Advancing river bathymetry mapping through physics-informed neural networks and SWOT satellite observations

Floods represent 45% of all natural disasters, affecting 2 billion people annually and causing over $40 billion in losses, but there are knowledge gaps affecting the accuracy of global flood models.

The challenge

Globally, rivers discharge ~37000km3 of water annually to oceans, representing two-fifths of terrestrial precipitation. Yet, river bathymetry—the submerged riverbed topography invisible to remote sensing and prohibitively expensive to survey—critically constrains hydrological modelling and flood simulation.

The absence of riverbed topography forces researchers to rely on oversimplified channel geometries and flow routing methods to represent complex river flow dynamics.

Consequently, even state-of-the-art global flood models exhibit significant discrepancies (60-70%) in hazard magnitude and spatial patterns, a costly uncertainty when on average floods cause over $40 billion in annual losses and water scarcity threatens ~2 billion people.  

What we're doing

This project addresses the critical river bathymetry data gap that fundamentally constrains flood modelling accuracy.

We will exploit the Surface Water and Ocean Topography (SWOT) satellite mission, which delivers unprecedented two-dimensional water surface elevation measurements at 10-60m resolution with 21-day repeat cycles, enabling global monitoring of rivers wider than ~50m.

We will develop a physics-informed machine learning framework that infers riverbed elevation from multi-temporal SWOT observations, enforcing hydraulic consistency with observed water surface profiles. Model validation employs UK Environment Agency bathymetric surveys across the rural River Severn and urban River Thames before global implementation. 

How it helps

This project advances river bathymetry reconstruction from costly, spatially limited field surveys toward continuous, global satellite-based monitoring.

Enhanced bathymetric accuracy reduces flood model uncertainties, mitigating losses through improved hazard modelling.

The methodology advances SWOT discharge estimation while enabling continental-scale flood monitoring for operational water resources management. 

Investigators

  • Youtong Rong, School of Geographical Sciences 
  • Dr Yanchen Zheng, School of Civil, Aerospace and Design Engineering 
  • Dr Jiangchao Qiu, Massachusetts Institute of Technology 
  • Prof Jeff Neal, School of Geographical Sciences 
  • Prof Paul Bates, School of Geographical Sciences