OpenFloodHub

OpenFloodHub, by Abhiram Mullapudi (of pystorms/open-storm lineage), is an open-source experiment in gauge-level flood forecasting: instead of one global model, it trains a compact 1D-CNN per river gauge on historical discharge, precipitation, temperature and soil moisture, and serves 12-hour-ahead streamflow predictions through a web interface — with a working deployment for Washington, DC gauges and pretrained checkpoints included.

It is a research prototype, and its own README warns against operational reliance. Its value is as scaffolding: a complete, inspectable, self-hostable forecasting stack (data fetching, training, serving) that a utility or research group can fork and point at their own gauges.