Forecasting Inundation Extent using REOF Analysis (FIER) for VIIRS Water Fraction Forecasting using Bias-corrected National Water Model Data

This study introduces a remote-sensing-based flood forecasting method, Forecasting Inundation Extent using Rotated Empirical Orthogonal Function (FIER), which leverages historical inundation extents extracted from satellite imageries to obtain flood-related spatial and temporal patterns recognized by correlation analysis with observed hydrological records such as streamflow or water level. Afterward, the satellite-driven temporal patterns and historical streamflow are coupled through a deep learning network to generate a regression model. This regression model is associated with forecasting hydrological data to synthesize the satellite imagery and forecast the flood inundation extent. We demonstrate the performance of FIER in water fraction map forecasting over two flood-prone regions, the Upper Mississippi Alluvial Plain and the Red River of the North Basin, using the VIIRS-derived water fraction product alongside the National Water Model (NWM) medium-range forecasting dataset. To address the common issue of underestimation or overestimation in flood inundation maps, we apply a quantile scale mapping bias correction method to the NWM streamflow forecasts. Additionally, a web application tool has been developed to provide users with an accessible online service for forecasting flood inundation extents.