Spatial and temporal variation and prediction of ecological carrying capacity based on machine learning and PLUS model

Ecological carrying capacity is key to evaluating sustainable development capacity, and land use change has varying degrees of impact on ecosystems. This study explores the ecological carrying capacity based on land use change, which is important for the conservation and scientific and rational use of ecosystems. Therefore, in this study, three machine learning classification 3algorithms, namely support vector machine, random forest and artificial neural network, were selected for accuracy comparison and analysis based on multi-source data, using Southwest Guangxi Karst-Beibu Gulf as the study area. The Patch-generating Land Use Simulation Model is used to dissect the spatial and temporal changes in ecological carrying capaciyt from 1990 to 2040 and future trends. The results showed that (1) Coupling multiple sources of data and machine learning classification algorithms can ultimately improve land use classification accuracy. The RF is compared and found to be the optimal classification algorithm for this study. (2) The area of arable and grassland gradually decreases between 1990 and 2040, mainly encroached upon by build land; the area of forested land gradually recovers, the watershed fluctuates up and down, and the area of build land gradually increases. (3) The ecological carrying capacity from 1990 to 2040 shows a trend of decreasing and then increasing, and the area of low and medium-low decreases from 2020 to 2040, with arable land and forest land having a stronger potential to increase ecological carrying capacity. Nanning City, Qinzhou City and Chongzuo City are severely damaged. The framework provides an effective tool for exploring future ecological carrying capacity to support the ecological conservation and future sustainable development of the region.