Abstract

Forecasting sandbank (চর) migration in braided river systems is notoriously difficult due to non-linear sediment transport, seasonal monsoon surges, and complex channel bifurcation dynamics.

We construct a Spatio-Temporal Graph Neural Network (ST-GNN) that treats sandbar landmasses as dynamic network nodes and connecting flow channels as directed, weighted edges. The model achieves a 31% improvement over traditional 2D hydrodynamic numerical solvers while reducing compute time from 18 hours to 4.2 seconds.

@article{logicdock2024deltaalluvial,
  title={Topological Neural Modeling for Alluvial Delta Morphology},
  author={Logicdock Studio},
  journal={Logicdock Working Papers},
  year={2024}
}