Examples¶

DDR ships with pre-trained weights and example notebooks for both supported geodatasets.

Directory Structure¶

examples/
├── merit/                      # MERIT-Hydro examples
│   ├── example_config.yaml     # Config pointing to v0.5.2 trained weights
│   ├── ddr-v0.5.2_merit_trained_weights.pt
│   └── plot_parameter_map.ipynb
├── lynker_hydrofabric/         # Lynker Hydrofabric v2.2 examples
│   ├── example_config.yaml     # Config pointing to v0.5.2 trained weights
│   ├── ddr-v0.5.2.lynker_hydrofabric_trained_weights.pt
│   └── plot_parameter_map.ipynb
├── eval/                       # Evaluation notebook (dataset-agnostic)
│   └── evaluate.ipynb
└── parameter_maps/             # Legacy v0.1.0a2 example (Lynker only)
    └── plot_parameter_map.ipynb

Parameter Map Notebooks¶

These notebooks visualize the spatial distribution of learned routing parameters (Manning's roughness, channel geometry) across CONUS.

Setup¶

1. Set DDR_DATA_DIR to your local data directory, or edit the example_config.yaml paths directly
2. Open the notebook for your dataset (examples/merit/ or examples/lynker_hydrofabric/)
3. Run all cells

Each example_config.yaml uses ${oc.env:DDR_DATA_DIR,./../../data} so paths resolve relative to the repo root's data/ folder by default.

What the Notebooks Show¶

1. Load config and trained weights — The v0.5.2 checkpoints use a 10-attribute KAN with hidden_size=21, grid=50, k=2
2. Predict spatial parameters — Run the KAN in eval mode to produce per-catchment parameter predictions
3. Map parameters — Plot Manning's n, q_spatial, and other learned parameters on the CONUS river network using GeoPandas and contextily basemaps

MERIT vs Lynker Differences¶

Model Evaluation¶

The examples/eval/evaluate.ipynb notebook demonstrates how to compare routed predictions against observations and the summed Q' baseline.