The prediction of regional agricultural carbon emissions is of great significance for regional environmental protection and sustainable development of regional agriculture. The article proposes a PLS-SA-AdaBoost prediction model combining partial least squares (PLS), simulated annealing algorithm (SA) and adaptive boosting algorithm (AdaBoost), which overcomes the shortcomings of a single model with insufficient prediction accuracy. Through an empirical study of agricultural carbon emissions in Fujian Province, China, the article validates the effectiveness of this combined forecasting model. The results show that the combined PLS-SA-AdaBoost forecasting model has higher accuracy and better performance than other models. The article predicts the future trend and range of agricultural carbon emissions in Fujian Province under five different scenarios.