A seamless global streamflow dataset for ~2.94 million rivers from 1980-2020.¶
Paper¶
Ji, H., Song, Y., Bindas, T. et al. Distinct hydrologic response patterns and trends worldwide revealed by physics-embedded learning. Nat Commun 16, 9169 (2025). https://doi.org/10.1038/s41467-025-64367-1
Results¶
| NSE | KGE | Bias | RMSE | FLV | FHV |
|---|---|---|---|---|---|
| 0.515 | 0.542 | -0.023 | 5.995 | 22.791 | -13.333 |
Data description¶
The global-scale seamless simulation from 1980-2020 zenodo is produced by a coupled High-resolution, multiscale, differentiable global water model that incorporates both a rainfall-runoff module (𝛿HBV2.0) and channel routing module (δMC2.0). To develop this high-resolution global dataset, we compiled 6,165 streamflow stations worldwide to train our model for the period 1980–2000 and conducted temporal validation from 2001 to 2015. The model achieved a median NSE of 0.721 and a median KGE of 0.725 during validation. Using the trained model, we then produced a seamless global simulation spanning 1980–2020. also, if you are intereseted in the near-real-time simulation, please let us know.
Code Release¶
The dHBV2.0UH code is available at mhpi/generic_deltaModel: High-resolution differentiable model, 𝛿HBV2.0. https://doi.org/10.5281/zenodo.14827983