Exploring the evolutionary dynamics of urban, agricultural, and ecological spaces is critical for regional sustainable development and spatial governance. However, traditional spatial simulation methods based on Cellular Automata often struggle to accommodate top-down spatial regulation, non-linear development patterns, and coordinated regional growth. The objective of this scientific research is to address these limitations by proposing a deep learning-based framework for simulating the future distribution of these three spaces. Utilizing the Unet++ model and integrating empirical data sources including multi-period remote sensing land-use mapping and prefecture-level socioeconomic statistical data, the framework predicts regional spatial patterns for the year 2030. Empirical results from the Yangtze River Economic Belt demonstrate that the model achieves high precision in large-scale spatial forecasting (with an average test accuracy of 99.32%) and effectively captures non-linear evolutionary characteristics. Predictions across various growth scenarios reveal that a moderate socioeconomic growth rate facilitates ecological preservation; controlling the expansion of urban space to approximately 20% by 2030 can prevent excessive resource depletion and regional imbalances. Consequently, it is recommended to implement the construction land increment targets outlined in current spatial planning to achieve a balance between economic growth and ecological protection.
Wei et al. (Thu,) studied this question.