Nonlinear threshold responses of ES supply–demand to configuration mediated by composition at cluster scale in the Yellow River Basin

Semi-arid river basins worldwide face pervasive imbalances between ecosystem services (ES) supply and demand, although how landscape composition mediates the nonlinear, threshold responses of these imbalances to landscape configuration remains unclear. Using the Yellow River Basin as a case study, we classified landscape composition contexts via k-means clustering and developed a triple-threshold identification framework integrating quadratic polynomial fitting, segmented regression, and Generalized Additive Models (GAMs). This framework was applied to quantify how landscape configuration affects the supply–demand ratio (ESDR) of water yield (WY), carbon sequestration (CS), soil conservation (SC), and food production (FP). We found that: (1) Landscape composition governs the divergence in configuration–ESDR relationships, with regulating services (e.g., CS) shifting from surpluses in vegetation-dominated clusters to deficits in urban and desert/bare-land clusters. (2) Within a given cluster, services respond inversely to the same configuration change: in a forest-dominated cluster, ESDR_CS increased beyond an Aggregation Index (AI) of approximately 65.24, whereas ESDR_FP decreased beyond AI≈62.84. (3) The same configuration–service relationship exhibited opposite threshold responses across clusters; for instance, beyond a critical Edge Density (ED), ESDR_CS declined in a forest-dominated cluster but increased in a desert/bare-land cluster. This study elucidates composition-mediated threshold dynamics in ES supply–demand and provides a transferable analytical framework for landscape management in large semi-arid river basins.