adaptation. Comparisons of proposed model results with other debris models applied to the study area m ight indicate which model most close ly estimate s
36 project worksheet recorded post hurricane tree debr is in Houston, Texas as a result of Hurricane Ike. The United States Army Corps of Engineers Hurricane Debris estimating model requires as inputs a base value for the number of households and multiplier values for the following: debris (in cy) as index ed by Saffir Simpson storm category, vegetation as indexed by density, commercial/industrial development as indexed by density, and storm precipitation as indexed by quantity. The number of households in the City of Houston, TX is roughly 718,000 according to United States 2000 Census data. The multiplier values are predetermined by the model but the user can select which of those values to use. Eight cubic yards was selected as the debris multiplier based on the designation of Hurricane Ike as a Category 2 hurricane on the Saffir Simpson scale. Median values of 1.3 and 1.2 were selected as multiplier values for vegetation and commercial/industrial densities, respectively, to account for variation amongst these categories across the Houston city landsca pe. A storm precipitation multiplier of 1.3 for medium high precipitation was selected based on documentation of rainfall amounts from Hurricane Ike (National Hurricane Center, 2008) The Storm Damage Assessment Protocol Florida Adaptation requires only the total number of street miles and a user defined designation of the storm events severity and subsequent debris generating capacity referred to as debris rate. The total number of street miles in Houston was determined to be roughly 8,107 according to geospatial roads data for the City of Houston, Texas obtained from the Texas General Land Office. The user defined debris rate was set to each possible value of medium low and high to compare with results of other models
37 Figure 2 1. Research area and 2008 re measurement UFORE plots
38 Fi gure 2 2. H*Wind model gridded points data for the greater H ouston area
39 Figure 2 3. Sampled land covers in the greater H ouston area including developed open space, developed low intensity, developed medium intensity, developed high intensity, woody wetlands, and other (note:other represents all unsampled land cover classes)
40 Figure 24. Scatter matrix of the dependent variable and a sample of explanatory variables. ln_debris is the logtransforme d dependent variable, debris. E xplanatory variables include: distance to the hurricane track = dist_to_track_mi_; biomass = bmlbs_plot; number of trees = not1; and d iameter = dbh1.
41 Table 21. Descriptions of sampled USGS land cover classifications USGS land cover classification Descriptions of USGS land cover classifications Developed open space Includes areas with a mixture of some constructed materials, but mostly vegetation in the form of lawn grasses. Impervious surfaces account for less than 20 percent of total cover. These areas most commonly include largelot single family housing units, parks, golf courses, and vegetation planted in developed settings for recreation, erosion control, or aesthetic purposes Developed low intensity Includes areas with a mixture of constructed materials and vegetation. Impervious surfaces account for 2049 percent of total cover. These areas most commonly include single family housing units. Developed medium intensity Includes areas with a mixture of constructed materials and vegetation. Impervious surfaces account for 5079 percent of the total cover. These areas most commonly include single family housing units. Developed high intensity Includes highly developed areas where people reside or work in high numbers. Examples include apartment complexes, row houses and commercial/industrial. Impervious surfaces account for 80 to100 percent of the total cover. Woody wetlands Areas where forest or shrubland vegetation accounts for greater than 20 percent of vegetation cover and the soil or substrate is periodically saturated with or covered with water.
42 Table 2 2 Descriptive statistics of variables selected for analysis of post storm effects from hurricane I ke on Houston T exas urban forests Statistic Measured post storm debris (cy/ac) # of trees DBH (in) Height (ft) Distance to hurricane track (mi) Pre storm tree biomass (cy/ac) Tree cover (%) Wind speed (mph) Wind steadi n ess (indexed value, 0 1) Minimum 0.00 1.00 6.18 11.00 0.03 12.08 5.00 64.37 0.11 Maximum 5007.98 32.00 29.60 83.00 49.88 42273.73 90.00 87.29 1.00 Mean 197.77 7.53 12.76 38.51 20.42 13035.25 37.97 78.23 0.33 Median 0.25 4.00 12.43 33.75 18.79 7300.83 35.00 79.60 0.34 Standard deviation 865.86 7.94 4.92 16.72 13.06 12686.00 26.59 5.96 0.17 *Per 0.6 acre Urban Forest Effects plot
43 CHAPTER 3 RESULTS Debris Models T hree statistical models of debris were developed. M odels are referred to hereafter as the Full, Final, and Alternate Model. The Full Model includes all variables selected for analysis including land cover ( e.g. categorical variable), distance to the hurricane track, number of trees, tree biomass, tree diameter, tree height, % tree cover, maximum wind speed, and wind steadiness. The Final Model was derived from the Full Model, selecting only those predictor variables that contributed significantly or contributed to model stability. The Alternate Model is a competing model which uses variables from each of the datasets to explain variability of debris. Variables tested are detailed in T able 31. Full Model Log of Deb ris = 0.64 + Land cover (DL= 3.0 7 ; DM= 2.8 4 ; DO= 4. 59 ; WW=0.00)) + Distance to hurricane track* 0.0 4 + Biomass* 0.00003 +# of trees*0.06 + Tree diameter* 0.13 + Tree height*0.06 + % Tree cover* 0.0 2 + Wind speed*0.05 + Wind steadiness*3. 70 This model produced an R2 value of 0.59; the independent variables explain 59% of the variation in the dependent variable (log of debris). This model also has a Root Mean Square Error (RMSE) of 1.78. According to measures of significance outlined in the Analysis of Variance table ( T able 3 2) for the Full Model, Land cover and height are the most significant variables (alpha<0.10) explaining variation in debris. Final Model Log of Debris = 5.08 + Land cover (DL= 3.31; DM= 3 00; DO= 4.89; WW=0.00) + Distance to hur ricane track* 0.04 + Tree diameter* 0.17 +Tree height*0.06
44 The Final model produced an R2 value of 0. 53; independent variables explain 5 3 % of the variation in the dependent variable (log of debris). This model has a Root Mean Square Error (RMSE) of 1.74, slightly less than that of the F ull M odel. According to measures of significance outlined in t he Analysis of Variance table (Table 33) for the Final Model, Land cover h eight and DBH (Alpha<0.10) are the most significant variables explaining variation in debris. Alternate Model Log of Debris = 1.45 + Land cover (DL= 2.5 9 ; DM= 2.5 9 ; DO= 3.91; WW=0.00) + Tree height*0.055 + Tree diameter* 0.16 + Wind speed*0.03 The Alternate model produced a n R2 value of 0.49; independent variables explain 4 9 % of the variation in the dependent variable (log of debris). This model has a Root Mean Square Error (RMSE) of 1.80, slightly more than that of the other two models. According to measures of significance outlined in t he Analysis of Variance table (Table 3 4) for the Alternate Model, like the ot her two models, Land cover DBH, and height (alpha <0.10) are the most significant variables explaining variation in debris. The residuals from the Full F inal and Al ternate Model s showed some pattern s of heteroscadasticity ( Fig ures 3 1 3 2 & 3 3 ), which is a violation of the assumptions necessary for testing a regression relationship. However, no alternate transformation of the data mitigated this problem in any of the three models Development of Debris Estimates The Alternate Model was applied to all 348 UFORE plots measured in 2001 to estimate debris within the study area. The DH land cover was initially aggregated with the DM land cover prior to statistical analysis a nd no additional aggregations of NLCD land covers were performed. Mean separation tests following model development
45 revealed that DM and DL we re not statistically different. The DO land cover was however statistically different from DM and DL, and WW wa s statistically different from all other land covers. Organization of NLCD data aggregates DM, DL, and DO into the Developed category. These land covers were not however aggregated as developed because of statistical differences. Parameter estimates generated for each land cover cannot be applied to untested land covers and therefore parameter estimates from sampled land covers were averaged and applied to unsampled land covers with associated UFORE plot data. Unsampled land covers without associated UFORE data (e.g. UFORE plots without trees) were assumed to have zero debris. Modeled debris values from all 348, 2001 UFORE plots were used to generate a spatial estimate using GIS (Figure 3 4). P redicted values for all modeled plots w ere spatially interpolated by Ordinary Kriging using the Gaussian semi variogram model Ordinary Kriging was used because it is the most common form of Kriging and is therefore most familiar to potential users of the methods and models outlined in this study. The cell size of the Kriging output raster was set to 49.2759 (m) so that each predicted cell would match the 0.6 acre size of the UFORE plots from which the predicted values were interpolated. R aster cell values within the Houston City Limits were then extracted using the city limits lay er as a mask. Cells were next converted to points so that an attribute table of debris values could be exported to calcul ate debris across the landscape. The spatial est imate resulted in an average of 1 5 28 cubic yards of debris per acre. Multiplied by the total acreage of Houston, TX, (~ 407 465 acres) this method estimates 6,227,691 total cubic yards of post storm debris were generated within
46 Houston, T exas city limits ( Figure 35 ) Forty Eight of the 2001 UFORE plots within Houston city limits were isolated to mathematically extrapolate model ed debris values. These plots demonstrated a calculated average of 14.05 cubic yards of debris per acre. When multiplied by the entire acreage in Houston, to tal debris was estimated to be 5,726,587.43 cubic yards. Project worksheet information provided by the City of Houston ( which is the data model results will be compared to) report 6,116,000 cubic yards of debris removed within the Houston City Limits (Victor Ayres, Deputy Director Solid Waste Management Department City of Houston, TX, 2 009, personal communication) Staff from the City of Houston estimate that 75% of that total, or 4,587,000 cubic yards, to be strictly vegetation debris ( Victor Ayres 2009, personal communication) The remaining 25% was categorized as mixed debris which includes unknown amounts of vegetation debris Debris volumes recorded as post Ike project worksheet data from the City of Houston and debris estimates from other debris models applied to the study area are compared with mathematical and spatial estimations of debris results from this study in T able 35. Results of Hypothesis Testing ( Hypothesis 1) A model of debris was dev e loped using plot level tree biomass values as the only explanatory variable to test significance of the biomass/debris correlation. This model had a root mean square error of 2.147 and an R2 value of 0.15; tree biomass explains roughly 15% of debris variation. The pvalue of 0.0256 shows the rel ationship to be statistically significant (Table 36). A n analysis of the biomass/debris relationship in Houston, TX can be made through comparisons between interpolated maps of prestorm tree biomass ( Figure 3 6) and post storm debris measurements ( Figur e 3 7). F igures 3 6 and 37 represent
47 respective variables in cubic yards per acre. The raster for prestorm tree biomass in cubic yards was derived by converting debris volumes from lbs./acre to cubic yards/acre using the conversion factor of 300lbs = 1 cubic yard of vegetation debris as outlined in FEMAs debris management guide (FEMA, 2007) A visual comparison of Figure 3 6 with Figure 3 7 supports Hypothesis 1 ; a reas of higher tree biomass North and East of Houston were the same regions that experienced greater debris. The positive correlation between prestorm tree biomass and post storm tree debris is further illustrated in Table 3 7 that shows the calculated difference between prestorm biomass and post storm debris for the 11 remeasurement UFORE plots that are within Houston city limits. Results of Hypothesis Testing ( Hypothesis 2) A model of tree debris was created using land cover as the only explanatory va riable to test significance of the variables relationship. This model was statistically significant ( Table 3 8 ) and had an R2 value of 0.39; land cover explains roughly 39% of debris variation. Results of this statistical analysis is supported by all s tatistical models, as they show land cover to be explain roughly between 30% and 40% of debris variation. Table 3 9 shows a detail of the estimates of biomass and debris by land cover based on the 34 remeasurement UFORE plots.
48 Figure 31. Plot of predicted versus residual values for full model residual -3 -2 -1 0 1 2 3 4 predicted -1 0 1 2 3 4 5 6 7 8
49 Figure 32. Plot of predicted versus residual values for final model residual -3 -2 -1 0 1 2 3 4 predicted -1 0 1 2 3 4 5 6 7
50 Figure 33. Plot of predicted versus residual values for alternate model residual -4 -3 -2 -1 0 1 2 3 4 predicted -1 0 1 2 3 4 5 6
51 Figure 34. Interpolated raster of post storm tree debris estimates based on alternate model predictions (n= 348) (cy/ac)
52 Figure 35. Interpolated raster of post storm tree debris estimates within Houston city l imits based on alternate model predictions (n= 348)
53 Fi gure 36. Interpolation raster of prestorm, standing tree biomass in cubic yards/acre (n=34)
54 Figure 37. Interpolation raster of measured post storm, downed tree debris in cubic yards per acr e (n=34)
55 Table 31. Explanatory variables t ested Description Units Variable Land cover associated with each plot 0.22 acre pixels L and cover Distance of plots to hurricane track Miles Dist ance to track Total tree biomass per plot Lbs Biomass Total number of trees per plot # trees Number of trees Average tree diameter per plot Inches DBH Average tree height per plot Feet H eight Percentage of tree cover per plot % T ree cover Maximum sustained surface wind speed Miles/hour Wind speed Relative wind steadiness Indexed value (0 1) Wind steadiness
56 Table 3 2. Analysis of v ariance (ANOVA) table for final model Source D F Type III SS Mean square F value Pr > f Land cover 3 26.898 8.966 2.91 0.0564 Distance to track 1 3.787 3.787 1.23 0.2793 Number of trees 1 0.517 0.517 0.17 0.6859 Dbh 1 5.392 5.392 1.75 0.1991 Height 1 9.082 9.082 2.94 0.0996 Tree cover 1 2.283 2.283 0.74 0.3985 Wind speed 1 1.916 1.916 0.62 0.4386 Steadiness 1 8.385 8.385 2.72 0.1128 Table 33. Analysis of variance (ANOVA) table for final model Source D F Type III SS Mean square F value Pr > f Land cover 3 48.781 16.260 5.36 0.0050 Distance to track 1 6.662 6.662 2.20 0.1500 D BH 1 10.882 10.882 3.59 0.0690 Height 1 18.513 18.513 6.10 0.0201 Table 34. Analysis of v ariance (ANOVA) table for alternate model Source D F Type III SS Mean square F value Pr > f Land cover 3 38.394 12.798 3.95 0.0187 Wind speed 1 1.003 1.003 0.31 0.5826 D BH 1 10.419 10.419 3.21 0.0843 Height 1 14.905 14.905 4.59 0.0412
57 Table 35. Modeled debris estimates based on alternate model predictions and comparison with other debris model estimates and official debris volumes as recorded on post Ike project worksheet data from the city of Houston. Debris assessment within H ouston city limits Debris value (average cubic yards/acre) Debri s value Total (1000 cy) Comparison with project worksheet data (vegetation & mixed)* Comparison with p roject worksheet data (estimated vegetation)* Project worksheet debris (mixed & vegetation) 15.01 6,116.00 0 N/a Project worksheet debris (estimated vegetation) 11.26 4,587.00 N/a 0 Modeled debris (mathematical) 14.05 5,726.43 6% +25% Modeled debris (spatial) 15.28 6,227.34 +2% +36% Hurricane D ebris E stimating M odel (USACE)1 8.58 3,494.60 43% 24% Storm Damage Assessment P rotocol (SDAPFL)2 4.65 1,896.00 69% 59% Storm Damage A ssessment P rotocol (SDAPFL)3 .81 329.00 95% 93% Storm Damage Assessment P rotocol (SDAPFL)4 25.66 10,457.00 +71% +123% C omparisons represent the % deviation of modeled debris estimates from project worksheet data recorded by the city of H ouston fol lowing hurricane I ke. Comparison with both the total recorded debris (vegetation & mixed) and vegetation debris (estimated) w as incorporated at the request of Houston city staff (Victor A yres, 2009, personal communication) 1. The Hurricane Debris Estimating M odel was developed by the United S tates Army Corps of Engineers (USACE) and represents the following values as model inputs: # of households = 718,000; debris multiplier = 8 cy; vegetation multiplier = 1.3; commercial/industrial multiplier = 1.2; precipitation multiplier = 1.3 2. The Storm Damage Assessment Protocol FL A daptation (SDAP FL ) was developed by the University of Florida and represents the following values as model inputs: total # of street miles = 8,107; debris rate = medium 3. The Storm Damage Assessment Protocol FL A daptation (SDAP FL) was devel oped by the University of Florida and represents the following values as model inputs: total # of street miles = 8,107; debris rate = low 4. The Storm Damage Assessment Protocol FL A daptation (SDAP FL) was deve loped by the University of Florida and represents the following values as model inputs: total # of street mi les = 8,107; debris rate = high
58 Table 36. ANOVA table for biomass debris model Source D F Type III SS Mean square F value Pr > f Biomass 1 25.289 25.289 5.48 0.0256 Table 37. Differences between prestorm biomass and post storm debris within Houston city limits using urban forest effects (UFORE) plots UFORE plot number Pre -storm biomass (cy/ac ) Post -storm measured debris1 (cy/ac) Post -storm modeled debris2 (cy/ac )) Pre -storm biomass minus post -storm measured debris Pre -storm biomass minus post -storm modeled debris 663 34.772 3.333 2.523 31.438 32.249 704 49.704 6.883 0.632 42.821 49.072 703 7.715 0.417 4.481 7.298 3.234 701 14.022 0.000 2.225 14.022 11.797 733 11.941 0.000 3.758 11.941 8.183 774 79.460 0.417 3.075 79.043 76.385 805 74.557 0.000 0.163 74.557 74.394 843 71.086 12.500 4.367 58.586 66.719 842 39.272 0.000 6.339 39.272 32.933 841 86.935 3.700 3.885 83.235 83.050 1027 113.864 0.000 19.727 113.864 94.137 Average 53.030 2.477 4.652 50.552 48.378 % o f biomass 100.000 4.600 8.700 95.400 91.300 1. M easured debris value s are those recorded from post I ke data collection of 2008 re measurement UFORE plots 2. Modeled d ebris values are those estimated for the same remeasurement UFORE plots based on alternate m odel predictions
59 Table 38. ANOVA table for land cover debris model Source D F Type III SS Mean square F value Pr > f Land cover 3 68.213 22.737 6.52 0.0016 Table 39 Averaged estimates of pre storm biomass and modeled post storm debris by land cover classification Land cover classification Pre -storm biomass (avg. cy/ac) Post -storm modeled debris (avg. cy/ac) % of Pre -storm biomass by land cover Developed low intensity 80.21 8.24 10 Developed medium & high intensity 39.47 5.11 13 Developed open space 75.13 3.56 5 Woody wetlands 137.72 169.42 123 Other (averaged values from sampled land cover classifications) 83.13 46.59 56
60 CHAPTER 4 DISCUSSION Review of Statistical Models and Comparison of Results T he Full Model explain ed the largest percentage of debris variation, however independent variables overlap i n explanations of variability contribute to high model error compared to the Final and Alternate models. The mean square error of the Full Model is higher than both the Final and Alternate Models (T ables 32, 33, & 3 4) indicating higher levels of error associated with estimates of debris. The Final and Alternate Models are therefore more realistic in terms of best fit. Independent variables used in the Final Model explain the greatest variation of debris estimates given this analysis T he Alternate Model has a lower R squared value and a higher error compared to the Final Model, however the Alternate Model may be more appropriate for estimating debris based on its inclusion of wind speed as a variable in place of the distance to track var iable of the Final Model. While the distance to hurricane track variable i s significant as detailed by the Final Model, it was dropped as a variable for the Alternate model due to presumed error that may be present due to Hurricane Ikes unique morphology and large radius of maximum winds (National Hurricane Center, 2008) Wind speed, as modeled throughout the temporal course of hurricane events, may therefore be better suited for inclusion as an independent, storm specific variable in a predictive model. T he Alternate model can be further justifi ed based on a slightly better distribution pattern of data spr ead when comparing graphs of residual versus predicted values generated by the model implying that the model is more consistent in its predictions than the Final Model
61 Analysis of predicted ve rsus residual graphs for any of the three models developed in this study reveals apparent heteroscadasticity. The wedgeshaped gap in the data spread of these graphs (located in the lower left corner of Figures 5, 6, & 7) indicates that any of these models have higher degrees of error when making low estimates of debris in co mparison to higher estimat es. Although data transformation can usually mitigate this issue, statistical analysis did not reveal any worthwhile transformation. This indicates the model could benefit from additional data sources not collected or included for analysis. M odels developed in this study are based on data that only i nclude prestorm trees greater than five inches in diameter It is therefore likely that the exclusion of small trees f rom the dataset is a contributing factor to heteroscadasticity of models results. A comparison of statistical and spatial debris estimates (using the Alternate Model) with reported project worksheet debris volumes indicate that model resul ts over p redicted d ebris generated as a result of Hurricane Ike Accuracy of project worksheet data however is suspect (Victor Ayres, 2009, personal communication) Debris removed in Houston could have been altered by homeowners blending yard debris wit h hurricane debris, debris remaining uncollected in natural areas or difficult to access sites, or private removal of debris (FEMA, 2007) The comparison with total debris removed (Table 35) is based on personal communications with Houston city staff who indicate that while only 75% of the total debris ( 4,587,000 cy) can be said to be strictly vegetati on, the impacts of the storm did not generate much building/construction debris and it is therefore likely that vegetation debris represents the overwhelming majority of total debris removed. This is in line with other southeastern US debris assessments
62 which have determined that vegetation consistently makes up the majority o f post hurricane debris (Escobedo et al ., 2009; FEM A, 2006). Despite Alternate Model over prediction, this model estimated debris within a small percentage of error (T able 35). Spati al estimates from t he Alternate Model suggest there were 6,227,000 cubic yards of debris that were generated by Hurricane Ike in Houston, TX; total debris recorded on project worksheets indicate 6,116, 000 cubic yards of debris or 2% less than the Alternate Model s spatial estimates for the same area. The perceived accu racy of the spatial estimate is encouraging because it lends validity to accuracy of modeled debris location in addition to debris volume. The Alternate models performance is not too surprising since project worksheet records of tree debris and data used to develop the Alternate Model are both based on post storm conditions resulting from the impacts of Hurricane Ike. Although the SDAP FL model is also based on post storm conditions from a sample of communities affected by hurricanes during the 2004200 5 Florida hurricane season (Escobedo et al., 2009; Staudhammer et al., 2009) this model did not provide accurate estimates of debr is compared with Alternate Model estimates and post Ike project worksheet records from Houston, TX The total debris estimated using the SDAPFL low debris rate is clearly far too small to make reasonable comparisons with rec orded debris volumes. Use of the medium debris rate would seem appropriate based on the designation of Hurricane Ike as a strong category two hurricane (National Hurricane Center, 2008) yet medium results under estimate d debris by roughly 60 70% compared wit h debris volumes recorded on post Ike project worksheets. The high debris rate from the SDAP FL over estimated debris by roughly 70120%,
63 clearly far above recorded volumes of debri s. The relatively poor performance of the SDAPFL model is not surprising considering the geographic, biological, and meteorological differences between the Houston study site and the data upon which this model is based. The Army Corps of Engineers Hurri cane Debris Estimating Model also produced relatively inaccurate estimates of debris compared with project worksheet data and results from the Alternate Model. One possible explanation for the poor results is that th is Hurricane Debris Estima ting Model is empirical in nature whereas the Alternate Model is based on scientific analys e s. Another possibility is related to the debris value multiplier of eight cubic yards which is assigned based on the designation of Hurricane Ike as a Saffir Simpson categor y two hurricane (National Hurricane Center, 2008) Hurricane Ike was a strong category two storm o n the verge of the category three wind speed threshold at landfall (National Hurricane Center, 2008) The relative strength of Hurricane Ike as a category two storm could have generated greater debris than is estimated by the Hurricane Debris Estimating Model using eight cubic yards as the debris rate multiplier. Other relatively poor outcomes of this study were related to the statistical contributions of storm related variables to estimates of debris. H urricane wind variables derived from the H*wind model made little contributions to explanations of debris variation. Wind is often assumed to be the primary driver of debris generation, y et as determined by this and other studies, the wind speed variable is a poor predictor of vegetati on debris (Kupfer et al 2008; Staudhammer et al 2009)
64 Hurricane v ariables such as steadiness explained even less variability of debris estimates when compared to wind speed. One possible explanation is that tree damage related to preexisting decay or weak tree stru c ture may occur at wind speeds far below maximum values modeled by H*Wind, and therefore the model may not capture this relationship using extreme values for wind as model inputs. Other studies of hurricaneforce winds on forested ecosystems su pport findi ngs that storm related variables explain less variation of damage or debris than nonstorm variables (Stanturf et al 2007; Kupfer et al 2008; Staudhammer et al 2009) and indicate that interactions between forest damage and hurricane winds are complex and difficult to analyze statistically Review of Hypothesis Testing The positive correlation and statistically significant relationship between prestorm urban forest biomass and post storm debris results in a failure to accept the null hypothesis of Hypothesis 1. Limitations and error associated with the data may be responsible for prestorm biomass explaining only 15% of debris variability and being therefore excluded as a model input through the stepwise process The positive correlation implies however that future development of a spatial debris estimation tool can utilize measures of prestorm biomass to partition post storm debris estimates as a percentage of total biomass. To better understand the biomass/de bris relationship in Houston, TX as a result of Hurricane Ike, Table 37 shows that total post storm tree debris in Houston represents between 4.6% and 8.7% of Houst ons 2001 total urban forest biomass Table 39 displays averaged estimates of debris per acre by land cover within the study area, and shows those estimates as a percentage of the average biomass in
65 cubic yards per acre by land cover The developed land covers show Hurricane Ike to have generated between three and nine cubic yards per acre which represent between five and thirteen % of total biomass per land cover in cubic yards per acre. The Woody Wetlands land cover however shows a dramatic increase to over 164 cubic yards of debris per acre representing 123% of the total biomass per acre. Debris estimates exceeding the t otal biomass by such a large margin is seemingly improbable and is likely a consequence of the extremely high debris values measured on plots with Woody Wetlands land cover during the post Ike data collection. Such high debris values from the data skew the models parameter estimate for Woody Wetlands UFORE plots of within the study area. Land cover is responsible for roughly 40% of the variability of debris estimated by the model s. The strength of this variable is demonstrated by its inclusion in all three models developed by this study While this cannot confirm nor disprove the hypothesis that land cover can be used as a pr oxy for variation of urban forest prestorm biomass and post storm debris ( Hypothesis 2), it does lend support to this concept and i s therefore a highly recommended for future research. Applicability of Findings Many results of this study are in line with scientific principles and related research (Kupfer et al., 2008; Oswalt and Oswalt, 2008; Escobedo et al., 2009; Staudhammer et al., 2009) A qualitative comparison of interpolated maps of measured bi omass and debris (Figures 3 6 & 3 7) reveals that areas of greater prestorm biomass to the north and east of the City of Houston also yield the greatest measures of post storm debris. These results can also be justified by methods of analysis used in the initial 2001 Houston UFORE study, where the N orthern regions of the study were analyzed
66 separate from Southern regions based on ecological differences between the type, percentage and density of tree and forest cover within those regions (Nowak, 2005). The Northern region is more rural and has greater forest density, particularly pine forest (Nowak, 2005). The southern region however, is more urbanized and therefore has lower tree density and percent canopy coverage (Nowak, 2005) which is likely responsible for lesser debris volumes Similarly, it was expected that greater debris would be measured east of t he hurricane track based on meteorological knowledge of hurricane wind intensities (Elsner and Birol Kara, 1999) Indeed, greater wind speeds we re generally found east of the hurricane track, where greater post storm debris was also measured by this study. This is supported by Figure 3 4 showing greater volumes of tree debris east of the hurricane track and lesser volumes to the west. It is possible however that the greater debris volumes to the North and East of Houston are related to both the path and morphology of hurricane Ike and the ecological landscape structure throughout the study area. A review of F igures 3 4 and 37 showing modeled and measured debris respectively, show s that less er debris volumes were recorded within the highly urbanized boundaries of the City of Houston. Greater debris volumes were observed in suburban and rural areas throughout the study region, with a bias toward more debris in the more densely forested Northern regions. Of the variables used, both tree height and diameter were demonstrated to be the strongest explanatory variables for debris variation compared with other plot level urban forest measurement s (Tables 32, 33, & 3 4) This is supported by related research of both biomass estimation (Jenkins et al
67 2003; Blackard et al 2008) and hurricane damage to forests (Everham and Brokaw, 1996; Kupfer et al 2008; Oswalt and Oswalt, 2008) Despite evidence from related research showing significance of forest density to forest damage or vegetation debris generation, measures of forest density used in this study (% tree cover and number of trees) showed little contribution to explanations of debris variati on. One possible reas on for this is that debris variation explained by measures of forest density may have been overshadowed by the categorical USGS land cover variable, which is modeled according to overlap of ancillary datasets including landscapescale estimates of percent tree cover derived from USDA Forest Service FIA measurements (Homer et al 2007) Statistical results of hurricane variables explanation of debris variation included in this study were similar to findings of the study by Kupfer et al. (2008). In both studies, the distance to track variable made similar contributions to over all debris or damage estimates and t he H*Wind variables explained very little v ariation of debris or damage. The Distance to track variable, as well as H*Wind variables may have been confounded by the morphology of Hurricane Ike. Hurricane Ike had an unusually large eye estimated to be 74 Kilometers in diameter (National Hurricane Center, 2008) which is 4 K m larger than the upper range of average hurricane eye diameters as outlined by Elsner and Kara (1999). Such a large diameter for this hurricane extends the radius of damaging winds across the landscape and thereby may be responsible for pat terns of damage observed, measured, and modeled by this study. As an example, 19 of 34, or roughly 56% of the selected post storm re measurement UFORE plots fell within the radius of Hurricane Ikes eye as it moved across the landscape and only two re-
68 measurement UFORE plots on the more damaging (east) side of the hurricane track were outside the eye wall of Hurricane Ike. Yet despite the intensity of Hurricane Ike, perhaps the most i mpor tant implication from results of this study is that debris generation in Houstons urban forest is mostly dependent on nonstorm variables from a statistical perspective Determination of nonstorm variables having significant influences over debris generation is supported by both Kupfer et al. (2 008) and Staudhammer et al. ( 2009) in their studies of hurricanerelated damage and debris of forested landscapes. Additional support comes from observations of Hurricane Andrews effec ts on vegetation and research by Duryea et al. (2007a), suggesting that hurricanecaused tree damage can be attributed to trees being improperly located, poorly maintained, or consisting of species with low tolerances to hurricane forces (Burban and Andresen, 1994; Duryea et al 2007a) The benefit of this finding is that managers of urban systems have relative control over nonstorm variables. Hurricanes cannot be prevented, whereas land cover and urban forest structure and condi tion can be manipulated using management strategies and governing policies. Limitations Due to financial and staffing constraints, post storm data collection did not take place until 8 October 2008, approximately twenty five days after landfall of Hurricane Ike. During this timeframe, much of the debris may have been moved, or removed from plots during any debris cleanup and disposal operations Lesser volumes of debris were therefore likely to have been recorded throughout the study area. T he timing of data collection may have contributed to this studys limitations however d ebris measurements relation to geographic, environmental, social, and
69 political attributes of the plots selected for remeasurement should also be taken into consideration. The selection of remeasurement UFORE plots was influenced by input from the Texas Forest Service (TFS). In consultation with data collectors, TFS staff advised exclusion of particular plots from the original tiered selections of remeasurement UFORE plots based on local knowledge of the study area. Plots with safety or access issues, plots severely damaged by storm surge, and plots experiencing major landuse change were eliminated from the dataset. Decisions to eliminate plots may have streamlined the data collection process but may have also introduced unquantifiable bias. It is also likely t hat t he full ext ent of interactions between debris dependent variables during hurricane events has not been fully quantified through scientific research Model results may be adversely affected by complexities of such interactions. Also, variables select ed for analysis may not accurately model tree debris generation in urban areas given the data available for this study Particularly notable is the seven year gap between pre and post storm data The 2001 data are only representative of that year and do not account for changes to urban forest health and structure that occurred between 2001 and 2008. Such changes include but are not limited to factors such as tree growth, tree pruning, tree mortality, tree planting, and/or tree removal. Another explanation for variables inability to accurately capture variation in debris estimates could be t he increased error associated with the use of modeled data as explanatory variables Both the H*Wind data and the USGS land cover data are modeled data products based on incorporation and analyses of related datasets. Indirect data sources will have higher error than direct measurements of the same
70 variables; this error therefore affects accuracies of any analyses utilizing them. Furthermore, spatial estimates of debris are affected by compounding the error of statistical analyses with error associated with interpolation of the data.
71 CHAPTER 5 CONCLUSIONS Results show that nonstorm related variables explain greater variation of debris estimates than storm related variables such as wind speed. This finding implies that tree failures and resulting debris are more closely associated with treeor site specific factors than meteorological variables. Another possibility may be that most tree failures occur at winds speeds below the modeled maximum values offered by the H*Wind dataset, and therefore the data does not accurately capture the full extent of relationships between wind speed and hurricanecaused tree debris generation. These results did however reinforce similar findings from related studies (Kupfer et al., 2008; Oswalt and Oswalt, 2008; Staudhammer et al., 2009) While the U.S. Army Corps of Engineers Hurricane Debris Estimating Model and the Storm Damage Assessment Protocol FL Adaptation produced relatively poor estimate s o f debris in Houston, TX, the results from the Al ternate Model developed by this study estimated tree debris within a relatively small margin of error compared with project worksheet data. R ationales for the performance of the SDAPFL and Hurricane Debris Estimating Model in Houston, TX, remain speculative until additional data collection, analysis, and research is performed to test debris estimations produced using these models Similarly, t he Alternate Model may be further verified using results and data from future post storm inventory of UFORE plots in other hurricaneaffected cites. Overall, the sample size and other limitations of the data hinder the accuracy of model s developed in this study These model s may be improved through the exploration and incorporation of yet untested variables. New models with greater
72 accuracy and precision may be developed in the future with the aid of additional data such as post storm debris measurements from future hurricane events and finer scale representations of both wind behavior and land cover variation. In the biomass debris model the tree biomass variable demonstrated a positive correlation with debris, implying that debris in creases as tree biomass increases. This result supports Hypothesis 1 ; however the relationship was not as strong as expected. Limitations of the dataset and i nteraction with other variables are likely responsible for th e relative weakness of this result. In addition, further e fforts to include biomass as a statistically significan t input variable in a debris estimation model were also likely hampered by limitations of and interactions among the data and therefore further testing of this hypothes is is recommended. In spite of a weak relationship between pre storm tree biomass and post storm debris the biomass debris relationship did not seem to adversely affect H ypothesis 2 that land cover is a proxy for capturing variability of pre storm tree biomass and post st orm tree debris. S tatistical analyse s, results of all three debris m odels and the land cover debris model show that land cover explains between roughly 38% and 49% of variability associated with debris estimates. This percentage, compared with R2 values of the Final and Alternate models suggest that land cover i s responsible for a disproportionally large contribution to variability of debris estimates. Further analysis and testing of this hypothesis with future datasets may reveal an even stronger relationship. The use of urban forest structure data as inputs to a post storm tree debris model highlights the opportunity for exploration of yet another new, practical, and scientific
73 application of UFORE data. Other unique implications of this research include the introduction of two concepts that (assuming verification t hough additional analyses and tests) may prove useful for generating vegetation debris estimates at large spatial scales in urban areas: 1.) Estimation of post storm tree debris as a percentage of prestorm tree biomass; and 2.) Estimation of debris using land cover data as a proxy variable for capturing variation of urban forest prestorm biomass and post storm debris. Hurricanecaused tree debris can be modeled in urban fores ts ; however analysis of future storm events may lead to more accurate verificati ons of results C ontinued refinement of both NLCD land cover data and H*Wind modeling by the USGS and the NHC respectively will facilitate availability of more accurate and/or precise data for analysis in future investigations of a similar nature. Future Research Opportunities The scope, scale, and original intentions of this study reveal numerous possibilities for continuing research. The presence of UFORE data in Houston, TX in addition to other Southeastern U.S. cities including Miami, Tampa, Gainesvil le, and Pensacola, FL, indicate that many opportunities exist for future data collection and analysis This is especially true considering that each of these cities is likely to experience future hurricane events. Therefore, future post hurricane debris data should ideally be collected and analyzed from any and all hurricane affected cities with UFORE data to test debris model results from this study and compare them with project worksheet data and estimates provided by other debris models. Additional analyses seeking explanations for low correlations between hurricane variables and hurricanecaused tree debris generation should also be pursued.
74 Furthermore, it is recommended that v ariables excluded from this analysis should be tested for correlations with post storm debris S uch variables include but are not limited to : geographic variables such as soil type, topography (elevation and aspect), ground surface cover percentages, and urban development patterns; storm variables such as rainfall amounts s torm size, forward storm speed, Saffir Simpson storm category, wind direction, radius of maximum winds and hurricane eye wall diameter; and tree and forest variables such as fore st edge effects, rooting space, measures of tree condition/health, and tree characteristics such as wood density, wood elasticity, plant allometry, tree crown density and dimensions, average tree lifespan, size at maturity, and categorical variables such as native vs. nonnative, and dec iduous vs. coniferous Many of the yet untested variables as described above represent a challenging but potentially important area of study to model s pecies hurricane response at landscape scales according to incorporation of results from Duryea et al (2007a, 2007b). F indings from Duryea et al (2007a, 2007b) and Jenkins et al. (2003) may also be used to model s pecies specific contributions to overall debris from partitioned components of tree structure such as leaves, stems, branches, and roots. Fu rthermore, development of biomass equations for tree species common to urbanized, hurricaneaffected cities of the Southeastern U.S. can also improve estimations of biomass and debris at the individual tree or species level. Although species specific vari ables may be difficult to extrapolate to varying spatial scales, results from such analyses could produce far more accurate debris estimations. In addition, modeling hurricane s capacities to alter urban forests through tree canopy cover or tree density c hanges may
75 be worth pursuing as such forest measures have been shown to have significant relationships with hurricane damage at landscape scales (Everham and Brokaw, 1996; Stanturf et al 2007) I mprovements to accuracy of landscapescale debris estimation will also yield rel ated improvements to landscapescale estimation of urban forest biomass. Further exploration of this relationship with future datasets may yield more robust results that can be used to develop more advanced and accurate models of debris (both statistical and spatial), including the development of a spatial GIS based debris estimation tool ( Appendix A). Incorporation of a statistical model such as those offered by this study, with the geospatial capabil ities of GIS can enable model estimates to be interpolated across landscapes and thereby generate spatially explicit estimations of debris which are not currently outputs of other debris models. Provided that methods outlined in this study lead to improvem ents in biomass estimation of urban forests, another opportunity for related research is in developing landscape scale estimation of urban forests stored carbon. Several articles reviewed for this study indicate the importance of the biomass/carbon relat ionship and the importance of carbon storage estimation to emerging trends in tree and forest ma nagement (Nowak, 1993; McPherson, 1994; McNulty, 2002; Jenkins et al 2003; Myeong et al 2006; Woodbury et al 2007; Hu and Wang, 2008) In this study, biomass was c alculated as a proportion of carbon storage values provided by UFORE model outputs. Therefore similar methods can be used to statistically analyze and spatially interpolat e UFORE values of carbon storage. Cities with UFORE data can estimate prehurricane values for carbon storage and biomass
76 based on UFORE model outputs. As such, debris model s produced with methods outli ned by this study can enable users to also estimate post hurricane tree debris and post hurricane changes to urban forest biomass and carbon storage at landscape scales Results of this project provide urban forest managers in Houston, Texas, with a verified debris estimation model that can be implemented for use and tested for effectiveness against future hurri cane events. Also, this project has identified another new and practical application of UFORE data. The design and outcomes of this research project reinforce evidence from the literature that methods for data collection and analysis of forest ecosystems can also be effectively applied to the study of urban ecosystems. Opportunities for continued research are numerous and results from this study will provide future researchers and urban forest managers with greater insights into disturbance responses of urban forests and related factors which drive hurricanecaused tree debris generation in urban ecosystems.
77 APPENDIX A A FRAMEWORK FOR DEVELOPING A GISBASED DEBRIS ESTIMATION TOOL Introduction This appendix describes a framework for integrating this studys results into a GIS based tool that relies on statistical models using data that are representative of the S outheastern United States. Incorporation of statistical models can allow the GIS t ool to be used to either predict or verify post storm debris volumes. Any model s or tools developed should meet the following criteria: It should be relatively easy to use The data should be readily available and come at little to no cost to the user. I t must cover the broad geographical scope of the Southeastern U.S. It must be temporally relevant The incorporation of GIS and other technologies for measurements of spatial relationships will play an importan t role in the development of a debris estimati on tool. However, methodologies for debris estimation by the enduser must remain simple and practical in order to ensure that they compliment rapidassessment protocols. The use of GIS technology provides flexibility for future research and incorporations of other data. For example, supervised classifications via remote sensing and GIS analysis may be found to have relevance or positive relationships to methods outlined in this project. If methods used in this study are found to contain commonalities f or debris estimation in conjunction with remote sensing, then this could lead to developments in methodologies that improve accuracy and ease of use for both statistical models and any GIS interface that may be developed. The following pages of this appendix describe a framework for a GIS based tool used to calculate hurricanecaused tree debris both spatially and volumetrically within a
78 predefined area for a selected storm and is based on statistical model s referenced in Chapter 3 The spatial analysis and calculations of debris will be performed within the model builder tool that is packaged with ArcGIS 9.3. This first iteration of the model provides a basic framework of programming language upon which future improvements may be made to finalize the tool. The framework for this tool has two major components: 1.) prediction of debris based on statistical and spatial analyses; and 2.) extraction of debris values according to user defined criteria and provision of outputs that describe debris volumes and locations. The work performed as outlined in this appendix focuses solely on the second component: Assessing, quantifying, and determining relative spatial locations of hurricanecaused tree debris. This was possible due to having debris data collected in Houston, TX in the wake of Hurricane Ike. As a result, the tool as described here utilizes the actual post storm data from Hurricane Ike. The two primary objectives for producing this component of the tool are: To return an output data layer to the map of the area of interest as a Points shapefile that expresses values of tree debris in terms of pounds/acre within a 60 meter buffer of all major roads in Houston, TX. To return output that reports the total amount of debris in terms of pounds/acre for the entire extent of the map of the area of interest. Methods The debris model was built in six general steps. A listing and breakdown of each step is as follows:
79 Step 1: Obtain and/or download the data There are two input data files for this project. The first file is a polyline shapefile of roads data available from the Texas General L and Office: ( http://www.glo.state.tx.us/gisdata/gisdata.html ). The second is a points shapefile of Urban Forest Effects (UFORE) plot data collected by the Texas Forest Service in 2001 This data was updated by researchers from the University of Florida in August of 2008 following Hurricane Ike. As described in Chapter x of this study, 348 UFORE plots were randomly located throughout the greater Houston Metro Area for the 2001 measurement; 34 of them were reinventoried following the storm (see Appendix A for description of how these plots were selected). These 34 plots are the foundation upon which the statisti cal model was built. The data are displayed in ArcGIS 9.3 using the NAD 1983 UTM Zone 15N coordinate system and Transverse Mercator projection. 6. Test and evaluate the model 5. Draft the VBA code that processes input data and produces outputs 4. Construct a form as a graphical user interface in Visual Basic Applications 3. String the necessarry GIS tools togther in ArcGIS Model Builder 2. Develop the model framework by first performing individual processes in sequence 1. Obtain and/or download the data
80 Step 2: Develop the model framework by first performing individual processes in sequence. A series of tools from ArcGIS 9.3 were used to produce the final output data layer. This process helped to determine exactly which tools were necessary and how those tools should be executed. For example, through iterative trials it was determined that a 60meter buffer o f the major roads was the optimal distance for capturing the values of interest (debris). A 60 meter buffer effectively captures data from the 30x30 meter cells that comprise the interpolated raster of debris in ArcGIS. Step 3: Stringing the necessary GIS tools together in ArcGIS 9.3 Model Builder A screen capture of the model and a breakdown of the specific tools (Fig A 1 ) are as follows where blue ovals = input data; yellow rectangles = ArcGIS tools performing functions on data; green ovals = data r esulting from ArcGIS tool processes; and P indicates which data are model paramet ers. Figure A 1. Model developed using ArcGIS Model Builder Kriging (Figure A 2): The plot layer is the input layer for this tool. Based on debris values in the UFORE plot layers attribute table, this operation constructs a rasterized prediction map of debris for unsampled portions of the map. The output cell size is set at 60 (m) to strike a
81 balance between detail of the output and efficiency of the tool. A larger cell size covers more land area and limits spatial detail whereas a smaller cell size, while providing greater detail, slow down the performance of the tool due to doubling the size of the data produced. Default settings were used in all other cases. Figure A 2. Screen capture of ArcGIS Kriging operation Buffer (Figure A 3): The roads layer is the input layer for this tool. A 60 meter buffer that integrates the data by dissolving ALL boundaries is designated as the optimal buffer. Default settings for kriging method, semivariogram model, search radius, and number of points were used in all other cases. The following tool creates the buffer.
82 Figure A 3. Screen Capt ure of ArcGIS Buffer operation Extract by Mask (Figure A 4): The results from the Kriging and Buffer operations are the input layers for this tool. The roads buffer is a mask used to extract the raster values for debris predicted by the Kriging result. Figure A 4. Screen capture of ArcGIS Extract by Mask operation Raster to Point (Figure A 5): The result from the extract by mask tool is the input layer to this raster to point tool that converts the raster cells to a points file and returns this final output layer to the map.
83 Figure A 5. Screen capture of ArcGIS Raster to Point operation Step 4 (Fig ure A 6 ): Construct a form as a graphical user i nterface in Visual Basic A pplications scripting language. The user interface for the tool prompts the user for a roads layer, a plot layer, and a filename to run the tool and produce the end result. Figure A 6. Screen capture of graphical user interface developed for this tool with ArcGIS. Step 5 ( Figure A 7 ): Draft the VBA code that processes input data and produces outputs. The following flowchart outlines the process of the VBA code:
84 Figure A 7. Process of ArcGIS programmed subroutines to execute the debris estimation tool. Step 6: Test and evaluate the model. Once all the components of the model were drafted, an iterative process was undertaken to search for mistakes and problems that hindered the models performance. Results This project generated the following results which are displayed below in two screencaptures of the final output layer (Figures A 8 and A 9) Pushing the button on the application actives the form (interface) The 'initialize' subroutine loops through the map layers and populates the form If selected, the 'cancel' subroutine quits the form and returns the user's attention to the application The 'run' subroutine passes input layers to model builder which runs the tools and produces the output layer The 'returnTotal' function loops through the attribute table of the output layer, sums the total debris, and returns this to the user
85 Figure A 8 Screen capture of a points file of tree debris values expressed in pounds/acre. Figure A 9 A n enlargement of Figure A 8 showing finer detail
86 As shown in the screen capture below (Figure A 10), the total hurricanecaus e d tree debris along major roads in Houston, Texas is 659065 lbs/acre (screen capture below is in error; it should read lbs/acre instead of cubic yards). The caveat is that this number is not divided by the acre age of the roads buffer which was used as a mask to extract the values and therefore is not specific to the actual land area of interest. This figure is helpful for determining total debris within the user designated area; however a per acre average cannot be determined without dividing the total debris by the acreage of the area. Figure A 10. ArcGIS 9.3 message box returned to user with model results This screencapture of the output layers attribute table (Figure A 11) shows the debris values in the column named GRID CODE. The model generated a total of 26,425 unique points along major roads in the City of Houston. The vast number of points generated demonstrates the explicit spatial nature of the output. For exampl e, the user can identify specific points within specific areas of interest to identify a unique debris volume that is estimated to be found there.
87 Figure A 11. Sample of interpolated debris values extracted to points layer Discussion: Despite limitations of the tool, t he results from this study and methods outlined in this framework can be used to develop a GIS based debris estimation tool Using this study and framework, the city of Houston now has a simple and user friendly means to spatially and vo lumetrically estimate post storm, hurricanecaused tree debris whereas no such method has previously existed. In general, the graphical interface of ArcGIS Model Builder facilitated smoother and more efficient development of this tool by
88 comparison to the rigorous and less intuitive script writing process in Visual Basic A pplication To determine accuracy of the tools result for Houston ( 659,065 lbs/acre), debris estimates produced by this tool may be tested and verified by comparing them with actual debris values recorded on project worksheet data submitted to FEMA by affected communities for reimbursement purposes. The most import test will be a comparison of the tools results with actual project worksheet data submitted by the City of Houston for r eimbursement of post Ike debris disposal. Limitations There are several limitations of the model, ways that it may be improved, and options for development of future models The value of debris reported is in lbs/acre, whereas measurements in cubic yards are a more desirable measure as these are the units requested by FEMA for submittal of project worksheets. Simple conversion factors can be programmed into the tool so that l bs/acre can be converted to cubic yards. In doing so, it is possible to have the tool report estimates of debris in both lbs/acre and cubic yards. The value of total debris reported doesnt consider the acreage of the buffer. The value returned to the user is in units of lbs/acre, however this number is not divided by the total acreage of the area of interest, which in this case is the buffer of major roads. Steps should be added to model builder that first convert the debris values from pounds of debri s/ 0 1 acre to pounds of debris/acre. This currently does not exist and can be done by using ArcGIS tools that add and calculate a new field within the UFORE Plot layer attribute table. The same process can be employed to convert the area of the roads buf fer from square meters to acres.
89 The next step is to build another Inquire Cursor in VBA that loops through the acreage of the buffer, sums it, and then divides the total debris/acre by the total acreage in order to return to the user a value of total debris for the area of interest. Ideally, this tool should be predictive. The true benefit is that it may predict debris quantities before a storm strikes; the input data used to determine debris was based on post storm measurements. At this time, the tool has no predictive capacity. The statistical analysis of available data ( Chapter 2) needs to be incorporated so the tool can estimate debris based on prestorm input variables. This will enable the model to be predictive and therefore be far more useful to the user. This tool does not rely on estimations of biomass derived from prestorm UFORE data. The nature of this relationship may prove useful to the ultimate development of an accurate, predictive, spatial model of hurricanecaused tree debris. F urther exploration and analysis of the relationships between prestorm urban forest biomass post storm tree debris may yield results that can be applied to the tools programming processes in order to produce more accurate spatial predictions of hurricanecaused tree debris. This model assumes the user will have performed all necessary data manipulation prior to importing data to the model as inputs which includes synchronizing the projections and coordinate systems and determining what specific roads sh all be considered major. In this case, the major roads were first extracted with a Select by Attributes and then constrained to the Houston City limits using the ArcGIS Clip tool. The lack of local knowledge of major roads in Houston may have resulted in the selection of inappropriate roads such as elevated Interstate Freeways that are outside the citys jurisdiction and are unlikely to have high amounts of debris based on their
90 physical characteristics. The assumption that the user possesses local k nowledge cannot be addressed; local knowledge is required. Ideally, a short companion document should be developed as a users manual for the debris model in order to address frequently asked questions and assumptions the model makes of the users ability to correctly manipulate the data prior to inclusion in the model. Development of such a document, while recommended, is beyond the scope of this project. This model is limited in its lack of flexibility to extract debris values by mask and by the cell size of the Kriging prediction mapthe mask in this model is hardcoded as a 60meter buffer of major roads with a prediction map cell size of 30. Flexibility in choosing an extraction mask (i.e. a layer of city properties or a layer of residential properties) and the prediction map cell size can be incorporated into the VBA interface as selectable parameters. While this model is simple in its design, there are ways that the user experience c ould be improved. Adjustments to the model interface such as the addition of a Browse command button to set the filepath of the output file will make the model more adaptable. In addition, the inclusion of an About command button and/or a related image may better define the intent and scope of the models capability. Conclusion In summary, this tool posed by this framework should be considered a pilot project that is foundational for the development of more sophisticated models and tools in the future. While this tool helps the user to better understand the generation and distribution of tree debris, its many assumptions and limitations hinder the tools results and applicability. The adoption of the aforementioned recommendations to address the assumptions and limitations is planned and will lead to the development of a relatively accurate and predictive debris estimation tool.
91 LIST OF REFERENCES Anderson, J.R., 1977. Landuse and land cover changes: A framework for monitoring. Journal of Research of the US Geological Survey 5, 143153. Balsillie, J.H., 2002. Expedient assessments of coastal storm and hurricane damage potential. In, Environmental Geosciences. Florida Geological Surv ey, Tallahassee, FL, p. 8. Batista, W.B., Platt, W.J., 2003. Tree population responses to hurricane disturbance: syndromes in a southeastern USA oldgrowth forest. Journal of Ecology 91, 197212. Blackard, J.A., Finco, M.V., Helmer, E.H., Holden, G.R., Ho ppus, M.L., Jacobs, D.M., Lister, A.J., Moisen, G.G., Nelson, M.D., Riemann, R., Ruefenacht, B., Salanjanu, D., Weyermann, D.L., Winterberger, K.C., Brandeis, T.J., Czaplewski, R.L., McRoberts, R.E., Patterson, P.L., Tymico, R.P., 2008. Mapping U.S. Forest Biomass using Nationwide Forest Inventory Data and Moderate Resolution Information. Remote Sensing of Environment 112, 19. Burban, Andresen, 1994. Storms over the Urban Forest. In. Burley, S., Robinson, S.L., Lundholm, J.T., 2008. Post hurricane vegetation recovery in an urban forest. Landscape and Urban Planning 85, 111122. Buyantuyev, A., Wu, J., Gries, C., 2007. Estimating vegetation cover in an urban environment based on Landsat ETM+ imagery: A case study in Phoenix, USA. International Journal of Remote Sensing 28, 269291. Cadenasso, M.L., Pickett, S.T.A., Schwarz, K., 2007. Spatial heterogeneity in urban ecosystems: reconceptualizing land cover and a framework for classification. Frontiers in Ecology and the Environment 5, 8088. Campbell, P., 1997. Current Population Reports, Population Projections 1995 2025. U.S. Census Bureau, 6. COES, 2005. Disaster Debris Management Training Manual. In. Response and Recovery Division, Technical Assistance Programs, Governor's Office of Emergency Services, State of California, p. 188. Duryea, M.L., Kampf, E., Littell, R.C., 2007a. Hurricanes and the urban forest 1: effects on southeastern United States coastal plain tree species. Journal of Arboriculture 33, 8397. Duryea, M.L., Kampf, E., Littell, R.C., Rodriguez Pedraza, C.D., 2007b. Hurricanes and the urban forest II: effects on tropical and subtropical tree species. Journal of Arboriculture 33, 98112.
92 Elsner, J.B., Birol Kara, A., 1999. Hurricanes of the North Atlantic: Climate and Society. Oxford University P ress, New York. Escobedo, F.J., Luley, C.J., Bond, J., Staudhammer, C., Bartel, C., 2009. Hurricane debris and damage assessment for Florida urban forests. Journal of Arboriculture and Urban Forestry 35, 100106. Everham, E.M., Brokaw, N.V.L., 1996. Forest damage and recovery from catastrophic wind. Botanical Review 62, 113185. FEMA, 2006. HAZUS MH MR3 Hurricane Model Technical Manual. < http://www.fema.gov/library/viewRecord.do?id=3504> FEMA, 2007. Debris Management Guide. In: Agency, F.E.M. (Ed.), Wasgington D.C. Foster, D.R., Boose, E.R., 1992. Patterns of forest damage resulting from catastrophic wind in central New England, USA. Journal of Ecology 80, 7998. Francis, J.K., Gillespie, A.J.R., 1993. Relating gust speed to tree damage in Hurricane Hugo, 1989. Journal of Arboriculture 19, 368373. Homer, C., Dewitz, J., Fry, J., Coan, M., Hossain, N., Larson, C., Herold, N., McKerrow, A., VanDriel, J.N., Wickham, J., 2007. Completion of th e 2001 National Land Cover Database for the Conterminous United States. Photogrammetric Engineering & Remote Sensing, 337341. Hu, H., Wang, G.G., 2008. Changes in Forest Biomass Carbon Storage in the South Carolina Piedmont between 1936 and 2005. Forest E cology and Management 225, 14001408. Jenkins, J.C., Chojnacky, D.C., Heath, L.S., Birdsey, R.A., 2003. National scale biomass estimators for United States tree species. Forest Science 49, 1235. Kupfer, J.A., Myers, A.T., McLane, S.E., Melton, G.N., 2008. Patterns of forest damage in a southern Mississippi landscape caused by Hurricane Katrina. Ecosystems 11, 4560. Margoles and Umpierre, 2005. Broward County's WebBased Hurricane Debris Estimation Tool. In, Broward County Emergency Management Agency, Brow ard County, FL. McNulty, S.G., 2002. Hurricane impacts on US forest carbon sequestration. Environmental Pollution 116, S17S24. McPherson, E.G., 1994. Chicago's urban forest ecosystem: results of the Chicago Urban Forest Climate Project. General Technical Report Northeastern Forest Experiment Station, USDA Forest Service, vi + 201 pp.
93 Merritt, M., 2009. Bayou Region Urban Forestry Coordinator, Texas Forest Service. In. Myeong, S., Nowak, D.J., Duggin, M.J., 2006. A temporal analysis of urban forest carbon storage using remote sensing. Remote Sensing of Environment 101, 277282. National Hurricane Center, 2008. Tropical Cyclone Report: Hurricane Ike. In. National Oceanic and Atmospheric Administration, p. 55. Nowak, D.J., 1993. Atmospheric Carbonreduction by urban trees. Journal of Environmental Management 37, 207217. Nowak, D.J., 2002. Brooklyn's Urban Forest. In. USDA Forest Service, p. 112. Nowak, D.J., 2006. UFORE Summary Paper. In, Syracuse, NY. Nowak, D.J., Rowntree, R.A., McPherson, E.G., Sisinni, S .M., Kerkmann, E.R., Stevens, J.C., 1996. Measuring and analyzing urban tree cover. Landscape and Urban Planning 36, 49 57. Nowak, D.J.e.a., 2005. Houston's Regional Forest. In. Oswalt, S.N., Oswalt, C.M., 2008. Relationships between common forest metrics and realized impacts of Hurricane Katrina on forest resources in Mississippi. Forest Ecology and Management 225, 16921700. Pielke, R.A., Landsea, C.W., 1998. Normalized hurricane damages in the United States: 192595. Weather and Forecasting 13, 621631. Powell, M.D., Bowman D, Gilhousen, D., Murillo, S., Carrasco, N., St. Fleur, R., 2004. Tropical Cyclone Winds at Landfall: The ASOS C MAN Wind Exposure Documentation Project. BAMS, American Meteorological Society Powell, M.D., Houston, S.H., 1996. Hurrica ne Andrews landfall in South Florida. Part II: surface wind fields and potential real time applications. Weather and Forecasting. Powell, M.D., Houston, S.H., Amat, L.R., Morisseau Leroy, N., 1998. The HRD real time hurricane wind analysis system. Journal of Wind Engineering and Industrial Aerodynamics 7778, 5364. Ridd, M.K., 1995. Exploring a V I S (vegetationimpervious surface soil) model for urban ecosystem analysis through remote sensing: comparative anatomy for cities. International Journal of Remote Sensing 16, 21652185. Sales, M.H., Souza, C.M., Kyriakidis, P.C., Roberts, D.A., Vidal, E., 2007. Improving spatial distribution estimation of forest biomass with geostatistics: A case study for Rondonia, Brazil. Ecological Modelling 205, 221230.
94 Stan turf, J.A., Goodrick, S.L., Outcalt, K.W., 2007. Disturbance and coastal forests: A strategic approach to forest management in hurricane impact zones. Forest Ecology and Management 250, 119135. Staudhammer, C., Escobedo, F.J., Luley, C.J., Bond, J., 2009. Patterns of urban forest debris from 2004 and 2005 Florida hurricane seasons. Southern Journal of Applied Forestry 33, 193196. Tanner, E.V.J., Bellingham, P.J., 2006. Less diverse forest is more resistant to hurricane disturbance: evidence from montane rain forests in Jamaica. Journal of Ecology 94, 10031010. Tanner, E.V.J., Kapos, V., Healey, J.R., 1991. Hurricane effects on forest ecosystems in the Caribbean. Biotropica 23, 513521. Woodbury, P.B., Smith, J.E., Heath, L.S., 2007. Carbon sequestration in the US forest sector from 1990 to 2010. Forest Ecology and Management 241, 1427. Xiao, Q., Ustin, S.L., McPherson, E.G., 2004. Using AVIRIS data and multiplemasking techniques to map urban forest tree species. International Journal of Remote Sensing 25, 56375654. Zeng, H.C., Talkkari, A., Peltola, H., Kellomaki, S., 2007. A GIS based decision support system for risk assessment of wind damage in forest management. Environmental Modelling & Software 22, 12401249. Zimmerman, J.K., Everham, E.M., Waide, R.B., Lodge, D.J., Taylor, C.M., Brokaw, N.V.L., 1994. Responses of tree species to hurricane winds in subtropical wet forest in Puerto Rico implications for tropical tree lifehistories. Journal of Ecology 82, 911922. Zipperer, W.C., Wu, J.G., Pouyat, R.V., Pickett, S.T.A., 2000. The application of ecological principles to urban and urbanizing landscapes. Ecological Applications 10, 685688.
95 BIOGRAPHICAL SKETCH Benjamin K. Thompson was born in Oxford, CT. He earned his Bachelors of Science degree in Urban and Community Forestry from Unity College in Maine in May of 2000. After five years of employment as the Community Assistance Forester for the Washington State Department of Natural Resources Urban and Community Forestry Pr ogram, Benjamin came to Gaines ville, FL to pursue his Master of Science degree in the School of Forest Resources and Conservation at the University of Florida. He received his M.S. from the University of Florida in December 2009.