Main content area

Improving drought predictability in Arkansas using the ensemble PDSI forecast technique

Liu, Yan, Hwang, Yeonsang
Stochastic environmental research and risk assessment 2015 v.29 no.1 pp. 79-91
climate, drought, financial economics, models, planning, prediction, temperature, water management, Arkansas
Drought prediction is important for improved water resources management and agriculture planning. Although Arkansas has suffered severe droughts and economic loss in recent years, no significant study has been done. This study proposes a local nonparametric autoregressive model with designed stochastic residual-resampling approach to produce ensemble drought forecasts with associated confidence. The proposed model utilizes historical climate records, including drought indices, temperature, and precipitation to improve the quality of the short-term forecast of drought indices. Monthly forecasts of Palmer Drought Severity Index (PDSI) in Arkansas climate divisions show remarkable skills with 2–3 month lead-time based on selected performance measure such as, Normalized Root Mean Square Error (NRMSE) and the Kuiper Skill Score (KSS). Rank histograms also show that the model captures the natural variability very well in the produced drought forecasts. The incorporation of categorical long-term precipitation prediction significantly enhances the performance of the monthly drought forecasts.