Flood Detection using Google Earth Engine: A Comparison of Optical and SAR-based Remote Sensing Methods

Floods are one of the most frequent and damaging disasters in India, particularly during the monsoon season. The urban settlement of Mahad, which lies along the Savitri River in Raigad district, Maharashtra, is highly susceptible to flooding. This requires reliable and scalable flood detection methods for disaster preparedness which current methods struggle with concerning issues such as false positives and cloud interference. We compare and contrast five Satellite based flood detection methods implemented in Google Earth Engine (GEE) - hybrid models, NDWI classification, and SAR ratio thresholding. We focus on accuracy, validation strategy, automation, scalability, and type of data (optical/SAR) as our evaluation criteria. The findings shows that NDWI approach are suits for water detection with high precision, but also under-perform in the presence of cloud cover, whereas SAR-based techniques are impervious to cloud cover, but prone to false positives. The effectiveness of ground validation across studies was only limited to one study, which was robust. In relation to these findings, we present a hybrid flood identification approach which incorporates visual validation using FCCs and thresholding with Sentinel-1 SAR VV ratio, as well as NDWI with Sentinel-2. The proposed methodology will enable accurate, cloud-independent and scalable flood mapping in areas affected by monsoon, such as Mahad.