Probabilistic drought characterization in the categorical form using ordinal regression

Abstract

Summary Drought is an insidious natural hazard that may cause tremendous losses to different sectors, including agriculture and ecosystems. Reliable drought monitoring and early warning are of critical importance for drought preparedness planning and mitigation to reduce potential impacts. Traditional drought monitoring is generally based on drought indices, such as Standardized Precipitation Index (SPI), that are computed from hydro-climatic variables. The U.S. Drought Monitor (USDM) classifies drought conditions into different drought categories to provide composite drought information by integrating multiple drought indices, which has been commonly used to aid decision making at the federal, state, and local levels. Characterizing drought in categories similar to USDM would be important for decision making for both research and operational purposes. However, drought monitoring, based on a variety of drought indices, is challenged by the classification of drought into categories used by USDM. In this study, an ordinal regression model is proposed to characterize droughts in USDM drought categories based on several drought indices, in which the probability of each drought category can be estimated. The proposed method is assessed by comparing with USDM in Texas and a satisfactory performance for estimating drought categories is revealed.