Basin-wide flood depth and exposure mapping from SAR images and machine learning models

Area of Science:

• Earth and Environmental Sciences
• Remote Sensing
• Hydrology

Background:

• Increasing frequency of high-intensity rainfall events in the Indian subcontinent leads to severe urban and riverine flash floods.
• Effective flood mitigation and disaster risk management require rapid assessments of flooded areas.
• The August 2019 southern India floods caused significant loss of life and displacement, highlighting the need for advanced mapping techniques.

Purpose of the Study:

• To develop and validate an integrated methodology for rapid flood extent and depth mapping using Synthetic Aperture Radar (SAR) images and a digital elevation model (DEM).
• To assess the reliability and accuracy of the proposed methodology for near real-time flood risk assessment.

Main Methods:

• Utilized Sentinel-1 SAR images (pre- and during flood) and the MERIT DEM for mapping flood extent and depth.
• Employed a random forest model with DEM-derived variables to map flood susceptibility in areas with limited SAR coverage.
• Generated flood extent and depth maps for inundation zones.

Main Results:

• The integrated methodology achieved high reliability, with accuracy values of 90% for training data and 86% for validation data.
• Successfully mapped flood extent and depth for riverine flooding in southern India.
• Demonstrated the capability to generate flood maps even with limited SAR scene coverage.

Conclusions:

• The proposed approach provides a precise, simple, and fast method for flood mapping.
• The methodology is highly reliable and can be applied to other flood-prone areas, especially those with incomplete historical flood data.
• This technique supports effective disaster risk reduction and policy-making for future flood events.