This paper presents a simulation model of surface change using GIS which combines many data together based on hierarchical fusion and ConvLSTM-CA coupling with attention. The model addresses three key limitations in existing approaches: inadequate semantic alignment in feature fusion, fixed neighborhood rules that fail to capture urban expansion anisotropy, and static weight allocation ignoring temporal evolution. By integrating spectral, topographic, climatic, and socioeconomic dimensions through adaptive channel attention and implementing distance decay neighborhoods, the framework achieves dynamic factor screening and uncertainty quantification via evidence theory. The model has been tested on 2020-2022 study data, identified as 92.7% accurate with 18.3% FoM improvement over FLUS, and therefore it is a powerful predictor of urban expansion.
Li et al. (Thu,) studied this question.