Mapping Floodwater from Radar Imagery

EasyChair Preprint 8587, version 2

Abstract

Flooding is aggravating every year as global warming exacerbating causing sea level rising. Recent flood in Bangladesh in 2022 broke all previous records of flooding in that area. Simultaneously, it submerged areas that have seen no flood in the last 100 years. As the situation is getting much worse year to year and sudden flooding occurring regularly throughout the world, it is becoming a necessity to forecast flood warning so that casualties can be lessened.Sentinel-1 mission has been launched to collect earths surface data and Microsoft’s new Planetary Computer made the data available for researchers. This short paper is based on the outcomes of a competition - ”STAC Overflow: Map Floodwater from Radar Imagery”, hosted by DrivenData and Microsoft AI for Earth, where the goal was to detect flood coverage areas in near realtime. Here, participants generated predictions as single-band 512×512 pixel images whether there is water or no water in every pixel. Jaccard index has been used as the performance metric and top performing models have achieved over 0.80.

Keyphrases: floodwater, image segmentation, satellite imagery

@booklet{EasyChair:8587,
  author    = {Tashin Ahmed},
  title     = {Mapping Floodwater from Radar Imagery},
  howpublished = {EasyChair Preprint 8587},
  year      = {EasyChair, 2022}}