Lee et al. (2025) Large‐Scale Drought Forecasting in the U.S. Southern Plains Through a Hybrid Cluster‐Based Wavelet‐Machine Learning Approach

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water Resources Research
• Year: 2025
• Date: 2025-11-01
• Authors: Sanghyun Lee, Ali Danandeh Mehr, Daniel N. Moriasi, Ali Mirchi
• DOI: 10.1029/2024wr039744

Research Groups

Not explicitly stated in the provided abstract.

Short Summary

This study developed a novel hybrid clustering-based machine learning approach, combining Discrete Wavelet Transform (DWT) and Multilayer Perceptrons (MLPs), to forecast the gridded Standardized Precipitation-Evapotranspiration Index (SPEI) across the U.S. Southern Plains, demonstrating effective capture of drought spatial variability for early warning systems.

Objective

• To develop and evaluate a novel hybrid machine learning approach for high-resolution gridded Standardized Precipitation-Evapotranspiration Index (SPEI) forecasting across the U.S. Southern Plains with 1-month and 3-month lead times.

Study Configuration

• Spatial Scale: U.S. Southern Plains (regional, high-resolution gridded data).
• Temporal Scale: Drought forecasting with 1-month and 3-month lead times.

Methodology and Data

• Models used: Multilayer Perceptrons (MLPs), Long Short-Term Memory (LSTM), Genetic Programming (GP). A clustering-based method was used for training. Discrete Wavelet Transform (DWT) was applied for temporal pattern capture.
• Data sources: High-resolution gridded data sets (implied observational/reanalysis data for SPEI, precipitation, and potential evapotranspiration).

Main Results

• The hybrid DWT-MLP models demonstrated superior performance compared to LSTM and GP, based on Nash-Sutcliffe efficiency and root-mean-square error.
• DWT significantly enhanced model accuracy by effectively capturing key temporal patterns in the SPEI series.
• Physical and hydrologic attributes strongly influenced input selections for the models.
• A 12-month lag period was effective in regions with weaker seasonality, while mutual information-based lag selection benefited areas with strong seasonality.
• For 3-month-ahead forecasts, including decomposed potential evapotranspiration as an input improved accuracy in drier regions but decreased it in humid areas.
• The DWT-MLP forecast maps effectively captured the spatial variability of drought, showing high correlations with observed values.

Contributions

• Introduction of a novel hybrid clustering-based DWT-MLP approach for high-resolution gridded drought forecasting.
• Demonstration of the effectiveness of spatially adaptive drought prediction using clustered grid cells.
• Identification of the influence of physical/hydrologic attributes on input selection and the benefits of DWT for capturing temporal patterns in SPEI.
• Validation of the approach for regional drought early warning systems, enhancing water resources management adaptations.

Funding

Not explicitly stated in the provided abstract.

Citation

@article{Lee2025LargeScale,
  author = {Lee, Sanghyun and Mehr, Ali Danandeh and Moriasi, Daniel N. and Mirchi, Ali},
  title = {Large‐Scale Drought Forecasting in the U.S. Southern Plains Through a Hybrid Cluster‐Based Wavelet‐Machine Learning Approach},
  journal = {Water Resources Research},
  year = {2025},
  doi = {10.1029/2024wr039744},
  url = {https://doi.org/10.1029/2024wr039744}
}