Authors: Dipsikha Devi – University of Alabama
Title: FIMbench: A Benchmark Database of Multi-Tier Flood Inundation Maps for Large-Scale Evaluation
Presentation Type: Poster Presentation
Abstract: Flood inundation maps (FIMs) are critical for model calibration and validation, flood risk assessment, and operational early warning dissemination. There is a lack of quality benchmark FIM with large temporal coverage and accuracy. We present FIMbench, a multi-tier benchmark database of FIM in a standardized format with a unified naming convention and quality metadata. The database includes five tiers based on the spatial resolution and quality: high-resolution NOAA’s Emergency Response Imagery (ERI) (Tier-1); PlanetScope Scene (Tier-2), Sentinel-1A (Tier-3), FEMA’s BLE 100 and 500-year flood (Tier-4), FIM generated from USGS high water marks (HWM).
The flood maps for Tier-1 were processed with both ML-based classification algorithms and hand-labelled digitization followed by an expert driven quality assurance through detailed GIS based visual inspection. For Tier-2 and Tier-3, the raw imageries were post-processed with a hydrologically guided gap-filling algorithm as optical and radar-based images are susceptible to gaps due to cloud cover and double bounced phenomenon respectively. Because gap-filling may introduce over-prediction biases, an additional quality controlled was implemented to retain only flooded areas where high-confidence flooded pixels outnumber low-confidence pixels. Tier-4 flood maps were directly post-processed and converted to binary maps consistent with their widespread use in model evaluation studies. The FIMbench database is also integrated with a python-based FIMeval framework enabling automated intercomparison of flood inundation maps across multiple data sources and spatial scales.