Ouyang et al. (2026) A python framework for differentiable hydrological modeling and research workflow automation

Identification

• Journal: Environmental Modelling & Software
• Year: 2026
• Date: 2026-01-24
• Authors: Wenyu Ouyang, Shuolong Xu, Yikai Chai, Laihong Zhuang, Zhihong Liu, Lei Ye, Xinzhuo Wu, Yong Peng, Chi Zhang
• DOI: 10.1016/j.envsoft.2026.106895

Research Groups

• School of Infrastructure Engineering, Dalian University of Technology, Dalian, China
• School of Computer Science and Technology, Dalian University of Technology, Dalian, China
• Ningbo Institute of Dalian University of Technology, Ningbo, China
• National Institute of Excellence Engineers, Zhejiang University, Hangzhou, China

Short Summary

This study introduces a Python-based modular framework for constructing differentiable hydrological models and automating research workflows. It demonstrates that differentiable models built with this framework achieve comparable streamflow simulation performance to traditional approaches in case studies.

Objective

• To introduce a Python-based modular framework for constructing differentiable hydrological models and automating research workflows, aiming to enhance reproducibility, scalability, and adaptability in hydrological modeling.

Study Configuration

• Spatial Scale: CAMELS (Catchment Attributes and Meteorology for Large-sample Studies) basins.
• Temporal Scale: Not explicitly stated in the provided text.

Methodology and Data

• Models used: Differentiable hydrological models (integrating neural networks via torchhydro), traditional process-based hydrological models (hydromodel).
• Data sources: Diverse datasets, including open-access and proprietary resources, observation data, processed by hydrodataset and hydrodatasource modules.

Main Results

• A Python-based modular framework was developed, integrating five key modules: hydrodataset and hydrodatasource for data preprocessing, hydromodel for traditional modeling, torchhydro for differentiable modeling, and HydroDHM for workflow orchestration.
• Differentiable models constructed using the framework achieved streamflow simulation performance comparable to traditional approaches in case studies conducted in CAMELS basins.
• The framework ensures reproducibility, scalability, and adaptability across diverse hydrological contexts by decoupling data handling from model development and providing uv-installable (and pip-compatible) modules.

Contributions

• Introduction of a novel, modular Python framework for building differentiable hydrological models and automating end-to-end research workflows.
• Integration of data preprocessing, traditional modeling, differentiable modeling, and workflow orchestration into a unified, installable system.
• Demonstration of comparable performance of differentiable models to traditional models, while offering enhanced reproducibility, scalability, and adaptability.
• Provides a flexible platform to facilitate the development and comparison of traditional, deep learning, and hybrid hydrological models.

Funding

• Not specified in the provided text.

Citation

@article{Ouyang2026python,
  author = {Ouyang, Wenyu and Xu, Shuolong and Chai, Yikai and Zhuang, Laihong and Liu, Zhihong and Ye, Lei and Wu, Xinzhuo and Peng, Yong and Zhang, Chi},
  title = {A python framework for differentiable hydrological modeling and research workflow automation},
  journal = {Environmental Modelling & Software},
  year = {2026},
  doi = {10.1016/j.envsoft.2026.106895},
  url = {https://doi.org/10.1016/j.envsoft.2026.106895}
}