Nonlinear Response of Carbon Use Efficiency to Driving Factors in the Poyang Lake Basin: Integration of XGBoost-SHAP and GeoDetector Models

Carbon use efficiency (CUE) is a critical indicator of ecosystem carbon sink capacity. However, the nonlinear response of CUE to complex environmental drivers remains poorly understood in subtropical humid regions. This study analyzed the spatiotemporal dynamics of vegetation CUE in the Poyang Lake Basin (PYLB) from 2001 to 2020 and quantified the influence of natural and anthropogenic factors using XGBoost-SHAP and GeoDetector models. The results showed that: (1) The average annual CUE in the PYLB was 0.455, exhibiting a declining trend, with a linear rate of −0.001524 yr−1. (2) SHAP analysis revealed that the association between LAI and CUE exhibited a non-monotonic transition at a threshold of approximately 1.80. Specifically, while lower LAI levels were positively correlated with CUE, this relationship shifted to a negative trend as LAI exceeded the threshold, demonstrating a phase-specific coupling pattern across the canopy density gradient. For hydrothermal drivers, CUE exhibited localized downward fluctuations when precipitation was between 1700 and 2700 mm and temperature ranged from 15.0 to 18.0 °C. (3) GeoDetector analysis indicated that LAI was the dominant individual factor controlling the spatial heterogeneity of CUE (q = 0.471), and its interaction with TEM exerted the strongest synergistic effect (q = 0.493). These results emphasize the necessity of considering nonlinear thresholds and factor interactions when evaluating ecosystem carbon budgets in changing climates.