Sharma et al. (2025) Leveraging Sentinel-2 Data and Machine Learning for Drought Detection in India: The Process of Ground Truth Construction and a Case Study

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

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-09-11
• Authors: Shubham Subhankar Sharma, Jit Mukherjee, Fabio Dell’Acqua
• DOI: 10.3390/rs17183159

Research Groups

Not specified in the provided text.

Short Summary

This study investigates the use of multispectral Sentinel-2 remote sensing indices and machine learning to detect drought conditions in three regions of India during the Rabi season. XGBoost, combined with a seasonal majority voting strategy, achieved 96.67% accuracy, precision, and recall, identifying Normalized Multi-band Drought Index (NMDI) and Day of Season (DOS) as the most influential features.

Objective

• To explore the use of multispectral Sentinel-2 remote sensing indices and machine learning techniques to detect drought conditions in Jodhpur, Amravati, and Thanjavur, India, during the Rabi season.

Study Configuration

• Spatial Scale: Three distinct regions of India: Jodhpur, Amravati, and Thanjavur (district level).
• Temporal Scale: Rabi season (October–April) over a ten-year period, with the remote sensing dataset organized from 2016–2025.

Methodology and Data

• Models used: Random Forest, XGBoost, Bagging Classifier, Gradient Boosting. A seasonal majority voting strategy was applied for final drought labeling. Shapley Additive Explanation (SHAP) analysis, Borda Count, and Weighted Sum analysis were used for feature importance.
• Data sources: Multispectral Sentinel-2 remote sensing indices (12 indices studied). Reference drought conditions were sourced from official government drought declarations and regional/national news publications providing seasonal drought maps.

Main Results

• Among initial machine learning models, XGBoost achieved the highest accuracy (84.80%), followed by Bagging Classifier (83.98%) and Random Forest (82.98%).
• With the seasonal majority voting strategy, XGBoost and Bagging Classifier achieved 96.67% accuracy, precision, and recall. Random Forest reached 90%, and Gradient Boosting reached 83.33% across all metrics.
• Normalized Multi-band Drought Index (NMDI) and Day of Season (DOS) consistently emerged as the most influential features in determining model predictions, supported by SHAP, Borda Count, and Weighted Sum analysis.
• Red-edge Chlorophyll Index (RECI), Normalized Difference Water Index (NDWI), Normalized Difference Moisture Index (NDMI), and Ratio Drought Index (RDI) were also identified as important features.
• A feature ablation study showed XGBoost demonstrated the best overall performance, particularly when using the Top 5 features.

Contributions

• Demonstrated the effective application of multispectral Sentinel-2 remote sensing indices combined with machine learning for timely drought detection in specific agricultural regions of India.
• Identified key remote sensing indices (NMDI, DOS) that are most influential for drought prediction in the studied regions, offering interpretable insights into model decisions.
• Developed and validated a seasonal majority voting strategy that significantly improved the accuracy, precision, and recall of drought detection models.
• Provided a robust framework for drought monitoring that can aid in implementing effective mitigation strategies.

Funding

Not specified in the provided text.

Citation

@article{Sharma2025Leveraging,
  author = {Sharma, Shubham Subhankar and Mukherjee, Jit and Dell’Acqua, Fabio},
  title = {Leveraging Sentinel-2 Data and Machine Learning for Drought Detection in India: The Process of Ground Truth Construction and a Case Study},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs17183159},
  url = {https://doi.org/10.3390/rs17183159}
}