Stochastic simulation of the severity of hydrological drought

Abstract This study seeks to identify a stochastic model (or models) that can be used for mathematical description, simulation and short‐term forecasting of severity of hydrological drought in the Wabi Shebele river basin, Ethiopia. The drought variables (run sum and run length) are extracted using NIZOWKA after defining a possible truncation level from monthly streamflow data. Hydrologically homogeneous pools in the basin are established using hierarchical clustering using the parameters mean rainfall, temperature, normalized digital vegetation index and stream flow. Thus, five prominent homogeneous pools are developed. Among the regions identified, one at the upstream semi‐humid portion, station 61004, and another at the downstream arid portion, station 63001, were selected for the analysis. Stochastic modelling of the severity of hydrological drought using the Box and Jenkins methodology was carried out at both sites and compared. Diagnostic checks were performed for all the models identified from the autocorrelation function (ACF) and partial autocorrelation function (PACFs) using automatic selection criteria. The stochastic component explains an appreciable proportion of the variance (82–89%) caused in the time series at both locations. The multiplicative seasonal autoregressive integrated moving average (SARIMA) (0, 1, 1) (0, 1, 1) 12 outperforms the candidate models considered in adequately capturing the severity of drought in the area. The implications of the analysis for future water management in the basin are also discussed.