Quantifying Urban Flood Extent Using Satellite Imagery and Random Forest: A Case Study in Southeastern Pennsylvania

Quantifying Urban Flood Extent Using Satellite Imagery and Random Forest: A Case Study in Southeastern Pennsylvania | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Quantifying Urban Flood Extent Using Satellite Imagery and Random Forest: A Case Study in Southeastern Pennsylvania Rebecca Composto, Mirela G Tulbure, Varun Tiwari, Mollie D. Gaines, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3882712/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 27 Jul, 2024 Read the published version in Natural Hazards → Version 1 posted 5 You are reading this latest preprint version Abstract The risk of floods from tropical storms is increasing due to climate change and human development. Maps of past flood extents can aid in planning and mitigation efforts to decrease flood risk. In 2021, Hurricane Ida slowed over the Mid-Atlantic and Northeast United States and released unprecedented rainfall. Satellite imagery and the Random Forest algorithm are a reliable combination to map flood extents. However, this combination is not usually applied to urban areas. We used Sentinel-2 imagery (10 m), along with derived indices, elevation, and land cover data, as inputs to a Random Forest model to make a new flood extent for southeastern Pennsylvania. The model was trained and validated with a dataset created with input from PlanetScope imagery (3 m) and social media posts related to the flood event. The overall accuracy of the model is 99%, and the flood class had a user’s and producer’s accuracy each over 99%. We then compared the flood extent to the Federal Emergency Management Agency (FEMA) flood zones at the county and tract level and found that more flooding occurred in the Minimal Hazard zone than in the 500-year flood zone. Our Random Forest model relies on publicly available data and software to efficiently and accurately make a flood extent map that can be deployed to other urban areas. Flood extent maps like the one developed here can help decision-makers focus efforts on recovery and resilience. Machine Learning Emergency Management Flooding Surface Water Hurricane Ida Climate Change Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction 1.1 Floods Cause Harm Flooding has become a devastating and destructive hazard due to human development in high-risk areas (CRED, 2015 ) and climate change. The World Bank estimated in 2022 that 1.81 billion people, or 23% of the world’s population, are at risk of intense floods (Rentschler et al., 2022 ). Floods cause many types of harm (de Bruijn et al., 2019 ; Rosser et al., 2017 ), including economic losses (Pinos & Quesada-Román, 2022 ), damages to private homes/assets, damages to public infrastructure (Goffi et al., 2020 ), involuntary displacement, impacts on mental health (Markhvida et al., 2020 ), and disruptions to daily life and traffic flow (Hosseiny et al., 2020 ). In the most dangerous circumstances, floods can lead to loss of life (Goffi et al., 2020 ), taking 146 lives in 2021 in the U.S. (US Department of Commerce, 2021). In the U.S., floods cause major economic losses, with an average yearly cost of $ 4.5B (Smith, 2023 ). Among flooding, the average cost of each tropical storm, including hurricanes, is $ 22.2B (which includes damages caused by floods and winds) (Smith, 2023 ). The economic and social costs of floods are high, and the risk of floods from hurricanes is expected to increase due in part to trends in climate change (Trenberth et al., 2018 ) and human development. 1.2 Hurricanes risk increasing Climate change is heating the atmosphere, allowing it to hold more moisture and increasing precipitation frequency and intensity (Van Oldenborgh et al., 2017 ), leading to higher flood risk (Ireland et al., 2015 ; Van Oldenborgh et al., 2017 ). In addition to more extreme precipitation, the U.S. is facing more intense hurricanes (Kossin et al., 2007 ). At the global scale, climate change is causing the speed of tropical cyclones to decrease and precipitation rates to increase (Kossin, 2018 ). In the U.S., this same trend applies with tropical storms stalling more often along the coast and increasing precipitation rates (Hall & Kossin, 2019 ). The intensity (Holland & Bruyère, 2014 ; Knutson et al., 2015 ), storm surge (Lin et al., 2012 ), and precipitation of these hurricane events are also expected to worsen due to climate change. 1.3 Unequal distribution of risk People in poverty are disproportionately at risk of floods around the globe (Garbutt et al., 2015 ; Kawasaki et al., 2020 ; Mtapuri et al., 2018 ; Winsemius et al., 2018 ). This trend signifies that flood risk is not equally distributed (Wing et al., 2022 ). This imbalance occurs globally because people in poverty are more likely to live in a floodplain due to the concentration of jobs and transportation (Mtapuri et al., 2018 ). In the U.S., the same trend of inequitable flood risk applies, where flood risk disproportionately impacts poorer communities (Wing et al., 2022 ). At the city level, one case study of Los Angeles found that poorer communities have disproportionately higher flood risk, but this trend varied by flood type (Sanders et al., 2022 ). These findings at the national and local scale show that it is important to investigate the distribution of flood impacts to help inform emergency response and recovery. 1.4 Hurricane Ida 2021 Hurricane Ida was a category 4 hurricane that landed in the U.S. on August 29, 2021, bringing catastrophic damage (Beven II et al., 2022 ). In the first days of September, Hurricane Ida stalled, becoming extratropical and bringing heavy rains with rates around 3 inches per hour to states in the Mid-Atlantic and Northeast (Beven II et al., 2022 ). The storm caused dozens of fatalities and damaged homes, businesses, vehicles and infrastructure (Smith, 2023 ). The National Oceanic and Atmospheric Association (NOAA) National Centers for Environmental Information (NCEI) estimates that the cost of Hurricane Ida is $ 80.2 Billion (Consumer Price Index-Adjusted) (Smith, 2023 ), the costliest hazard of 2021. In Pennsylvania, Hurricane Ida brought precipitation and floods that caused damage (Beven II et al., 2022 ). In the aftermath of Hurricane Ida, individuals and households in Pennsylvania received $ 124M in funding from the Federal Emergency Management Agency (FEMA) to cover damages (Cooper et al., 2022 ). Even with this influx in funding, many counties in the state are still recovering from the damages of Hurricane Ida, including Philadelphia, Montgomery, Delaware and Bucks counties (Cooper et al., 2022 ). Given the trends in intensifying hurricanes, it is imperative to plan for future events using insights from past flood events (Brandt et al., 2021 ). 1.5 Flood extent from satellite imagery Satellite imagery is a reliable data source (Hermas et al., 2021 ) that is regularly used to map surface water and flood dynamics across various scales (Ayanu et al., 2012 ; Jones, 2019 ; Pekel et al., 2016 ; Tulbure et al., 2016 ). Machine learning is an effective method for classifying floods in satellite imagery (Tulbure et al., 2022 ). It has higher classification accuracy than parametric strategies (Maxwell et al., 2018 ), such as with a single water index (Goffi et al., 2020 ). Supervised machine learning classification of satellite imagery relies on accurate training and validation data (Olofsson et al., 2014 ). A known flood extent from a reliable dataset (Hondula et al., 2021 ) or created from aerial photography (Rosser et al., 2017 ; Schnebele et al., 2014 ) is ideal training data. However, it is not available for every flood event. In the case of Hurricane Ida, aerial imagery was not collected in the study area of southeastern Pennsylvania (US Department of Commerce, 2022 ). Since an accurate flood extent was unavailable as training and validation data, we created one from satellite imagery and social media (Ireland et al., 2015 ; Perin et al., 2022 ). Urban environments have complex waterways, shallow and ephemeral flooding, and ponding, meaning the flood extent is discontinuous (Tanim et al., 2022 ; Woznicki et al., 2019 ). Flood models can generate flood maps in urban areas (Knighton et al., 2021 ; Liu et al., 2015 ), however, flood maps generated from satellite imagery are considered more aligned with ground conditions. Synthetic Aperture Radar (SAR) data can collect data through clouds, but is limited in urban areas due to its side-looking nature (Mason et al., 2014 ) and can have gaps due to tall buildings causing radar shadowing or layover (Clement et al., 2018 ). Optical imagery cannot permeate through clouds, but does not face the same challenges with tall buildings. For our study area after Hurricane Ida, there was optical imagery, Sentinel-2, collected, but no freely available SAR data collected. In this study we used optical data and combined it with several datasets in our machine learning model to address the complexity of urban environments. 1.6 Gaps and objectives There is currently no Hurricane Ida flood extent in Pennsylvania, including the city of Philadelphia and surrounding counties. The floods resulting from Hurricane Ida dissipated slowly and coincidentally overlapped with Sentinel-2 imagery collection, providing a unique opportunity for testing flood detection methods in an urban environment. Previous research has predominantly used satellite imagery and machine learning to detect floods using a vetted flood extent (from satellite imagery or a flood model) and rarely applies these methods to an urban area. Therefore we focused on the urban area of Philadelphia and surrounding counties after Hurricane Ida, an event without a vetted flood extent available, to fill these gaps. The objectives of this research were to: (1) combine Sentinel-2 imagery and other data in a Random Forest (RF) machine learning algorithm to create a novel flood extent in southeastern Pennsylvania after Hurricane Ida; (2) compare the flood extent to FEMA’s (Federal Emergency Management Agency) flood zones; (3) use the flood extent to calculate flood exposure and determine the equality of its distribution. Since hurricanes and floods are expected to increase, methods for accurate and timely flood extent maps in urban areas are fundamental for improving recovery and mitigation efforts. 2. Methodology 2.1 Study site Philadelphia, located in southeastern Pennsylvania, is the sixth largest city by population in the U.S., with over 1.5 million people (Census, 2021 ). Over the past hundreds of years, the city has grown and developed, building over the existing streams. Floods occur in Philadelphia 12 days per year, and this frequency is expected to increase (Sweet et al., 2019 ). While the city is regularly flooded, it has made considerable efforts to reduce floods through multiple avenues, including its stormwater management program and strict design requirements (Hosseiny et al., 2020 ). On September 1, 2021, Philadelphia was directly in the path of Hurricane Ida (Fig. 1 ) and received 2.37 inches of precipitation, significantly more than its annual monthly mean precipitation of 0.12 inches (NOAA, 2021). The impacts of Hurricane Ida in Philadelphia and the resulting floods were wide-ranging from disrupting daily life to damaging personal assets (Pulcinella et al., 2021 ). In addition to Philadelphia, surrounding less urban counties, including Delaware, Bucks, and Montgomery counties, were impacted by Hurricane Ida flooding and incorporated into our study area (Cooper et al., 2022 ). According to the U.S. 2020 Census, Philadelphia (both the name of the city and county) is the most urban at 100%, then Delaware and Montgomery counties at 88% and 76%, respectively, and lastly, Bucks County at 44% (U.S. Census Bureau, 2020 ). 2.2 Datasets 2.2.1 Satellite Imagery We selected Sentinel-2 imagery because it aligned with peak Hurricane Ida flooding in our study area and was the highest spatial resolution of publicly available imagery. Sentinel-2 is a mission run by the European Space Agency and produces publicly available satellite imagery of the globe at a 10–60 m spatial resolution and a temporal resolution of ~ 5 days. We used Sentinel-2 imagery that was collected a day after Hurricane Ida passed through the study area (September 2, 2021) with little (< 1%) cloud cover, making it an optimal data source. In Google Earth Engine (GEE), we obtained the imagery, filtered it temporally and spatially, and removed cloud and cirrus pixels using the quality assessment band (QA60) (Tiwari et al., 2024 ). Other imagery that was considered included Synthetic Aperture Radar (SAR) data, PlanetScope imagery and aerial imagery. The Sentinel-1 imagery dates did not align with the peak flooding in the study area. We considered PlanetScope imagery as the basis for the flood extent but instead used it to create the training and validation data for the RF model since it is best practice to use a higher resolution image for training and validation data than the model input (Olofsson et al., 2014 ). We did not consider aerial imagery because it was not collected by the National Geodetic Survey after Hurricane Ida in the study area (US Department of Commerce, 2022 ). 2.2.2 Sentinel-2 Indices The RF inputs included all Sentinel-2 surface reflectance bands, two vegetation indices and six water indices, all previously shown to be important when mapping floods with satellite data (Goffi et al., 2020 ; Tulbure et al., 2016 , 2022 ) (Table 1 ). We produced all the indices in GEE. The vegetation indices are used to help classify NotWater pixels by identifying areas of vegetation. We used several water indices, each with different strengths, to help categorize Water (permanent) and Flood pixels in the model. The Normalized Difference Water Index (NDWI) is the standard for classifying water using green and near-infrared (NIR) bands (McFeeters, 1996 ). The Modified Normalized Difference Water Index (MNDWI) is a variation of NDWI that uses shortwave infrared (SWIR) instead of NIR and is more suitable in built-up areas than NDWI (Xu, 2006 ). The Automated Water Extraction Index (AWEI) uses five spectral bands to improve water classification by decreasing the environmental noise of shadows and dark surfaces (Feyisa et al., 2014 ). Two variations of the AWEI formulas (AWEI nsh and AWEI sh ) have different effectiveness in urban areas. The AWEI nsh formula is more equipped for urban areas because it effectively eliminates built surfaces. The AWEI sh formula is more equipped for filtering out shadows, but is less equipped for urban areas because it tends to misclassify reflective roofs as water. We also used Linear Spectral Unmixing (LSU) to produce three inputs, each with the percent of three different “endmembers” (water, urban and vegetation) or classes for each pixel. Pixels have mixed spectral signatures because the underlying land cover is mixed and highly variable (C. Yang et al., 2007 ). LSU addresses this heterogeneity by using all bands to estimate each pixel's “endmember” percent (C. Yang et al., 2007 ). LSU is helpful in the context of floods because it can be used to determine the fraction of water in each pixel and produce flood maps (Bangira et al., 2017 ; Gómez-Palacios et al., 2017 ). Table 1 Sentinel-2 derived indices used in our RF classification. Category Index Name Formula Reference Vegetation Normalized Difference Vegetation Index (NDVI) NDVI = (NIR - RED / NIR + RED) Kriegler et al., 1969 Enhanced Vegetation Index (EVI) EVI = 2.5 * ( NIR - RED) / (NIR + 6 * RED − 7.5 * BLUE + 1) Huete, 1997 Water Normalized Difference Water Index (NDWI) NDWI = (GREEN - NIR) / (GREEN + NIR) McFeeters, 1996 Modified Normalized Difference Water Index (MNDWI) MNDWI = (GREEN - MIR) / (GREEN + MIR) Xu, 2006 Automated Water Extraction Index (AWEI nsh ) AWEI nsh = 4 * (GREEN - SWIR1) - (0.25 * NIR + 2.75 * SWIR2) Feyisa et al., 2014 Automated Water Extraction Index (AWEI sh ) AWEI nsh = BLUE + 2.5 * GREEN − 1.5 * (NIR + SWIR1) = 0.25 * SWIR2 Feyisa et al., 2014 Water Ratio Index (WRI) WRI = (GREEN + RED) / (NIR + SWIR1) Shen & Li, 2010 Normalized Difference Moisture Index (NDMI) NDMI = (NIR - SWIR1) / (NIR + SWIR1) Gao, 1996 Flood Normalized Difference Flow Index (NDFI) NDFI = (RED - SWIR2) / (RED + SWIR2) Boschetti et al., 2014 Multiple Fractional Coefficients (water, urban and vegetation) R mix = (F water * R water ) + (F urban * R urban ) + (F vegetation * R vegetation ) Settle & Drake, 1993 2.2.3 Additional Inputs In addition to Sentinel-2 surface reflectance bands and derived indices, several other datasets readily available in GEE were incorporated into the RF model (Table 2 ). A digital elevation model (DEM) can be used to derive data (e.g., slope) that influences where floods occur (Tulbure et al., 2016 ). In our model, we used the United States Geological Survey (USGS) 3DEP 10 m National Map (U.S. Geological Survey, 2023 ) in GEE to calculate slope, aspect, and hillshade. We also used the USGS National Land Cover Database at 30 m resolution in the model resampled to 10 m, because land cover and impervious surface contribute to flood extent (Apel et al., 2016 ; Blum et al., 2020 ). In GEE we also incorporated datasets of 30 m resolution from the European Commission’s Joint Research Centre (JRC), including surface water occurrence and surface water classification (water, seasonal, permanent) (Pekel et al., 2016 ). We downscaled the JRC datasets and the USGS National Land Cover Database in GEE using the resample function and bilinear method to match the other RF inputs at 10 m resolution. In GEE, we combined the non-Sentinel-2 inputs and reprojected them to match the Sentinel-2 inputs. Next, all the inputs were combined, and a 3x3 window was created for each band. Table 2 The RF model used 62 features derived from satellite imagery, digital elevation, land cover and water datasets that were preprocessed in Google Earth Engine (GEE). Category Source Year Original Resolution Name Reflectance, Vegetation, Water, Moisture, Linear Spectral Unmixing (LSU) Sentinel-2 2021 10–60 m 12 Reflectance Bands 9 Derived Indices 3 Fractional coefficients (water, urban, vegetation) DEM USGS 3DEP 10m National Map Seamless (1/3 Arc-Second) 2017 10 m Slope Aspect Hillshade Land Cover USGS National Land Cover Database (NLCD) 2019 30 m Landcover Impervious Water European Commission Joint Research Centre (JRC) 2021 30 m Global Surface Water - Occurrence Yearly Water Classification History Window All above rows Varies 10–60 m 3x3 Window 2.2.4 Training and Validation Data In QGIS, we created the training and validation dataset by hand using several datasets (Fig. 2 ) (Perin et al., 2022 ). We used PlanetScope (3 m) imagery, which is higher resolution than our model inputs (Maxwell et al., 2018 ; Olofsson et al., 2014 ), and the National Hydrography Dataset (NHD), as reference to create Flood and Water polygons at the resolution of Sentinel-2 (10 m) imagery (Ireland et al., 2015 ). We also used social media posts from the Global Food Monitor project, a publicly available database of flood-related tweets (de Bruijn et al., 2019 ). We used ~ 3,000 tweets, including text and photos, and ~ 140 unique points to guide the creation of Flood polygons (Akhtar et al., 2021 ; Schnebele et al., 2014 ). The Google basemaps in QGIS provided high-resolution imagery for drawing NotWater polygons (Perin et al., 2022 ). Random points in each county were used to guide the location of NotWater polygons. Before drawing the NotWater polygons, we ensured that the polygon was outside the NHD and that the basemap imagery was clear of pools and other surface water. For all three classes (NotWater, Water, Flood), every layer of data was checked to verify the accuracy of the polygon class. The training and validation dataset was 433 polygons consisting of 164 NotWater, 131 Water and 138 Flood polygons. Next, we uploaded the training and validation dataset into GEE and converted it into 10 m pixels. We randomly oversampled the rarer classes of Water and Flood and used all NotWater polygons (Maxwell et al., 2018 ; Olofsson et al., 2014 ). Then, we split the dataset into two stratified random samples, with 70% for training and 30% for validation (Table 3 ). Table 3 Training and validation dataset (pixels 10 m) consists of three classes, oversampling the rarer Water and Flood classes, split into two stratified random samples with ~ 48,000 pixels to train and ~ 20,000 pixels to validate the RF model. Type Percent Total Pixels Pixels NotWater Water Flood Training 70% 48,352 29,473 14,721 4,158 Validation 30% 20,648 12,527 6,279 1,842 Total 100% 69,000 42,000 21,000 6,000 2.2.5 FEMA Flood Zones The National Flood Hazard Layer (NFHL) is a database of flood zones and flood insurance requirements maintained by FEMA in support of the National Flood Insurance Program (NFIP) (FEMA, 2023). The NFHL is available for the entire study area, therefore, every flooded pixel will occur in a FEMA flood zone. The flood zones we focus on in this study are the 100-yr (100-year), 500-yr (500-year), and Minimal Hazard zones because they are the primary risk classifications. The 100-yr and 500-yr flood zones have a 1% and 0.2% likelihood of flooding yearly (FEMA, 2020). We combined all other FEMA flood zones (floodway, 1% annual chance flood hazard contained in channel, area with reduced flood risk due to levee, and 1% depth less than 1 foot) into an “Other” category that encapsulates areas that are less common. The Minimal Hazard zone is outside the 500-yr flood zone and at higher elevations. Once we created the Hurricane Ida flood extent, we used the FEMA flood zones to determine the area and percent area of the Hurricane Ida flood that occurred in the different zones at the county and tract level. 2.2.6 Vulnerability and Demographic Datasets The Centers for Disease Control and Prevention (CDC) and Agency for Toxic Substances and Disease Registry (ATSDR) Social Vulnerability Index (SVI, hereafter) is a vulnerability index at the tract and county level in the United States. Vulnerability is a community’s ability to prevent suffering and financial loss due to a disaster (Fielding, 2018 ). The CDC’s SVI dataset estimates an overall vulnerability score using four themes (socioeconomic status, household characteristics, racial & ethnic minority status, and housing type & transportation) (CDC, 2020 ). From the CDC’s SVI 2020 dataset, we used the sum of the four themes as the SVI index and the number of people below the 150% federal poverty level. The poverty data in the CDC’s SVI dataset came from the 2016–2020 American Community Survey (ACS). These two attributes were used to determine if Hurricane Ida floods disproportionately impacted people in poverty and vulnerable populations. 2.3 Random Forest Model The RF machine learning algorithm is an ensemble classifier that uses a large number of decision trees that each use different random samples and a subset of features to assign a class, then the majority vote of all the trees classifies the data (Breiman, 2001 ; Maxwell et al., 2018 ). The RF algorithm effectively classifies surface water (Phan et al., 2020 ; Tulbure & Broich, 2013 ) and floods (Tulbure et al., 2016 , 2022 ). While the algorithm is a ‘black box’, meaning you cannot visualize all trees, it still has higher accuracy than parametric techniques, and it has the added benefit of being robust to smaller and lower quality training datasets (Maxwell et al., 2018 ). We executed the RF algorithm in GEE with the steps outlined in Fig. 3 . We used the ee.Classifier.smileRandomForest function in GEE to train the model on the training data, then classify the entire study area and determine feature importance. The number of pixels used to train and validate are outlined in Table 3 . We used the validation data to create a confusion matrix. After the first run of the algorithm, we ran the model with the Sentinel-2 features and indices and different combinations of additional features and parameters in order to select the features and parameters (number of trees, number of features at each split) that produce the highest overall accuracy. The final parameters chosen for the algorithm were the default numbers, 100 trees and eight features per split. These parameters align with those chosen in other research using RF classification in GEE (Phan et al., 2020 ). After these parameters were selected, the algorithm was run again with the optimal parameters and all the features. 2.4 Assessment of flood impact After we created a flood extent with our RF model, we exported it from GEE and converted it from a raster to a vector file to compare to the vector file of FEMA flood zones. Then, the flood polygons were clipped to different FEMA flood zones. In this study, we calculated the area and percent flood in four different FEMA flood zones: 100-yr, 500-yr, Minimal Hazards, and Other. The Other category consists of areas, including floodways, that are likely to be flooded; therefore, we expected most flooding to occur in the 100-yr and Other zone. After the flood extent was clipped to the different zones, the area (km 2 ) and the percent of the flood that occurred in each FEMA flood zone were calculated at the county and census tract level. We assessed the flood exposure equality by plotting Lorenz Curves and calculating the Gini index. The Gini index is typically used to study income inequality (Lorenz, 1905 ), but can also be applied to studying flood exposure inequalities (Sanders et al., 2022 ). The Gini index ranges from − 1 to 1, 0 representing perfect equality. To determine the equality of flooding, we first merged the CDC’s SVI data and flood data by tract, then calculated flood exposure per tract. In this instance, the flood exposure is the tract population multiplied by the percent of the tract flooded. To measure equality in the different FEMA flood zones, flood exposure is the tract population multiplied by the percent of a given FEMA zone, then multiplied by the percent of the zone flooded. Then, we sorted the table by desired variables (population below the poverty line, SVI score) and plotted the Lorenz Curve with the cumulative percent of flood exposure on the y axis and the cumulative percent of population on the x axis. Then, we calculated the Gini index to determine if there was a disproportionate impact on people in poverty and vulnerable populations, and if this impact varied by flood zone. 3. Results We quantified the flood extent after Hurricane Ida in southeastern Pennsylvania using Sentinel-2 satellite imagery, derived indices, linear unmixing, land cover and surface water data in a RF model trained with polygons of three classes: NotWater, Water and Flood. The training data used PlanetScope imagery and incorporated crowdsourced social media and permanent water data. When creating a flood extent, this approach proved highly accurate (> 99% overall accuracy). The resulting flood extent compared to the FEMA flood zones also reveals that, in this event, there was more than double the area of floods in the Minimal Hazard zones than in the 500-yr flood zone. 3.1 Flood Extent Generated with Random Forest Model Our methods produced a new flood extent map after Hurricane Ida of three classes: NotWater, Water and Flood for the study area of four counties in southeastern Pennsylvania, including the urban area of Philadelphia County, where no prior flood extent existed. The result is a map of flood extent for September 2, 2021, a day after Hurricane Ida passed through the study area and the day the Sentinel-2 mission captured data (Fig. 4 ). When zooming into different land uses in the study area, the RF model Flood classification visually aligns with Sentinel-2 false color imagery (Fig. 5 ). We created a confusion matrix for our RF Model using the validation dataset. The overall accuracy was 99.86%, and for the Flood class, the producer’s and the user’s accuracy were over 99% (Table 4 ). Table 4 Confusion Matrix (pixel count) of Random Forest (RF) model with an overall accuracy of 99.87%. Reference Data NotWater Water Flood Total User’s accuracy (%) Classified Data NotWater 12,527 0 0 12,527 > 99% Water 4 6,261 14 6,279 > 99% Flood 6 3 1,833 1,842 > 99% Total 12,537 6,264 1,847 20,648 Producer’s accuracy (%) > 99% > 99% > 99% We found the relative feature importance using the explain function in GEE (Table 5 ). The most important feature for classification came from the Sentinel-2 Band 1, aerosols 3x3 window and aerosols; the next important feature was slope. The most important water index was NDFI, although in previous runs, it was MNDWI 3x3 window. The 3x3 window of the features tended to have lower importance in our RF model than the regular features. Consistently, the least important features were Water and Water 3x3 window from the JRC dataset. Table 5 Relative feature importance for the RF model. Top Features Middle Features Lower Features Aerosols 2.74 NDMI 1.76 MNDWI 1.50 Aerosols (3x3) 2.35 NDVI (3x3) 1.75 JRC Water Occurrence 1.47 Slope 2.25 Blue (3x3) 1.73 WRI 1.46 Vegetation fraction (3x3) 2.19 Red Edge 3 1.71 Impervious 1.46 Red Edge 1 2.17 Red Edge 2 (3x3) 1.65 Aspect (3x3) 1.45 Landcover (3x3) 2.10 Blue 1.65 EVI 1.44 SWIR 2 1.99 NDWI 1.64 Aspect 1.43 Green 1.98 NIR 1.64 AWEI sh (3x3) 1.37 Landcover 1.95 JRC Water Occurrence (3x3) 1.60 NDMI (3x3) 1.36 SWIR 2 (3x3) 1.91 NDFI (3x3) 1.58 EVI (3x3) 1.30 NDFI 1.91 AWEI nsh 1.58 SWIR 1 (3x3) 1.28 Red (3x3) 1.90 SWIR 1 1.55 Impervious (3x3) 1.17 Water vapor (3x3) 1.90 Water fraction (3x3) 1.54 NIR (3x3) 1.16 NDVI 1.84 AWEI nsh (3x3) 1.54 Hillshade 1.14 Vegetation fraction 1.83 WRI (3x3) 1.52 Red Edge 4 (3x3) 1.12 Red Edge 1 (3x3) 1.82 NDWI (3x3) 1.52 Red Edge 4 1.07 Water vapor 1.81 Urban fraction 1.52 Hillshade (3x3) 1.02 Green (3x3) 1.80 Red Edge 2 1.51 JRC Water (3x3) 0.92 Water fraction 1.79 AWEI sh 1.51 Slope (3x3) 0.82 Urban fraction (3x3) 1.79 MNDWI (3x3) 1.50 JRC Water 0.81 Red 1.76 Red Edge 3 (3x3) 1.50 *The (3x3) signifies it is the mean value of the variable created with a 3 by 3 window 3.2 Comparison to FEMA Flood zones After we created a flood extent for Hurricane Ida using our RF model, we calculated the area and percent of floods that occurred in the different FEMA flood zones (Table 6 ). The FEMA flood zones we investigated were the 100-yr, 500-yr, Minimal Hazard zones and Other (a combination of rarer zones). All counties experienced less than half a percent of the total county area being flooded. Bucks County had the most flooding by area, while Philadelphia and Bucks counties had the highest percentage of the area flooded, with 0.51% and 0.43%, respectively. Table 6 The total classified flood, both area (km 2 ) and percent, per county in the study area and within FEMA flood zones. Every area calculation comes with a margin of error of at least +/-1%, given the uncertainty in the RF model. County Total Floods Floods within FEMA Flood Zones Other 100-yr 500-yr Minimal Hazard % km² % km² % km² % km² % km² Bucks 0.43 6.94 48.0 3.33 34.7 2.41 2.4 0.17 14.6 1.01 Delaware 0.20 0.98 30.3 0.30 58.1 0.57 0.9 0.01 7.8 0.08 Montgomery 0.41 5.15 51.5 2.65 40.1 2.06 1.5 0.08 6.6 0.34 Philadelphia 0.51 1.89 44.9 0.85 51.5 0.97 0.4 0.01 3.2 0.06 Most flooding occurred in the 100-yr and Other FEMA flood zones. All counties received some amount of flooding in Minimal Hazard zones and more flooding by area and percent in the Minimal Hazard zone than the 500-yr zone. In Bucks County, ~ 14% +/- 1.4% of total floods occurred in the Minimal Hazard zone. The percentage of FEMA zones flooded told a slightly different story with the Other zone receiving the highest proportion of flooding, then the 100-yr, then 500-yr, then the Minimal Hazard zone (Table 7 ). Table 7 Percentage of FEMA flood zones flooded per county. County FEMA Flood Zones % Flooded Other 100-yr 500-yr Minimal Hazard Bucks 8.23 2.18 0.58 0.58 Delaware 3.53 1.27 0.10 0.10 Montgomery 7.88 3.68 0.16 0.16 Philadelphia 10.24 2.36 0.05 0.05 Tract-level flood information can reveal more local patterns in flooding. Figure 6 shows the percentage of floods that occurred in a subset of FEMA flood zones (100-yr, 500-yr, Minimal Hazard). The tract-level shows a similar pattern to the county information, showing that the highest percent of floods occurs in the 100-yr and Minimal Hazard zones and the smallest percent of floods occurs in the 500-yr zone. 3.3 Vulnerability and Poverty Distribution In addition to investigating the distribution of Hurricane Ida floods across FEMA flood zones, we also investigated the distribution of floods across tract poverty and SVI scores. We found that flood exposure did not disproportionately impact people below the poverty line or vulnerable populations (higher SVI scores) (Fig. 7 ). People below the poverty line were underexposed to floods overall (G = -0.205), as well as in FEMA’s Other (G = -0.218) and Minimal Hazard zone (G = -0.363). Vulnerable populations (i.e. higher SVI scores) were underexposed to floods overall (G = -0.274), as well as in FEMA’s Other (G = -0.287), 100-yr (G = -0.248) and Minimal Hazard zones (G = -0.389). For both variables studied, there was a weak underexposure to floods for people below the poverty line and vulnerable populations in FEMA’s 500-yr zones (-0.2 < G < -0.1). 4. Discussion We developed reproducible methods for creating a flood extent after a major flood event in an urban area using the RF algorithm and mostly freely available data and software. While the RF algorithm and satellite imagery has been previously used to detect floods (Schaffer-Smith et al., 2020 ; Tulbure et al., 2018 , 2022 ), these methods are rarely applied in an urban environment. We successfully applied these methods to the dense urban county of Philadelphia and three surrounding, less urban counties. We then used this novel flood extent of Hurricane Ida to investigate its distribution in FEMA flood zones and across the population. 4.1 Flood Extent The flood extent for the study area had an overall accuracy of 99% and the Flood class had a users and producer's accuracy both over 99%. These metrics demonstrate that our methods are accurate when producing a flood extent in urban and suburban areas. A visual inspection also showed that in Philadelphia our RF model accurately classified floods in the frequently flooded neighborhood of Manayunk, the highway Interstate-676, and along a popular path, the Schuylkill Banks. Our RF model accuracy is relatively high, compared to the accuracies of other studies that used remotely sensed imagery to classify water. Feng et al. (2015) and Rosser et al. ( 2017 ) focused on urban flooding in China (87.3% accuracy) and the United Kingdom (95% accuracy), respectively. Studies in non-urban areas also had relatively high accuracies. A study mapping hurricane flooding in North Carolina, USA had an accuracy of 91% (Schaffer-Smith et al., 2020 ) and a study mapping surface water in a semi-arid river basin in Australia had an overall accuracy of 99% with 80% as the user’s accuracy of the water class (Tulbure et al., 2022 ). Some of these variations in accuracy likely come from the classification method and type of remote sensing imagery used. Most of the studies used RF methods, similar to our study (Feng et al., 2015; Schaffer-Smith et al., 2020 ; Tulbure et al., 2022 ), but others used the Otsu method (Rosser et al., 2017 ). Additionally, each of these studies used a different type of remote sensing imagery, including Unmanned Aerial Vehicle (Feng et al., 2015), Landsat-8 (Rosser et al., 2017 ), SAR (Schaffer-Smith et al., 2020 ), and Harmonized Landsat Sentinel-2 (Tulbure et al., 2022 ). One possibility of the higher accuracy of this RF model is the relatively small study area and use of several datasets in addition to the satellite imagery. While the overall accuracies of the flood extent and the Flood class were high, there were undetected Water and Flood areas. Since optical imagery cannot permeate obstructions (e.g., bridges, trees), the RF model did not detect water or floods under these areas. Optical imagery also cannot permeate clouds, therefore to create a flood map for an event with heavy cloud cover, these methods can be applied with SAR imagery. The RF model can be adapted for a cloudy flood event and use SAR imagery, which can permeate clouds, as the main input along with supporting DEM and land cover data, to create a flood extent map. The RF model had difficulty detecting narrow rivers or creeks. When this occurred, the creek was usually categorized as NotWater or if the river’s banks flooded, then it was categorized as Flood. This was the case for most creeks, for example, Pennypack Creek and Wissahickon Creek, both ~ 20 miles long crossing multiple counties. This study used Sentinel-2 imagery in part because it is publicly accessible, but applying this RF model with higher-resolution imagery from the private sector (e.g. PlanetScope data) may improve classification, especially for smaller water bodies (Cooley et al., 2017 ). Our RF model's feature importance varied from other models detecting surface water. For instance, in our RF model the highest-ranked feature was aerosols (Band 1 of Sentinel-2, central wavelength 443 nm, 60 m resolution), and we have not found other models with a similar pattern. Slope was also a highly ranked feature, likely because it is a good predictor of floods since it influences water pooling. The AWEI indices (AWEI sh and AWEI nsh ) are useful for mapping surface water, along with MNDWI and NDWI (Feyisa et al., 2014 ; Pickens et al., 2020 ; Tulbure et al., 2016 ). In our RF model, the highest-ranked water index was NDFI, although on earlier model runs, MNDWI ranked in the top ten of features. While AWEI sh is usually helpful for mapping surface water, it may have been less important in this model because it tends to misclassify highly reflective surfaces (such as skyscrapers) as floods (Feyisa et al., 2014 ). The JRC Water input, which is a dataset created using Landsat imagery and classifying pixels into permanent, seasonal or not water, (Pekel et al., 2016 ) was consistently the least important feature. It is likely the least important because the JRC Yearly Water Classification History dataset does not include bodies of water smaller than 30 m by 30 m (Pekel et al., 2016 ) and the spatial resolution of our study is 10 m. Despite this drawback, when experimenting with different feature combinations, this dataset still slightly increased accuracy and was included in the final RF model. Overall, it is difficult to compare our RF model and other RF models detecting floods and surface water because each uses different satellite imagery and input features. Our study area also tends to be smaller and more urban than other studies, which is potentially the reason our feature importance differs from other research. While the indices and datasets for this RF model were carefully selected, there is room for model simplification and decreasing the number of features. In addition to decreasing the number of features, experimentation can be done to determine which features are more effective in urban areas versus suburban areas to tailor the RF model to each county. These methods created an accurate flood extent for an urban area with GEE and free imagery as inputs. While the training data took time to compile, the methods are accessible. They can be quickly deployed to find the flood extent in another urban area when satellite imagery is available. One limitation in making the training and validation dataset is that the curated flood-related tweets obtained from the Global Flood Monitor project are unavailable after February 2023 due to a change in Twitter’s web scraping policies (Calma, 2023 ). These accurate flood extents can be used to improve and validate flood models and calculate flood depth (Bangira et al., 2017 ; Woznicki et al., 2019 ). Outside of modeling, accurate flood maps created with satellite imagery can help governments at the local and state level with preparedness and mitigation efforts (Akhtar et al., 2021 ). It can also help governments address the impacts of floods through adaptation projects and fixing zoning regulations (Wing et al., 2022 ). 4.2 FEMA Flood Zones In our study area, the Hurricane Ida’s flood extent, by area and percent, predominantly occurred in the 100-yr and Other FEMA flood zones. These flood zones also had the highest proportion of floods compared to the other zones. This result aligns with FEMA’s flood zone descriptions because 100-yr flood zones have a 1% chance of flooding every year and the Other category includes zones that are designed to flood. Since our study compared a singular event to the FEMA flood zones, which represent the probability of yearly flooding, no conclusions can be drawn about the effectiveness of the FEMA zones. In every county twice the amount of floods, by area and percent, occurred in the Minimal Hazard zone than the 500-yr zone. Since we did not quantify the flood depth or damages, this is not necessarily cause for concern, more a reflection of where water pooled. One county that stood out with a high proportion of flooding in the Minimal Hazard zone was Bucks County. There may be more floods in the Minimal Hazard zone because the 100-yr and 500-yr zones are substantially smaller. When investigating the percentage of the FEMA zones flooded, a higher percentage of the 500-yr flood zone was flooded (0.05% − 0.58%) than the Minimal Hazard zone (0.02% − 0.07%) in all counties. There are two patterns occurring: by area, more floods occurred in the Minimal Hazard zone than the 500-yr zone, whereas by percentage, the 500-yr flood zone had more flooding than the Minimal Hazard zone. Our research shows that flooding can and does occur in the Minimal Hazard zone, which is important given the misconception that living outside the FEMA flood zone means there is no flood risk (Billings et al., 2022 ; Wing et al., 2022 ). FEMA flood zones underestimate flood exposure (Wing et al., 2018 ) and do not account for pluvial floods (U.S. Government Accountability Office, 2021 ). Floods and resulting damages regularly occur outside the FEMA 100-yr flood zone (Collins et al., 2022 ). Our Hurricane Ida flood map can reveal areas with a high proportion of floods in minimal-risk areas that may benefit from recovery assistance and future mitigation. Future research could expand the time scale and determine the rate of floods and the overall effectiveness of different FEMA flood zones in this study area. 4.3 Flood Distribution We used the flood extent of Hurricane Ida to investigate if flood exposure was equally distributed in the study area. We found that total flood exposure did not disproportionately impact people in poverty and vulnerable populations. While this finding does not align with previous research showing that vulnerable populations are more exposed to floods (Tate et al., 2021 ), there are a few caveats. Firstly, we looked at a subset of counties impacted by Hurricane Ida. Secondly, we used flood exposure at the Census tract level, when block group level or land parcel level can capture more detailed spatial heterogeneity (Brelsford et al., 2017 ). Thirdly, for the flood exposure calculation, we used flood area and not flood depth or damages. Still, this result demonstrates an efficient method to get a snapshot of flood distribution and can be deployed again in conjunction with flood depth for a more accurate flood exposure calculation. Too often, flooding disproportionately affects vulnerable populations with the fewest resources to recover (Schaffer-Smith et al., 2020 ; Tate et al., 2021 ). Therefore, it is important to research this trend and highlight vulnerable areas that received high amounts of flooding and could benefit from additional support and resources to recover after flooding. 5. Conclusion Accurate methods for quantifying flood extent can provide insights into the affected area and damages (Rosser et al., 2017 ), and inform response (Akhtar et al., 2021 ). Satellite imagery and the RF algorithm are a reliable combination to create a flood extent but are not often applied in urban areas. Our methods combine GEE, satellite imagery and other freely available data to create a RF model. This model created a new flood extent of Hurricane Ida in southeastern Pennsylvania with 99% accuracy and can be applied to other urban areas. This flood extent can be used to validate flood models and view patterns of flooding at the tract level. We investigated the distribution of this event in FEMA flood zones and found that most flooding occurred in the 100-yr and Other zones. In our study area, more floods occurred in the Minimal Hazard zone than the 500-yr zone, affirming previous research that consistently found flooding outside FEMA’s flood zones. This research refined methods for creating an accurate flood extent in urban areas and created a new one for our study area. Flood extent maps can serve stakeholders such as land use managers (Sofia et al., 2017 ), emergency planners (Goffi et al., 2020 ), city planners and residents (Hosseiny et al., 2020 ). The maps can help inform recovery efforts, prioritize mitigation efforts (Hosseiny et al., 2020 ) and plan for future development (Sofia et al., 2017 ), making our cities more resilient to the increasing risk of floods. Declarations 6. Acknowledgements We appreciate the thorough comments and suggestions from the anonymous reviewers and editor. We also appreciate the European Space Agency for Sentinel-2A data, Planet Labs for PlanetScope data and Google for the Google Earth Engine (GEE) platform. 9.1 Funding The authors declare that no funds, grants, or other support were received during the preparation of this manuscript. 9.2 Competing Interests The authors have no relevant financial or non-financial interests to disclose. 9.3 Author Contributions Rebecca Composto: Conceptualization, Investigation, Methodology, Formal Analysis, Visualization, Writing - original draft, Writing - review & editing. Mirela Tulbure: Supervision, Conceptualization, Methodology, Writing - review & editing. Varun Tiwari: Methodology (supporting), Formal Analysis (supporting GEE code), Writing - review & editing. Mollie D. Gaines: Methodology (supporting), Formal Analysis (supporting Python code), Writing - review & editing. Júlio Caineta: Methodology (supporting), Writing - Review & editing. References Akhtar, Z., Ofli, F., & Imran, M. (2021). Towards Using Remote Sensing and Social Media Data for Flood Mapping. ISCRAM 2021 Conference Proceedings–18th International Conference on Information Systems for Crisis Response and Management, 536–551. Apel, H., Martínez Trepat, O., Hung, N. N., Chinh, D. T., Merz, B., & Dung, N. V. (2016). Combined fluvial and pluvial urban flood hazard analysis: Concept development and application to Can Tho city, Mekong Delta, Vietnam. Natural Hazards and Earth System Sciences, 16(4), 941–961. https://doi.org/10.5194/nhess-16-941-2016 Ayanu, Y. Z., Conrad, C., Nauss, T., Wegmann, M., & Koellner, T. (2012). Quantifying and Mapping Ecosystem Services Supplies and Demands: A Review of Remote Sensing Applications. Environmental Science & Technology, 46(16), 8529–8541. https://doi.org/10.1021/es300157u Bangira, T., Alfieri, S. M., Menenti, M., Van Niekerk, A., & Vekerdy, Z. (2017). A Spectral Unmixing Method with Ensemble Estimation of Endmembers: Application to Flood Mapping in the Caprivi Floodplain. Remote Sensing, 9(10), Article 10. https://doi.org/10.3390/rs9101013 Beven II, J., Hagen, A., & Berg, R. (2022). National Hurricane Center Tropical Cyclone Report: Hurricane Ida. NOAA. https://www.nhc.noaa.gov/data/tcr/AL092021_Ida.pdf Billings, S. B., Gallagher, E. A., & Ricketts, L. (2022). Let the rich be flooded: The distribution of financial aid and distress after hurricane harvey. Journal of Financial Economics, 146(2), 797–819. https://doi.org/10.1016/j.jfineco.2021.11.006 Blum, A. G., Ferraro, P. J., Archfield, S. A., & Ryberg, K. R. (2020). Causal Effect of Impervious Cover on Annual Flood Magnitude for the United States. Geophysical Research Letters, 47(5), e2019GL086480. https://doi.org/10.1029/2019GL086480 Boschetti, M., Nutini, F., Manfron, G., Brivio, P. A., & Nelson, A. (2014). Comparative Analysis of Normalised Difference Spectral Indices Derived from MODIS for Detecting Surface Water in Flooded Rice Cropping Systems. PLOS ONE, 9(2), e88741. https://doi.org/10.1371/journal.pone.0088741 Brandt, S. A., Lim, N. J., Colding, J., & Barthel, S. (2021). Mapping Flood Risk Uncertainty Zones in Support of Urban Resilience Planning. Urban Planning, 6(3), 258–271. https://doi.org/10.17645/up.v6i3.4073 Breiman, L. (2001). Random Forests. Machine Learning, 45(1), 5–32. https://doi.org/10.1023/A:1010933404324 Brelsford, C., Lobo, J., Hand, J., & Bettencourt, L. M. A. (2017). Heterogeneity and scale of sustainable development in cities. Proceedings of the National Academy of Sciences, 114(34), 8963–8968. https://doi.org/10.1073/pnas.1606033114 C. Yang, J. H. Everitt, & J. M. Bradford. (2007). Using Multispectral Imagery and Linear Spectral Unmixing Techniques for Estimating Crop Yield Variability. Transactions of the ASABE, 50(2), 6676–674. https://doi.org/10.13031/2013.22658 Calma, J. (2023, May 31). Scientists say they can’t rely on Twitter anymore. The Verge. https://www.theverge.com/2023/5/31/23739084/twitter-elon-musk-api-policy-chilling-academic-research CDC. (2020). CDC/ATSDR Social Vulnerability Index [Database State]. https://www.atsdr.cdc.gov/placeandhealth/svi/data_documentation_download.html Census. (2021). City and Town Population Totals: 2020-2021. Census.Gov. https://www.census.gov/data/tables/time-series/demo/popest/2020s-total-cities-and-towns.html Clement, M. a., Kilsby, C. g., & Moore, P. (2018). Multi-temporal synthetic aperture radar flood mapping using change detection. Journal of Flood Risk Management, 11(2), 152–168. https://doi.org/10.1111/jfr3.12303 Collins, E. L., Sanchez, G. M., Terando, A., Stillwell, C. C., Mitasova, H., Sebastian, A., & Meentemeyer, R. K. (2022). Predicting flood damage probability across the conterminous United States. Environmental Research Letters, 17(3), 034006. https://doi.org/10.1088/1748-9326/ac4f0f Cooley, S. W., Smith, L. C., Stepan, L., & Mascaro, J. (2017). Tracking Dynamic Northern Surface Water Changes with High-Frequency Planet CubeSat Imagery. Remote Sensing, 9(12), Article 12. https://doi.org/10.3390/rs9121306 Cooper, K., Rizzo, E., & Schmidt, S. (2022, September 1). One year after Hurricane Ida, Pa. Residents are still paying the price. WHYY. https://whyy.org/articles/pensylvannia-hurricane-ida-one-year-anniversary/ CRED. (2015). The Human Cost of Natural Disasters: A global perspective. Centre for Research on the Epidemiology of Disasters (CRED). http://repo.floodalliance.net/jspui/44111/1165 de Bruijn, J. A., de Moel, H., Jongman, B., de Ruiter, M. C., Wagemaker, J., & Aerts, J. C. J. H. (2019). A global database of historic and real-time flood events based on social media. Scientific Data, 6(1), Article 1. https://doi.org/10.1038/s41597-019-0326-9 FEMA. (2020, July 8). Flood Zones. https://www.fema.gov/glossary/flood-zones FEMA. (2023, October 23). Laws and Regulations. https://www.fema.gov/flood-insurance/rules-legislation/laws Feyisa, G. L., Meilby, H., Fensholt, R., & Proud, S. R. (2014). Automated Water Extraction Index: A new technique for surface water mapping using Landsat imagery. Remote Sensing of Environment, 140, 23–35. https://doi.org/10.1016/j.rse.2013.08.029 Fielding, J. L. (2018). Flood risk and inequalities between ethnic groups in the floodplains of England and Wales. Disasters, 42(1), 101–123. https://doi.org/10.1111/disa.12230 Gao, B. (1996). NDWI—A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sensing of Environment, 58(3), 257–266. https://doi.org/10.1016/S0034-4257(96)00067-3 Garbutt, K., Ellul, C., & Fujiyama, T. (2015). Mapping social vulnerability to flood hazard in Norfolk, England. Environmental Hazards, 14(2), 156–186. https://doi.org/10.1080/17477891.2015.1028018 Goffi, A., Stroppiana, D., Brivio, P. A., Bordogna, G., & Boschetti, M. (2020). Towards an automated approach to map flooded areas from Sentinel-2 MSI data and soft integration of water spectral features. International Journal of Applied Earth Observation and Geoinformation, 84, 101951. https://doi.org/10.1016/j.jag.2019.101951 Gómez-Palacios, D., Torres, M. A., & Reinoso, E. (2017). Flood mapping through principal component analysis of multitemporal satellite imagery considering the alteration of water spectral properties due to turbidity conditions. Geomatics, Natural Hazards and Risk, 8(2), 607–623. https://doi.org/10.1080/19475705.2016.1250115 Hall, T. M., & Kossin, J. P. (2019). Hurricane stalling along the North American coast and implications for rainfall. Npj Climate and Atmospheric Science, 2(1), Article 1. https://doi.org/10.1038/s41612-019-0074-8 Hermas, E., Gaber, A., & El Bastawesy, M. (2021). Application of remote sensing and GIS for assessing and proposing mitigation measures in flood-affected urban areas, Egypt. The Egyptian Journal of Remote Sensing and Space Science, 24(1), 119–130. https://doi.org/10.1016/j.ejrs.2020.03.002 Holland, G., & Bruyère, C. L. (2014). Recent intense hurricane response to global climate change. Climate Dynamics, 42(3), 617–627. https://doi.org/10.1007/s00382-013-1713-0 Hondula, K. L., DeVries, B., Jones, C. N., & Palmer, M. A. (2021). Effects of Using High Resolution Satellite‐Based Inundation Time Series to Estimate Methane Fluxes From Forested Wetlands. Geophysical Research Letters, 48(6). https://doi.org/10.1029/2021GL092556 Hosseiny, H., Crimmins, M., Smith, V. B., & Kremer, P. (2020). A Generalized Automated Framework for Urban Runoff Modeling and Its Application at a Citywide Landscape. Water, 12(2), 357. https://doi.org/10.3390/w12020357 Huete, A. (1997). A comparison of vegetation indices over a global set of TM images for EOS-MODIS. Remote Sensing of Environment, 59(3), 440–451. https://doi.org/10.1016/S0034-4257(96)00112-5 Ireland, G., Volpi, M., & Petropoulos, G. P. (2015). Examining the Capability of Supervised Machine Learning Classifiers in Extracting Flooded Areas from Landsat TM Imagery: A Case Study from a Mediterranean Flood. Remote Sensing, 7(3), Article 3. https://doi.org/10.3390/rs70303372 Jones, J. W. (2019). Improved Automated Detection of Subpixel-Scale Inundation—Revised Dynamic Surface Water Extent (DSWE) Partial Surface Water Tests. Remote Sensing, 11(4), Article 4. https://doi.org/10.3390/rs11040374 Kawasaki, A., Kawamura, G., & Zin, W. W. (2020). A local level relationship between floods and poverty: A case in Myanmar. International Journal of Disaster Risk Reduction, 42, 101348. https://doi.org/10.1016/j.ijdrr.2019.101348 Knighton, J., Hondula, K., Sharkus, C., Guzman, C., & Elliott, R. (2021). Flood risk behaviors of United States riverine metropolitan areas are driven by local hydrology and shaped by race. Proceedings of the National Academy of Sciences, 118(13), e2016839118. https://doi.org/10.1073/pnas.2016839118 Knutson, T. R., Sirutis, J. J., Zhao, M., Tuleya, R. E., Bender, M., Vecchi, G. A., Villarini, G., & Chavas, D. (2015). Global Projections of Intense Tropical Cyclone Activity for the Late Twenty-First Century from Dynamical Downscaling of CMIP5/RCP4.5 Scenarios. Journal of Climate, 28(18), 7203–7224. https://doi.org/10.1175/JCLI-D-15-0129.1 Kossin, J. P. (2018). A global slowdown of tropical-cyclone translation speed. Nature, 558(7708), Article 7708. https://doi.org/10.1038/s41586-018-0158-3 Kossin, J. P., Knapp, K. R., Vimont, D. J., Murnane, R. J., & Harper, B. A. (2007). A globally consistent reanalysis of hurricane variability and trends. Geophysical Research Letters, 34(4). https://doi.org/10.1029/2006GL028836 Kriegler, F. J., Malila, W. A., Nalepka, R. F., & Richardson, W. (1969). Preprocessing Transformations and Their Effects on Multispectral Recognition (p. 97). https://ui.adsabs.harvard.edu/abs/1969rse..conf...97K Lin, N., Emanuel, K., Oppenheimer, M., & Vanmarcke, E. (2012). Physically based assessment of hurricane surge threat under climate change. Nature Climate Change, 2(6), Article 6. https://doi.org/10.1038/nclimate1389 Liu, L., Liu, Y., Wang, X., Yu, D., Liu, K., Huang, H., & Hu, G. (2015). Developing an effective 2-D urban flood inundation model for city emergency management based on cellular automata. Natural Hazards and Earth System Sciences, 15(3), 381–391. https://doi.org/10.5194/nhess-15-381-2015 Lorenz, M. O. (1905). Methods of Measuring the Concentration of Wealth. Publications of the American Statistical Association, 9(70), 209. https://doi.org/10.2307/2276207 Markhvida, M., Walsh, B., Hallegatte, S., & Baker, J. (2020). Quantification of disaster impacts through household well-being losses. Nature Sustainability, 3(7), Article 7. https://doi.org/10.1038/s41893-020-0508-7 Mason, D. C., Giustarini, L., Garcia-Pintado, J., & Cloke, H. L. (2014). Detection of flooded urban areas in high resolution Synthetic Aperture Radar images using double scattering. International Journal of Applied Earth Observation and Geoinformation, 28, 150–159. https://doi.org/10.1016/j.jag.2013.12.002 Maxwell, A. E., Warner, T. A., & Fang, F. (2018). Implementation of machine-learning classification in remote sensing: An applied review. International Journal of Remote Sensing, 39(9), 2784–2817. https://doi.org/10.1080/01431161.2018.1433343 McFeeters, S. K. (1996). The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. International Journal of Remote Sensing, 17(7), 1425–1432. https://doi.org/10.1080/01431169608948714 Mtapuri, O., Dube, E., & Matunhu, J. (2018). Flooding and poverty: Two interrelated social problems impacting rural development in Tsholotsho district of Matabeleland North province in Zimbabwe. Jamba : Journal of Disaster Risk Studies, 10(1), 1–7. https://doi.org/10.4102/jamba.v10i1.455 NOAA. (2021, September). National Weather Service Forecast Office. https://www.weather.gov/wrh/climate?wfo=phi Olofsson, P., Foody, G. M., Herold, M., Stehman, S. V., Woodcock, C. E., & Wulder, M. A. (2014). Good practices for estimating area and assessing accuracy of land change. Remote Sensing of Environment, 148, 42–57. https://doi.org/10.1016/j.rse.2014.02.015 Pekel, J.-F., Cottam, A., Gorelick, N., & Belward, A. S. (2016). High-resolution mapping of global surface water and its long-term changes. Nature, 540(7633), 418–422. https://doi.org/10.1038/nature20584 Perin, V., Tulbure, M. G., Gaines, M. D., Reba, M. L., & Yaeger, M. A. (2022). A multi-sensor satellite imagery approach to monitor on-farm reservoirs. Remote Sensing of Environment, 270, 112796. https://doi.org/10.1016/j.rse.2021.112796 Phan, T. N., Kuch, V., & Lehnert, L. W. (2020). Land Cover Classification using Google Earth Engine and Random Forest Classifier—The Role of Image Composition. Remote Sensing, 12(15), Article 15. https://doi.org/10.3390/rs12152411 Pickens, A. H., Hansen, M. C., Hancher, M., Stehman, S. V., Tyukavina, A., Potapov, P., Marroquin, B., & Sherani, Z. (2020). Mapping and sampling to characterize global inland water dynamics from 1999 to 2018 with full Landsat time-series. Remote Sensing of Environment, 243, 111792. https://doi.org/10.1016/j.rse.2020.111792 Pinos, J., & Quesada-Román, A. (2022). Flood Risk-Related Research Trends in Latin America and the Caribbean. Water, 14(1), Article 1. https://doi.org/10.3390/w14010010 Pulcinella, M., Meyer, K., & Cooper, K. (2021, September 1). Long recovery ahead as Ida’s remnants lead to historic flooding, tornadoes in Philly region. WHYY. https://whyy.org/articles/philly-says-to-shelter-in-place-as-schuylkill-river-expected-to-rise-to-major-flood-stage/ Rentschler, J., Salhab, M., & Jafino, B. A. (2022). Flood exposure and poverty in 188 countries. Nature Communications, 13(1), Article 1. https://doi.org/10.1038/s41467-022-30727-4 Rosser, J. F., Leibovici, D. G., & Jackson, M. J. (2017). Rapid flood inundation mapping using social media, remote sensing and topographic data. Natural Hazards, 87(1), 103–120. https://doi.org/10.1007/s11069-017-2755-0 Sanders, B. F., Schubert, J. E., Kahl, D. T., Mach, K. J., Brady, D., AghaKouchak, A., Forman, F., Matthew, R. A., Ulibarri, N., & Davis, S. J. (2022). Large and inequitable flood risks in Los Angeles, California. Nature Sustainability, 1–11. https://doi.org/10.1038/s41893-022-00977-7 Schaffer-Smith, D., Myint, S. W., Muenich, R. L., Tong, D., & DeMeester, J. E. (2020). Repeated Hurricanes Reveal Risks and Opportunities for Social-Ecological Resilience to Flooding and Water Quality Problems. Environmental Science & Technology, 54(12), 7194–7204. https://doi.org/10.1021/acs.est.9b07815 Schnebele, E., Cervone, G., & Waters, N. (2014). Road assessment after flood events using non-authoritative data. Natural Hazards and Earth System Sciences, 14(4), 1007–1015. https://doi.org/10.5194/nhess-14-1007-2014 Settle, J. J., & Drake, N. A. (1993). Linear mixing and the estimation of ground cover proportions. International Journal of Remote Sensing, 14(6), 1159–1177. https://doi.org/10.1080/01431169308904402 Shen, L., & Li, C. (2010). Water body extraction from Landsat ETM+ imagery using adaboost algorithm. 2010 18th International Conference on Geoinformatics, 1–4. https://doi.org/10.1109/GEOINFORMATICS.2010.5567762 Smith, A. B. (2023). U.S. Billion-dollar Weather and Climate Disasters, 1980—Present (NCEI Accession 0209268) [dataset]. NOAA National Centers for Environmental Information. https://doi.org/10.25921/STKW-7W73 Sofia, G., Roder, G., Dalla Fontana, G., & Tarolli, P. (2017). Flood dynamics in urbanised landscapes: 100 years of climate and humans’ interaction. Scientific Reports, 7(1), Article 1. https://doi.org/10.1038/srep40527 Sweet, W., Dusek, G. (Gregory P. ), Marcy, D. C., Greg (Gregory W.), C., & Marra, J. (2019). 2018 State of U.S. High Tide Flooding with a 2019 Outlook. https://doi.org/10.25923/RBV9-TH19 Tanim, A. H., McRae, C. B., Tavakol-Davani, H., & Goharian, E. (2022). Flood Detection in Urban Areas Using Satellite Imagery and Machine Learning. Water, 14(7), Article 7. https://doi.org/10.3390/w14071140 Tate, E., Rahman, M. A., Emrich, C. T., & Sampson, C. C. (2021). Flood exposure and social vulnerability in the United States. Natural Hazards, 106(1), 435–457. https://doi.org/10.1007/s11069-020-04470-2 Tiwari, V., Tulbure, M. G., Caineta, J., Gaines, M. D., Perin, V., Kamal, M., Krupnik, T. J., Aziz, M. A., & Islam, A. T. (2024). Automated in-season rice crop mapping using Sentinel time-series data and Google Earth Engine: A case study in climate-risk prone Bangladesh. Journal of Environmental Management, 351, 119615. https://doi.org/10.1016/j.jenvman.2023.119615 Trenberth, K. E., Cheng, L., Jacobs, P., Zhang, Y., & Fasullo, J. (2018). Hurricane Harvey Links to Ocean Heat Content and Climate Change Adaptation. Earth’s Future, 6(5), 730–744. https://doi.org/10.1029/2018EF000825 Tulbure, M. G., & Broich, M. (2013). Spatiotemporal dynamic of surface water bodies using Landsat time-series data from 1999 to 2011. ISPRS Journal of Photogrammetry and Remote Sensing, 79, 44–52. https://doi.org/10.1016/j.isprsjprs.2013.01.010 Tulbure, M. G., Broich, M., Ju, J., Masek, J. G., & Wearne, J. (2018). Quantifying surface water extent and flooding in a dynamic dryland river system using the Harmonized Landsat/Sentinel-2 Reflectance Product. 2018, H21E-08. Tulbure, M. G., Broich, M., Perin, V., Gaines, M., Ju, J., Stehman, S. V., Pavelsky, T., Masek, J. G., Yin, S., Mai, J., & Betbeder-Matibet, L. (2022). Can we detect more ephemeral floods with higher density harmonized Landsat Sentinel 2 data compared to Landsat 8 alone? ISPRS Journal of Photogrammetry and Remote Sensing, 185, 232–246. https://doi.org/10.1016/j.isprsjprs.2022.01.021 Tulbure, M. G., Broich, M., Stehman, S. V., & Kommareddy, A. (2016). Surface water extent dynamics from three decades of seasonally continuous Landsat time series at subcontinental scale in a semi-arid region. Remote Sensing of Environment, 178, 142–157. https://doi.org/10.1016/j.rse.2016.02.034 U.S. Census Bureau. (2020). County-level Urban and Rural information for the 2020 Census [dataset]. https://www.census.gov/programs-surveys/geography/guidance/geo-areas/urban-rural.html US Department of Commerce. (2022, October 26). Hurricane Ida Emergency Response Imagery. https://oceanservice.noaa.gov/news/aug21/ngs-storm-imagery-ida.html US Department of Commerce, N. (2021). NWS Preliminary US Flood Fatality Statistics. NOAA’s National Weather Service. https://www.weather.gov/arx/usflood U.S. Geological Survey. (2023). 3D Elevation Program 10-Meter Resolution Digital Elevation Model. [dataset]. https://www.usgs.gov/the-national-map-data-delivery U.S. Government Accountability Office. (2021). FEMA Flood Maps: Better Planning and Analysis Needed to Address Current and Future Flood Hazards. https://www.gao.gov/assets/gao-22-104079.pdf Van Oldenborgh, G. J., Van Der Wiel, K., Sebastian, A., Singh, R., Arrighi, J., Otto, F., Haustein, K., Li, S., Vecchi, G., & Cullen, H. (2017). Attribution of extreme rainfall from Hurricane Harvey, August 2017. Environmental Research Letters, 12(12), 124009. https://doi.org/10.1088/1748-9326/aa9ef2 Wing, O. E. J., Bates, P. D., Smith, A. M., Sampson, C. C., Johnson, K. A., Fargione, J., & Morefield, P. (2018). Estimates of present and future flood risk in the conterminous United States. Environmental Research Letters, 13(3), 034023. https://doi.org/10.1088/1748-9326/aaac65 Wing, O. E. J., Lehman, W., Bates, P. D., Sampson, C. C., Quinn, N., Smith, A. M., Neal, J. C., Porter, J. R., & Kousky, C. (2022). Inequitable patterns of US flood risk in the Anthropocene. Nature Climate Change, 12(2), Article 2. https://doi.org/10.1038/s41558-021-01265-6 Winsemius, H. C., Jongman, B., Veldkamp, T. I. E., Hallegatte, S., Bangalore, M., & Ward, P. J. (2018). Disaster risk, climate change, and poverty: Assessing the global exposure of poor people to floods and droughts. Environment and Development Economics, 23(3), 328–348. https://doi.org/10.1017/S1355770X17000444 Woznicki, S. A., Baynes, J., Panlasigui, S., Mehaffey, M., & Neale, A. (2019). Development of a spatially complete floodplain map of the conterminous United States using random forest. Science of The Total Environment, 647, 942–953. https://doi.org/10.1016/j.scitotenv.2018.07.353 Xu, H. (2006). Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery. International Journal of Remote Sensing, 27(14), 3025–3033. https://doi.org/10.1080/01431160600589179 Cite Share Download PDF Status: Published Journal Publication published 27 Jul, 2024 Read the published version in Natural Hazards → Version 1 posted Editorial decision: Major revisions 03 Jun, 2024 Reviewers agreed at journal 27 Feb, 2024 Reviewers invited by journal 19 Feb, 2024 Editor assigned by journal 23 Jan, 2024 First submitted to journal 22 Jan, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3882712","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":273977331,"identity":"4977d637-5454-4879-aa9b-41ea3183fdf8","order_by":0,"name":"Rebecca Composto","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAqUlEQVRIiWNgGAWjYBADOSjNTLwWY9K1JDYQrcW8/fDRzQUVdenb2Y8/k2CosIbpxQ1kzqSl3Z5x5nDuzp4cMwmGM+mEtUhI8Jjd5m07kLvhBg+bBGPbYaK11KUb3GB/JsH4j3gtzAkGNxjMJBgbiNHCA/GL4YYzOcYWCcfSjQlrYT987DYwxOQNjh9/eONDjbUsQS0ggIiLBGKUo2oZBaNgFIyCUYANAAAsDjpLaahycgAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0009-0000-6869-4158","institution":"North Carolina State University College of Natural Resources","correspondingAuthor":true,"prefix":"","firstName":"Rebecca","middleName":"","lastName":"Composto","suffix":""},{"id":273977332,"identity":"f828893c-adf3-4ddb-afe6-6b1c2e3f0aac","order_by":1,"name":"Mirela G Tulbure","email":"","orcid":"","institution":"NC State University College of Natural Resources","correspondingAuthor":false,"prefix":"","firstName":"Mirela","middleName":"G","lastName":"Tulbure","suffix":""},{"id":273977333,"identity":"5c28a9bf-a2f2-47b5-bb94-ca859329d786","order_by":2,"name":"Varun Tiwari","email":"","orcid":"","institution":"NC State University College of Natural Resources","correspondingAuthor":false,"prefix":"","firstName":"Varun","middleName":"","lastName":"Tiwari","suffix":""},{"id":273977334,"identity":"c2b57a86-67a9-47de-ab3f-d68e8fb5fff1","order_by":3,"name":"Mollie D. Gaines","email":"","orcid":"","institution":"NC State University College of Natural Resources","correspondingAuthor":false,"prefix":"","firstName":"Mollie","middleName":"D.","lastName":"Gaines","suffix":""},{"id":273977335,"identity":"591be164-a788-47bd-8b9e-8386c03354d8","order_by":4,"name":"Júlio Caineta","email":"","orcid":"","institution":"NC State University College of Natural Resources","correspondingAuthor":false,"prefix":"","firstName":"Júlio","middleName":"","lastName":"Caineta","suffix":""}],"badges":[],"createdAt":"2024-01-20 22:14:43","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3882712/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3882712/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s11069-024-06817-5","type":"published","date":"2024-07-27T16:15:51+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":51978476,"identity":"c1d6ee6b-f779-451e-aeb5-7b35a8006f25","added_by":"auto","created_at":"2024-03-04 20:31:58","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1404249,"visible":true,"origin":"","legend":"\u003cp\u003eFrom left to right: the Hurricane Ida track in 2021 in the U.S. from NOAA’s National Hurricane Center and Central Pacific Hurricane Center, the study area of four counties (Delaware, Montgomery, Bucks, Philadelphia) in southeastern Pennsylvania all impacted by flooding with permanent water from the USGS National Hydrography Dataset, and Sentinel-2 false color imagery (SWIR2, NIR, Red as RGB) on August 13, 2021 (pre-flood) and September 2, 2021 (post-flood).\u003c/p\u003e","description":"","filename":"fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-3882712/v1/d33e9b5fcffebc65d34fe9f8.png"},{"id":51978477,"identity":"0aab6120-44d9-4d7f-a321-4ab4a4fa5976","added_by":"auto","created_at":"2024-03-04 20:31:58","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":9560155,"visible":true,"origin":"","legend":"\u003cp\u003eThe training and validation polygons were drawn by hand in QGIS for the RF model using several layers to corroborate the class (NotWater, Water, Flood). These layers included Google roads and satellite basemaps, PlanetScope (3 m) false color imagery (Blue, NIR, Red as RGB) from September 2, 2021, Sentinel-2 (10 m) false color imagery (SWIR2, NIR, Red as RGB) from September 2, 2021, and the USGS National Hydrography Dataset (NHD) for permanent water and flood-related tweets curated by Global Flood Monitor.\u003c/p\u003e","description":"","filename":"fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-3882712/v1/7e9248faa996b1f66988027e.png"},{"id":51978480,"identity":"124937ad-7d21-4d29-9832-473419dac73e","added_by":"auto","created_at":"2024-03-04 20:31:58","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":107731,"visible":true,"origin":"","legend":"\u003cp\u003eFlow chart of methodology.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-3882712/v1/d07822f3d26a31341189c174.png"},{"id":51979365,"identity":"c10ddac1-cc93-4971-8628-ada9f4cba05c","added_by":"auto","created_at":"2024-03-04 20:39:58","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1340052,"visible":true,"origin":"","legend":"\u003cp\u003eFlood extent map after Hurricane Ida on September 2, 2021, created using our RF model.\u003c/p\u003e","description":"","filename":"fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-3882712/v1/04e7d8b6141190b851f6a995.png"},{"id":51978482,"identity":"d0a04df1-ee4f-42c2-b663-7570b6890884","added_by":"auto","created_at":"2024-03-04 20:31:59","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":835039,"visible":true,"origin":"","legend":"\u003cp\u003eExamples of flood extent in different land uses (Farm, Neighborhood, Highway/urban) created using our RF model after Hurricane Ida on September 2, 2021. From left to right, Sentinel-2 false color imagery (SWIR2, NIR, Red as RGB) on August 13, 2023 (pre-flood), on September 2, 2021 (post-flood), and RF classification of flood extent overlaid on September 2, 2021 imagery.\u003c/p\u003e","description":"","filename":"fig5.png","url":"https://assets-eu.researchsquare.com/files/rs-3882712/v1/f3893608bda6f77189c23685.png"},{"id":51978478,"identity":"20aa884b-d397-482e-8340-8e7d30418122","added_by":"auto","created_at":"2024-03-04 20:31:58","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":308666,"visible":true,"origin":"","legend":"\u003cp\u003eFlood percent and area (km\u003csup\u003e2\u003c/sup\u003e) per tract that occurred in a subset of FEMA flood zones.\u003c/p\u003e","description":"","filename":"fig6.png","url":"https://assets-eu.researchsquare.com/files/rs-3882712/v1/e2fafcd3a007b0b7610876e5.png"},{"id":51978483,"identity":"0609d056-0a57-401e-abab-e9ffb2a4531f","added_by":"auto","created_at":"2024-03-04 20:31:59","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":87128,"visible":true,"origin":"","legend":"\u003cp\u003eLorenz Curves of flood exposure for people in poverty (top row) and vulnerable populations (bottom row), with Gini coefficients (G) for total flood exposure and flood exposure within FEMA flood zones (Other, 100-yr, 500-yr, Minimal Hazard).\u003c/p\u003e","description":"","filename":"fig7.png","url":"https://assets-eu.researchsquare.com/files/rs-3882712/v1/1929502519022e791dfe5534.png"},{"id":61596118,"identity":"94539ec6-ab6e-46ed-af9e-f7a16ba535fd","added_by":"auto","created_at":"2024-08-01 17:24:56","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":16245095,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3882712/v1/a1ac75f0-fc30-4d47-838d-3a02f74169f7.pdf"}],"financialInterests":"","formattedTitle":"Quantifying Urban Flood Extent Using Satellite Imagery and Random Forest: A Case Study in Southeastern Pennsylvania","fulltext":[{"header":"1. Introduction","content":"\u003cdiv id=\"Sec2\" class=\"Section2\"\u003e \u003ch2\u003e1.1 Floods Cause Harm\u003c/h2\u003e \u003cp\u003eFlooding has become a devastating and destructive hazard due to human development in high-risk areas (CRED, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2015\u003c/span\u003e) and climate change. The World Bank estimated in 2022 that 1.81\u0026nbsp;billion people, or 23% of the world\u0026rsquo;s population, are at risk of intense floods (Rentschler et al., \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Floods cause many types of harm (de Bruijn et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Rosser et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), including economic losses (Pinos \u0026amp; Quesada-Rom\u0026aacute;n, \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), damages to private homes/assets, damages to public infrastructure (Goffi et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), involuntary displacement, impacts on mental health (Markhvida et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), and disruptions to daily life and traffic flow (Hosseiny et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). In the most dangerous circumstances, floods can lead to loss of life (Goffi et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), taking 146 lives in 2021 in the U.S. (US Department of Commerce, 2021). In the U.S., floods cause major economic losses, with an average yearly cost of \u003cspan\u003e$\u003c/span\u003e4.5B (Smith, \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Among flooding, the average cost of each tropical storm, including hurricanes, is \u003cspan\u003e$\u003c/span\u003e22.2B (which includes damages caused by floods and winds) (Smith, \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). The economic and social costs of floods are high, and the risk of floods from hurricanes is expected to increase due in part to trends in climate change (Trenberth et al., \u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) and human development.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e1.2 Hurricanes risk increasing\u003c/h2\u003e \u003cp\u003eClimate change is heating the atmosphere, allowing it to hold more moisture and increasing precipitation frequency and intensity (Van Oldenborgh et al., \u003cspan citationid=\"CR83\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), leading to higher flood risk (Ireland et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Van Oldenborgh et al., \u003cspan citationid=\"CR83\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). In addition to more extreme precipitation, the U.S. is facing more intense hurricanes (Kossin et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). At the global scale, climate change is causing the speed of tropical cyclones to decrease and precipitation rates to increase (Kossin, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). In the U.S., this same trend applies with tropical storms stalling more often along the coast and increasing precipitation rates (Hall \u0026amp; Kossin, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). The intensity (Holland \u0026amp; Bruy\u0026egrave;re, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Knutson et al., \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2015\u003c/span\u003e), storm surge (Lin et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2012\u003c/span\u003e), and precipitation of these hurricane events are also expected to worsen due to climate change.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e1.3 Unequal distribution of risk\u003c/h2\u003e \u003cp\u003ePeople in poverty are disproportionately at risk of floods around the globe (Garbutt et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Kawasaki et al., \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Mtapuri et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Winsemius et al., \u003cspan citationid=\"CR86\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). This trend signifies that flood risk is not equally distributed (Wing et al., \u003cspan citationid=\"CR85\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). This imbalance occurs globally because people in poverty are more likely to live in a floodplain due to the concentration of jobs and transportation (Mtapuri et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). In the U.S., the same trend of inequitable flood risk applies, where flood risk disproportionately impacts poorer communities (Wing et al., \u003cspan citationid=\"CR85\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). At the city level, one case study of Los Angeles found that poorer communities have disproportionately higher flood risk, but this trend varied by flood type (Sanders et al., \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). These findings at the national and local scale show that it is important to investigate the distribution of flood impacts to help inform emergency response and recovery.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e1.4 Hurricane Ida 2021\u003c/h2\u003e \u003cp\u003eHurricane Ida was a category 4 hurricane that landed in the U.S. on August 29, 2021, bringing catastrophic damage (Beven II et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). In the first days of September, Hurricane Ida stalled, becoming extratropical and bringing heavy rains with rates around 3 inches per hour to states in the Mid-Atlantic and Northeast (Beven II et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). The storm caused dozens of fatalities and damaged homes, businesses, vehicles and infrastructure (Smith, \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). The National Oceanic and Atmospheric Association (NOAA) National Centers for Environmental Information (NCEI) estimates that the cost of Hurricane Ida is \u003cspan\u003e$\u003c/span\u003e80.2\u0026nbsp;Billion (Consumer Price Index-Adjusted) (Smith, \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), the costliest hazard of 2021.\u003c/p\u003e \u003cp\u003eIn Pennsylvania, Hurricane Ida brought precipitation and floods that caused damage (Beven II et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). In the aftermath of Hurricane Ida, individuals and households in Pennsylvania received \u003cspan\u003e$\u003c/span\u003e124M in funding from the Federal Emergency Management Agency (FEMA) to cover damages (Cooper et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Even with this influx in funding, many counties in the state are still recovering from the damages of Hurricane Ida, including Philadelphia, Montgomery, Delaware and Bucks counties (Cooper et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Given the trends in intensifying hurricanes, it is imperative to plan for future events using insights from past flood events (Brandt et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e1.5 Flood extent from satellite imagery\u003c/h2\u003e \u003cp\u003eSatellite imagery is a reliable data source (Hermas et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) that is regularly used to map surface water and flood dynamics across various scales (Ayanu et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Jones, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Pekel et al., \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Tulbure et al., \u003cspan citationid=\"CR77\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). Machine learning is an effective method for classifying floods in satellite imagery (Tulbure et al., \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). It has higher classification accuracy than parametric strategies (Maxwell et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2018\u003c/span\u003e), such as with a single water index (Goffi et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSupervised machine learning classification of satellite imagery relies on accurate training and validation data (Olofsson et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). A known flood extent from a reliable dataset (Hondula et al., \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) or created from aerial photography (Rosser et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Schnebele et al., \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2014\u003c/span\u003e) is ideal training data. However, it is not available for every flood event. In the case of Hurricane Ida, aerial imagery was not collected in the study area of southeastern Pennsylvania (US Department of Commerce, \u003cspan citationid=\"CR79\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Since an accurate flood extent was unavailable as training and validation data, we created one from satellite imagery and social media (Ireland et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Perin et al., \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eUrban environments have complex waterways, shallow and ephemeral flooding, and ponding, meaning the flood extent is discontinuous (Tanim et al., \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Woznicki et al., \u003cspan citationid=\"CR87\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Flood models can generate flood maps in urban areas (Knighton et al., \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Liu et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2015\u003c/span\u003e), however, flood maps generated from satellite imagery are considered more aligned with ground conditions. Synthetic Aperture Radar (SAR) data can collect data through clouds, but is limited in urban areas due to its side-looking nature (Mason et al., \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2014\u003c/span\u003e) and can have gaps due to tall buildings causing radar shadowing or layover (Clement et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Optical imagery cannot permeate through clouds, but does not face the same challenges with tall buildings. For our study area after Hurricane Ida, there was optical imagery, Sentinel-2, collected, but no freely available SAR data collected. In this study we used optical data and combined it with several datasets in our machine learning model to address the complexity of urban environments.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e1.6 Gaps and objectives\u003c/h2\u003e \u003cp\u003eThere is currently no Hurricane Ida flood extent in Pennsylvania, including the city of Philadelphia and surrounding counties. The floods resulting from Hurricane Ida dissipated slowly and coincidentally overlapped with Sentinel-2 imagery collection, providing a unique opportunity for testing flood detection methods in an urban environment. Previous research has predominantly used satellite imagery and machine learning to detect floods using a vetted flood extent (from satellite imagery or a flood model) and rarely applies these methods to an urban area. Therefore we focused on the urban area of Philadelphia and surrounding counties after Hurricane Ida, an event without a vetted flood extent available, to fill these gaps.\u003c/p\u003e \u003cp\u003eThe objectives of this research were to: (1) combine Sentinel-2 imagery and other data in a Random Forest (RF) machine learning algorithm to create a novel flood extent in southeastern Pennsylvania after Hurricane Ida; (2) compare the flood extent to FEMA\u0026rsquo;s (Federal Emergency Management Agency) flood zones; (3) use the flood extent to calculate flood exposure and determine the equality of its distribution. Since hurricanes and floods are expected to increase, methods for accurate and timely flood extent maps in urban areas are fundamental for improving recovery and mitigation efforts.\u003c/p\u003e \u003c/div\u003e"},{"header":"2. Methodology","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Study site\u003c/h2\u003e \u003cp\u003ePhiladelphia, located in southeastern Pennsylvania, is the sixth largest city by population in the U.S., with over 1.5\u0026nbsp;million people (Census, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Over the past hundreds of years, the city has grown and developed, building over the existing streams. Floods occur in Philadelphia 12 days per year, and this frequency is expected to increase (Sweet et al., \u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). While the city is regularly flooded, it has made considerable efforts to reduce floods through multiple avenues, including its stormwater management program and strict design requirements (Hosseiny et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eOn September 1, 2021, Philadelphia was directly in the path of Hurricane Ida (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) and received 2.37 inches of precipitation, significantly more than its annual monthly mean precipitation of 0.12 inches (NOAA, 2021). The impacts of Hurricane Ida in Philadelphia and the resulting floods were wide-ranging from disrupting daily life to damaging personal assets (Pulcinella et al., \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). In addition to Philadelphia, surrounding less urban counties, including Delaware, Bucks, and Montgomery counties, were impacted by Hurricane Ida flooding and incorporated into our study area (Cooper et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). According to the U.S. 2020 Census, Philadelphia (both the name of the city and county) is the most urban at 100%, then Delaware and Montgomery counties at 88% and 76%, respectively, and lastly, Bucks County at 44% (U.S. Census Bureau, \u003cspan citationid=\"CR78\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Datasets\u003c/h2\u003e \u003cdiv id=\"Sec11\" class=\"Section3\"\u003e \u003ch2\u003e2.2.1 Satellite Imagery\u003c/h2\u003e \u003cp\u003eWe selected Sentinel-2 imagery because it aligned with peak Hurricane Ida flooding in our study area and was the highest spatial resolution of publicly available imagery. Sentinel-2 is a mission run by the European Space Agency and produces publicly available satellite imagery of the globe at a 10\u0026ndash;60 m spatial resolution and a temporal resolution of ~\u0026thinsp;5 days. We used Sentinel-2 imagery that was collected a day after Hurricane Ida passed through the study area (September 2, 2021) with little (\u0026lt;\u0026thinsp;1%) cloud cover, making it an optimal data source. In Google Earth Engine (GEE), we obtained the imagery, filtered it temporally and spatially, and removed cloud and cirrus pixels using the quality assessment band (QA60) (Tiwari et al., \u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eOther imagery that was considered included Synthetic Aperture Radar (SAR) data, PlanetScope imagery and aerial imagery. The Sentinel-1 imagery dates did not align with the peak flooding in the study area. We considered PlanetScope imagery as the basis for the flood extent but instead used it to create the training and validation data for the RF model since it is best practice to use a higher resolution image for training and validation data than the model input (Olofsson et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). We did not consider aerial imagery because it was not collected by the National Geodetic Survey after Hurricane Ida in the study area (US Department of Commerce, \u003cspan citationid=\"CR79\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section3\"\u003e \u003ch2\u003e2.2.2 Sentinel-2 Indices\u003c/h2\u003e \u003cp\u003eThe RF inputs included all Sentinel-2 surface reflectance bands, two vegetation indices and six water indices, all previously shown to be important when mapping floods with satellite data (Goffi et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Tulbure et al., \u003cspan citationid=\"CR77\" class=\"CitationRef\"\u003e2016\u003c/span\u003e, \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). We produced all the indices in GEE. The vegetation indices are used to help classify NotWater pixels by identifying areas of vegetation. We used several water indices, each with different strengths, to help categorize Water (permanent) and Flood pixels in the model.\u003c/p\u003e \u003cp\u003eThe Normalized Difference Water Index (NDWI) is the standard for classifying water using green and near-infrared (NIR) bands (McFeeters, \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e1996\u003c/span\u003e). The Modified Normalized Difference Water Index (MNDWI) is a variation of NDWI that uses shortwave infrared (SWIR) instead of NIR and is more suitable in built-up areas than NDWI (Xu, \u003cspan citationid=\"CR88\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). The Automated Water Extraction Index (AWEI) uses five spectral bands to improve water classification by decreasing the environmental noise of shadows and dark surfaces (Feyisa et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). Two variations of the AWEI formulas (AWEI\u003csub\u003ensh\u003c/sub\u003e and AWEI\u003csub\u003esh\u003c/sub\u003e) have different effectiveness in urban areas. The AWEI\u003csub\u003ensh\u003c/sub\u003e formula is more equipped for urban areas because it effectively eliminates built surfaces. The AWEI\u003csub\u003esh\u003c/sub\u003e formula is more equipped for filtering out shadows, but is less equipped for urban areas because it tends to misclassify reflective roofs as water.\u003c/p\u003e \u003cp\u003eWe also used Linear Spectral Unmixing (LSU) to produce three inputs, each with the percent of three different \u0026ldquo;endmembers\u0026rdquo; (water, urban and vegetation) or classes for each pixel. Pixels have mixed spectral signatures because the underlying land cover is mixed and highly variable (C. Yang et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). LSU addresses this heterogeneity by using all bands to estimate each pixel's \u0026ldquo;endmember\u0026rdquo; percent (C. Yang et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). LSU is helpful in the context of floods because it can be used to determine the fraction of water in each pixel and produce flood maps (Bangira et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; G\u0026oacute;mez-Palacios et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2017\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSentinel-2 derived indices used in our RF classification.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCategory\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIndex Name\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFormula\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eReference\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eVegetation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNormalized Difference Vegetation Index (NDVI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNDVI = (NIR - RED / NIR\u0026thinsp;+\u0026thinsp;RED)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eKriegler et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e1969\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEnhanced Vegetation Index (EVI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eEVI\u0026thinsp;=\u0026thinsp;2.5 * ( NIR - RED) / (NIR\u0026thinsp;+\u0026thinsp;6 * RED \u0026minus;\u0026thinsp;7.5 * BLUE\u0026thinsp;+\u0026thinsp;1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHuete, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e1997\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"5\" rowspan=\"6\"\u003e \u003cp\u003eWater\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNormalized Difference Water Index (NDWI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNDWI = (GREEN - NIR) / (GREEN\u0026thinsp;+\u0026thinsp;NIR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMcFeeters, \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e1996\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eModified Normalized Difference Water Index (MNDWI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMNDWI = (GREEN - MIR) / (GREEN\u0026thinsp;+\u0026thinsp;MIR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eXu, \u003cspan citationid=\"CR88\" class=\"CitationRef\"\u003e2006\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAutomated Water Extraction Index (AWEI\u003csub\u003ensh\u003c/sub\u003e )\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAWEI\u003csub\u003ensh\u003c/sub\u003e = 4 * (GREEN - SWIR1) - (0.25 * NIR\u0026thinsp;+\u0026thinsp;2.75 * SWIR2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eFeyisa et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2014\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAutomated Water Extraction Index (AWEI\u003csub\u003esh\u003c/sub\u003e )\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAWEI\u003csub\u003ensh\u003c/sub\u003e = BLUE\u0026thinsp;+\u0026thinsp;2.5 * GREEN \u0026minus;\u0026thinsp;1.5 * (NIR\u0026thinsp;+\u0026thinsp;SWIR1)\u0026thinsp;=\u0026thinsp;0.25 * SWIR2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eFeyisa et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2014\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWater Ratio Index (WRI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWRI = (GREEN\u0026thinsp;+\u0026thinsp;RED) / (NIR\u0026thinsp;+\u0026thinsp;SWIR1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eShen \u0026amp; Li, \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2010\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNormalized Difference Moisture Index (NDMI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNDMI = (NIR - SWIR1) / (NIR + SWIR1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGao, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e1996\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFlood\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNormalized Difference Flow Index (NDFI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNDFI = (RED - SWIR2) / (RED\u0026thinsp;+\u0026thinsp;SWIR2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eBoschetti et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2014\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMultiple\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFractional Coefficients (water, urban and vegetation)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eR\u003csub\u003emix\u003c/sub\u003e = (F\u003csub\u003ewater\u003c/sub\u003e * R\u003csub\u003ewater\u003c/sub\u003e) + (F\u003csub\u003eurban\u003c/sub\u003e * R\u003csub\u003eurban\u003c/sub\u003e) + (F\u003csub\u003evegetation\u003c/sub\u003e * R\u003csub\u003evegetation\u003c/sub\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSettle \u0026amp; Drake, \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e1993\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section3\"\u003e \u003ch2\u003e2.2.3 Additional Inputs\u003c/h2\u003e \u003cp\u003eIn addition to Sentinel-2 surface reflectance bands and derived indices, several other datasets readily available in GEE were incorporated into the RF model (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). A digital elevation model (DEM) can be used to derive data (e.g., slope) that influences where floods occur (Tulbure et al., \u003cspan citationid=\"CR77\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). In our model, we used the United States Geological Survey (USGS) 3DEP 10 m National Map (U.S. Geological Survey, \u003cspan citationid=\"CR81\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) in GEE to calculate slope, aspect, and hillshade. We also used the USGS National Land Cover Database at 30 m resolution in the model resampled to 10 m, because land cover and impervious surface contribute to flood extent (Apel et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Blum et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn GEE we also incorporated datasets of 30 m resolution from the European Commission\u0026rsquo;s Joint Research Centre (JRC), including surface water occurrence and surface water classification (water, seasonal, permanent) (Pekel et al., \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). We downscaled the JRC datasets and the USGS National Land Cover Database in GEE using the resample function and bilinear method to match the other RF inputs at 10 m resolution. In GEE, we combined the non-Sentinel-2 inputs and reprojected them to match the Sentinel-2 inputs. Next, all the inputs were combined, and a 3x3 window was created for each band.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe RF model used 62 features derived from satellite imagery, digital elevation, land cover and water datasets that were preprocessed in Google Earth Engine (GEE).\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCategory\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSource\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eYear\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eOriginal Resolution\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eName\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eReflectance, Vegetation, Water, Moisture,\u003c/p\u003e \u003cp\u003eLinear Spectral Unmixing (LSU)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eSentinel-2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e2021\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e10\u0026ndash;60 m\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e12 Reflectance Bands\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e9 Derived Indices\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e3 Fractional coefficients (water, urban, vegetation)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eDEM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eUSGS 3DEP 10m National Map Seamless (1/3 Arc-Second)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e2017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e10 m\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSlope\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAspect\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHillshade\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eLand Cover\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eUSGS National Land Cover Database (NLCD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e2019\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e30 m\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLandcover\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eImpervious\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eWater\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eEuropean Commission Joint Research Centre (JRC)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e2021\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e30 m\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGlobal Surface Water - Occurrence\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eYearly Water Classification History\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWindow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAll above rows\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eVaries\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e10\u0026ndash;60 m\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e3x3 Window\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section3\"\u003e \u003ch2\u003e2.2.4 Training and Validation Data\u003c/h2\u003e \u003cp\u003eIn QGIS, we created the training and validation dataset by hand using several datasets (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) (Perin et al., \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). We used PlanetScope (3 m) imagery, which is higher resolution than our model inputs (Maxwell et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Olofsson et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2014\u003c/span\u003e), and the National Hydrography Dataset (NHD), as reference to create Flood and Water polygons at the resolution of Sentinel-2 (10 m) imagery (Ireland et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). We also used social media posts from the Global Food Monitor project, a publicly available database of flood-related tweets (de Bruijn et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). We used\u0026thinsp;~\u0026thinsp;3,000 tweets, including text and photos, and ~\u0026thinsp;140 unique points to guide the creation of Flood polygons (Akhtar et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Schnebele et al., \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2014\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe Google basemaps in QGIS provided high-resolution imagery for drawing NotWater polygons (Perin et al., \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Random points in each county were used to guide the location of NotWater polygons. Before drawing the NotWater polygons, we ensured that the polygon was outside the NHD and that the basemap imagery was clear of pools and other surface water. For all three classes (NotWater, Water, Flood), every layer of data was checked to verify the accuracy of the polygon class. The training and validation dataset was 433 polygons consisting of 164 NotWater, 131 Water and 138 Flood polygons.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eNext, we uploaded the training and validation dataset into GEE and converted it into 10 m pixels. We randomly oversampled the rarer classes of Water and Flood and used all NotWater polygons (Maxwell et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Olofsson et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). Then, we split the dataset into two stratified random samples, with 70% for training and 30% for validation (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eTraining and validation dataset (pixels 10 m) consists of three classes, oversampling the rarer Water and Flood classes, split into two stratified random samples with ~\u0026thinsp;48,000 pixels to train and ~\u0026thinsp;20,000 pixels to validate the RF model.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eType\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003ePercent\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eTotal Pixels\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c6\" namest=\"c4\"\u003e \u003cp\u003ePixels\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNotWater\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eWater\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eFlood\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTraining\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e70%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e48,352\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e29,473\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e14,721\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4,158\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eValidation\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e30%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e20,648\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e12,527\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e6,279\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1,842\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTotal\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e100%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e69,000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e42,000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e21,000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e6,000\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section3\"\u003e \u003ch2\u003e2.2.5 FEMA Flood Zones\u003c/h2\u003e \u003cp\u003eThe National Flood Hazard Layer (NFHL) is a database of flood zones and flood insurance requirements maintained by FEMA in support of the National Flood Insurance Program (NFIP) (FEMA, 2023). The NFHL is available for the entire study area, therefore, every flooded pixel will occur in a FEMA flood zone. The flood zones we focus on in this study are the 100-yr (100-year), 500-yr (500-year), and Minimal Hazard zones because they are the primary risk classifications. The 100-yr and 500-yr flood zones have a 1% and 0.2% likelihood of flooding yearly (FEMA, 2020). We combined all other FEMA flood zones (floodway, 1% annual chance flood hazard contained in channel, area with reduced flood risk due to levee, and 1% depth less than 1 foot) into an \u0026ldquo;Other\u0026rdquo; category that encapsulates areas that are less common. The Minimal Hazard zone is outside the 500-yr flood zone and at higher elevations. Once we created the Hurricane Ida flood extent, we used the FEMA flood zones to determine the area and percent area of the Hurricane Ida flood that occurred in the different zones at the county and tract level.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section3\"\u003e \u003ch2\u003e2.2.6 Vulnerability and Demographic Datasets\u003c/h2\u003e \u003cp\u003eThe Centers for Disease Control and Prevention (CDC) and Agency for Toxic Substances and Disease Registry (ATSDR) Social Vulnerability Index (SVI, hereafter) is a vulnerability index at the tract and county level in the United States. Vulnerability is a community\u0026rsquo;s ability to prevent suffering and financial loss due to a disaster (Fielding, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). The CDC\u0026rsquo;s SVI dataset estimates an overall vulnerability score using four themes (socioeconomic status, household characteristics, racial \u0026amp; ethnic minority status, and housing type \u0026amp; transportation) (CDC, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). From the CDC\u0026rsquo;s SVI 2020 dataset, we used the sum of the four themes as the SVI index and the number of people below the 150% federal poverty level. The poverty data in the CDC\u0026rsquo;s SVI dataset came from the 2016\u0026ndash;2020 American Community Survey (ACS). These two attributes were used to determine if Hurricane Ida floods disproportionately impacted people in poverty and vulnerable populations.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Random Forest Model\u003c/h2\u003e \u003cp\u003eThe RF machine learning algorithm is an ensemble classifier that uses a large number of decision trees that each use different random samples and a subset of features to assign a class, then the majority vote of all the trees classifies the data (Breiman, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2001\u003c/span\u003e; Maxwell et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). The RF algorithm effectively classifies surface water (Phan et al., \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Tulbure \u0026amp; Broich, \u003cspan citationid=\"CR74\" class=\"CitationRef\"\u003e2013\u003c/span\u003e) and floods (Tulbure et al., \u003cspan citationid=\"CR77\" class=\"CitationRef\"\u003e2016\u003c/span\u003e, \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). While the algorithm is a \u0026lsquo;black box\u0026rsquo;, meaning you cannot visualize all trees, it still has higher accuracy than parametric techniques, and it has the added benefit of being robust to smaller and lower quality training datasets (Maxwell et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2018\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eWe executed the RF algorithm in GEE with the steps outlined in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. We used the \u003cem\u003eee.Classifier.smileRandomForest\u003c/em\u003e function in GEE to train the model on the training data, then classify the entire study area and determine feature importance. The number of pixels used to train and validate are outlined in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. We used the validation data to create a confusion matrix. After the first run of the algorithm, we ran the model with the Sentinel-2 features and indices and different combinations of additional features and parameters in order to select the features and parameters (number of trees, number of features at each split) that produce the highest overall accuracy. The final parameters chosen for the algorithm were the default numbers, 100 trees and eight features per split. These parameters align with those chosen in other research using RF classification in GEE (Phan et al., \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). After these parameters were selected, the algorithm was run again with the optimal parameters and all the features.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Assessment of flood impact\u003c/h2\u003e \u003cp\u003eAfter we created a flood extent with our RF model, we exported it from GEE and converted it from a raster to a vector file to compare to the vector file of FEMA flood zones. Then, the flood polygons were clipped to different FEMA flood zones. In this study, we calculated the area and percent flood in four different FEMA flood zones: 100-yr, 500-yr, Minimal Hazards, and Other. The Other category consists of areas, including floodways, that are likely to be flooded; therefore, we expected most flooding to occur in the 100-yr and Other zone. After the flood extent was clipped to the different zones, the area (km\u003csup\u003e2\u003c/sup\u003e) and the percent of the flood that occurred in each FEMA flood zone were calculated at the county and census tract level.\u003c/p\u003e \u003cp\u003eWe assessed the flood exposure equality by plotting Lorenz Curves and calculating the Gini index. The Gini index is typically used to study income inequality (Lorenz, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e1905\u003c/span\u003e), but can also be applied to studying flood exposure inequalities (Sanders et al., \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). The Gini index ranges from \u0026minus;\u0026thinsp;1 to 1, 0 representing perfect equality. To determine the equality of flooding, we first merged the CDC\u0026rsquo;s SVI data and flood data by tract, then calculated flood exposure per tract. In this instance, the flood exposure is the tract population multiplied by the percent of the tract flooded. To measure equality in the different FEMA flood zones, flood exposure is the tract population multiplied by the percent of a given FEMA zone, then multiplied by the percent of the zone flooded. Then, we sorted the table by desired variables (population below the poverty line, SVI score) and plotted the Lorenz Curve with the cumulative percent of flood exposure on the y axis and the cumulative percent of population on the x axis. Then, we calculated the Gini index to determine if there was a disproportionate impact on people in poverty and vulnerable populations, and if this impact varied by flood zone.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cp\u003eWe quantified the flood extent after Hurricane Ida in southeastern Pennsylvania using Sentinel-2 satellite imagery, derived indices, linear unmixing, land cover and surface water data in a RF model trained with polygons of three classes: NotWater, Water and Flood. The training data used PlanetScope imagery and incorporated crowdsourced social media and permanent water data. When creating a flood extent, this approach proved highly accurate (\u0026gt;\u0026thinsp;99% overall accuracy). The resulting flood extent compared to the FEMA flood zones also reveals that, in this event, there was more than double the area of floods in the Minimal Hazard zones than in the 500-yr flood zone.\u003c/p\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Flood Extent Generated with Random Forest Model\u003c/h2\u003e \u003cp\u003eOur methods produced a new flood extent map after Hurricane Ida of three classes: NotWater, Water and Flood for the study area of four counties in southeastern Pennsylvania, including the urban area of Philadelphia County, where no prior flood extent existed. The result is a map of flood extent for September 2, 2021, a day after Hurricane Ida passed through the study area and the day the Sentinel-2 mission captured data (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). When zooming into different land uses in the study area, the RF model Flood classification visually aligns with Sentinel-2 false color imagery (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eWe created a confusion matrix for our RF Model using the validation dataset. The overall accuracy was 99.86%, and for the Flood class, the producer\u0026rsquo;s and the user\u0026rsquo;s accuracy were over 99% (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eConfusion Matrix (pixel count) of Random Forest (RF) model with an overall accuracy of 99.87%.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c5\" namest=\"c3\"\u003e \u003cp\u003eReference Data\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNotWater\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eWater\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFlood\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eTotal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cem\u003eUser\u0026rsquo;s accuracy (%)\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003eClassified Data\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNotWater\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e12,527\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e12,527\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cem\u003e\u0026gt;\u0026thinsp;99%\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWater\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e6,261\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e6,279\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cem\u003e\u0026gt;\u0026thinsp;99%\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFlood\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1,833\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1,842\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cem\u003e\u0026gt;\u0026thinsp;99%\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTotal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e12,537\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e6,264\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1,847\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e20,648\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cem\u003eProducer\u0026rsquo;s accuracy (%)\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cem\u003e\u0026gt;\u0026thinsp;99%\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003e\u0026gt;\u0026thinsp;99%\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003e\u0026gt;\u0026thinsp;99%\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eWe found the relative feature importance using the \u003cem\u003eexplain\u003c/em\u003e function in GEE (Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). The most important feature for classification came from the Sentinel-2 Band 1, aerosols 3x3 window and aerosols; the next important feature was slope. The most important water index was NDFI, although in previous runs, it was MNDWI 3x3 window. The 3x3 window of the features tended to have lower importance in our RF model than the regular features. Consistently, the least important features were Water and Water 3x3 window from the JRC dataset.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eRelative feature importance for the RF model.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eTop Features\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eMiddle Features\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eLower Features\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAerosols\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNDMI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMNDWI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.50\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAerosols (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNDVI (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eJRC Water Occurrence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.47\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSlope\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBlue (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.73\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eWRI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.46\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVegetation fraction (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRed Edge 3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eImpervious\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.46\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRed Edge 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRed Edge 2 (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAspect (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.45\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLandcover (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBlue\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eEVI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.44\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSWIR 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNDWI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAspect\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.43\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGreen\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNIR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAWEI\u003csub\u003esh\u003c/sub\u003e (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.37\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLandcover\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eJRC Water Occurrence (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNDMI (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.36\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSWIR 2 (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNDFI (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eEVI (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.30\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNDFI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAWEI\u003csub\u003ensh\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSWIR 1 (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.28\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRed (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSWIR 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eImpervious (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.17\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWater vapor (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWater fraction (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNIR (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.16\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNDVI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAWEI\u003csub\u003ensh\u003c/sub\u003e (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHillshade\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.14\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVegetation fraction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWRI (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eRed Edge 4 (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.12\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRed Edge 1 (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNDWI (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eRed Edge 4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.07\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWater vapor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eUrban fraction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHillshade (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.02\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGreen (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRed Edge 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eJRC Water (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWater fraction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAWEI\u003csub\u003esh\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSlope (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUrban fraction (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMNDWI (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eJRC Water\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.81\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRed Edge 3 (3x3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003e*The (3x3) signifies it is the mean value of the variable created with a 3 by 3 window\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Comparison to FEMA Flood zones\u003c/h2\u003e \u003cp\u003eAfter we created a flood extent for Hurricane Ida using our RF model, we calculated the area and percent of floods that occurred in the different FEMA flood zones (Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). The FEMA flood zones we investigated were the 100-yr, 500-yr, Minimal Hazard zones and Other (a combination of rarer zones). All counties experienced less than half a percent of the total county area being flooded. Bucks County had the most flooding by area, while Philadelphia and Bucks counties had the highest percentage of the area flooded, with 0.51% and 0.43%, respectively.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe total classified flood, both area (km\u003csup\u003e2\u003c/sup\u003e) and percent, per county in the study area and within FEMA flood zones. Every area calculation comes with a margin of error of at least +/-1%, given the uncertainty in the RF model.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"11\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eCounty\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" morerows=\"1\" nameend=\"c3\" namest=\"c2\" rowspan=\"2\"\u003e \u003cp\u003eTotal Floods\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"8\" nameend=\"c11\" namest=\"c4\"\u003e \u003cp\u003eFloods within FEMA Flood Zones\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003e\u003cem\u003eOther\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c7\" namest=\"c6\"\u003e \u003cp\u003e\u003cem\u003e100-yr\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e \u003cp\u003e\u003cem\u003e500-yr\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c11\" namest=\"c10\"\u003e \u003cp\u003e\u003cem\u003eMinimal\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003eHazard\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e%\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ekm\u0026sup2;\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e%\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ekm\u0026sup2;\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e%\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003ekm\u0026sup2;\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003e%\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003ekm\u0026sup2;\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003e%\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e \u003cp\u003ekm\u0026sup2;\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBucks\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.43\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e6.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e48.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3.33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e34.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e2.41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e2.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e14.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e1.01\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDelaware\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e30.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e58.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e7.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.08\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMontgomery\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e51.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e40.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e2.06\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e6.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.34\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePhiladelphia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e44.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e51.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e3.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.06\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eMost flooding occurred in the 100-yr and Other FEMA flood zones. All counties received some amount of flooding in Minimal Hazard zones and more flooding by area and percent in the Minimal Hazard zone than the 500-yr zone. In Bucks County, ~\u0026thinsp;14% +/- 1.4% of total floods occurred in the Minimal Hazard zone. The percentage of FEMA zones flooded told a slightly different story with the Other zone receiving the highest proportion of flooding, then the 100-yr, then 500-yr, then the Minimal Hazard zone (Table\u0026nbsp;\u003cspan refid=\"Tab7\" class=\"InternalRef\"\u003e7\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab7\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 7\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePercentage of FEMA flood zones flooded per county.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eCounty\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"4\" nameend=\"c5\" namest=\"c2\"\u003e \u003cp\u003eFEMA Flood Zones % Flooded\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cem\u003eOther\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cem\u003e100-yr\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003e500-yr\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003eMinimal\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cem\u003eHazard\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBucks\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e8.23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.58\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDelaware\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.53\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMontgomery\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e7.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.16\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePhiladelphia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e10.24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2.36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.05\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTract-level flood information can reveal more local patterns in flooding. Figure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e shows the percentage of floods that occurred in a subset of FEMA flood zones (100-yr, 500-yr, Minimal Hazard). The tract-level shows a similar pattern to the county information, showing that the highest percent of floods occurs in the 100-yr and Minimal Hazard zones and the smallest percent of floods occurs in the 500-yr zone.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Vulnerability and Poverty Distribution\u003c/h2\u003e \u003cp\u003eIn addition to investigating the distribution of Hurricane Ida floods across FEMA flood zones, we also investigated the distribution of floods across tract poverty and SVI scores. We found that flood exposure did not disproportionately impact people below the poverty line or vulnerable populations (higher SVI scores) (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e). People below the poverty line were underexposed to floods overall (G = -0.205), as well as in FEMA\u0026rsquo;s Other (G = -0.218) and Minimal Hazard zone (G = -0.363). Vulnerable populations (i.e. higher SVI scores) were underexposed to floods overall (G = -0.274), as well as in FEMA\u0026rsquo;s Other (G = -0.287), 100-yr (G = -0.248) and Minimal Hazard zones (G = -0.389). For both variables studied, there was a weak underexposure to floods for people below the poverty line and vulnerable populations in FEMA\u0026rsquo;s 500-yr zones (-0.2\u0026thinsp;\u0026lt;\u0026thinsp;G \u0026lt; -0.1).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eWe developed reproducible methods for creating a flood extent after a major flood event in an urban area using the RF algorithm and mostly freely available data and software. While the RF algorithm and satellite imagery has been previously used to detect floods (Schaffer-Smith et al., \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Tulbure et al., \u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e2018\u003c/span\u003e, \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), these methods are rarely applied in an urban environment. We successfully applied these methods to the dense urban county of Philadelphia and three surrounding, less urban counties. We then used this novel flood extent of Hurricane Ida to investigate its distribution in FEMA flood zones and across the population.\u003c/p\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Flood Extent\u003c/h2\u003e \u003cp\u003eThe flood extent for the study area had an overall accuracy of 99% and the Flood class had a users and producer's accuracy both over 99%. These metrics demonstrate that our methods are accurate when producing a flood extent in urban and suburban areas. A visual inspection also showed that in Philadelphia our RF model accurately classified floods in the frequently flooded neighborhood of Manayunk, the highway Interstate-676, and along a popular path, the Schuylkill Banks.\u003c/p\u003e \u003cp\u003eOur RF model accuracy is relatively high, compared to the accuracies of other studies that used remotely sensed imagery to classify water. Feng et al. (2015) and Rosser et al. (\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) focused on urban flooding in China (87.3% accuracy) and the United Kingdom (95% accuracy), respectively. Studies in non-urban areas also had relatively high accuracies. A study mapping hurricane flooding in North Carolina, USA had an accuracy of 91% (Schaffer-Smith et al., \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) and a study mapping surface water in a semi-arid river basin in Australia had an overall accuracy of 99% with 80% as the user\u0026rsquo;s accuracy of the water class (Tulbure et al., \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Some of these variations in accuracy likely come from the classification method and type of remote sensing imagery used. Most of the studies used RF methods, similar to our study (Feng et al., 2015; Schaffer-Smith et al., \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Tulbure et al., \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), but others used the Otsu method (Rosser et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Additionally, each of these studies used a different type of remote sensing imagery, including Unmanned Aerial Vehicle (Feng et al., 2015), Landsat-8 (Rosser et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), SAR (Schaffer-Smith et al., \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), and Harmonized Landsat Sentinel-2 (Tulbure et al., \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). One possibility of the higher accuracy of this RF model is the relatively small study area and use of several datasets in addition to the satellite imagery.\u003c/p\u003e \u003cp\u003eWhile the overall accuracies of the flood extent and the Flood class were high, there were undetected Water and Flood areas. Since optical imagery cannot permeate obstructions (e.g., bridges, trees), the RF model did not detect water or floods under these areas. Optical imagery also cannot permeate clouds, therefore to create a flood map for an event with heavy cloud cover, these methods can be applied with SAR imagery. The RF model can be adapted for a cloudy flood event and use SAR imagery, which can permeate clouds, as the main input along with supporting DEM and land cover data, to create a flood extent map.\u003c/p\u003e \u003cp\u003eThe RF model had difficulty detecting narrow rivers or creeks. When this occurred, the creek was usually categorized as NotWater or if the river\u0026rsquo;s banks flooded, then it was categorized as Flood. This was the case for most creeks, for example, Pennypack Creek and Wissahickon Creek, both ~\u0026thinsp;20 miles long crossing multiple counties. This study used Sentinel-2 imagery in part because it is publicly accessible, but applying this RF model with higher-resolution imagery from the private sector (e.g. PlanetScope data) may improve classification, especially for smaller water bodies (Cooley et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2017\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eOur RF model's feature importance varied from other models detecting surface water. For instance, in our RF model the highest-ranked feature was aerosols (Band 1 of Sentinel-2, central wavelength 443 nm, 60 m resolution), and we have not found other models with a similar pattern. Slope was also a highly ranked feature, likely because it is a good predictor of floods since it influences water pooling. The AWEI indices (AWEI\u003csub\u003esh\u003c/sub\u003e and AWEI\u003csub\u003ensh\u003c/sub\u003e) are useful for mapping surface water, along with MNDWI and NDWI (Feyisa et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Pickens et al., \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Tulbure et al., \u003cspan citationid=\"CR77\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). In our RF model, the highest-ranked water index was NDFI, although on earlier model runs, MNDWI ranked in the top ten of features. While AWEI\u003csub\u003esh\u003c/sub\u003e is usually helpful for mapping surface water, it may have been less important in this model because it tends to misclassify highly reflective surfaces (such as skyscrapers) as floods (Feyisa et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2014\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe JRC Water input, which is a dataset created using Landsat imagery and classifying pixels into permanent, seasonal or not water, (Pekel et al., \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2016\u003c/span\u003e) was consistently the least important feature. It is likely the least important because the JRC Yearly Water Classification History dataset does not include bodies of water smaller than 30 m by 30 m (Pekel et al., \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2016\u003c/span\u003e) and the spatial resolution of our study is 10 m. Despite this drawback, when experimenting with different feature combinations, this dataset still slightly increased accuracy and was included in the final RF model.\u003c/p\u003e \u003cp\u003eOverall, it is difficult to compare our RF model and other RF models detecting floods and surface water because each uses different satellite imagery and input features. Our study area also tends to be smaller and more urban than other studies, which is potentially the reason our feature importance differs from other research. While the indices and datasets for this RF model were carefully selected, there is room for model simplification and decreasing the number of features. In addition to decreasing the number of features, experimentation can be done to determine which features are more effective in urban areas versus suburban areas to tailor the RF model to each county.\u003c/p\u003e \u003cp\u003eThese methods created an accurate flood extent for an urban area with GEE and free imagery as inputs. While the training data took time to compile, the methods are accessible. They can be quickly deployed to find the flood extent in another urban area when satellite imagery is available. One limitation in making the training and validation dataset is that the curated flood-related tweets obtained from the Global Flood Monitor project are unavailable after February 2023 due to a change in Twitter\u0026rsquo;s web scraping policies (Calma, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThese accurate flood extents can be used to improve and validate flood models and calculate flood depth (Bangira et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Woznicki et al., \u003cspan citationid=\"CR87\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Outside of modeling, accurate flood maps created with satellite imagery can help governments at the local and state level with preparedness and mitigation efforts (Akhtar et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). It can also help governments address the impacts of floods through adaptation projects and fixing zoning regulations (Wing et al., \u003cspan citationid=\"CR85\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec25\" class=\"Section2\"\u003e \u003ch2\u003e4.2 FEMA Flood Zones\u003c/h2\u003e \u003cp\u003eIn our study area, the Hurricane Ida\u0026rsquo;s flood extent, by area and percent, predominantly occurred in the 100-yr and Other FEMA flood zones. These flood zones also had the highest proportion of floods compared to the other zones. This result aligns with FEMA\u0026rsquo;s flood zone descriptions because 100-yr flood zones have a 1% chance of flooding every year and the Other category includes zones that are designed to flood. Since our study compared a singular event to the FEMA flood zones, which represent the probability of yearly flooding, no conclusions can be drawn about the effectiveness of the FEMA zones.\u003c/p\u003e \u003cp\u003eIn every county twice the amount of floods, by area and percent, occurred in the Minimal Hazard zone than the 500-yr zone. Since we did not quantify the flood depth or damages, this is not necessarily cause for concern, more a reflection of where water pooled. One county that stood out with a high proportion of flooding in the Minimal Hazard zone was Bucks County. There may be more floods in the Minimal Hazard zone because the 100-yr and 500-yr zones are substantially smaller. When investigating the percentage of the FEMA zones flooded, a higher percentage of the 500-yr flood zone was flooded (0.05% \u0026minus;\u0026thinsp;0.58%) than the Minimal Hazard zone (0.02% \u0026minus;\u0026thinsp;0.07%) in all counties. There are two patterns occurring: by area, more floods occurred in the Minimal Hazard zone than the 500-yr zone, whereas by percentage, the 500-yr flood zone had more flooding than the Minimal Hazard zone.\u003c/p\u003e \u003cp\u003eOur research shows that flooding can and does occur in the Minimal Hazard zone, which is important given the misconception that living outside the FEMA flood zone means there is no flood risk (Billings et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Wing et al., \u003cspan citationid=\"CR85\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). FEMA flood zones underestimate flood exposure (Wing et al., \u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) and do not account for pluvial floods (U.S. Government Accountability Office, \u003cspan citationid=\"CR82\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Floods and resulting damages regularly occur outside the FEMA 100-yr flood zone (Collins et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Our Hurricane Ida flood map can reveal areas with a high proportion of floods in minimal-risk areas that may benefit from recovery assistance and future mitigation. Future research could expand the time scale and determine the rate of floods and the overall effectiveness of different FEMA flood zones in this study area.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec26\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Flood Distribution\u003c/h2\u003e \u003cp\u003eWe used the flood extent of Hurricane Ida to investigate if flood exposure was equally distributed in the study area. We found that total flood exposure did not disproportionately impact people in poverty and vulnerable populations. While this finding does not align with previous research showing that vulnerable populations are more exposed to floods (Tate et al., \u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), there are a few caveats. Firstly, we looked at a subset of counties impacted by Hurricane Ida. Secondly, we used flood exposure at the Census tract level, when block group level or land parcel level can capture more detailed spatial heterogeneity (Brelsford et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Thirdly, for the flood exposure calculation, we used flood area and not flood depth or damages. Still, this result demonstrates an efficient method to get a snapshot of flood distribution and can be deployed again in conjunction with flood depth for a more accurate flood exposure calculation. Too often, flooding disproportionately affects vulnerable populations with the fewest resources to recover (Schaffer-Smith et al., \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Tate et al., \u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Therefore, it is important to research this trend and highlight vulnerable areas that received high amounts of flooding and could benefit from additional support and resources to recover after flooding.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eAccurate methods for quantifying flood extent can provide insights into the affected area and damages (Rosser et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), and inform response (Akhtar et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Satellite imagery and the RF algorithm are a reliable combination to create a flood extent but are not often applied in urban areas. Our methods combine GEE, satellite imagery and other freely available data to create a RF model. This model created a new flood extent of Hurricane Ida in southeastern Pennsylvania with 99% accuracy and can be applied to other urban areas. This flood extent can be used to validate flood models and view patterns of flooding at the tract level. We investigated the distribution of this event in FEMA flood zones and found that most flooding occurred in the 100-yr and Other zones. In our study area, more floods occurred in the Minimal Hazard zone than the 500-yr zone, affirming previous research that consistently found flooding outside FEMA\u0026rsquo;s flood zones. This research refined methods for creating an accurate flood extent in urban areas and created a new one for our study area. Flood extent maps can serve stakeholders such as land use managers (Sofia et al., \u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), emergency planners (Goffi et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), city planners and residents (Hosseiny et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). The maps can help inform recovery efforts, prioritize mitigation efforts (Hosseiny et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) and plan for future development (Sofia et al., \u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), making our cities more resilient to the increasing risk of floods.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch3\u003e6. Acknowledgements\u003c/h3\u003e\n\u003cp\u003eWe appreciate the thorough comments and suggestions from the anonymous reviewers and editor. We also appreciate the European Space Agency for Sentinel-2A data, Planet Labs for PlanetScope data and Google for the Google Earth Engine (GEE) platform.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e9.1 Funding\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e9.2 Competing Interests\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;The authors have no relevant financial or non-financial interests to disclose.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e9.3 Author Contributions\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eRebecca Composto: Conceptualization, Investigation, Methodology, Formal Analysis, Visualization, Writing - original draft, Writing - review \u0026amp; editing. Mirela Tulbure: Supervision, Conceptualization, Methodology, Writing - review \u0026amp; editing. Varun Tiwari: Methodology (supporting), Formal Analysis (supporting GEE code), Writing - review \u0026amp; editing. Mollie D. Gaines: Methodology (supporting), Formal Analysis (supporting Python code), Writing - review \u0026amp; editing. J\u0026uacute;lio Caineta: Methodology (supporting), Writing - Review \u0026amp; editing.\u003c/p\u003e"},{"header":"References","content":"\u003cp\u003eAkhtar, Z., Ofli, F., \u0026amp; Imran, M. (2021). Towards Using Remote Sensing and Social Media Data for Flood Mapping. ISCRAM 2021 Conference Proceedings\u0026ndash;18th International Conference on Information Systems for Crisis Response and Management, 536\u0026ndash;551.\u003c/p\u003e\n\u003cp\u003eApel, H., Mart\u0026iacute;nez Trepat, O., Hung, N. N., Chinh, D. T., Merz, B., \u0026amp; Dung, N. V. (2016). Combined fluvial and pluvial urban flood hazard analysis: Concept development and application to Can Tho city, Mekong Delta, Vietnam. Natural Hazards and Earth System Sciences, 16(4), 941\u0026ndash;961. https://doi.org/10.5194/nhess-16-941-2016\u003c/p\u003e\n\u003cp\u003eAyanu, Y. Z., Conrad, C., Nauss, T., Wegmann, M., \u0026amp; Koellner, T. (2012). Quantifying and Mapping Ecosystem Services Supplies and Demands: A Review of Remote Sensing Applications. Environmental Science \u0026amp; Technology, 46(16), 8529\u0026ndash;8541. https://doi.org/10.1021/es300157u\u003c/p\u003e\n\u003cp\u003eBangira, T., Alfieri, S. M., Menenti, M., Van Niekerk, A., \u0026amp; Vekerdy, Z. (2017). A Spectral Unmixing Method with Ensemble Estimation of Endmembers: Application to Flood Mapping in the Caprivi Floodplain. Remote Sensing, 9(10), Article 10. https://doi.org/10.3390/rs9101013\u003c/p\u003e\n\u003cp\u003eBeven II, J., Hagen, A., \u0026amp; Berg, R. (2022). National Hurricane Center Tropical Cyclone Report: Hurricane Ida. NOAA. https://www.nhc.noaa.gov/data/tcr/AL092021_Ida.pdf\u003c/p\u003e\n\u003cp\u003eBillings, S. B., Gallagher, E. A., \u0026amp; Ricketts, L. (2022). Let the rich be flooded: The distribution of financial aid and distress after hurricane harvey. Journal of Financial Economics, 146(2), 797\u0026ndash;819. https://doi.org/10.1016/j.jfineco.2021.11.006\u003c/p\u003e\n\u003cp\u003eBlum, A. G., Ferraro, P. J., Archfield, S. A., \u0026amp; Ryberg, K. R. (2020). Causal Effect of Impervious Cover on Annual Flood Magnitude for the United States. Geophysical Research Letters, 47(5), e2019GL086480. https://doi.org/10.1029/2019GL086480\u003c/p\u003e\n\u003cp\u003eBoschetti, M., Nutini, F., Manfron, G., Brivio, P. A., \u0026amp; Nelson, A. (2014). Comparative Analysis of Normalised Difference Spectral Indices Derived from MODIS for Detecting Surface Water in Flooded Rice Cropping Systems. PLOS ONE, 9(2), e88741. https://doi.org/10.1371/journal.pone.0088741\u003c/p\u003e\n\u003cp\u003eBrandt, S. A., Lim, N. J., Colding, J., \u0026amp; Barthel, S. (2021). Mapping Flood Risk Uncertainty Zones in Support of Urban Resilience Planning. Urban Planning, 6(3), 258\u0026ndash;271. https://doi.org/10.17645/up.v6i3.4073\u003c/p\u003e\n\u003cp\u003eBreiman, L. (2001). Random Forests. Machine Learning, 45(1), 5\u0026ndash;32. https://doi.org/10.1023/A:1010933404324\u003c/p\u003e\n\u003cp\u003eBrelsford, C., Lobo, J., Hand, J., \u0026amp; Bettencourt, L. M. A. (2017). Heterogeneity and scale of sustainable development in cities. Proceedings of the National Academy of Sciences, 114(34), 8963\u0026ndash;8968. https://doi.org/10.1073/pnas.1606033114\u003c/p\u003e\n\u003cp\u003eC. Yang, J. H. Everitt, \u0026amp; J. M. Bradford. (2007). Using Multispectral Imagery and Linear Spectral Unmixing Techniques for Estimating Crop Yield Variability. Transactions of the ASABE, 50(2), 6676\u0026ndash;674. https://doi.org/10.13031/2013.22658\u003c/p\u003e\n\u003cp\u003eCalma, J. (2023, May 31). Scientists say they can\u0026rsquo;t rely on Twitter anymore. The Verge. https://www.theverge.com/2023/5/31/23739084/twitter-elon-musk-api-policy-chilling-academic-research\u003c/p\u003e\n\u003cp\u003eCDC. (2020). CDC/ATSDR Social Vulnerability Index [Database State]. https://www.atsdr.cdc.gov/placeandhealth/svi/data_documentation_download.html\u003c/p\u003e\n\u003cp\u003eCensus. (2021). City and Town Population Totals: 2020-2021. Census.Gov. https://www.census.gov/data/tables/time-series/demo/popest/2020s-total-cities-and-towns.html\u003c/p\u003e\n\u003cp\u003eClement, M. a., Kilsby, C. g., \u0026amp; Moore, P. (2018). Multi-temporal synthetic aperture radar flood mapping using change detection. Journal of Flood Risk Management, 11(2), 152\u0026ndash;168. https://doi.org/10.1111/jfr3.12303\u003c/p\u003e\n\u003cp\u003eCollins, E. L., Sanchez, G. M., Terando, A., Stillwell, C. C., Mitasova, H., Sebastian, A., \u0026amp; Meentemeyer, R. K. (2022). Predicting flood damage probability across the conterminous United States. Environmental Research Letters, 17(3), 034006. https://doi.org/10.1088/1748-9326/ac4f0f\u003c/p\u003e\n\u003cp\u003eCooley, S. W., Smith, L. C., Stepan, L., \u0026amp; Mascaro, J. (2017). Tracking Dynamic Northern Surface Water Changes with High-Frequency Planet CubeSat Imagery. Remote Sensing, 9(12), Article 12. https://doi.org/10.3390/rs9121306\u003c/p\u003e\n\u003cp\u003eCooper, K., Rizzo, E., \u0026amp; Schmidt, S. (2022, September 1). One year after Hurricane Ida, Pa. Residents are still paying the price. WHYY. https://whyy.org/articles/pensylvannia-hurricane-ida-one-year-anniversary/\u003c/p\u003e\n\u003cp\u003eCRED. (2015). The Human Cost of Natural Disasters: A global perspective. Centre for Research on the Epidemiology of Disasters (CRED). http://repo.floodalliance.net/jspui/44111/1165\u003c/p\u003e\n\u003cp\u003ede Bruijn, J. A., de Moel, H., Jongman, B., de Ruiter, M. C., Wagemaker, J., \u0026amp; Aerts, J. C. J. H. (2019). A global database of historic and real-time flood events based on social media. Scientific Data, 6(1), Article 1. https://doi.org/10.1038/s41597-019-0326-9\u003c/p\u003e\n\u003cp\u003eFEMA. (2020, July 8). Flood Zones. https://www.fema.gov/glossary/flood-zones\u003c/p\u003e\n\u003cp\u003eFEMA. (2023, October 23). Laws and Regulations. https://www.fema.gov/flood-insurance/rules-legislation/laws\u003c/p\u003e\n\u003cp\u003eFeyisa, G. L., Meilby, H., Fensholt, R., \u0026amp; Proud, S. R. (2014). Automated Water Extraction Index: A new technique for surface water mapping using Landsat imagery. Remote Sensing of Environment, 140, 23\u0026ndash;35. https://doi.org/10.1016/j.rse.2013.08.029\u003c/p\u003e\n\u003cp\u003eFielding, J. L. (2018). Flood risk and inequalities between ethnic groups in the floodplains of England and Wales. Disasters, 42(1), 101\u0026ndash;123. https://doi.org/10.1111/disa.12230\u003c/p\u003e\n\u003cp\u003eGao, B. (1996). NDWI\u0026mdash;A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sensing of Environment, 58(3), 257\u0026ndash;266. https://doi.org/10.1016/S0034-4257(96)00067-3\u003c/p\u003e\n\u003cp\u003eGarbutt, K., Ellul, C., \u0026amp; Fujiyama, T. (2015). Mapping social vulnerability to flood hazard in Norfolk, England. Environmental Hazards, 14(2), 156\u0026ndash;186. https://doi.org/10.1080/17477891.2015.1028018\u003c/p\u003e\n\u003cp\u003eGoffi, A., Stroppiana, D., Brivio, P. A., Bordogna, G., \u0026amp; Boschetti, M. (2020). Towards an automated approach to map flooded areas from Sentinel-2 MSI data and soft integration of water spectral features. International Journal of Applied Earth Observation and Geoinformation, 84, 101951. https://doi.org/10.1016/j.jag.2019.101951\u003c/p\u003e\n\u003cp\u003eG\u0026oacute;mez-Palacios, D., Torres, M. A., \u0026amp; Reinoso, E. (2017). Flood mapping through principal component analysis of multitemporal satellite imagery considering the alteration of water spectral properties due to turbidity conditions. Geomatics, Natural Hazards and Risk, 8(2), 607\u0026ndash;623. https://doi.org/10.1080/19475705.2016.1250115\u003c/p\u003e\n\u003cp\u003eHall, T. M., \u0026amp; Kossin, J. P. (2019). Hurricane stalling along the North American coast and implications for rainfall. Npj Climate and Atmospheric Science, 2(1), Article 1. https://doi.org/10.1038/s41612-019-0074-8\u003c/p\u003e\n\u003cp\u003eHermas, E., Gaber, A., \u0026amp; El Bastawesy, M. (2021). Application of remote sensing and GIS for assessing and proposing mitigation measures in flood-affected urban areas, Egypt. The Egyptian Journal of Remote Sensing and Space Science, 24(1), 119\u0026ndash;130. https://doi.org/10.1016/j.ejrs.2020.03.002\u003c/p\u003e\n\u003cp\u003eHolland, G., \u0026amp; Bruy\u0026egrave;re, C. L. (2014). Recent intense hurricane response to global climate change. Climate Dynamics, 42(3), 617\u0026ndash;627. https://doi.org/10.1007/s00382-013-1713-0\u003c/p\u003e\n\u003cp\u003eHondula, K. L., DeVries, B., Jones, C. N., \u0026amp; Palmer, M. A. (2021). Effects of Using High Resolution Satellite‐Based Inundation Time Series to Estimate Methane Fluxes From Forested Wetlands. Geophysical Research Letters, 48(6). https://doi.org/10.1029/2021GL092556\u003c/p\u003e\n\u003cp\u003eHosseiny, H., Crimmins, M., Smith, V. B., \u0026amp; Kremer, P. (2020). A Generalized Automated Framework for Urban Runoff Modeling and Its Application at a Citywide Landscape. Water, 12(2), 357. https://doi.org/10.3390/w12020357\u003c/p\u003e\n\u003cp\u003eHuete, A. (1997). A comparison of vegetation indices over a global set of TM images for EOS-MODIS. Remote Sensing of Environment, 59(3), 440\u0026ndash;451. https://doi.org/10.1016/S0034-4257(96)00112-5\u003c/p\u003e\n\u003cp\u003eIreland, G., Volpi, M., \u0026amp; Petropoulos, G. P. (2015). Examining the Capability of Supervised Machine Learning Classifiers in Extracting Flooded Areas from Landsat TM Imagery: A Case Study from a Mediterranean Flood. Remote Sensing, 7(3), Article 3. https://doi.org/10.3390/rs70303372\u003c/p\u003e\n\u003cp\u003eJones, J. W. (2019). Improved Automated Detection of Subpixel-Scale Inundation\u0026mdash;Revised Dynamic Surface Water Extent (DSWE) Partial Surface Water Tests. Remote Sensing, 11(4), Article 4. https://doi.org/10.3390/rs11040374\u003c/p\u003e\n\u003cp\u003eKawasaki, A., Kawamura, G., \u0026amp; Zin, W. W. (2020). A local level relationship between floods and poverty: A case in Myanmar. International Journal of Disaster Risk Reduction, 42, 101348. https://doi.org/10.1016/j.ijdrr.2019.101348\u003c/p\u003e\n\u003cp\u003eKnighton, J., Hondula, K., Sharkus, C., Guzman, C., \u0026amp; Elliott, R. (2021). Flood risk behaviors of United States riverine metropolitan areas are driven by local hydrology and shaped by race. Proceedings of the National Academy of Sciences, 118(13), e2016839118. https://doi.org/10.1073/pnas.2016839118\u003c/p\u003e\n\u003cp\u003eKnutson, T. R., Sirutis, J. J., Zhao, M., Tuleya, R. E., Bender, M., Vecchi, G. A., Villarini, G., \u0026amp; Chavas, D. (2015). Global Projections of Intense Tropical Cyclone Activity for the Late Twenty-First Century from Dynamical Downscaling of CMIP5/RCP4.5 Scenarios. Journal of Climate, 28(18), 7203\u0026ndash;7224. https://doi.org/10.1175/JCLI-D-15-0129.1\u003c/p\u003e\n\u003cp\u003eKossin, J. P. (2018). A global slowdown of tropical-cyclone translation speed. Nature, 558(7708), Article 7708. https://doi.org/10.1038/s41586-018-0158-3\u003c/p\u003e\n\u003cp\u003eKossin, J. P., Knapp, K. R., Vimont, D. J., Murnane, R. J., \u0026amp; Harper, B. A. (2007). A globally consistent reanalysis of hurricane variability and trends. Geophysical Research Letters, 34(4). https://doi.org/10.1029/2006GL028836\u003c/p\u003e\n\u003cp\u003eKriegler, F. J., Malila, W. A., Nalepka, R. F., \u0026amp; Richardson, W. (1969). Preprocessing Transformations and Their Effects on Multispectral Recognition (p. 97). https://ui.adsabs.harvard.edu/abs/1969rse..conf...97K\u003c/p\u003e\n\u003cp\u003eLin, N., Emanuel, K., Oppenheimer, M., \u0026amp; Vanmarcke, E. (2012). Physically based assessment of hurricane surge threat under climate change. Nature Climate Change, 2(6), Article 6. https://doi.org/10.1038/nclimate1389\u003c/p\u003e\n\u003cp\u003eLiu, L., Liu, Y., Wang, X., Yu, D., Liu, K., Huang, H., \u0026amp; Hu, G. (2015). Developing an effective 2-D urban flood inundation model for city emergency management based on cellular automata. Natural Hazards and Earth System Sciences, 15(3), 381\u0026ndash;391. https://doi.org/10.5194/nhess-15-381-2015\u003c/p\u003e\n\u003cp\u003eLorenz, M. O. (1905). Methods of Measuring the Concentration of Wealth. Publications of the American Statistical Association, 9(70), 209. https://doi.org/10.2307/2276207\u003c/p\u003e\n\u003cp\u003eMarkhvida, M., Walsh, B., Hallegatte, S., \u0026amp; Baker, J. (2020). Quantification of disaster impacts through household well-being losses. Nature Sustainability, 3(7), Article 7. https://doi.org/10.1038/s41893-020-0508-7\u003c/p\u003e\n\u003cp\u003eMason, D. C., Giustarini, L., Garcia-Pintado, J., \u0026amp; Cloke, H. L. (2014). Detection of flooded urban areas in high resolution Synthetic Aperture Radar images using double scattering. International Journal of Applied Earth Observation and Geoinformation, 28, 150\u0026ndash;159. https://doi.org/10.1016/j.jag.2013.12.002\u003c/p\u003e\n\u003cp\u003eMaxwell, A. E., Warner, T. A., \u0026amp; Fang, F. (2018). Implementation of machine-learning classification in remote sensing: An applied review. International Journal of Remote Sensing, 39(9), 2784\u0026ndash;2817. https://doi.org/10.1080/01431161.2018.1433343\u003c/p\u003e\n\u003cp\u003eMcFeeters, S. K. (1996). The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. International Journal of Remote Sensing, 17(7), 1425\u0026ndash;1432. https://doi.org/10.1080/01431169608948714\u003c/p\u003e\n\u003cp\u003eMtapuri, O., Dube, E., \u0026amp; Matunhu, J. (2018). Flooding and poverty: Two interrelated social problems impacting rural development in Tsholotsho district of Matabeleland North province in Zimbabwe. Jamba : Journal of Disaster Risk Studies, 10(1), 1\u0026ndash;7. https://doi.org/10.4102/jamba.v10i1.455\u003c/p\u003e\n\u003cp\u003eNOAA. (2021, September). National Weather Service Forecast Office. https://www.weather.gov/wrh/climate?wfo=phi\u003c/p\u003e\n\u003cp\u003eOlofsson, P., Foody, G. M., Herold, M., Stehman, S. V., Woodcock, C. E., \u0026amp; Wulder, M. A. (2014). Good practices for estimating area and assessing accuracy of land change. Remote Sensing of Environment, 148, 42\u0026ndash;57. https://doi.org/10.1016/j.rse.2014.02.015\u003c/p\u003e\n\u003cp\u003ePekel, J.-F., Cottam, A., Gorelick, N., \u0026amp; Belward, A. S. (2016). High-resolution mapping of global surface water and its long-term changes. Nature, 540(7633), 418\u0026ndash;422. https://doi.org/10.1038/nature20584\u003c/p\u003e\n\u003cp\u003ePerin, V., Tulbure, M. G., Gaines, M. D., Reba, M. L., \u0026amp; Yaeger, M. A. (2022). A multi-sensor satellite imagery approach to monitor on-farm reservoirs. Remote Sensing of Environment, 270, 112796. https://doi.org/10.1016/j.rse.2021.112796\u003c/p\u003e\n\u003cp\u003ePhan, T. N., Kuch, V., \u0026amp; Lehnert, L. W. (2020). Land Cover Classification using Google Earth Engine and Random Forest Classifier\u0026mdash;The Role of Image Composition. Remote Sensing, 12(15), Article 15. https://doi.org/10.3390/rs12152411\u003c/p\u003e\n\u003cp\u003ePickens, A. H., Hansen, M. C., Hancher, M., Stehman, S. V., Tyukavina, A., Potapov, P., Marroquin, B., \u0026amp; Sherani, Z. (2020). Mapping and sampling to characterize global inland water dynamics from 1999 to 2018 with full Landsat time-series. Remote Sensing of Environment, 243, 111792. https://doi.org/10.1016/j.rse.2020.111792\u003c/p\u003e\n\u003cp\u003ePinos, J., \u0026amp; Quesada-Rom\u0026aacute;n, A. (2022). Flood Risk-Related Research Trends in Latin America and the Caribbean. Water, 14(1), Article 1. https://doi.org/10.3390/w14010010\u003c/p\u003e\n\u003cp\u003ePulcinella, M., Meyer, K., \u0026amp; Cooper, K. (2021, September 1). Long recovery ahead as Ida\u0026rsquo;s remnants lead to historic flooding, tornadoes in Philly region. WHYY. https://whyy.org/articles/philly-says-to-shelter-in-place-as-schuylkill-river-expected-to-rise-to-major-flood-stage/\u003c/p\u003e\n\u003cp\u003eRentschler, J., Salhab, M., \u0026amp; Jafino, B. A. (2022). Flood exposure and poverty in 188 countries. Nature Communications, 13(1), Article 1. https://doi.org/10.1038/s41467-022-30727-4\u003c/p\u003e\n\u003cp\u003eRosser, J. F., Leibovici, D. G., \u0026amp; Jackson, M. J. (2017). Rapid flood inundation mapping using social media, remote sensing and topographic data. Natural Hazards, 87(1), 103\u0026ndash;120. https://doi.org/10.1007/s11069-017-2755-0\u003c/p\u003e\n\u003cp\u003eSanders, B. F., Schubert, J. E., Kahl, D. T., Mach, K. J., Brady, D., AghaKouchak, A., Forman, F., Matthew, R. A., Ulibarri, N., \u0026amp; Davis, S. J. (2022). Large and inequitable flood risks in Los Angeles, California. Nature Sustainability, 1\u0026ndash;11. https://doi.org/10.1038/s41893-022-00977-7\u003c/p\u003e\n\u003cp\u003eSchaffer-Smith, D., Myint, S. W., Muenich, R. L., Tong, D., \u0026amp; DeMeester, J. E. (2020). Repeated Hurricanes Reveal Risks and Opportunities for Social-Ecological Resilience to Flooding and Water Quality Problems. Environmental Science \u0026amp; Technology, 54(12), 7194\u0026ndash;7204. https://doi.org/10.1021/acs.est.9b07815\u003c/p\u003e\n\u003cp\u003eSchnebele, E., Cervone, G., \u0026amp; Waters, N. (2014). Road assessment after flood events using non-authoritative data. Natural Hazards and Earth System Sciences, 14(4), 1007\u0026ndash;1015. https://doi.org/10.5194/nhess-14-1007-2014\u003c/p\u003e\n\u003cp\u003eSettle, J. J., \u0026amp; Drake, N. A. (1993). Linear mixing and the estimation of ground cover proportions. International Journal of Remote Sensing, 14(6), 1159\u0026ndash;1177. https://doi.org/10.1080/01431169308904402\u003c/p\u003e\n\u003cp\u003eShen, L., \u0026amp; Li, C. (2010). Water body extraction from Landsat ETM+ imagery using adaboost algorithm. 2010 18th International Conference on Geoinformatics, 1\u0026ndash;4. https://doi.org/10.1109/GEOINFORMATICS.2010.5567762\u003c/p\u003e\n\u003cp\u003eSmith, A. B. (2023). U.S. Billion-dollar Weather and Climate Disasters, 1980\u0026mdash;Present (NCEI Accession 0209268) [dataset]. NOAA National Centers for Environmental Information. https://doi.org/10.25921/STKW-7W73\u003c/p\u003e\n\u003cp\u003eSofia, G., Roder, G., Dalla Fontana, G., \u0026amp; Tarolli, P. (2017). Flood dynamics in urbanised landscapes: 100 years of climate and humans\u0026rsquo; interaction. Scientific Reports, 7(1), Article 1. https://doi.org/10.1038/srep40527\u003c/p\u003e\n\u003cp\u003eSweet, W., Dusek, G. (Gregory P. ), Marcy, D. C., Greg (Gregory W.), C., \u0026amp; Marra, J. (2019). 2018 State of U.S. High Tide Flooding with a 2019 Outlook. https://doi.org/10.25923/RBV9-TH19\u003c/p\u003e\n\u003cp\u003eTanim, A. H., McRae, C. B., Tavakol-Davani, H., \u0026amp; Goharian, E. (2022). Flood Detection in Urban Areas Using Satellite Imagery and Machine Learning. Water, 14(7), Article 7. https://doi.org/10.3390/w14071140\u003c/p\u003e\n\u003cp\u003eTate, E., Rahman, M. A., Emrich, C. T., \u0026amp; Sampson, C. C. (2021). Flood exposure and social vulnerability in the United States. Natural Hazards, 106(1), 435\u0026ndash;457. https://doi.org/10.1007/s11069-020-04470-2\u003c/p\u003e\n\u003cp\u003eTiwari, V., Tulbure, M. G., Caineta, J., Gaines, M. D., Perin, V., Kamal, M., Krupnik, T. J., Aziz, M. A., \u0026amp; Islam, A. T. (2024). Automated in-season rice crop mapping using Sentinel time-series data and Google Earth Engine: A case study in climate-risk prone Bangladesh. Journal of Environmental Management, 351, 119615. https://doi.org/10.1016/j.jenvman.2023.119615\u003c/p\u003e\n\u003cp\u003eTrenberth, K. E., Cheng, L., Jacobs, P., Zhang, Y., \u0026amp; Fasullo, J. (2018). Hurricane Harvey Links to Ocean Heat Content and Climate Change Adaptation. Earth\u0026rsquo;s Future, 6(5), 730\u0026ndash;744. https://doi.org/10.1029/2018EF000825\u003c/p\u003e\n\u003cp\u003eTulbure, M. G., \u0026amp; Broich, M. (2013). Spatiotemporal dynamic of surface water bodies using Landsat time-series data from 1999 to 2011. ISPRS Journal of Photogrammetry and Remote Sensing, 79, 44\u0026ndash;52. https://doi.org/10.1016/j.isprsjprs.2013.01.010\u003c/p\u003e\n\u003cp\u003eTulbure, M. G., Broich, M., Ju, J., Masek, J. G., \u0026amp; Wearne, J. (2018). Quantifying surface water extent and flooding in a dynamic dryland river system using the Harmonized Landsat/Sentinel-2 Reflectance Product. 2018, H21E-08.\u003c/p\u003e\n\u003cp\u003eTulbure, M. G., Broich, M., Perin, V., Gaines, M., Ju, J., Stehman, S. V., Pavelsky, T., Masek, J. G., Yin, S., Mai, J., \u0026amp; Betbeder-Matibet, L. (2022). Can we detect more ephemeral floods with higher density harmonized Landsat Sentinel 2 data compared to Landsat 8 alone? ISPRS Journal of Photogrammetry and Remote Sensing, 185, 232\u0026ndash;246. https://doi.org/10.1016/j.isprsjprs.2022.01.021\u003c/p\u003e\n\u003cp\u003eTulbure, M. G., Broich, M., Stehman, S. V., \u0026amp; Kommareddy, A. (2016). Surface water extent dynamics from three decades of seasonally continuous Landsat time series at subcontinental scale in a semi-arid region. Remote Sensing of Environment, 178, 142\u0026ndash;157. https://doi.org/10.1016/j.rse.2016.02.034\u003c/p\u003e\n\u003cp\u003eU.S. Census Bureau. (2020). County-level Urban and Rural information for the 2020 Census [dataset]. https://www.census.gov/programs-surveys/geography/guidance/geo-areas/urban-rural.html\u003c/p\u003e\n\u003cp\u003eUS Department of Commerce. (2022, October 26). Hurricane Ida Emergency Response Imagery. https://oceanservice.noaa.gov/news/aug21/ngs-storm-imagery-ida.html\u003c/p\u003e\n\u003cp\u003eUS Department of Commerce, N. (2021). NWS Preliminary US Flood Fatality Statistics. NOAA\u0026rsquo;s National Weather Service. https://www.weather.gov/arx/usflood\u003c/p\u003e\n\u003cp\u003eU.S. Geological Survey. (2023). 3D Elevation Program 10-Meter Resolution Digital Elevation Model. [dataset]. https://www.usgs.gov/the-national-map-data-delivery\u003c/p\u003e\n\u003cp\u003eU.S. Government Accountability Office. (2021). FEMA Flood Maps: Better Planning and Analysis Needed to Address Current and \u0026nbsp;Future Flood Hazards. https://www.gao.gov/assets/gao-22-104079.pdf\u003c/p\u003e\n\u003cp\u003eVan Oldenborgh, G. J., Van Der Wiel, K., Sebastian, A., Singh, R., Arrighi, J., Otto, F., Haustein, K., Li, S., Vecchi, G., \u0026amp; Cullen, H. (2017). Attribution of extreme rainfall from Hurricane Harvey, August 2017. Environmental Research Letters, 12(12), 124009. https://doi.org/10.1088/1748-9326/aa9ef2\u003c/p\u003e\n\u003cp\u003eWing, O. E. J., Bates, P. D., Smith, A. M., Sampson, C. C., Johnson, K. A., Fargione, J., \u0026amp; Morefield, P. (2018). Estimates of present and future flood risk in the conterminous United States. Environmental Research Letters, 13(3), 034023. https://doi.org/10.1088/1748-9326/aaac65\u003c/p\u003e\n\u003cp\u003eWing, O. E. J., Lehman, W., Bates, P. D., Sampson, C. C., Quinn, N., Smith, A. M., Neal, J. C., Porter, J. R., \u0026amp; Kousky, C. (2022). Inequitable patterns of US flood risk in the Anthropocene. Nature Climate Change, 12(2), Article 2. https://doi.org/10.1038/s41558-021-01265-6\u003c/p\u003e\n\u003cp\u003eWinsemius, H. C., Jongman, B., Veldkamp, T. I. E., Hallegatte, S., Bangalore, M., \u0026amp; Ward, P. J. (2018). Disaster risk, climate change, and poverty: Assessing the global exposure of poor people to floods and droughts. Environment and Development Economics, 23(3), 328\u0026ndash;348. https://doi.org/10.1017/S1355770X17000444\u003c/p\u003e\n\u003cp\u003eWoznicki, S. A., Baynes, J., Panlasigui, S., Mehaffey, M., \u0026amp; Neale, A. (2019). Development of a spatially complete floodplain map of the conterminous United States using random forest. Science of The Total Environment, 647, 942\u0026ndash;953. https://doi.org/10.1016/j.scitotenv.2018.07.353\u003c/p\u003e\n\u003cp\u003eXu, H. (2006). Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery. International Journal of Remote Sensing, 27(14), 3025\u0026ndash;3033. https://doi.org/10.1080/01431160600589179\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"natural-hazards","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"nhaz","sideBox":"Learn more about [Natural Hazards](https://www.springer.com/journal/11069)","snPcode":"11069","submissionUrl":"https://submission.nature.com/new-submission/11069/3","title":"Natural Hazards","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Machine Learning, Emergency Management, Flooding, Surface Water, Hurricane Ida, Climate Change","lastPublishedDoi":"10.21203/rs.3.rs-3882712/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3882712/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe risk of floods from tropical storms is increasing due to climate change and human development. Maps of past flood extents can aid in planning and mitigation efforts to decrease flood risk. In 2021, Hurricane Ida slowed over the Mid-Atlantic and Northeast United States and released unprecedented rainfall. Satellite imagery and the Random Forest algorithm are a reliable combination to map flood extents. However, this combination is not usually applied to urban areas. We used Sentinel-2 imagery (10 m), along with derived indices, elevation, and land cover data, as inputs to a Random Forest model to make a new flood extent for southeastern Pennsylvania. The model was trained and validated with a dataset created with input from PlanetScope imagery (3 m) and social media posts related to the flood event. The overall accuracy of the model is 99%, and the flood class had a user\u0026rsquo;s and producer\u0026rsquo;s accuracy each over 99%. We then compared the flood extent to the Federal Emergency Management Agency (FEMA) flood zones at the county and tract level and found that more flooding occurred in the Minimal Hazard zone than in the 500-year flood zone. Our Random Forest model relies on publicly available data and software to efficiently and accurately make a flood extent map that can be deployed to other urban areas. Flood extent maps like the one developed here can help decision-makers focus efforts on recovery and resilience.\u003c/p\u003e","manuscriptTitle":"Quantifying Urban Flood Extent Using Satellite Imagery and Random Forest: A Case Study in Southeastern Pennsylvania","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-03-04 20:31:53","doi":"10.21203/rs.3.rs-3882712/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Major revisions","date":"2024-06-04T03:06:30+00:00","index":"","fulltext":""},{"type":"reviewerAgreed","content":"","date":"2024-02-27T05:47:42+00:00","index":0,"fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-02-19T23:25:21+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-01-23T12:11:32+00:00","index":"","fulltext":""},{"type":"submitted","content":"Natural Hazards","date":"2024-01-22T10:41:12+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"natural-hazards","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"nhaz","sideBox":"Learn more about [Natural Hazards](https://www.springer.com/journal/11069)","snPcode":"11069","submissionUrl":"https://submission.nature.com/new-submission/11069/3","title":"Natural Hazards","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"cfb91d84-2668-4c49-942f-71df0aada223","owner":[],"postedDate":"March 4th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2024-08-01T17:06:33+00:00","versionOfRecord":{"articleIdentity":"rs-3882712","link":"https://doi.org/10.1007/s11069-024-06817-5","journal":{"identity":"natural-hazards","isVorOnly":false,"title":"Natural Hazards"},"publishedOn":"2024-07-27 16:15:51","publishedOnDateReadable":"July 27th, 2024"},"versionCreatedAt":"2024-03-04 20:31:53","video":"","vorDoi":"10.1007/s11069-024-06817-5","vorDoiUrl":"https://doi.org/10.1007/s11069-024-06817-5","workflowStages":[]},"version":"v1","identity":"rs-3882712","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3882712","identity":"rs-3882712","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}