Authors: Dipsikha Devi – The University of Alabama
Title: Evaluating Flood Inundation Mapping Predictions Using Large-Scale Benchmark Datasets
Presentation Type: Poster
Abstract: Accurate Flood Inundation Mapping (FIM) is critical for assessing the extent of flooding and severity. Evaluating model-predicted FIM (M-FIM) against high-quality benchmark FIM (B-FIM) is imperative for the operational flood inundation forecasting model. The FIM evaluation can be conducted qualitatively (expert visual inspection) and quantitatively (statistical analysis of the using the confusion matrix). However, in diverse case studies, manual processing of the rasters can be repetitive and prone to human error. To address this, we developed a Flood Inundation Mapping Predictions Evaluation Framework (FIMPEF) for automating quantitative assessments of M-FIM against B-FIM. FIMPEF removes the Permanent Water Bodies from the evaluation domain using a user-input file and features automated extraction of flood extent using two built-in algorithms. It also supports the building hits evaluations using a user-input building footprint file or global datasets available via Earth Engine. High-quality B-FIM is essential for FIM evaluation. We generated high-quality binary benchmark datasets from (a) Planet’s satellite enhanced with a hydrologically guided region-grow algorithm (b) high-resolution aerial imagery of NOAA’s Emergency Response Imagery. Using FIMPEF, we automatically assessed the NOAA Office of Weather Prediction Height Above the Nearest Drainage (OWP-HAND) FIM against the B-FIMs at different locations demonstrating FIMPEF’s capability in diverse evaluations.