Authors: Saide Zand – The University of Alabama
Title: GEE-FMF: A Google Earth Engine-Based Machine Learning Framework for Efficient Regional Flood Mapping
Presentation Type: Poster
Abstract: Accurate hindcast flood mapping is critical for disaster response, risk assessment, and resilience planning, particularly in data-scarce regions. Previous studies used traditional hydrodynamic models, which are computationally intensive, slow to deploy, and require time-consuming calibration. Others used machine learning-based models combined with remote sensing data, involving complex preprocessing steps to integrate multi-source geospatial datasets. We developed the Google Earth Engine-Based Flood Mapping Framework (GEE-FMF), an approach that leverages GEE and a Random Forest classifier to map flooded and non-flooded areas without requiring complex data preprocessing or local computing resources. GEE-FMF integrates diverse geospatial datasets, such as multispectral Landsat imagery, dual-polarized SAR data, coastal digital elevation models, and hindcast flood maps from the DFLOW-FM model. We employed a transfer learning approach by training and validating the model on a portion of the Galveston Bay region in Texas and testing it on the remainder of the region of interest impacted by Hurricane Harvey. We evaluated model performance using classification metrics: accuracy (0.92), precision (0.92), recall (0.66), F1-score (0.77), and Kappa coefficient (0.73). By leveraging GEE’s cloud infrastructure, GEE-FMF offers a scalable, efficient, and accessible solution for large-scale post-event flood mapping, supporting rapid disaster response and long-term risk management.