Tobin et al. (2025) Validation of Downscaled SoilMERGE with NDVI and Storm-Event Analysis in Oklahoma and Kansas

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

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-12-18
• Authors: Kenneth Tobin, Aaron Sanchez, Alejandro X. Alaniz, Stephanie Hernandez, Adriana Perez, Deepak Ganta, Marvin Bennett
• DOI: 10.3390/rs17244058

Research Groups

Not specified in the provided text.

Short Summary

This study evaluated a prototype 500 m downscaled version of the 0.125-degree SoilMERGE root zone soil moisture product using machine learning, demonstrating that downscaled products, particularly those using Extreme Gradient Boosting, significantly outperform the default product in predicting vegetation greenness and storm-event streamflow.

Objective

• To evaluate the performance of a prototype 500 m downscaled version of the 0.125-degree SoilMERGE (SMERGE) root zone soil moisture product (0 to 40 cm depth) using machine learning techniques.

Study Configuration

• Spatial Scale: Contiguous United States (product coverage); most of Oklahoma and Kansas (study area); 0.125 degrees (original SMERGE resolution); 500 meters (downscaled resolution); watershed scale (for streamflow analysis).
• Temporal Scale: Warm season from 2008 to 2019.

Methodology and Data

• Models used: Extreme Gradient Boosting (XGB), Gradient Boost, Random Forest (for downscaling and evaluation).
• Data sources: SoilMERGE (SMERGE) root zone soil moisture product; Normalized Difference Vegetation Index (NDVI) datasets; United States Geological Survey (USGS) streamflow data.

Main Results

• All three machine learning downscaled products (ranked correlation R2 values of 0.52 to 0.59) outperformed the default SMERGE (R2 = 0.44) in ranked correlation against Normalized Difference Vegetation Index (NDVI) datasets.
• Extreme Gradient Boosting (XGB) and Gradient Boost recorded a higher ranked correlation (R2 = 0.59) with NDVI than Random Forest (R2 = 0.52).
• For the most intense storm events (>35 mm/day), antecedent XGB downscaled SMERGE (ranked correlation R2 = 0.64) significantly outperformed antecedent streamflow (R2 = 0.43) and all other SMERGE versions (R2 values of 0.52 to 0.56) as a predictor of storm event response.

Contributions

• Demonstrated the broad-scale benefits of Machine Learning-assisted downscaling for root zone soil moisture products.
• Provided proof of concept for the development of high-resolution, state-based SMERGE products across the US Great Plains.

Funding

Not specified in the provided text.

Citation

@article{Tobin2025Validation,
  author = {Tobin, Kenneth and Sanchez, Aaron and Alaniz, Alejandro X. and Hernandez, Stephanie and Perez, Adriana and Ganta, Deepak and Bennett, Marvin},
  title = {Validation of Downscaled SoilMERGE with NDVI and Storm-Event Analysis in Oklahoma and Kansas},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs17244058},
  url = {https://doi.org/10.3390/rs17244058}
}