Zhang et al. (2026) A framework for identifying discriminative model, key factors, and precipitation blocking threshold on triggering drought propagation in the Xijiang River Basin (XRB)

Identification

• Journal: Climate Dynamics
• Year: 2026
• Date: 2026-02-05
• Authors: S Z Zhang, Qingxia Lin, Wenjuan Chang, Zhiyong Wu, Tao Peng, Jiali Guo, Xinzhi Wang
• DOI: 10.1007/s00382-026-08078-3

Research Groups

• Engineering Research Center of Eco-Environment in Three Gorges Reservoir Region, Ministry of Education, China Three Gorges University
• College of Hydraulic and Environmental Engineering, China Three Gorges University
• College of Hydrology and Water Resources, Hohai University

Short Summary

This study developed a GANs-enhanced machine learning framework to identify key factors and precipitation thresholds for meteorological-to-agricultural drought propagation in the Xijiang River Basin, finding that non-effective precipitation days, drought duration, and spatial complexity are critical, and daily precipitation exceeding 3 mm can mitigate propagation.

Objective

• To construct a machine learning-based framework to identify drought triggering conditions in the Xijiang River Basin (XRB).
• To quantify the precipitation modulation effect on triggering drought propagation.
• To explore the potential of this framework for predicting agricultural drought in the XRB under future climate scenarios.

Study Configuration

• Spatial Scale: Xijiang River Basin (XRB), covering an area of 353,100 square kilometers, with data resampled to 0.25° × 0.25° grids (416 grids).
• Temporal Scale: Historical analysis from 1961 to 2020; future projections from 2025 to 2099, divided into medium-term (2025–2060) and long-term (2061–2099) periods. Drought events were identified using a 120-day Standardized Precipitation Evapotranspiration Index (SPEI-120).

Methodology and Data

• Models used:
  • Hydrological Model: Variable Infiltration Capacity (VIC) model (version 4.2).
  • Drought Indices: Standardized Precipitation Evapotranspiration Index (SPEI) and Standardized Soil Moisture Index (SSMI).
  • Machine Learning Algorithms: Backpropagation (BP), Particle Swarm Optimization-Backpropagation (PSO-BP), Random Forest (RF), Extreme Gradient Boosting (XGBoost), K-Nearest Neighbors (KNN), Decision Tree (DT), Ridge Regression, Lasso, and Elastic Net Regression.
  • Data Augmentation: Generative Adversarial Networks (GANs).
  • Climate Model: MRI-ESM2-0 (from CMIP6) for future projections.
• Data sources:
  • Reanalysis Data: ERA5 (hourly precipitation, temperature, wind speed from 1961–2020 at 0.25° × 0.25° resolution).
  • Topographic Data: Digital Elevation Model (DEM) from Geospatial Data Cloud (90 m resolution).
  • Land Cover Data: UMD 1 km global land-cover dataset.
  • Soil Data: Harmonized World Soil Database (HWSD) by FAO.
  • Observed Runoff Data: Daily runoff observations (1961–2009) from Wuzhou Station, Ministry of Water Resources, China.
  • Future Climate Data: MRI-ESM2-0 (daily precipitation, wind speed, maximum and minimum temperature for 2025–2099 under SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5 scenarios, downscaled to 0.25° × 0.25°).

Main Results

• The XRB experienced 18 non-propagating and 27 propagating meteorological droughts between 1961 and 2020.
• Key factors triggering meteorological-to-agricultural drought propagation are non-effective precipitation days (NEPD), meteorological drought duration (DD), the comprehensive effect of meteorological drought area and spatial complexity (A_GAM), and meteorological drought area (DA).
• Daily precipitation exceeding 3 mm is identified as a threshold that begins to mitigate drought propagation.
• The GANs-enhanced K-nearest neighbors (KNN) model is the optimal discriminative model, achieving 100% classification accuracy for drought propagation triggering with 4-5 input factors.
• Sensitivity analysis revealed that increasing days with precipitation exceeding 3 mm could reduce propagating drought events by 11.1%, and increasing days with precipitation exceeding 5 mm could reduce them by 18.5%.
• Future projections (2025–2099) indicate that drought propagation ratios will exceed 73.0% across all SSP scenarios (ranging from 73.9% to 81.5%), surpassing the historical baseline of 60.0%.
• The mitigation potential of precipitation for drought propagation is projected to decline during 2061–2099 under all scenarios except SSP1-2.6.

Contributions

• Proposes a novel machine learning-based framework for identifying drought triggering conditions and quantifying precipitation mitigation effects, offering a direct predictive approach distinct from traditional probabilistic (Copula, Bayesian) frameworks.
• Identifies specific key factors (non-effective precipitation days, drought duration, drought area, and spatial complexity) and a quantitative precipitation blocking threshold (daily precipitation exceeding 3 mm) for meteorological-to-agricultural drought propagation in the XRB.
• Provides a highly accurate discriminative model (GANs-enhanced KNN) that offers actionable insights for water resource managers to implement targeted drought mitigation measures.
• Applies the framework to future climate scenarios, projecting increased drought propagation and a declining precipitation mitigation potential, thereby informing adaptive watershed management strategies for the XRB.
• The developed methodology is broadly applicable to other humid subtropical regions facing similar compound drought challenges.

Funding

• National Natural Science Foundation of China (Grant numbers 52479018, 52009065, 52179018)
• 111 Project of Hubei Province for Water Conservancy Engineering (China Three Gorges University)
• Discipline Innovation and Talent Introduction Base of Hydraulic Engineering

Citation

@article{Zhang2026framework,
  author = {Zhang, S Z and Lin, Qingxia and Chang, Wenjuan and Wu, Zhiyong and Peng, Tao and Guo, Jiali and Wang, Xinzhi},
  title = {A framework for identifying discriminative model, key factors, and precipitation blocking threshold on triggering drought propagation in the Xijiang River Basin (XRB)},
  journal = {Climate Dynamics},
  year = {2026},
  doi = {10.1007/s00382-026-08078-3},
  url = {https://doi.org/10.1007/s00382-026-08078-3}
}