Saeedi et al. (2025) Introducing a new clustering-based method for regionalization framework for continental-scale rainfall estimates from soil moisture dynamics using machine learning methods

Identification

• Journal: Agricultural and Forest Meteorology
• Year: 2025
• Date: 2025-08-22
• Authors: Mohammad Saeedi, Hyunglok Kim, Venkataraman Lakshmi
• DOI: 10.1016/j.agrformet.2025.110766

Research Groups

• Department of Civil and Environmental Engineering, University of Virginia, VA, USA
• Department of Environment and Energy Engineering, Gwangju Institute of Science and Technology, Gwangju, South Korea

Short Summary

This study introduces a novel calibration-free regionalization framework for continental-scale rainfall estimation from soil moisture dynamics, combining unsupervised (K-means) and supervised (rainfall-intensity classification) clustering with a genetic algorithm. The framework, demonstrated with the SM2RAIN-Net Water Flux (NWF) algorithm over the contiguous United States (CONUS), significantly outperforms classical SM2RAIN methods by achieving a 20 % improvement in Nash–Sutcliffe efficiency and a 10 % reduction in root mean square error.

Objective

• To develop a calibration-free regionalization framework for estimating rainfall from soil moisture dynamics, eliminating the need for dedicated calibration phases and location-specific fine-tuning.

Study Configuration

• Spatial Scale: Continental-scale, specifically demonstrated over the contiguous United States (CONUS).
• Temporal Scale: Not explicitly defined for the demonstration, but the method aims to reduce reliance on extended observation periods required by traditional calibration.

Methodology and Data

• Models used: SM2RAIN-Net Water Flux (NWF) algorithm, K-means clustering (unsupervised), rainfall-intensity classification (supervised), Genetic Algorithm.
• Data sources: Soil moisture dynamics (input for SM2RAIN-NWF). Specific source for CONUS demonstration not detailed.

Main Results

• The proposed calibration-free regionalization framework, integrating K-means, a genetic algorithm, and rainfall clustering, successfully estimates rainfall from soil moisture dynamics.
• The SM2RAIN–NWF algorithm, within this framework, performs particularly well in areas with higher rainfall intensity.
• The new method significantly outperforms classical SM2RAIN approaches, achieving a 20 % improvement in Nash–Sutcliffe efficiency.
• It also demonstrates a 10 % reduction in root mean square error compared to classical methods.

Contributions

• Introduction of the first calibration-free regionalization framework for rainfall estimation from soil moisture dynamics, eliminating the need for extensive calibration data and location-specific adjustments.
• Novel integration of K-means clustering, a genetic algorithm, and rainfall clustering for automatic determination of model parameters in rainfall estimation.
• Demonstrated significant performance improvements (20 % increase in Nash–Sutcliffe efficiency and 10 % reduction in root mean square error) over existing classical SM2RAIN methods.
• Provides a robust "bottom-up" strategy for rainfall estimation that reduces reliance on traditional ground-based gauges and remote-sensing products.

Funding

Not specified in the provided text.

Citation

@article{Saeedi2025Introducing,
  author = {Saeedi, Mohammad and Kim, Hyunglok and Lakshmi, Venkataraman},
  title = {Introducing a new clustering-based method for regionalization framework for continental-scale rainfall estimates from soil moisture dynamics using machine learning methods},
  journal = {Agricultural and Forest Meteorology},
  year = {2025},
  doi = {10.1016/j.agrformet.2025.110766},
  url = {https://doi.org/10.1016/j.agrformet.2025.110766}
}