Wang et al. (2025) Meta-learning-driven intelligent ensemble approach for robust drought evaluation across China

Identification

• Journal: Atmospheric Research
• Year: 2025
• Date: 2025-09-22
• Authors: Chunchen Wang, Zice Ma, Peng Sun, Ronghao Yang, Chongyang Zhang
• DOI: 10.1016/j.atmosres.2025.108492

Research Groups

• School of Computer and Information Engineering, Chuzhou University, Chuzhou 239000, China
• School of Geography and Tourism, Anhui Normal University, Wuhu 241002, China
• State Key Laboratory of Earth Surface Processes and Resource Response in the Yangze-Huaihe River Basin, Anhui Normal University, Wuhu 241002, China
• Engineering Technology Research Center of Resources Environment and GIS, Anhui Normal University, Wuhu 241002, China

Short Summary

This study develops a Comprehensive Drought Monitoring Model based on a Meta-learning Ensemble Algorithm (CDMMMLEA) that integrates multi-source remote sensing and geospatial data to enhance drought monitoring accuracy and robustness across China from 2001 to 2023, demonstrating superior performance over benchmark models and revealing spatiotemporal drought evolution patterns.

Objective

• To develop a Comprehensive Drought Monitoring Model based on a Meta-learning Ensemble Algorithm (CDMMMLEA) that integrates multi-source remote sensing and geospatial data to overcome limitations of conventional methods and improve the accuracy and robustness of drought monitoring across China.

Study Configuration

• Spatial Scale: Across China, with specific analysis in the Songliao River Basin (SLRB), Yellow River Basin (YRB), Loess Plateau, and western SLRB.
• Temporal Scale: 2001 to 2023.

Methodology and Data

• Models used: Comprehensive Drought Monitoring Model based on a Meta-learning Ensemble Algorithm (CDMMMLEA), ensemble learning framework, five benchmark machine learning models (for comparison).
• Data sources: Multi-source remote sensing data (canopy temperature, vegetation indices, soil moisture, canopy water content), meteorological data, auxiliary geospatial data.

Main Results

• CDMMMLEA significantly outperforms five benchmark machine learning models across China, achieving the highest correlation with the Standardized Precipitation Evapotranspiration Index (SPEI) and lowest error, particularly in the Songliao River Basin (SLRB) and Yellow River Basin (YRB).
• The model more effectively captures the spatial propagation of drought, including boundary shifts, intensity gradients, and temporal persistence, compared to SPEI, maintaining errors below 10% even in regions with sparse station data.
• Drought evolution from 2001 to 2023 reveals a phased pattern: high-frequency, high-intensity, and long-duration droughts dominated during 2001–2010, followed by a mitigation phase after 2011.
• During the mitigation phase (post-2011), drought intensity decreased by 25%, and duration shortened by 0.5 to 1.3 months per event.
• Autumn droughts were identified as most severe, affecting 62.3% of the Loess Plateau and western SLRB.

Contributions

• Development of CDMMMLEA, a novel meta-learning ensemble algorithm for comprehensive drought monitoring, addressing limitations of conventional methods (sparse station coverage, poor spatial continuity, inadequate representation of nonlinear interactions).
• Integration of diverse multi-source remote sensing, meteorological, and geospatial data into a robust monitoring framework.
• Provision of a reliable tool for high-resolution spatiotemporal drought assessment, supporting operational early warning systems, optimized water resource allocation, and region-specific drought adaptation strategies across China.
• Detailed analysis of spatiotemporal drought evolution patterns across China from 2001 to 2023, including intensity, duration, and seasonal impacts.

Funding

• Not specified in the provided text.

Citation

@article{Wang2025Metalearningdriven,
  author = {Wang, Chunchen and Ma, Zice and Sun, Peng and Yang, Ronghao and Zhang, Chongyang},
  title = {Meta-learning-driven intelligent ensemble approach for robust drought evaluation across China},
  journal = {Atmospheric Research},
  year = {2025},
  doi = {10.1016/j.atmosres.2025.108492},
  url = {https://doi.org/10.1016/j.atmosres.2025.108492}
}