The first continuous floodplain depth maps for the United States

We are developing the first ever spatially continuous floodplain inundation depth maps of the continental United States (CONUS) at 10-m resolution. With the 10-m National Elevation Dataset digital elevation model (DEM) and the 30-m U.S. Environmental Protection Agency EnviroAtlas floodplain extent dataset as inputs, we applied a topography-based geostatistical approach to calculate maximum potential inundation depth in the floodplains. Our input floodplain extent dataset was derived through the Machine Learning (ML) of DEM-based hydrogeomorphic variables, soil characteristics, and land cover data. Specifically, the ML model used the existing FEMA 100-year floodplain maps as the training dataset in a Random Forest algorithm, and subsequently classifying the landscape as floodplain or non-floodplain. The geostatistical approach of our depth analysis estimated water depth for any given floodplain extent by first identifying the elevations at the edge of the extent boundaries from the input DEM, assigning floodwater surface elevation by calculating the minimum statistic in the neighborhood around each cell, and finally calculating the floodplain inundation depth by subtracting floodwater surface elevation from the input DEM. Correspondingly, the output inundation depth maps from this analysis essentially refers to potential maximum inundation depth and not the actual inundation depth in floodplains. The U.S. floodplain inundation depth maps derived from this work will provide floodplain managers a ready-made dataset to calibrate and validate their flood models, assist climate resilience studies by offering critical information for depth-to-damage analysis, and above all let policy makers re-envision traditional flood insurance and buyout programs.