Cheng et al. (2025) Spatially distinct drought patterns and influencing factors across China: a machine learning approach with a comprehensive index

Identification

• Journal: Ecological Indicators
• Year: 2025
• Date: 2025-09-08
• Authors: Yongming Cheng, Qiang An, Liu Liu, Hao Li, Guanhua Huang
• DOI: 10.1016/j.ecolind.2025.114170

Research Groups

• State Key Laboratory of Efficient Utilization of Agricultural Water Resources, China Agricultural University, Beijing, China
• College of Water Resources and Civil Engineering, China Agricultural University, Beijing, China
• Center for Agricultural Water Research in China, China Agricultural University, Beijing, China
• Hydro-Climate Extremes Lab, Ghent University, Ghent, Belgium

Short Summary

This study validated the Combined Climatologic Deviation Index (CCDI) for drought monitoring in China and assessed spatiotemporal drought patterns and their driving factors, revealing intensified drought in arid and plateau regions, and varied impacts of vegetation greening across different climatic zones.

Objective

• To validate the applicability of the Combined Climatologic Deviation Index (CCDI) across China’s three major natural zones (Eastern Monsoon, Northwestern Arid, Tibetan Plateau).
• To quantify spatiotemporal and seasonal drought trends in different climatic zones using CCDI.
• To characterize the dynamic relationship between drought and its drivers (e.g., soil moisture, vapor pressure deficit, vegetation) and to elucidate the differences in the mechanisms by which vegetation greening affects drought in humid, arid, and alpine regions.

Study Configuration

• Spatial Scale: China, divided into three major natural zones (Eastern Monsoon Zone, Northwestern Arid Zone, Tibetan Plateau Zone). Data aggregated/interpolated to 0.25° × 0.25° spatial resolution.
• Temporal Scale: Monthly data from April 2002 to December 2018 (17 years).

Methodology and Data

• Models used:
  • Combined Climatologic Deviation Index (CCDI) for drought assessment.
  • Theil-Sen estimator with Mann-Kendall (MK) trend analysis.
  • Random Forest (RF) regression algorithm for simulating CCDI changes and sensitivity analysis.
  • Shapley additive explanations (SHAP) framework for model interpretation.
• Data sources:
  • Vegetation: Monthly Normalized Difference Vegetation Index (NDVI) from Moderate Resolution Imaging Spectroradiometer (MODIS) (MOD13C2 Version 6), 0.05° spatial resolution.
  • Climate Factors:
    • Precipitation and Air Temperature (Temp) from China Meteorological Forcing Dataset (CMFD).
    • Evapotranspiration (ET) and Potential Evapotranspiration (PET) from Global Land Evaporation Amsterdam Model version 3.6a (GLEAM 3.6a), 0.25° × 0.25° spatial resolution.
    • Vapor Pressure Deficit (VPD) from TerraClimate, ~4 km spatial resolution.
  • Hydrological Data:
    • Total Water Storage Anomaly (TWSA) from GRACE RL06 Mascon product (Center for Space Research), 0.25° × 0.25° spatial resolution.
    • Surface Soil Moisture (SM) (0–10 cm) from GLEAM 3.6a.
  • Auxiliary Data:
    • Standard Precipitation Evaporation Index (SPEI).
    • Self-calibrated Palmer Drought Severity Index (sc-PDSI).
    • China Flood and Drought Disaster Bulletin.

Main Results

• The CCDI accurately captured drought conditions in China, showing high Spearman's correlation coefficients with sc-PDSI (positive correlation across 91.7% of the area) and SPEI (positive correlation across 97.9% of the area).
• From April 2002 to December 2018, 58.3% of China's area exhibited a decreasing trend in CCDI, indicating a general shift towards drier conditions.
  • Drought conditions intensified significantly in the Northwestern Arid Zone (Zone 2) and the Tibetan Plateau Zone (Zone 3), with trend slopes of -4.8 × 10⁻⁴ and -3.0 × 10⁻⁴ per month, respectively. Monthly CCDI values in these zones consistently showed negative trends throughout the year.
  • The Eastern Monsoon Zone (Zone 1) showed a non-significant overall trend (slope = 3.6 × 10⁻⁵ per month), but its hotspot, the North China Plain, experienced significant drought intensification (rate of 4.5 × 10⁻⁴ per month). Drought in Zone 1 intensified in spring and summer, with alleviation in autumn and winter.
• Soil moisture (SM) and vapor pressure deficit (VPD) were identified as the two most important factors affecting drought.
  • An increase in ΔSM consistently alleviated drought conditions across all zones, with sensitivities of 0.106, 0.148, and 0.721 per 0.1 m³/m³ in Zone 1, Zone 2, and Zone 3, respectively.
  • Increases in ΔVPD exacerbated drought in Zone 2 and Zone 3, while reduced ΔVPD mitigated drought stress in Zone 1. The sensitivities of ΔCCDI to ΔVPD were -0.773, -0.146, and -0.053 per 0.5 hPa for Zone 1, Zone 2, and Zone 3, respectively.
  • Increases in ΔET exacerbated drought in all zones.
• Sustained vegetation greening exhibited divergent effects across zones:
  • In Zone 1, greening alleviated drought (sensitivity of 0.029 per 0.1 NDVI), possibly through increased precipitation recycling.
  • In Zone 2 and Zone 3, greening intensified drought (sensitivities of -0.003 and -0.597 per 0.1 NDVI, respectively) due to exacerbated water deficits from evapotranspiration.
• Random Forest models explained 42–86% of the variance in the CCDI trend, with mean absolute SHAP values for ΔVPD ranging from 0.17 to 0.85 and for ΔSM from 0.19 to 0.65.

Contributions

• Provided the first comprehensive assessment of drought and its driving factors, including vegetation greenness, across China's three major natural zones using the Combined Climatologic Deviation Index (CCDI).
• Validated the applicability and superior performance of the CCDI in capturing drought evolution compared to single-variable indices, demonstrating its effectiveness in monitoring complex drought events.
• Quantified the spatiotemporal and seasonal drought trends and their significant heterogeneity across China's diverse climatic zones.
• Established an interpretable machine learning framework (Random Forest with SHAP) to systematically analyze and quantify the dynamic, non-linear relationships and marginal effects of multiple hydroclimatic drivers (soil moisture, vapor pressure deficit, evapotranspiration, vegetation) on long-term drought trends.
• Elucidated the distinct, regionally-dependent mechanisms by which vegetation greening affects drought in humid, arid, and alpine regions, highlighting its dual role (mitigation in humid areas, exacerbation in arid/alpine areas).
• Offered crucial insights for regionalized water resource management and drought mitigation strategies in China, emphasizing the importance of integrating vegetation dynamics and land-atmosphere interactions.

Funding

• National Natural Science Foundation of China (Grant numbers: 52379054 and 52079138)
• 2115 Talent Development Program of China Agricultural University (Grant Number: 00109019)

Citation

@article{Cheng2025Spatially,
  author = {Cheng, Yongming and An, Qiang and Liu, Liu and Li, Hao and Huang, Guanhua},
  title = {Spatially distinct drought patterns and influencing factors across China: a machine learning approach with a comprehensive index},
  journal = {Ecological Indicators},
  year = {2025},
  doi = {10.1016/j.ecolind.2025.114170},
  url = {https://doi.org/10.1016/j.ecolind.2025.114170}
}