Bayesian vine copulas improve agricultural drought prediction for long lead times

Abstract

Drought prediction models generally focus on shorter lead times (1–3-months) as their performance drastically declines at longer lead times (> 3 months). However, reliable agricultural drought prediction model with longer lead times is fundamental for reducing the impacts on agriculture sector, ecosystem, environment, and water resources. We propose a novel agricultural drought prediction model for long lead times by integrating vine copulas with Bayesian model averaging (hereafter, BVC model). Considering the previous meteorological drought, antecedent hot condition, and agricultural drought persistence as three predictors, the BVC model predicts agricultural drought with 1–6-month lead times. Here we focus on summer season (e.g., August) drought in China where agricultural drought impacts are more pronounced. Compared with optimal vine copula (OVC), average vine copula (AVC), and persistence-based models, the proposed BVC model performs better for 1–6-month lead times. Our findings can improve agricultural drought management, food security assessment, and early drought warning.