ML/AI Based Flood Mapping in Swat WatershedUsing Sentinel-I and Sentinel-II Data

This research uses Sentinel-1 and Sentinel-2 data for flood monitoring and mapping, with a focus on the accuracy and reliability of these remote sensing techniques in identifying flood inundation areas. The objectives of this study revolved around the accuracy and reliability of these techniques in detecting and mapping floodwaters. Water indices, namely NDWI and WRI, were utilized to extract floodwater areas and generate flood inundation maps. Additionally, flood extent maps were generated using Sentinel-1 data to complement the findings from Sentinel-2 data. The study implemented a multi-sensor and multi-index approach, considering both optical and radar data, to provide a comprehensive analysis of flood events. Image selection based on low cloud cover was employed to ensure high-quality and cloud-free imagery for accurate flood extent estimation. The selected images were processed using water indices, NDWI and WRI, which effectively captured the spatial distribution of floodwaters. The results revealed insights into the temporal variation and spatial distribution of flood extents, allowing for the identification of most affected areas. The analysis of Sentinel-2 imagery for July 2022 showcased a progressive intensification of the flood event, with the most affected regions being Charbagh, Mangora, Saidu Sharif, and Chakdara. The flood extents increased in August 2022, affecting areas such as Mangora, Saidu Sharif, Charbagh, Manglor, Barikot, and Chakdara. Furthermore, the flood extent in September 2022 indicated the persistence of floodwaters in areas with relatively fewer sloping surfaces. The integration of Sentinel-1 data provided enhanced comprehension into flood extents, particularly in challenging conditions such as high cloud cover or dense vegetation. The flood inundation maps generated from Sentinel-1 data complemented the findings from Sentinel-2 data, enhancing the accuracy and reliability of flood extent assessments. It is important to note that the high areas observed in the Sentinel-1 flood inundation maps are due to the mosaic of all the images acquired during the respective months. This approach includes all the water detected by Sentinel-1 from the 15 images, resulting in a larger affected area being shown. The flood inundation areas derived from Sentinel-1 data for July, August, and September were 129 km², 431 km², and 66 km², respectively. The analysis of Sentinel-1 data reveals that Kalam, Bahrain, and Madyan are highly vulnerable to intense flooding, as indicated by the high flood levels observed in these regions. The steep terrain, narrow valleys, and high rainfall intensity contribute to the heightened flood risk in these areas. The flood extents in Mangora, Saidu Sharif, and Barikot also reached significant levels, indicating widespread inundation in these regions. Overall, the study demonstrated the effectiveness of Sentinel-1 and Sentinel-2 data in flood monitoring and mapping. The multi-sensor and multi-index approach enhanced the reliability and robustness of the flood extent assessments, enabling better-informed decision-making processes for emergency response planning, resource allocation, and the implementation of effective flood mitigation strategies. The findings highlighted the importance of considering multiple indices and satellite data sources to obtain a comprehensive understanding of flood dynamics, while acknowledging the influence of cloud cover and other factors on the accuracy of the results.