Abstract
In recent years, lifestyle changes and urbanization of societies, as well as macro-environmental changes, i.e. climate changes (CCs), have caused changes in the land spatial structure and the transfer of resources between different economic sectors of the land. The development of long-term spatial development plans (SDPs) needs to be compatible with CCs, especially in hyperarid areas with low supplies and high demands. In this research, machine learning methods; including Cellular Automata (CA), Random Forest (RF) and regression models through PLUS model were used to simulate the amount of supplies and demands based on land cover (LC) maps during the years 2000, 2010 and 2020 in the hyperarid areas of Kerman, Iran. Then, the best predicted model (Kappa = 0.94, overall accuracy = 0.98) was used to simulate changes in LC classes under climate change scenarios (CCSs) for 2050. The results showed the efficiency of machine learning in simulating land cover changes (LCCs) under CCSs. Findings revealed that SDPs of these areas are not compatible under any possible consideration of CCSs. The modeling results showed that spatial development plans under CCSs is not environmentally efficient and there is no compatibility between supplies, based on agricultural lands, and demands, based on increased population, by 2050. Overall, under the scenario of RCP 8.5, man-made, agriculture and natural LC classes with 106.9, 2.9, and 18.6% changes, respectively, showed the greatest changes compared to 2020. Population control, adjustment of infrastructures, and changes in LC plans can reduce socio-economical and socio-environmental problems in the future of hyperarid areas to some extent.