Hydra Temporal
Improving National Streamflow Forecasts (Honors Thesis)
Problem
National water models often struggle with local precision, leaving communities with uncertain flood warnings.
Why it matters
Achieved 26–54% RMSE reduction over baselines, directly improving decision support for water management.
Approach
- Developed advanced Transformer-based models to catch residuals that physical models miss.
- Integrated multi-source data (NWM v3.0+, ERA5, USGS) for a comprehensive view.
- Designed leakage-safe evaluation to ensure performance holds up in the future.
- Preparing manuscript for submission to Water Resources Research (AGU).
Stack
PyTorchTransformerHydraMLflowxarrayAWS Batch
Results
- 26–54% reduction in prediction error (RMSE) on held-out basins.
- Consistent improvement across varying forecast horizons.
- Demonstrated stability under seasonal distribution shifts.
What I learned
- Hybrid architectures can correct physical model biases without overfitting.
- Rigorous evaluation design is the difference between a demo and a product.