<p>Crop models are extensively utilized in research and applications related to regional crop yield estimation. Sensitivity analysis plays a crucial role in the localization of model parameters and is significant for the calibration and application of these models. This study gathered meteorological, soil, and crop yield data for major grain crops rice, wheat, maize, and soybean from 35 sites in southwest China between 2007 and 2016. The Extended Fourier Amplitude Sensitivity Test (EFAST) was employed to assess the spatiotemporal sensitivity and interactive effects of eight key crop parameters that influence yields in the AquaCrop-OSPy model. Results resultsrevealed that standardized water productivity (WP), crop coefficient before canopy senescence (K_cb), and harvest index (HI) are critical sensitive parameters affecting the yields of major grain crops in Southwest China with first-order (S_i) and total-order (S_Ti) sensitivity indices exceeding 0.25). The sensitivity ranking remained consistent over time and space, with WP &gt; HI &gt; K_cb. WP demonstrated the greatest stability. K_cb demonstrated the most significant interaction effects, with its S_Ti showing the largest increase relative to S_i (an increase of 21.74%). The spatiotemporal characteristics of parameter sensitivity were significantly influenced by climatic and environmental conditions. The sensitivity of WP showed significant water dependence, with response intensity controlled by regional waterstress gradients. K_cb exhibited the highest spatial variability, heavily reliant on environmental conditions, and was particularly sensitive to missing/uncertain input data. HI remained relatively stable annually, but its spatial sensitivity was constrained by the distribution of heat resources and increased with the effective temperature gradient, which was particularly significant in high-latitude/high-altitude regions. By fixing the parameters with low sensitivity at constant values, only those parameters significantly influencing the yield were calibrated. This approach effectively simplified the model and enhanced its calibration accuracy. The AquaCrop-OSPy model demonstrated strong applicability to rice, wheat, corn and soybean. The coefficient of determination (R²) between the model simulation and the actual observed yield was between 0.817 and 0.920, the normalized root mean square error (nRMSE) was between 0.31% and 11.65%, and the coefficient of residual mass (CRM) was between −0.0112 and 0.0656. The research results improve the simulation efficiency and accuracy of the AquaCrop-OSPy model, providing a reference for localization and regional application.</p>