Kim et al. (2025) Hyperspectral Remote Sensing and Artificial Intelligence for High-Resolution Soil Moisture Prediction

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

Identification

• Journal: Water
• Year: 2025
• Date: 2025-10-27
• Authors: Ki-Sung Kim, Seung‐Jun Lee, Jeongjun Park, Gigwon Hong, Kicheol Lee
• DOI: 10.3390/w17213069

Research Groups

Information not provided in the paper text.

Short Summary

This study developed a drone-based hyperspectral approach using visible and near-infrared reflectance to estimate gravimetric soil water content. An artificial neural network model achieved a high coefficient of determination (0.9557), demonstrating accurate and reproducible mapping suitable for operational decision support.

Objective

• To develop and evaluate a drone-based hyperspectral approach that accurately maps visible and near-infrared reflectance to gravimetric soil water content using machine learning methods.

Study Configuration

• Spatial Scale: Field-scale, localized (implied by "drone-based" and "field estimation").
• Temporal Scale: Single acquisition for model development, with a framework designed for larger, multi-date acquisitions.

Methodology and Data

• Models used: Simple regression, multiple regression, tree-based ensembles, Gradient Boosting, Artificial Neural Network (ANN) with three hidden layers, rectified linear unit activations, and dropout.
• Data sources: Drone-based hyperspectral reflectance (visible and near-infrared bands) paired with ground-truth gravimetric water content measured by oven drying. A dataset of 1000 matched samples was used.

Main Results

• Conventional machine learning methods (simple/multiple regression, tree-based ensembles) were limited by band collinearity and piecewise approximations, failing to meet accuracy targets.
• Gradient boosting reached the accuracy target but showed different trade-offs in variable sensitivity.
• An Artificial Neural Network, trained with feature count sweep and early stopping, achieved a coefficient of determination (R²) of 0.9557 using ten predictors.
• The ANN model demonstrated accurate mapping from hyperspectral reflectance to gravimetric water content.

Contributions

• Provides a reproducible framework for accurate field estimation of gravimetric soil water content using drone-based hyperspectral data and an Artificial Neural Network.
• Demonstrates the superior performance of ANNs over conventional machine learning methods for this specific application, overcoming challenges like band collinearity.
• Establishes a foundation suitable for larger, multi-date acquisitions and integration into operational decision support systems for hydrology and water resources management.

Funding

Information not provided in the paper text.

Citation

@article{Kim2025Hyperspectral,
  author = {Kim, Ki-Sung and Lee, Seung‐Jun and Park, Jeongjun and Hong, Gigwon and Lee, Kicheol},
  title = {Hyperspectral Remote Sensing and Artificial Intelligence for High-Resolution Soil Moisture Prediction},
  journal = {Water},
  year = {2025},
  doi = {10.3390/w17213069},
  url = {https://doi.org/10.3390/w17213069}
}