A predictive model for spatial and temporal evolution of tidal flats based on STIAU, improved vision transformer and cellular automata

Due to the multiple influences of hydrodynamic and meteorological environments, the existing predictive models for spatial and temporal evolution of tidal flats fail to adequately take into account the nonlinear relationship between the driving factors and tidal flats changes, and lack the time-dependent effective modeling. Therefore, this paper proposes a new predictive model for the spatial and temporal evolution of tidal flats, namely, the Spatial and Temporal Interaction Aggregation Update-Improved Vision Transformer-Cellular Automata (SV-CA) model. First, K-Means clustering analysis was used to obtain the tidal flat elevation class distribution map. After that, the spatial-temporal features of the tidal flat elevation class distribution map were captured by introducing the STIAU module, which was used to improve the modeling ability of the model in terms of time dependence. Meanwhile, after linear mapping of the image blocks in Vision Transformer, a cross-attention fusion module is used to capture the nonlinear features of each driving factor with respect to the spatial-temporal evolution of the tidal flats, aiming to enhance the nonlinear representation of the model. The experimental results show that the OA and Kappa coefficients are improved by 2.04%~ 3.59% and 1.57%~ 3.44%, respectively, compared with Random Forest-CA, traditional CNN-CA, Artificial Neural Network-CA and Vision Transformer-CA.