Reproduce “A physics-guided neural network for flooding area detection using SAR imagery and local river gauge observations” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Monika Gierszewska, Tomasz Berezowski

A physics-guided neural network, FPGNN, maps water extent using Sentinel-1 SAR imagery and river gauge data without relying on labeled training examples. The model integrates physical principles into its loss function, demonstrating robust performance across diverse hydrological conditions and often surpassing traditional unsupervised methods in accuracy.

Key insights

  • Embedding the physical principle of a monotonic relationship between river water elevation and flooded area directly into a neural network's loss function enables robust unsupervised water extent mapping.
  • The use of convolutional layers in the FPGNN allows for effective extraction of contextual features, which significantly improves discrimination between water bodies and spectrally similar land cover, such as smooth agricultural fields.
  • This physics-guided approach maintains high detection accuracy across a wide spectrum of hydrological conditions, including challenging low water levels, where traditional unsupervised methods often exhibit reduced performance.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.