Multi-Sensor Deep Learning Fusion for Spatiotemporal Water Surface Monitoring in the Yasothon Province's Chi River Basin, Thailand

            Water security in the Chi River Basin is critical for the agricultural economy and ecosystem stability of Yasothon Province, Thailand. However, effective spatiotemporal monitoring of water surface dynamics is frequently hindered by persistent cloud cover during the monsoon season, limiting the utility of traditional optical remote sensing. This study addresses this challenge by developing a robust Multi-Sensor Deep Learning Fusion system that integrates Synthetic Aperture Radar (SAR) and optical satellite imagery to ensure continuous observation capabilities.

            We employ a U-Net convolutional neural network architecture, selected for its high boundary precision and efficiency with limited training datasets. The model is trained on a fused six-channel input configuration, combining Sentinel-1 SAR data (weather-independent) with Sentinel-2 optical bands (RGB), augmented by the Normalized Difference Water Index (NDWI) and Normalized Difference Vegetation Index (NDVI). This multi-modal approach enhances feature extraction, allowing for the accurate differentiation of open water from floating vegetation and flooded agricultural lands in complex transition zones.

            The study analyzes the hydrological cycle of 2022, capturing distinct drought, flood, and post-flood conditions. To ensure hydrological validity, the model’s segmentation outputs are not merely visually assessed but are quantitatively validated against ground-truth water level data from the E.20A gauge station in Kham Khuean Kaeo District. By establishing a precise Stage-Area Relationship, this research demonstrates a scalable, cost-effective framework for flood risk assessment and water capital estimation, offering a resilient solution for river basin management in cloud-prone tropical regions.