 The study proposes a CNNGRU model to simulate land use changes at the regional level, by extracting spatiotemporal neighborhood features and long-term dependence in time series data. The model achieved the highest simulation accuracy of 0.93 for 6, with an increased Kappa coefficient of 1.22% to 4.57%, indicating that spatial features should be initially extracted during the SNF Deep Learning extraction process, and variable neighborhood unit sizes influence the simulation results. This article was authored by Binxiao, Jimmy Niu, Zhizeng Zhao, and others.