Multi-scale driving mechanisms of landscape patterns on the supply-demand matching of water yield service in the Middle Reaches of the Yellow River

Studying the scale-dependent effects of the driving mechanisms of landscape patterns on water yield service in the Middle Reaches of the Yellow River is of great significance for supporting landscape pattern regulation and ecological restoration in the Yellow River Basin. Based on the multi-scale analysis framework of the multilandr package in R, we set 10 circular analysis windows with radii increasing at equal intervals from 3 to 30 km. Integrating the InVEST model, complex networks, machine learning, and the SHapley Additive exPlanations (SHAP) tool, we systematically elucidated the multi-scale driving mechanisms of landscape patterns on the supply-demand matching of water yield service from two aspects: network structure and driving thresholds. The results showed that: (1) As the scale increased, the patch area proportions of cropland and construction land decreased, while those of forest and grassland increased; the patch shape index mainly showed a trend of first rising and then falling; patch cohesion and Shannon’s diversity index continued to increase; the circular window with a radius of 9 km was the critical scale for the scale-specific differentiation of landscape patterns and supply-demand matching. (2) The network structure between landscape patterns and the water yield supply-demand ratio was stable and insensitive to scale changes, among which Shannon’s diversity index and the area proportion of each landuse type were always at the core of the network with high node strength; the area proportion of construction land showed a strong negative correlation with the water yield supply-demand ratio across all scales. (3) The supply-demand imbalance was more significantly exacerbated when the patch area proportions of cropland and construction land exceeded thresholds of 40%-50% and 5%, respectively, and when their patch cohesion indices surpassed thresholds of 90 and 75; the forest patch area proportion exceeding 40% was more conducive to water yield surplus at larger scales; the driving patterns associated with the patch area proportions and cohesion indices of cropland and construction land showed little variation with scale, whereas the effects of the shape and cohesion indices of forest and grassland changed significantly with scale. Therefore, the associated network between landscape patterns and the water yield service supply-demand ratio in the Middle Reaches of the Yellow River exhibits cross-scale stability. However, the driving directions and thresholds of landscape patterns on the water yield service supply-demand ratio are characterized by scale differentiation as the dominant feature, with scale similarity serving as a supplementary one.