Calibration of cultivar parameters is essential in ensuring the reliability of process-based crop models. Although local optimization methods have been widely used for parameter calibration, they can be sensitive to initial parameter values. Here we propose Local Optimization Global Oriented (LOGO) framework, which integrates a local search algorithm with posterior distribution assessment derived from the global search algorithm. In the present study, we examined the effects of initial values and site diversity on calibration outcomes with pseudo-observation datasets generated across six sites representing rice production environments in Asia. Our results indicated that estimated phenology-related parameters were relatively robust, whereas yield-related parameters had greater variability depending on the choice of initial values. Application of the LOGO framework reduced the distance between true and estimated parameter values. These findings provide a systematic basis for developing more robust calibration protocols and highlight the importance of integrating multiple optimization strategies.