Wang et al. (2025) Investigating the meteorological causes of hydrological drought through the integration of spatiotemporal cubes and interpretable machine learning: A case study of the Yangtze River Basin

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2025
• Date: 2025-09-25
• Authors: Hongxiang Wang, Sheng-tong Cheng, Lintong Huang, Wenxian Guo
• DOI: 10.1016/j.ejrh.2025.102796

Research Groups

School of Water Conservancy, North China University of Water Resources and Electric Power, Zhengzhou, China

Short Summary

This study investigates the meteorological causes of hydrological drought in the Yangtze River Basin (1980-2019) using spatiotemporal cubes and interpretable machine learning (XGBoost-SHAP, CNN-SHAP). It found that climate change is the dominant driver in the upper reaches (85% contribution), while human activities have a larger influence in the downstream (climate change 55% contribution), with specific meteorological factors driving drought in each region.

Objective

• To evaluate the impact of meteorological factors (temperature, precipitation, potential evapotranspiration) on hydrological drought in the Yangtze River Basin from 1980 to 2019.
• To thoroughly analyze the spatiotemporal characteristics of meteorological factors and their effects on hydrological drought using spatiotemporal cubes and SHAP models.
• To systematically explain the formation mechanisms and quantitatively analyze the driving factors (climate change and human activities) of hydrological drought in the Yangtze River Basin.

Study Configuration

• Spatial Scale: Yangtze River Basin, China, covering approximately 18,000 square kilometers across 11 provinces and municipalities. The basin was divided into upper/middle reaches and lower reaches for regional analysis.
• Temporal Scale: 1980 to 2019 (40 years), with daily data aggregated to monthly scale for analysis.

Methodology and Data

• Models used:
  • Spatiotemporal cube (for clustering and hotspot analysis of meteorological factors).
  • Machine Learning models for natural runoff reconstruction: Support Vector Machine (SVM), Random Forest (RF), Long Short-Term Memory (LSTM), XGBoost, and Convolutional Neural Network (CNN). (XGBoost was optimal for upstream, CNN for downstream).
  • Interpretable Machine Learning: SHAP (Shapley Additive exPlanations) model, integrated with XGBoost and CNN.
  • Hydrological drought index: Standardized Runoff Index (SRI) based on gamma distribution.
  • Statistical methods: K-nearest neighbors, K-means clustering, Getis-Ord Gi statistic, Mann-Kendall (M-K) test.
• Data sources:
  • Meteorological data: Daily records from 102 stations in the Yangtze River Basin (1980-2019), including temperature, precipitation, relative humidity, atmospheric pressure, sunshine hours, and wind speed. Sourced from the National Meteorological Information Center (http://data.cma.cn/).
  • Flow data: Daily average flow from 11 hydrological stations (1980-2019). Sourced from the Yangtze River Water Resources Commission (http://www.cjw.gov.cn/).
  • DEM elevation data: Sourced from HydroSHEDS (https://www.hydrosheds.org).

Main Results

• Spatial Heterogeneity of Meteorological Factors: Monthly average temperature in the basin ranges from -23.9 °C to 37.1 °C. Annual precipitation varies from less than 300 mm in upstream areas to over 1300 mm in downstream regions. Potential evapotranspiration ranges from 5.2 mm to 301.2 mm. The upper reaches are characterized by cold and dry conditions, while the lower reaches are hot and humid.
• Meteorological Trends: Temperature shows a significant upward trend across the basin. Precipitation exhibits significant fluctuations with an overall non-significant trend. Potential evapotranspiration generally shows an increasing trend.
• Optimal Runoff Reconstruction Models: The XGBoost model demonstrated the best performance for reconstructing natural runoff in the upper/middle Yangtze River Basin (R²=0.93, Nash-Sutcliffe Efficiency (NSE)=0.90). The CNN model was optimal for the lower Yangtze River Basin (R²=0.87, NSE=0.86).
• Attribution of Hydrological Drought Drivers:
  • Upper Reaches: Climate change is the primary driver, contributing 74% annually to hydrological drought, with summer contributions reaching 85%. Human activities have a relatively minor influence.
  • Lower Reaches: Climate change contributes 55% annually, but human activities exert a larger influence, particularly in spring (73% contribution).
• SHAP Model Analysis of Meteorological Factor Importance:
  • Upper Reaches: Measured runoff (SHAP value 0.38), relative humidity (0.27), and precipitation (0.22) are the most important meteorological factors influencing hydrological drought.
  • Lower Reaches: Measured runoff (0.58), average temperature (0.25), and relative humidity (0.03) are the most important factors.
• Interaction of Meteorological Factors: In the upstream, precipitation shows strong positive correlations with observed runoff, average temperature, and relative humidity. High temperature or high evapotranspiration conditions exponentially increase the contribution of climate factors to drought intensification. In the downstream, observed runoff strongly correlates with precipitation and average temperature. Complex nonlinear interactions exist, where increasing evapotranspiration significantly enhances the importance of precipitation and average temperature to hydrological drought.

Contributions

• Addresses the gap in existing literature by thoroughly exploring the uneven spatiotemporal distribution of meteorological factors and their complex interactions in hydrological drought.
• Integrates spatiotemporal cubes and interpretable machine learning (XGBoost-SHAP and CNN-SHAP) to systematically analyze and quantify the drivers of hydrological drought.
• Provides a regionally specific attribution of hydrological drought to climate change and human activities, offering new insights into their relative contributions across the upper and lower Yangtze River Basin.
• Offers a theoretical foundation for drought monitoring and prediction systems and provides scientific evidence for targeted water resource management and policy development in the Yangtze River Basin.

Funding

• Basic Research Project of Key Scientific Research Projects of Colleges and Universities of Henan Province (24ZX007).

Citation

@article{Wang2025Investigating,
  author = {Wang, Hongxiang and Cheng, Sheng-tong and Huang, Lintong and Guo, Wenxian},
  title = {Investigating the meteorological causes of hydrological drought through the integration of spatiotemporal cubes and interpretable machine learning: A case study of the Yangtze River Basin},
  journal = {Journal of Hydrology Regional Studies},
  year = {2025},
  doi = {10.1016/j.ejrh.2025.102796},
  url = {https://doi.org/10.1016/j.ejrh.2025.102796}
}