Land use spatial optimization with CARS model and watershed hydrological analysis

Accurately extracting the spatiotemporal evolution characteristics of land use and effectively integrating hydrological prior information, while achieving spatial pattern optimization under natural laws and policy constraints, is a key challenge in land use spatial optimization. This study puts forward an integrated system combining 3D Convolutional Neural Network-Bidirectional Long Short-Term Memory Network with Cellular Automata with Random Seeds. This system extracts local spatiotemporal features and long-term sequential dependencies of land and hydrology, and uses the cellular automata model with random seeds to simulate spatial pattern evolution, thus constructing a complete technical solution from potential mining to spatial optimization. Experimental results show that the area under the receiver operating characteristic curve of the proposed algorithm reaches 0.89, and the F1 score improves to 0.93. In simulation of the main urban area of Harbin, the integrated system predicts urban development boundary with an error fluctuation less than ±400 km2, runoff simulation error as low as 2.1%, and the land use retention rate reaches 89%, all outperforming comparison models. This study provides precise and effective technical support for the spatial optimization of land use that incorporates hydrological prior characteristics, improving the scientific and practical value of spatial optimization.