Duan et al. (2026) Spatiotemporal variation of extreme precipitation and its terrain modulation in the Hengduan Mountains: machine learning and interpretable analysis

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-08-22
• Authors: Qiyan Duan, Guokun Chen, Xingwu Duan, Qingke Wen, Haijuan Zhao, Fengyuya Jing, Zhiyuan Chen
• DOI: 10.1016/j.jhydrol.2026.136293

Research Groups

• Faculty of Land and Resources Engineering, Kunming University of Science and Technology
• Yunnan Key Laboratory of Quantitative Remote Sensing, Kunming University of Science and Technology
• Yunnan International Joint Laboratory for Integrated Sky-Ground Intelligent Monitoring of Mountain Hazards
• Institute of International Rivers and Eco-security, Yunnan University
• Yunnan Key Laboratory of Soil Erosion Prevention and Green Development, Yunnan University
• Aerospace Information Research Institute, Chinese Academy of Sciences
• National Engineering Research Center for Geomatics (NCG), Chinese Academy of Sciences

Short Summary

This study analyzes the spatiotemporal evolution of extreme precipitation in the Hengduan Mountains from 2005 to 2024 and utilizes interpretable machine learning to quantify the nonlinear influence of topographic factors on these events.

Objective

• To systematically examine the spatiotemporal variation, propagation characteristics, and spatial organization of extreme precipitation in the Hengduan Mountains.
• To investigate the nonlinear modulation effects of terrain (elevation, slope, curvature) on different intensities of extreme precipitation.

Study Configuration

• Spatial Scale: Regional (Hengduan Mountains, China).
• Temporal Scale: 2005–2024 (Daily resolution).

Methodology and Data

• Models used: XGBoost regression, SHAP (SHapley Additive exPlanations), CP-PD (Partial Dependence), Permutation Feature Importance (PFI), and ten ETCCDI indices.
• Data sources: High-resolution daily precipitation data.

Main Results

• Trends: Extreme precipitation exhibits gradual evolution with a significant intensity increase during the rainy season; indices such as PRCPTOT, Rx5day, and R95p show positive trends with slopes between 0.35 and 1.71.
• Spatial Distribution: A clear southeast-northwest contrast exists, with higher intensities concentrated in low latitudes and orographically uplifted areas.
• Propagation: Extreme precipitation events follow a stable south-to-north propagation pattern, aligning with background monsoon circulation and regional topography.
• Terrain Modulation: Elevation is the primary driver of extreme precipitation distribution, with SHAP scores reaching 0.70 for the R99p index. While slope and curvature influence moderate-intensity events, the predictive importance concentrates almost exclusively on elevation for the most extreme events (R99p).

Contributions

• Advances the understanding of the spatial organization and propagation paths of extreme precipitation in complex mountainous terrain.
• Demonstrates the utility of interpretable machine learning (XGBoost + SHAP/PFI) to decouple the nonlinear effects of various topographic factors on different levels of precipitation extremity.

Funding

Not specified in the provided text.

Citation

@article{Duan2026Spatiotemporal,
  author = {Duan, Qiyan and Chen, Guokun and Duan, Xingwu and Wen, Qingke and Zhao, Haijuan and Jing, Fengyuya and Chen, Zhiyuan},
  title = {Spatiotemporal variation of extreme precipitation and its terrain modulation in the Hengduan Mountains: machine learning and interpretable analysis},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136293},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136293}
}