•  
  •  
 

Alt Text Acknowledgement

1

Final Abstract

Floods are the most common natural disaster globally, causing thousands of deaths every year. Timely and accurate flood forecasts are crucial for developing early warning systems, evacuation plans, and effective strategies to enhance community resilience. Soil moisture is an important parameter in flood forecasting. Unfortunately, existing soil moisture sensor networks and optical remote sensing–based techniques for soil moisture estimation lack the spatial resolution needed for accurate flood forecasting. This study presents an approach for flood prediction using Sentinel-1 C-band synthetic aperture radar (SAR) data. SAR offers high spatial resolution and operates day and night under all weather conditions, making it advantageous for flood modeling. Furthermore, Sentinel-1’ s six-day revisit cycle allows for relatively continuous soil moisture monitoring, which is essential for flash flood modeling. This study uses Hurricane Helene, which impacted the southeastern United States between September 24 and September 27, 2024, as a case study. Flood analyses are performed in parts of North Carolina, Tennessee, and Georgia. Ten USGS streamgages were selected, and flood extents around each streamgage are mapped using stream stage data from the USGS National Water Information System (NWIS) mapper. For each streamgage, Sentinel-1 images ranging from two weeks before to two weeks after the storm were gathered using the Alaska Satellite Facility’s Vertex Data Search. A support vector machine (SVM) classifier is trained to relate SAR backscatter parameters to flooding. The model accuracy was improved by also including precipitation rate, land use/land cover, and elevation data as inputs. This approach is promising for improving flood prediction accuracy and spatial resolution, especially in areas with limited ground-based soil moisture sensor networks.

Download

Share

COinS