Identifying trade-offs and synergies among land use functions using an XGBoost-SHAP model: A case study of Kunming, China

• Reveal the nonlinear features of land use function trade-offs/synergies. • Identified the dominant factors impacting the land use function trade-offs/synergies. • Visualized the nonlinear features and optimal threshold range for dominant factors. • Targeted management measures have been proposed for each land use function clusters. Exploring the spatial non-stationarity and driving mechanisms of trade-offs/synergies among land use functions(LUFs), which are crucial for effectively alleviating human-land conflicts and enhancing the overall benefits and sustainable development of regional territorial space. While most existing studies have analyzed the spatio-temporal patterns and influencing factors of LUF trade-offs/synergies from a macro scale, these studies often fail to accurately capture the multivariate interactions and complex nonlinear relationships of the geographical system within the man-earth areal system. The study area is the Kunming city on the Yunnan-Guizhou Plateau, where rapid urbanization is occurring. First, geographic weighted regression (GWR) and constrained line methods were applied at the grid unit to examine the spatial heterogeneity and nonlinear characteristics of LUF trade-offs/synergies. Then, an interpretable machine learning model (XGBoost-SHAP) was utilized to provide an intuitive explanation of the nonlinear response mechanism of LUF trade-offs/synergies. Finally, a self-organizing feature mapping network (SOM) was developed to identify LUF clusters. The findings are summarized as follows. (1) From 2000 to 2020, significant spatial heterogeneity was observed in LUF trade-offs/synergies. The interaction between ecological function (EF) and production function (PF), as well as between living function (LF) and PF, showed a convex function relationship with evident boundary effects. The interaction between EF and LF displayed a concave trade-off. (2) Elevation, slope, precipitation, distance to the city center, distance to the county center, distance to the county road, distance to river, and land use degree were the dominant factors influencing LUF trade-offs/synergies in Kunming. (3) The process of the dominant factors affects on the LUF trade-offs/synergies demonstrated strong nonlinear characteristics, and there was a significant threshold effect. (4) Based on five identified LUF clusters and the distribution of trade-offs/synergies within these clusters, differentiated LUF management measures are proposed. These results are helpful in understanding the internal mechanism of LUF system and provide technical support for the multifunctional land development, rational utilization and scientific management of land resources.