[A Comprehensive Comparative Analysis of Land Use Change Prediction Models Based on Land Types]

Increasing human activities and climate change have intensified the shift of land use, and the construction of land use models has become a hot spot in the study of land use change. Taking Shaanxi Province as the study area, the prediction performances of four models, namely, CA-Markov, CLUES, FLUS, and PLUS, were comprehensively compared, and the contribution of driving factors was analyzed by combining the LightGBM model. The model accuracy was verified based on the land use data from 2017 to 2022 to explore the applicability and differences of different models in predicting land use changes. The results show that: ① From 2017 to 2022, the areas of forest land, cultivated land, and construction land in Shaanxi Province increased, while the area of grassland decreased significantly by 4 085.63 km². ② The LightGBM model verified the high explanatory power of driving factors on land use patterns and showed that natural factors dominated the distribution of ecological land (forest land, grassland, and water), while socio-economic factors significantly affected the changes in construction land. ③ The OA of CA-Markov, CLUES, FLUS, and PLUS were 0.90, 0.92, 0.92, and 0.93, respectively, and the Kappa coefficients were 0.86, 0.88, 0.88, and 0.90, respectively. The prediction accuracy of PLUS was higher, and the prediction performance of PLUS in the main land types was relatively accurate. The total deviation of FLUS in different land types was the smallest, followed by the PLUS model. ④ For the spatial distribution results, CA-Markov had a single model, but its operation was the most convenient. The patch distribution of CLUES was relatively clustered, and the patches predicted by the FLUS model were distributed in patches. The expansion pattern was closer to the actual situation. The patch distribution in the PLUS result was the most dispersed and fragmented. Different models had different usage conditions and parameter requirements, and the setting of driving factors may also have affected the prediction effect. This study can provide important methodological support and data reference for land use change prediction and provide a scientific basis for future spatial optimization management and sustainable development planning.