HYDRA
Hybrid Deep-learning for Residual Analysis
A transformer-based machine learning system for correcting errors in National Water Model streamflow predictions across the Appalachian region.
Model Pipeline
Stage 1
NWM Forecast
Hourly short-range discharge predictions
Stage 2
Residual Calculator
Compute model error against observations
Stage 3
HYDRA Transformer
Temporal attention with hydrologic constraints
Stage 4
Corrected Streamflow
Bias-corrected hydrograph at target sites
LegendInputs and intermediate residuals are transformed into a site-specific corrected discharge signal while preserving physical plausibility.
Comparative Analysis
Multi-Site Evaluation
Compare model performance across 4 USGS gauging stations in the New River and Watauga watersheds.
Ablation Studies
Experiment Grid
Evaluate causal masking, physics constraints, and architecture variants using consistent metrics.
Interactive Insights
Chart-Driven Diagnostics
Inspect hydrographs, site-level improvements, and residual error distributions for each configuration.
4
Study Sites
21%
Avg RMSE Reduction
6
Experiments
2010-2020
Study Period