Estimation of Flood Inundation Area Using Soil Moisture Active Passive Fractional Water Data with an LSTM Model

Area of Science:

• Earth and Environmental Sciences
• Remote Sensing
• Artificial Intelligence

Background:

• Accurate flood monitoring and forecasting are crucial for disaster preparedness and mitigation.
• Satellite observations combined with deep learning offer advanced capabilities for detecting flood patterns.

Purpose of the Study:

• To develop and validate a deep learning model for estimating flood inundation areas using satellite data.
• To assess the efficacy of the Soil Moisture Active Passive (SMAP) fractional water (FW) as a reference in flood modeling.

Main Methods:

• A long short-term memory (LSTM) model was employed, integrating soil moisture, rainfall forecasts, and topography.
• Datasets were resampled to a 30 m spatial resolution using bicubic interpolation for flood modeling.
• The LSTM model's flood inundation estimates were validated against Sentinel-1 SAR images and confusion matrix metrics.

Main Results:

• The LSTM model achieved a high accuracy estimation of SMAP FW, with an average Area Under the Curve (AUC) of 0.93.
• Flood inundation area validation yielded a high-performance accuracy of approximately 0.9.
• Optimal performance was observed in regions with low vegetation cover, seasonal water variations, and flat topography.

Conclusions:

• The proposed framework demonstrates methodological promise for enhanced flood disaster preparedness and resilience.
• The study highlights the effectiveness of deep learning models, specifically LSTM, in utilizing satellite data for flood estimation.
• SMAP FW proved to be a valuable reference for accurate flood area estimation in diverse topographical conditions.