Das et al. (2026) Short‐Term Agricultural Drought Prediction Using Explainable Data‐Driven Models and Quantified Uncertainty

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

Identification

• Journal: International Journal of Climatology
• Year: 2026
• Date: 2026-10-07
• Authors: K. Saranya Das, N. R. Chithra, Venkataramana Sridhar
• DOI: 10.1002/joc.70628

Research Groups

Not explicitly stated in the abstract.

Short Summary

This study evaluates machine learning and deep learning models for short-term agricultural drought prediction in Palakkad, India, using lagged hydroclimatic indices. It finds that Artificial Neural Networks (ANN) provide the highest accuracy, primarily driven by local lagged variables, and integrates uncertainty quantification and explainable AI for enhanced operational reliability.

Objective

• Evaluate the effectiveness of the Standardized Precipitation Evapotranspiration Index (SPEI) as a key lagged predictor of agricultural drought.
• Develop and evaluate multiple machine learning and deep learning models (ANN, CNN) to predict the Standardized Soil Moisture Index (SSI) at 1–3-month lead times using lagged SPEI and SSI values.
• Assess the potential influence of global climatic indices on regional drought dynamics.
• Quantify predictive uncertainty using Quantile Regression (QR) and evaluate model interpretability through SHapley Additive exPlanations (SHAP) to identify primary drought drivers.

Study Configuration

• Spatial Scale: Palakkad district of Kerala, India.
• Temporal Scale: Short-term prediction (1-month, 2-month, and 3-month lead times).

Methodology and Data

• Models used: Artificial Neural Network (ANN), Convolutional Neural Network (CNN), Quantile Regression (QR), SHapley Additive exPlanations (SHAP).
• Data sources: Lagged values of Standardized Precipitation Evapotranspiration Index (SPEI), Standardized Soil Moisture Index (SSI), and global climatic indices.

Main Results

• The Artificial Neural Network (ANN) consistently achieved the highest predictive accuracy for SSI:
  • R² values: 0.84 at 1-month lead time, 0.53 at 2-month, and 0.26 at 3-month.
  • Root Mean Square Error (RMSE): 0.50 to 0.59 at 1-month lead time, increasing at longer lead times.
• The Convolutional Neural Network (CNN) showed comparable performance, with R² values ranging from 0.77 to 0.82 across different grid points at the 1-month lead time.
• Uncertainty analysis demonstrated high coverage reliability (Prediction Interval Coverage Probability (PICP) ≈ 0.88–0.89) for ANN models, with increasing interval width at longer lead times.
• SHAP analysis confirmed that local lagged hydroclimatic variables are the dominant drivers for near-term drought prediction.
• Global climatic indices provided only marginal improvements to prediction accuracy.

Contributions

• Jointly integrates predictive uncertainty quantification (using Quantile Regression) and explainable artificial intelligence (using SHAP) into short-term drought studies, enhancing operational reliability and scientific robustness.
• Provides a transparent and robust framework for short-term agricultural drought prediction, early warning systems, and climate-resilient agricultural management.
• Evaluates the effectiveness of SPEI as a key lagged predictor for agricultural drought in a specific drought-prone region of India.

Funding

Not explicitly stated in the abstract.

Citation

@article{Das2026ShortTerm,
  author = {Das, K. Saranya and Chithra, N. R. and Sridhar, Venkataramana},
  title = {Short‐Term Agricultural Drought Prediction Using Explainable Data‐Driven Models and Quantified Uncertainty},
  journal = {International Journal of Climatology},
  year = {2026},
  doi = {10.1002/joc.70628},
  url = {https://doi.org/10.1002/joc.70628}
}