Quantifying future drought change and associated uncertainty in southeastern Australia with multiple potential evapotranspiration models

Abstract

Projection of drought under a changing climate is important for drought risk assessment. Changes in precipitation (P) and potential evapotranspiration (ETp) are expected to influence future drought occurrence. Thus, it is important to include both factors to accurately quantify change in drought frequency under future climate scenarios. Standardized precipitation evapotranspiration index (SPEI) is a widely used index in drought assessment because it considers the influence of both P and ETp on drought. Thus, in this study we used SPEI to quantify change in drought frequency under two different emission scenarios (RCP4.5 and RCP8.5) in the wheat belt of southeastern Australia with climatic data downscaled from 34 global climate models (GCMs). We also investigated whether differences ETp models would make a difference on drought projection. Therefore, we employed five different traditional ETp models (Penman, Jensen-Haise, Makkink, Abtew, Hargreaves) and three random forest (RF)-based models to calculate SPEI in this study. Results showed that drought, especially moderate and severe drought, would occur more frequently under future climate scenarios and the increased frequency was generally greater in spring and winter than in summer and autumn. Severe drought occurring in spring would increase by 3.1%–21.7% under RCP4.5 and 5.2%–41.0% under RCP8.5. In autumn, the likely mean increase of severe drought frequency was 0.7%–13.0% under RCP4.5 and 2.7%–27.9% under RCP8.5. Differences in the projected increase of drought frequency were found among the different ETp models. In general, RF-based ETp models, which projected larger increases in ETp, generally also projected larger increases in drought occurrence. A multilinear regression relationship was built between changes in drought frequency and changes in ETp and P. The regression showed that the increased drought frequency was a combined result of the increasing ETp and decreasing P, and that the increasing ETp might be the more dominant factor. The contribution of GCMs, RCPs, different ETp models, and their interaction to the uncertainty in drought projection was quantified with the use of analysis of variance. Results showed that GCMs and their interaction with RCPs were the dominant factors influencing uncertainty in drought projection.