Abstract Improving the prediction of crop production is critical for strategy development associated with global food security, particularly as the climate continues to change. Process-based ecosystem models are increasingly used for simulating global agricultural production. However, such simulations often use a single crop variety in global assessments, implying that major crops are identical across all regions of the world. To address this limitation, we applied a Bayesian approach to calibrate regional types of maize (Zea mays L), capturing the aggregated traits of local varieties, for DayCent ecosystem model simulations, using global crop production data from 2001 to 2013. We selected major cropping regions from the FAO Global Agro-Environmental Stratification as a basis for the regionalization and identified the most important model parameters through a global sensitivity analysis. We calibrated DayCent using the sampling importance resampling algorithm and found significant improvement in DayCent simulations of maize yields with the calibrated regional varieties. Compared to a single type of maize for the world, the regionalization of maize leads to reductions in root mean squared error of 11%, 31%, 27%, 30%, 19%, and 27% and reductions in bias of 59%, 59%, 50%, 81%, 32%, and 56% for Africa, East Asia, Europe, North America, South America, and South and Southeast Asia, respectively. We also found the optimum parameter values of radiation use efficiency are positively correlated with the income level of different regions, which indicates that breeding has enhanced the photosynthetic efficiency of maize in developed countries. There may also be opportunities for expanding crop breeding programs in developing countries to enhance photosynthesis efficiency and reduce the yield gap in these regions. This study highlights the importance of representing regional variation in crop types for achieving accurate predictions of crop yields.