Proposal and evaluation of nonstationary versions of SPEI and SDDI based on climate covariates for regional drought analysis

Multivariate standardized drought indices (MSDIs) are helpful in effectively characterizing, monitoring, and forecasting drought events over various spatial-temporal scales. Currently, there is a need to develop nonstationary versions of such indices for use in climate change scenarios to account for nonstationarities in different hydrometeorological variables (e.g., precipitation, evapotranspiration) affecting droughts. This article contributes nonstationary versions of two drought indices: SPEI (Standardized Precipitation-Evapotranspiration Index) and SDDI (Standardized Deficit Distance Index). They consider time-varying probability distribution whose location parameter is determined using climate covariates. It is suggested to delineate homogeneous drought regions using a fuzzy dynamic clustering approach to characterize regional droughts. Furthermore, a multi-site block bootstrap-based modified seasonal Mann-Kendall test is proposed to detect regional trends in drought analysis. For the first time, the performances and agreement of different pairs of stationary and nonstationary versions of three MSDIs, namely SPEI, SDDI, RDI (Reconnaissance Drought Index), are analyzed at at-site and regional scales through a case study on Karnataka state, India. The strength of agreement between different pairs of MSDIs is moderate to fair, and climate indices chosen for their estimation differed across regions and months. In analysis with nonstationary MSDIs, (i) at-site mean intensity and frequency of moderate/severe/extreme drought events were higher, whereas (ii) at-site mean duration and mean/maximum severity, and regional drought severity and intensity (discerned using severity-area-frequency and intensity-area-duration relationships) were generally lower relative to their stationary counterparts. Furthermore, nonstationary and stationary versions of RDI yielded higher regional drought intensity values for higher accumulation periods.

Automated summary

This study contributes nonstationary versions of drought indices and proposes a method for detecting regional trends. There is moderate to fair agreement between different pairs of stationary and nonstationary versions, with different climate indices chosen varying by region and month. Analysis with nonstationary versions shows higher intensity and frequency of drought events at site level, while duration and severity at regional level are generally lower compared to their stationary counterparts.