LAUSR

The uncertainties in climate projections in arid regions are quite high due to the large variability of climate and the lack of high-quality climate observations. In this study, an ensemble of four Coupled Model Intercomparison Project Phase 5 (CMIP5) General Circulation Model (GCM) namely GISS-E2-H, HadGEM2-ES, MIROC5, and NorESM1-M simulations was downscaled for the assessment of the spatiotemporal changes in precipitation in the data-scarce arid province (Balochistan) of Pakistan for four Representative Concentration Pathway (RCP) scenarios. The gauge-based gridded precipitation data of the Global Precipitation Climatology Centre (GPCC) having a spatial resolution of 0.5° was used for this purpose. Support Vector Machine (SVM) was used for the development of non-local model output statistics (MOS) downscaling models for each grid by linking the GPCC precipitation with the GCM simulated precipitation across a spatial domain (latitudes 03°–45° N and longitudes 42°–92° E). Then, Random Forest (RF) algorithm was used to develop the multi-model ensemble (MME) of downscaled precipitation projections. The performances of the models were assessed in terms of normalized root mean square error (NRMSE), percentage of bias (PBIAS), and modified index of agreement (md). The results indicated that the non-local SVM-based MOS models coupled with RF MME can simulate historical precipitation over the region quite well. The MME of GCMs projected changes in the annual, monsoon, and winter precipitation in the range of − 30% to 30% for different RCPs. Overall, the MME of GCMs indicated an increase in precipitation in the monsoon-dominated wetter regions in the east, while a decrease in winter precipitation dominated arid region in the west. A decrease in annual precipitation over the majority of the southeast, east, and northeastern arid regions was projected which may increase the aridity in the region.