Ghosh et al. (2025) Quantifying rainfall-induced climate risk in rainfed agriculture: A volatility-based time series study from semi-arid India

Identification

• Journal: Agricultural Water Management
• Year: 2025
• Date: 2025-09-08
• Authors: Soham Ghosh, Sujay Mukhoti, Pritee Sharma
• DOI: 10.1016/j.agwat.2025.109775

Research Groups

• Department of Economics, School of Humanities and Social Sciences, Indian Institute of Technology Indore, Indore, Madhya Pradesh, India
• Mehta Family School of Sustainability, Indian Institute of Technology Indore, Indore, Madhya Pradesh, India
• Jaya Prakash Narayan National Centre of Excellence in the Humanities, Indian Institute of Technology Indore, Indore, Madhya Pradesh, India
• Operations Management and Quantitative Techniques Area, Indian Institute of Management Indore, Indore, Madhya Pradesh, India

Short Summary

This study develops a volatility-in-mean time series framework to quantify rainfall-induced climate risk on rice yield forecasts in semi-arid Maharashtra, India, finding that GARCH-type models, particularly eGARCH and gjrGARCH with log-differenced rainfall measures, significantly improve forecast accuracy and robustness.

Objective

• To develop and evaluate a volatility-in-mean time series modeling framework to quantify how intra-annual rainfall variability (rainfall risk) influences rice yield forecasts in Maharashtra, India.

Study Configuration

• Spatial Scale: Maharashtra, India (a semi-arid, monsoon-dependent region).
• Temporal Scale: Monthly rainfall data from January 1962 to December 2021 (60 years); Annual rice production data from 1962 to 2021.

Methodology and Data

• Models used:
  • ARIMA (AutoRegressive Integrated Moving Average)
  • ARIMAX (ARIMA with exogenous variables)
  • GARCH (Generalized Autoregressive Conditional Heteroskedasticity) variants:
    • sGARCH (symmetric GARCH)
    • eGARCH (exponential GARCH)
    • gjrGARCH (threshold GARCH)
    • iGARCH (integrated GARCH)
  • GARCH-ARIMA (GARCH in mean with ARIMA component)
  • GARCH-ARIMAX (GARCH in mean with ARIMAX component and exogenous variables)
  • Four distinct measures of intra-seasonal rainfall variability (RV1, RV2, RV3, RV4) based on realized rainfall variance.
• Data sources:
  • Monthly rainfall data: India Water Resources Information System (India-WRIS), Government of India.
  • Annual rice production data: EPWRF India Time Series database, Economic and Political Weekly Research Foundation.

Main Results

• Volatility-based models (e.g., eGARCH and gjrGARCH variants) consistently deliver superior forecast accuracy and greater robustness compared to simpler ARIMAX or iGARCH configurations.
• Rainfall risk measures based on higher-order, first-difference (log-differenced) monthly rainfall (RV3 and RV4) are the most successful predictors, yielding lower Akaike Information Criterion (AIC) and Mean Absolute Error (MAE) values.
• Models using contemporaneous rainfall volatility measures outperform those using lagged measures, indicating an immediate impact of seasonal climate anomalies on current-year crop results.
• The eGARCH_ARIMAX model achieved the best in-sample performance (lowest AIC) and out-of-sample MAE among models incorporating exogenous regressors.
• The gjrGARCH_ARIMA model recorded the lowest uncertainty ratio (1.11) based on the ratio of out-of-sample to in-sample Mean Squared Error (MSE), indicating stable generalization.
• Sensitivity analysis showed that sGARCH and eGARCH configurations, particularly with RV3 and RV4, exhibited small and symmetric changes (typically below 10%) under ±10% perturbations to rainfall variability measures, demonstrating stable generalization. In contrast, ARIMAX and iGARCH models showed higher susceptibility (30-40% changes).

Contributions

• Introduces a novel volatility-in-mean time series modeling framework to quantify rainfall-induced climate risk in rainfed agriculture.
• Develops and evaluates four distinct measures of intra-seasonal rainfall variability (realized rainfall variance) for enhanced agricultural forecasting.
• Demonstrates that GARCH-type models, especially eGARCH and gjrGARCH, significantly improve rice yield forecast accuracy and robustness by integrating dynamic meteorological risk indicators.
• Highlights the critical importance of contemporaneous rainfall volatility for crop yield prediction, informing early warning systems.
• Provides evidence-based tools for strengthening early warning systems, supporting adaptive policy design, and promoting resilient cropping systems in monsoon-dependent regions.
• Complements existing agricultural water management policies by offering tools for anticipatory planning and efficient resource allocation based on high-frequency rainfall risk.

Funding

• Jaya Prakash Narayan National Centre of Excellence in the Humanities, Indian Institute of Technology Indore, India (IITI/JPN/PRJ/2025/152/P001).
• PhD Fellowship provided by the DST-INSPIRE programme, Department of Science and Technology, Government of India (190728).
• Indian Institute of Technology Indore.

Citation

@article{Ghosh2025Quantifying,
  author = {Ghosh, Soham and Mukhoti, Sujay and Sharma, Pritee},
  title = {Quantifying rainfall-induced climate risk in rainfed agriculture: A volatility-based time series study from semi-arid India},
  journal = {Agricultural Water Management},
  year = {2025},
  doi = {10.1016/j.agwat.2025.109775},
  url = {https://doi.org/10.1016/j.agwat.2025.109775}
}