Understanding the coordination between ecosystem service value (ESV) and landscape ecological risk (ERI) is essential for ecological management in arid inland river basins. However, their spatiotemporal coupling and nonlinear driving mechanisms remain poorly understood. Using the main stem of the Tarim River as a case study, this study integrated multi-source data from 1990 to 2020 with a coupling coordination degree (CCD) model and the XGBoost-SHAP interpretable machine learning framework to investigate the evolution and drivers of ESV–ERI coordination. The results revealed a significant and intensifying spatial clustering of CCD, with Global Moran’s I increasing from 0.741 to 0.793. The basin showed a distinct spatial gradient, with high-value clustering in the upper reaches, transitional differentiation in the middle reaches, and low-value clustering in the lower reaches. Land use intensity (LUI) was identified as the dominant driver, explaining 68.9% of CCD variation. The driving mechanisms varied by river section: the upper reaches were mainly influenced by socioeconomic disturbances, the middle reaches were jointly controlled by LUI and deep soil moisture, and the lower reaches were more sensitive to climate variability and land-use change. Nonlinear analysis further showed an inverted U-shaped relationship between LUI and CCD, with a threshold near 1.53. These findings provide new insights into the spatially heterogeneous and nonlinear mechanisms of ESV–ERI coordination and offer scientific support for ecological risk control and land–water resource management in arid inland river basins.