The suitability of differentiable, physics-informed machine learning hydrologic models for ungauged regions and climate change impact assessment

Feng, Dapeng; Beck, Hylke; Lawson, Kathryn; Shen, Chaopeng

doi:https://doi.org/10.5194/hess-27-2357-2023

Articles | Volume 27, issue 12

https://doi.org/10.5194/hess-27-2357-2023

© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/hess-27-2357-2023

© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 27, issue 12

Research article

|

30 Jun 2023

Research article |

| 30 Jun 2023

The suitability of differentiable, physics-informed machine learning hydrologic models for ungauged regions and climate change impact assessment

Dapeng Feng, Hylke Beck, Kathryn Lawson, and Chaopeng Shen

Data sets

A large-sample watershed-scale hydrometeorological dataset for the contiguous USA A. J. Newman, K. Sampson, M. P. Clark, A. Bock, R. J. Viger, and D. Blodgett https://doi.org/10.5065/D6MW2F4D

CAMELS Extended Maurer Forcing Data F. Kratzert https://doi.org/10.4211/hs.17c896843cf940339c3c3496d0c1c077

CAMELS benchmark models F. Kratzert https://doi.org/10.4211/hs.474ecc37e7db45baa425cdb4fc1b61e1

MOD16A2 MODIS/Terra Net Evapotranspiration 8-Day L4 Global 500\,m SIN Grid V006 S. Running, Q. Mu, and M. Zhao https://doi.org/10.5067/MODIS/MOD16A2.006

Model code and software

Differentiable hydrologic models: dPL+evolved HBV D. Feng, C. Shen, J. Liu, K. Lawson, and H. Beck https://doi.org/10.5281/zenodo.7091334

mhpi/hydroDL: MHPI-hydroDL (v2.0) K. Fang, C. Shen, and D. Feng https://doi.org/10.5281/zenodo.5015120

Short summary

Powerful hybrid models (called δ or delta models) embrace the fundamental learning capability of AI and can also explain the physical processes. Here we test their performance when applied to regions not in the training data. δ models rivaled the accuracy of state-of-the-art AI models under the data-dense scenario and even surpassed them for the data-sparse one. They generalize well due to the physical structure included. δ models could be ideal candidates for global hydrologic assessment.