This code is for an adjoint-based differentiable model to enable implicit ODE solutions for large-scale hydrological modeling, reducing the distortion of fluxes and physical parameters caused by numerical errors in previous models that used explicit and operation-splitting schemes. Please follow the provided examples to use this code. If you find this code useful for your research, please cite the papers listed below.
Song, Yalan, Wouter JM Knoben, Martyn P. Clark, Dapeng Feng, Kathryn E. Lawson, and Chaopeng Shen. "When ancient numerical demons meet physics-informed machine learning: adjoint-based gradients for implicit differentiable modeling." Hydrology and Earth System Sciences Discussions 2023 (2023): 1-35.
Feng, D., Liu, J., Lawson, K., & Shen, C. (2022). Differentiable, learnable, regionalized process-based models with multiphysical outputs can approach state-of-the-art hydrologic prediction accuracy. Water Resources Research, 58, e2022WR032404. https://doi.org/10.1029/2022WR032404