Characterization of drought over Botswana using multivariate statistics

Abstract

Droughts pose a significant challenge to water resources, causing several socio-economic consequences. The growing economy requires improved assessments of drought-related impacts in the water sector, particularly under semi-arid environments where its climate is getting drier and warmer. This study proposes a probabilistic model (copula approach) that is intended to contribute to the drought risk assessment by providing an essential information in drought prediction, decision making and management of the limited water resources available during drought events. The three key pillars (i. Drought monitoring and early warning systems, ii.vulnerability, and impact assessments, and iii. mitigation and response measures) are recommended as the basis of national drought policy and management plans, providing a practical way to organize multiple actions and activities that the country need to implement to better prepare and respond to drought. In the study, drought events are characterized by duration, severity and intensity, and the Standardized Precipitation Index (SPI) and the Standardized Precipitation Evapotranspiration Index (SPEI) were used to analyse hydrological drought based on gridded rain gauge and potential evapotranspiration data referred to as Climatic Research Unit (CRU) covering a period of 1901-2018 at a time scale of 12 months. Both SPI and SPEI were able to detect the spatial and temporal variation of drought. But SPEI was able to identify more droughts in the severe to moderate categories over wider areas in the country than SPI did. A set of seven homogeneous drought regimes based on spatial characteristics of SPEI were obtained. Region 1 and 7 are relatively wet regions, followed by region 2 and 6, while region 2 and 4 are relatively dry regions which borders the Kgalagadi Basin. The optimal marginal distribution for drought duration, severity and intensity were identified by employing the Akaike Information Criterion (AIC). The drought duration was best described by the Generalized Extreme Value distribution while the drought severity and intensity were both found to optimally fit with Weibull distribution. Nine copula distributions, namely-Normal, Student’s t, Gumbel-Hougaard, Rotated Gumbel, Clayton, Rotated Clayton, Joe Clayton, Frank, and Plackett copula distributions were applied to construct the bivariate and trivariate distributions. The most appropriate copula functions were determined also based on AIC. The joint distribution of the best marginal cumulative density functions of duration and severity is found to optimally fit Normal copula while Clayton copula distribution is the best copula function that describes joint distribution of duration and intensity as well as the joint distribution of severity and intensity over most grids in Botswana. The trivariate distribution of the univariate marginals vi of duration, severity and intensity is best fitted by Normal copula. The conditional return period of drought of different categories was also determined in a multivariate context by coupling duration, severity, intensity of drought based on copula distribution and cumulative density functions. Most of the historical drought events over homogenous drought regimes in Botswana have short duration, low severity, low intensity and short return period, with drought regimes 1, 3, 4 and 7 having longer drought return periods than the other three zones (2, 5 and 6). The risk of having long and severe droughts within the 10-year design lifetime of any hydrological system was low in drought events with longer duration and high severity across all the regions in Botswana. Improved information on drought characterization can be useful in evaluating the water-supply capability and the needed supplementary water resources during severe drought conditions for a specific water-supply system.