143-4 Rapid Flood Monitoring Using Sentinel-2A Satellite Data and Machine Learning: Insights from the 2025 Kerr County Flooding Event, Texas

Session: A Showcase of Student Research in Geoinformatics and Data Science (Posters)

Presenting Author:

Authors:

Abstract:

Flood events represent a growing threat to human life, infrastructure, and agriculture worldwide. Devastating flash floods swept across Central Texas on July 4, 2025, severely impacting Kerr County and surrounding regions. According to the National Weather Service, water levels in the Guadalupe River at Kerrville rose precipitously, with some areas experiencing a 6 m increase in less than three hours, quickly reaching major flood stage. The catastrophic floods have tragically claimed more than 130 lives. Such swift flooding events highlight the critical need for timely flood monitoring and detection, underscoring the importance of this research.

 In response to this event, this study developed [TC1] [MS2] an automated flood detection methodology using Sentinel-2A satellite imagery combined with spectral indices including the Normalized Difference Water Index (NDWI), Modified Normalized Difference Water Index (MNDWI), Automated Water Extraction Index (AWEI), Enhanced Water Index (EWI), Modified Bare-soil Water Index (MBWI), Automated Normalized Difference Water Index (ANDWI), and Normalized Channel Index for Water Identification (NCIWI)  alongside machine learning classifiers such as Random Forest (RF), Gradient Boosting Machine (GBM), Support Vector Machine (SVM), and k-Nearest Neighbor (kNN). Implemented on the Google Earth Engine platform, the approach was designed to rapidly map and quantify flood extents, specifically highlighting affected urban and residential areas. The feature importance analysis further identified MBWI and AWEI as the most influential indices in detecting flooded regions. Accuracy assessment demonstrated that the RF and GBM classifiers achieved superior performance, with overall accuracy of 98.6% and 97.0%, respectively. Additionally, the total flooded area estimated by RF and GBM was approximately 5.31 and 4.88 km², respectively. These findings highlight the utility of integrating Sentinel-2A data and advanced machine learning techniques for timely, accurate flood mapping, thus providing essential insights for ongoing disaster response, recovery operations, and future flood-risk mitigation planning.

Geological Society of America Abstracts with Program. Vol. 57, No. 6, 2025

doi: 10.1130/abs/2025AM-8348

© Copyright 2025 The Geological Society of America (GSA), all rights reserved.