Mapping Floodwater Using High-Resolution Satellite Imagery and Machine Learning: Insights from the STAC Overflow Challenge (Short Essay for Writing Demonstration)

EasyChair Preprint 10749

Abstract

Floods, being the most frequently occurring and economically detrimental natural disasters on a global scale, require precise monitoring in order to facilitate efficient response and risk evaluation. This research paper presents a comprehensive analysis of the STAC Overflow: Map Floodwater from Radar Imagery competition, which is a worldwide endeavor that seeks to enhance flood mapping by utilizing machine learning techniques on high-resolution synthetic-aperture radar (SAR) imagery. The utilization of the Sentinel-1 mission's C-band Synthetic Aperture Radar (SAR) data was employed in this challenge, taking advantage of its ability to provide imaging capabilities unaffected by weather conditions and operational during both day and night. The dataset used in the competition was curated by Cloud to Street and Microsoft AI for Earth. It comprised satellite photos from the years 2016 to 2020, enabling the creation and assessment of flood mapping algorithms. The competition received contributions from over 660 individuals globally, resulting in a total of more than 1,400 entries.

Keyphrases: Floodwater Detection, Jaccard Index Evaluation, Sentinel-1 Imagery, Synthetic Aperture Radar (SAR), machine learning

@booklet{EasyChair:10749,
  author    = {Tashin Ahmed},
  title     = {Mapping Floodwater Using High-Resolution Satellite Imagery and Machine Learning: Insights from the STAC Overflow Challenge (Short Essay for Writing Demonstration)},
  howpublished = {EasyChair Preprint 10749},
  year      = {EasyChair, 2023}}