Spatially estimating flooding depths from damage reports

It is important that a sustainable community better prepare for and design mitigation processes for major flooding events, particularly as the climate is non-stationary. In recent years, there have been major storm events in the USA with record amounts of rainfall that some refer to as stalled storms. These stalled storms frequently result in flooding of urban areas which are not subject to riverine or storm surge flooding. This research focuses on using flood damage reports in conjunction with contour maps, geographical information systems, Light Detection and Ranging (LiDAR) data, photographs, and spatial averaging to develop total (high) flood elevation data sets for two neighborhoods for Tropical Storm Imelda in Beaumont, Texas. When various sources of data such as insurance or Federal Emergency Management Agency damage reports, updated LiDAR elevation sets, and coordination data are readily available, this may be an economical method of estimating maximum flood elevations. High-water marks are frequently collected by various agencies as soon as possible after a flooding event, but these data sets might provide even additional information and validation many months or years post an event.


Introduction
It is important to better prepare for and design mitigation processes for major flooding events for a more resilient and sustainable community, particularly as the climate is non-stationary (Bayazit 2015). Many researchers, modelers and decision-makers rely on 1 3 historical flooding data to accomplish this. An assessment of flood frequency estimation guidelines from different countries showed that the value of historical data is generally recognized, but practical methods for systematic and routine inclusion of this type of data into risk analyses are in many cases unavailable (Kjeldsen et al. 2014). In the USA, three of the federal agencies, the US Geological Survey (USGS), the National Weather Service (NWS) and the US Army Corps of Engineers (USACE) work together to gather information post major flood events such as hurricanes. At locations where there are no real-time monitoring devices, they use high-water marks to measure the maximum elevation of a flood or high-water event (USGS 2021a, b). Much of the collected information is then made available on a USGS website (USGS 2021a, b). However, there are many locations where highwater marks have not been collected particularly, in areas that have not frequently flooded or are not in established flood plains.
In recent years there have been some major storm events in the USA with record amounts of rainfall that some refer to as stalled storms. Two of note are Hurricane Harvey in 2017 and Tropical Storm Imelda in 2019. These events caused massive flooding with damages to properties and lives in areas not typically prone to flooding. According to a National Hurricane Center Report (Blake and Zelinsky 2018), from when reliable records were available from around the 1880s to September 2017, Hurricane Harvey "was the most significant tropical cyclone rainfall event in the United States history, both in scope and peak rainfall amounts." Record rainfalls amounts of over 60 inches were recorded just a few miles south of Beaumont, in Jefferson County, Texas. Just two years later, according to another report by the National Hurricane Center (Latto and Berg 2020) Tropical Storm Imelda brought more devastation to Jefferson County, Texas in September 2019, with possibly the highest peak 12-h intensity ever recorded near Fannett, Texas just west of Beaumont, as 31 inches fell in 12 hours. Figure 1 depicts the track and rainfall from Tropical Storm Imelda as it stalled over Southeast Texas (NCEP 2022).
Much of the flooding from both Hurricane Harvey and Tropical Storm Imelda was pluvial (rain based) and occurred in areas not subject to riverine (fluvial) or storm surge flooding. High-water mark or real-time monitoring device data were not available to aid in providing information to model and prepare for a more sustainable future due to these previously unexpected changes to the regional climate. However, there are other potential sources of information on flood levels. In the USA, the National Flood Insurance Program (NFIP) is managed by the Federal Emergency Management Agency (FEMA) (FEMA 2021a, b). Interested parties can purchase flood insurance and by following FEMA's procedures, which include inspections within affected structures, may claim reimbursement for damages. Damage reports are also filed in communities for many structures that do not have insurance but are eligible for other disaster compensation. These damage reports are frequently available from zoning, building, and other local code enforcement groups with a stake in protecting their communities (FEMA 2021a, b).
Many researchers studied how to estimate flood damage or future impacts based on flood depths, but few have used flood depths developed from damage reports. A study across 29 countries in Europe emphasized the importance of mapping river floods and their associated hazards and risk. The key difference between a flood hazard map and a flood risk map is that hazard maps contain information about the probability and/or magnitude of an event whereas flood risk maps contain additional information about the consequences (e.g., economic damage, number of people affected) of such an event (de Moel et al. 2009). Using methods such as Analytical Hierarchy Procedures (AHP), and Geographical Information Systems (GIS), Ouma and Tateishi (2014) designed flood vulnerability maps. GIS-based spatial analysis was used to examine drainage networks susceptible to flooding events in Athens, Greece (Diakakis 2014). Some researchers estimated flood damage in the urban environment based on other sources and remote sensing data (Gerl et al. 2014). In these models and maps, the accuracy and applicability of the data are also very important (Molinari and Scorzini 2017). Synthetic Aperture Radar (SAR) data have also been used to map flood and damages from typhoon Hagibis in Japan (Tay et al. 2020).
Researchers in various countries generated or used high-water marks for past flooding events using widely available resources. Kruger and Farmer (2012) used highwater marks in conjunction with digital elevation models (DEM)s to create a storm surge map for southeast Texas after Hurricane Ike. Vallimeena et al. (2018) from India used crowd-sourced images of people during the flooding event to estimate the flood depth using images uploaded into the FloodEvac, an emergency management system, by flood victims. Machine learning techniques were used to analyze the crowd-sourced images to estimate the flood depth, damages, and accessibility of the affected area. Schmidt et al. (2019) used Cycle-Consistent Adversarial Networks, a machine learning technique, to generate visual representation of the potential effects of climatic events. Meng et al. (2019) from the University of Wisconsin used web images to estimate flood depth: they used R-CNN (region-based convolutional neural network), a machine learning software, and the Face++ Emotional Application Programming Interface to analyze human body objects from floods and used the data to estimate the flood depth. Park et al. (2021) from Korea used crowd-sourced flooded images of cars to estimate 1 3 flood depth. The study used another version of R-CNN to detect the flooded images of cars and using 3D models of the vehicles to estimate the flood depth. Similarly, the application of CNN and images from camera footage in an open channel flow system was used to determine the instantaneous flow by measuring the surface velocity and water level (Meier et al. 2022).
The effect of flooding in urban areas can be devastating. The ability to use predictive tools to identify the path and trajectory is crucial in minimizing damages and fatalities. A typical architecture for urban flood forecast systems is presented around four main components: inputs, data collection, decision support and modeling tools, and warning process in an article by Henonin et al. (2013). A key component of flood forecasting is data which at times are not readily available. An investigation by René et al. (2014) suggested that depending on the level of accuracy required, and the purpose of the model, urban flood managers may have the means to make pluvial flood forecasts. Other methods of mitigating urban flash flood events include: (1) the use of radar sensor networks (X band and S band) to produce high-resolution quantitative precipitation estimations (QPE) (Chen and Chandrasekar 2015); (2) the use of cameras as a visual sensor to acquire spatiotemporal information for remote analysis on urban floods (Lo et al. 2015); (3) a combination of sensors, surveillance cameras, and modeling in a controlled environment to improve flood risk mapping (Moy de Vitry et al. 2017); and (4) using large scale particle image velocimetry based methods to measure surface flow velocity of overland runoff captured by surveillance and traffic cameras (Leitão et al. 2018).
A recent study of flooding in Southeast Texas used damage reports from the NFIP based on claims dating back to 1976 to estimate flood risks based on damage and referred to the areas mapped as damage plains. This study did not estimate associated flood depths (Mobley et al. 2021). Another recent study in Texas also developed a method to estimate first-floor elevations in neighborhoods on Galveston Island to compare to estimated flood elevations to aid in assessing flood risk. The method used drones to photograph building exteriors to estimate the first-floor elevations based on exterior first-floor door locations (Diaz 2020).
There are various methods used to estimate the extent of flooding, but spatially using compiled damage reports to estimate Total Flood Elevations (TFE) in a damage plain during the events is not one of them. This research focuses on using damage reports in conjunction with contour maps, GIS, Light Detection and Ranging (LiDAR) data and exterior photographs or other visual inspections of affected structures to develop additional total (high) flood elevation (TFE) data sets based on estimated elevations of a damage plain. Spatial analyses may then provide decision-makers and modelers with supplemental information on the depth of flooding, particularly from these stalled storms that are increasing in frequency in many areas, or other areas with increased urban flooding not related to rivers or storm surge. This method could also be used when there is a combination of flooding from storms. In the case of fluvial or storm surge flooding, using a damage plain from affected structures may perhaps supplement the data arrived from other high-water mark techniques that are frequently used. Of course, it is also useful in the urban areas most affected from the pluvial portion of the event.
This study focused on two neighborhoods in Beaumont, Texas which had devastating urban flooding during Tropical Storm Imelda. This effort also used spatial averaging to provide a more statistically viable data set (Brakewood and Grasso 2000).

Materials and methods
The method proposed in this study was carried out in three main steps. The first step was collecting information on the reported residential structures with major damage due to Tropical Storm Imelda and the photographs of the front face of these structures (Sect. 2.1). The second step involved using visual analysis of the photographs and LiDAR elevation information to determine first-floor elevations (FFEs) for each household (Sect. 2.2). The third step was using the first-floor elevation for each household together with the damage report information on inundation above the first-floor elevation to estimate the Total Flood Elevation (TFE) of the flood (damage) plain of the neighborhood and to apply spatially averaging (Sect. 2.3).

Data collection
Information regarding households in two neighborhoods that were flooded during Tropical Storm Imelda was collected from the City of Beaumont. The damage reports generated after the flooding events contain information regarding the location and extent of damage of the structures. The damage depths of these reports convey the interior flood damage. Each house was then given a unique referral identification based on street number and name, zip code, depth of damage, seriousness of damage (major, minor, affected) and longitude and latitude. In the report, all water depths were given in an exact value in inches. Therefore, the error is estimated to be 1 inch (2.54 cm).
For this project only the households listed as having major damage were evaluated. The Flood Damage Depth (FDD) within each household above the First Floor Elevation (FFE) was acquired from the damage reports. Photographs were collected for each of these households in December 2020 and were mainly focused on the front door, front wall and the front lawn area. Today's smart phones are equipped with high-definition cameras and were used for these images. Next, the approximate ground level coordinates at the edge of the overhang in front of the front door of each household were estimated (Front Overhang Coordinate-FOC). A combination of satellite and map views on Google Maps (Google Map 2020); 2D and 3D views on Google Earth (Google Earth 2020) were used to determine these approximate coordinates for each household.
Finally, LiDAR data were used to estimate the elevation of the ground at the front overhang (FOC) of the household. LiDAR elevation data for Jefferson County were collected from the Texas Natural Resources Information System (TNRIS) Data Hub. The datums of the LiDAR data are NAD83 for horizontal and NAVD88 for vertical respectively. For this LiDAR dataset, the vertical and horizontal accuracies are 19.584 cm and 2.45 m, respectively (both at a 95% confidence level), and the estimated spacing is 50 cm (TNRIS 2017). For each affected neighborhood, the FOC data and the LiDAR data were projected on a Jefferson County based street map, as noted in an example in Fig. 2.
The following steps describe the process used to obtain the approximate ground elevation of the front overhang for each household. 2. Geographic coordinates of the front overhang (FOC) of homes with damage reports are obtained and plotted on the DEM using a program such as ArcMap. The approximate front overhang coordinates (FOCs) are obtained from Google Maps. 3. With the coordinates overlaid on the DEM, geospatial analysis is done using the sample spatial analyst tool. The tool is used to extract the nearest corresponding ground elevations at the front overhang. The result of this analysis is a table containing the nearest ground elevation data and corresponding location. 4. The elevation data are plotted on the map using the location coordinates from the table.
These ground elevations are assigned to the nearest front overhang of each house plotted on the map. This ground elevation is further used to assess the extent of flooding experienced in a particular home as reported in the damage reports.

Data analysis
The data collected in the first step were organized and analyzed to determine the approximate Total Flood Elevation (TFE) in the damage plain for each of the collected households. Each house had various elements from the ground at the overhang to the front door such as steps, a sloped porch, and a threshold. The photographs were visually analyzed to determine the height of these various elements for each house, and the error of such height estimates is thought to be 2 inches (5.08 cm) based on the experience of one of the authors in land development site analyses from photographs. These elements were combined to estimate the Elevation Difference (ED) between the ground level at the Front Overhang Coordinates (FOC) and the First Floor Elevation (FFE). Note that since the error of house element heights is estimated to be 2 inches, the error of ED is 2 inches (5.08 cm). The ED and the elevation at the FOC were added to calculate the FFE. After calculating the FFE, the Flood Damage Depth (FDD) obtained from the damage report provided by the City of Beaumont was added to determine the Total Flood Elevation (TFE) for each house. Each house was also given a House Identification (HID) for the purposes of calculations and to maintain anonymity. Examples are provided in Table 1. As mentioned in Sect. 2.1, there are three errors: (1) the error of the FDD from the damage report is 1 inch (2.54 cm); (2) the FOC has the error of 19.584 cm (at 95% confidence level); and (3) the ED has an estimated error of 2 inches (5.08 cm). Therefore, the error of FFE is ± 24.9 cm (at 95% confidence level), and the error of TFE is ± 27.4 cm (at 95% confidence level). Note that the error estimate of LiDAR data has high accuracy, but the error estimates for ED (from photographs taken onsite) and the FDD (from the damage report) are relatively rougher. Therefore, the total error of TFE is estimated at ± 0.3 m or ± 1 ft.

Neighborhood analyses
The analyzed datasets for estimating the Total Flood Elevations (TFEs) were segregated into two neighborhoods based on their locations in Beaumont, Texas. Neighborhood BD is in and around Blossom Drive. Neighborhood SH contains houses to the south of Highway 69 around Chambless Drive and Theresa Avenue. For each neighborhood, the average and standard deviation of the TFEs of all affected houses were calculated. If any TFE was plus or minus three standard deviations, it was considered an outlier and removed from further calculation. Only one such household was found to be an outlier in the Blossom Drive neighborhood and removed from the calculations.
To increase the validity of the dataset for each neighborhood, spatial averaging of the TFEs was performed (Brakewood and Grasso 2000). This helps modulate errors and allows the results to focus on the neighborhood instead of individual homes, allowing for more anonymity. The houses evaluated in each neighborhood were triangulated and the average of the three triangulated TFEs was calculated for each triangle to determine the Triangulated Total Flood Elevations (TTFEs) in the damage plain and placed in the center of each triangle. The triangulation was performed by drawing lines from a household (node) to its nearest neighbors without any of the lines intersecting with other lines except at the nodes. When more than one option was available within a set of nodes, the options with the shortest lines were chosen.

Results
A summary of the two neighborhoods and the averaged values of the various damaged plain elevations are listed in Table 2. Note that the average of the TFE of the individual homes, and the average of the Triangulated TFE within each neighborhood are nearly identical, but the standard deviations are reduced by more than 50% in each case. For the selected houses in Neighborhoods BD and SH, the standard deviations of TFE are 1.32 ft; and 0.95 ft, respectively. These standard deviations are quite close to the estimated error of TFE, as analyzed in Sect. 2.2.
For each neighborhood, the spatially averaged TTFE values were mapped along with LiDAR elevation contours. The TTFEs for Neighborhood BD superimposed with LiDAR contours are shown in Fig. 3 for the Blossom Drive neighborhood and Fig. 4 for Neighborhood SH. In these figures, the contour elevations are shown as the highlighted whole numbers, and the TTFEs are shown as the unhighlighted elevations clustered in the middle.

Discussion and conclusions
It is recognized that the climate is non-stationary and many recent record storms may provide information that is useful for preparing for and designing for future storms. This work represents a process by which the information collected from flood damage reports can be used to estimate high flood elevations from a stalled storm event in an urban area. Damage depths in this damage plain method consider the damage extent on the interior walls of the building, where the layers inside the walls act as barriers against jumps and variations that occur on exterior water surface due to winds and other disturbances. In this case study only the structures with major damage were included. Using information on other less damaged structures might also supplement the data set. However, the most damaged structures are more likely to be officially inspected and then recorded by a city or FEMA, as structures with less damage may not be as likely to meet deductibles or have claims filed. Potential uses of these data are for validating hydrological modeling and providing additional information to entities involved in flood mitigation decisions. There are several advantages to this proposed process. Most of the resources required for the process may be readily available with additional data needed on the buildings added months or years post the event. The process is customizable based on the future resources. More technical elements can be added to the process such as using surveying equipment for more accurate estimation of ground and first-floor elevations. Artificial Intelligence (AI) techniques might be also used to estimate the depths of flood from the collected photographs, to further validate the results.
There are also some limitations to the proposed process. First, this method is prone to errors as it relies on visually estimating depths from photographs and reliance on the accuracy of the interior damage levels provided by the inspectors, and that error estimate is approximate. Second, the vertical error of LiDAR data is still relatively large (about 20 cm, i.e., 2/3 ft). Especially, the LiDAR data used in this study are outdated-they were collected in 2017, and thus some elevation changes may exist. Spatial averaging techniques aid in modulating these potential errors. compiling damage information data and Dr. Reda Amer for his help in training the authors in some GIS techniques.
Author contributions LH and XW contributed to the study conception and design and contributed to the error analyses. MA and NM performed material preparation, data collection and analysis and prepared the first draft of the manuscript. All authors commented on previous versions of the manuscript and read and approved the final manuscript.
Funding This research was funded by The Lower Neches Valley Authority and the Sabine River Authority of Texas. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
Availability of data and materials Data were collected from public sources, and therefore, no informed consent was needed. However, since household addresses were linked to the damaged structures, each household was provided with a unique identifier for internal data analysis. In addition, the locations of the modeled damage plain elevations as presented in the paper are not mapped at the aforementioned households, but instead represent an average of three households to provide for additional anonymity.

Declarations
Competing interests The authors declare that they have no competing interests.

Ethics approval and consent to participate
The authors have no relevant financial or non-financial interests to disclose.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.