Artificial Intelligence-Assisted Segmentation of Flood Water from a Drone Imagery: A Use Case

As a proof of concept, this paper demonstrates Artificial Intelligence-based segmenta-tion and inundation of flood water on a drone imagery captured at Kachulu Trading Centre along Lake Chilwa, Malawi. Kachulu experienced lakeshore flooding due to heavy rains from Tropical Cyclone Freddy in March 2023. Leveraging recent advance-ments in Artificial Intelligence (AI) and a high spatial resolution drone imagery, flood water at Kachulu is detected, and its extent estimated using the segment-geospatial (samgeo), which is a Segment Anything Model (SAM) image encoder in Python. The results show that samgeo performed reasonably well in extracting about 84.1% (80,276 sq.m) of flood water from 95, 399 sq.m of flooded area in a 3 m spatial resolution im-agery. Rapid estimation of flood water extent is vital for damage assessment, disaster response and, more importantly, future disaster preparedness in climate change sensi-tive and vulnerable regions.