Drought is a natural phenomenon in which the natural amount of water in an area is below the normal level. It has negative impacts on production in numerous industries and people’s lives, especially in the context of climate change. Investigating the spatial–temporal variation of drought is of great importance in water resource allocation and management. For a better understanding of how drought has changed in China from 1961 to 2020 and will change in the future period of this century (2021–2100), a spatial–temporal assessment of drought based on the standardized precipitation evapotranspiration index (SPEI) was carried out. The trends and characteristics (number, duration, and severity) of historical and future droughts in China were evaluated based on 12-month SPEI by employing the Mann–Kendall test, Sen’s slope and run theory. The similarities, differences, and spatial–temporal evolution of droughts in these two periods were analyzed. The results showed that in the historical period the number of droughts decreased gradually from the south of China to the north. Less frequent drought but with longer duration and stronger severity occurred in the northeast and the northern areas. In the future period, most parts of China are projected to suffer more severe droughts with longer duration, especially for Northeast China, North China, Qinghai–Tibetan Plateau, and Southwest China. The likely increasing severity and duration of droughts in most areas of China in the future makes it very necessary to formulate the corresponding drought prevention and relief strategies to reduce the possible losses caused by droughts.

General circulation models (GCMs) are developed to simulate the past climate and generate future climate predictions. In the context of global warming, their important roles in identifying possible solutions to water resources planning/management are recognized by the world. However, in actual and regional implementation, due to many factors like initial and boundary conditions, parameters and model structures and so forth, there are great variabilities and uncertainties across the future climate projections of GCMs outputs, which has attracted criticism from water resources planners. Thus, the GCMs usually must be evaluated for assessing their performances in simulating the historical observations. Currently, there are many different conclusions and opinions as to whether the optimal individual model is more advantageous or whether the combined consideration of MME works better. The purpose of this paper is to compare the advantages and disadvantages between the selection of the optimal single GCM and multi‐model ensemble (MME) based on one of the more objective selection methods called TOPSIS. The results show that the performances of GCMs in simulating precipitation and temperature in different climate subregions over China are not identical. For CMIP6 GCMs simulations, the optimal precipitation GCM is CMCC‐CM2‐SR5 for EC, MIROC6 for SC, SWC and NWC, CESM2‐WACCM for NC and NEC, and FGOALS‐g3 for QTP. As for temperature, only NESM3 and BCC‐ESM1 are the optimal GCM for QTP and NWC, respectively; in other subregions, MME is better than single GCM. In general, a simple arithmetic averaging approach employed to generate the MME model is not superior to the optimal GCM, although the error metric with the observed data is reduced, at the cost of a severe compression of the interannual variability.