Faria et al. (2026) Predicting Irrigated Rice Soil–Water Conditions Using Multispectral Remote Sensing and Machine Learning in Semi-Arid Australia

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

Identification

• Journal: Remote Sensing
• Year: 2026
• Date: 2026-08-01
• Authors: Brenno Tondato de Faria, Gustavo Magalhães Tercete, Rodrigo Filev Maia, John Hornbuckle
• DOI: 10.3390/rs18152504

Research Groups

Not specified

Short Summary

The study evaluates the use of multispectral remote sensing and machine learning to classify soil-water conditions in Australian rice fields, finding that while "Dry" and "Flooded" states are distinguishable, "Saturated" and "Flooded" states are difficult to differentiate.

Objective

• To identify an optimal set of multispectral remote sensing indices and a machine learning model to predict three specific soil–water conditions in irrigated rice: "Flooded", "Saturated", and "Dry".

Study Configuration

• Spatial Scale: Semi-Arid Australia (irrigated rice fields)
• Temporal Scale: Not specified

Methodology and Data

• Models used: Machine Learning (ML), Minimum Redundancy Maximum Relevance (mRMR) algorithm, and SHAP (SHapley Additive exPlanations) analysis.
• Data sources: Multispectral remote sensing indices.

Main Results

• Model 2 (utilizing mRMR-selected variables) outperformed Model 1 (utilizing standard literature indices) with an accuracy of 0.64 and a kappa of 0.46.
• ROC-AUC values for Model 2 were 0.87 for "Flooded", 0.62 for "Saturated", and 0.83 for "Dry".
• A high confusion rate was observed between "Flooded" and "Saturated" conditions, indicating that multispectral data alone is insufficient to distinguish these two states.
• SHAP analysis identified vegetation-sensitive indices (related to crop biomass, plant moisture, and senescence) as the primary drivers for the predictions.

Contributions

• Demonstrates the effectiveness of the mRMR algorithm in selecting relevant multispectral indices for soil-water classification.
• Highlights the inherent limitations of multispectral remote sensing in differentiating between saturated and fully ponded (flooded) conditions in rice cultivation.

Funding

Not specified

Citation

@article{Faria2026Predicting,
  author = {Faria, Brenno Tondato de and Tercete, Gustavo Magalhães and Maia, Rodrigo Filev and Hornbuckle, John},
  title = {Predicting Irrigated Rice Soil–Water Conditions Using Multispectral Remote Sensing and Machine Learning in Semi-Arid Australia},
  journal = {Remote Sensing},
  year = {2026},
  doi = {10.3390/rs18152504},
  url = {https://doi.org/10.3390/rs18152504}
}