Zubair et al. (2025) Agricultural drought forecasting using remote sensing: A hybrid modeling framework by integrating wavelet transformation and machine learning techniques

Identification

• Journal: Agricultural Water Management
• Year: 2025
• Date: 2025-10-27
• Authors: Muhammad Zubair, Zeeshan Zafar, Shenjun Yao, Zhongyang Guo, Adeel Ahmad Nadeem, Shah Fahd
• DOI: 10.1016/j.agwat.2025.109922

Research Groups

• Key Laboratory of Geographic Information Science (Ministry of Education), East China Normal University, Shanghai, China
• School of Geographic Sciences, East China Normal University, Shanghai, China
• State Key Laboratory of Water Resources Engineering and Management, Wuhan University, Wuhan, China
• College of Urban and Environmental Sciences, Northwest University, Xi’an, China

Short Summary

This study developed an innovative hybrid modeling framework integrating wavelet transform preprocessing with ensemble machine learning models (XGBoost, AdaBoost, Random Forest) to enhance agricultural drought prediction. The framework significantly improved prediction accuracy, with XGBoost achieving the highest performance (R² = 0.964) in forecasting the Vegetation Health Index (VHI).

Objective

• To enhance the accuracy of agricultural drought assessment and forecasting by proposing and evaluating an innovative hybrid modeling framework that integrates wavelet transform preprocessing with ensemble-based machine learning models.

Study Configuration

• Spatial Scale: Sindh Province, Pakistan (approximately 14.1 million hectares, with 4.9 million hectares designated agricultural land).
• Temporal Scale: Data extracted from 2001 to 2023.

Methodology and Data

• Models used: Hybrid modeling framework integrating Discrete Wavelet Transform (DWT) with ensemble machine learning models: XGBoost, AdaBoost, and Random Forest. DWT used the Coiflet–1 (coif1) mother wavelet with a four-level decomposition.
• Data sources:
  • CHIRPS (Climate Hazards Group InfraRed Precipitation with Station data) precipitation data (0.05° spatial precision).
  • MODIS-derived environmental indicators: Land Surface Temperature (LST), Normalized Difference Vegetation Index (NDVI), Modified Normalized Difference Water Index (MNDWI), Vegetation Condition Index (VCI), Temperature Condition Index (TCI), and Vegetation Health Index (VHI).
  • Data extracted and preprocessed using Google Earth Engine (GEE).

Main Results

• Wavelet preprocessing significantly improved the prediction accuracy of agricultural drought.
• Among the tested models, XGBoost achieved the highest predictive performance (R² = 0.964, Root Mean Square Error (RMSE) = 0.021, Mean Absolute Error (MAE) = 0.023).
• AdaBoost followed with strong performance (R² = 0.946, RMSE = 0.030, MAE = 0.024).
• Random Forest also showed commendable results (R² = 0.926, RMSE = 0.035, MAE = 0.027).
• Feature importance analysis revealed NDVI as the most pivotal variable in the XGBoost model for VHI prediction.
• Significant correlations were identified between drought-related variables: a strong negative correlation between LST and TCI (r = -0.96), and LST and VHI (r = -0.90), highlighting the impact of temperature stress on vegetation health.
• DWT decomposition and reconstruction demonstrated high fidelity for all drought-related variables, with RMSE values ranging from 0.04 to 0.67 and R² values between 0.77 and 0.92.

Contributions

• Proposed a novel hybrid framework that uniquely combines discrete wavelet transformation for multi-resolution analysis of non-stationary remote sensing data with a comparative ensemble of advanced machine learning models (XGBoost, AdaBoost, and Random Forest).
• Enhanced feature extraction and predictive accuracy for complex vegetation health dynamics by systematically decomposing input signals and utilizing wavelet coherence to identify time-frequency relationships among drought indicators.
• Established a robust benchmark for selecting the most effective algorithm for drought forecasting through a thorough comparison of various boosted and tree-based models for VHI prediction.
• Offers a robust decision-support tool for identifying drought-prone areas, enhancing agricultural resilience, and informing policy responses to climate-related risks.

Funding

• National Natural Science Foundation of China (grant number W2412140).

Citation

@article{Zubair2025Agricultural,
  author = {Zubair, Muhammad and Zafar, Zeeshan and Yao, Shenjun and Guo, Zhongyang and Nadeem, Adeel Ahmad and Fahd, Shah},
  title = {Agricultural drought forecasting using remote sensing: A hybrid modeling framework by integrating wavelet transformation and machine learning techniques},
  journal = {Agricultural Water Management},
  year = {2025},
  doi = {10.1016/j.agwat.2025.109922},
  url = {https://doi.org/10.1016/j.agwat.2025.109922}
}