Author(s): Chenlu Yu; Dong Wang

Linked Author(s): Dong Wang

Keywords: Agricultural drought; Information theory; Prediction; Soil moisture; Vine copula

Abstract: An accurate and reliable prediction of agricultural drought (AD) is crucial for ensuring food security and supporting water resources management. This study presents a probabilistic AD prediction framework that integrates information theory with vine copula model. Partial Information Correlation (PIC) is first used for feature selection and pre-processing of hydro-meteorological variables. Monthly soil moisture (SM) is then predicted using a C-vine copula quantile regression model, and probabilistic predictions are generated through Monte Carlo (MC) simulations. Based on the predicted SM, the Standardized Soil Moisture Index (SSMI) at multiple time scales (SSMI-1, SSMI-3, SSMI-6, and SSMI-12) is calculated for indirect AD prediction, while SSMI is also directly predicted using PIC-based predictors. A case study conducted at four stations in Yunnan Province, China, demonstrates that: (1) the PIC-based feature selection effectively identifies key hydro-meteorological drivers with clear physical interpretability; (2) despite inter-monthly variability, SM predictions perform strongly across all stations, with Nash-Sutcliffe efficiency (NSE) values exceeding 0.91 and prediction interval coverage percentage (PICP) above 0.86; and (3) AD predictability improves with increasing time scales, and direct prediction generally outperforms indirect prediction. The proposed framework provides valuable support for agricultural irrigation planning and drought disaster mitigation.

DOI: https://doi.org/10.64697/iahr.proc.hic2026.106

Year: 2026