Building a new machine learning-based model to estimate county-level climatic yield variation for maize in Northeast China

China is one of the top maize exporting countries in the world. In China, maize is the most important staple food crop, and Northeast China is one of the main maize-growing regions. Regional maize production in Northeast China is critical for national or even global food security. In order to take preventive measures in combating the potential impacts of climate change on maize yield, it is imperative to evaluate historical climatic maize yield variation at finer scales in Northeast China. Previous scholars have achieved some good results in crop yield projection by using machine learning methods. However, climatic crop yield variation has rarely been addressed by considering the geographical factors (i.e., elevation, latitude, and longitude) of the meteorological stations used. In this study, based on 18 climatic elements during the period of 2003–2016, we compared the performance of three machine leaching methods, including neural network, support vector machine, and random forest, with multiple linear regression in evaluating climatic maize yield variation. Overall, machine learning methods are superior to traditional multilinear regression, particularly the neural network (with an R2 of 0.43 and an annual average MAE of 1.22 ton·ha−1). We selected the best machine learning methods and used the Bayesian method to integrate them as a new model. The Bayesian method reduced the evaluation error (i.e., RMSE and MAE) and improved the estimation accuracy. By addressing the uncertainty in the meteorological stations, our newly built model could be extended to other regions for evaluating climatic crop yield variation.