[Scenario Prediction of Carbon Peak in Provincial Agriculture of China Based on Machine Learning]

Accurately predicting the peak of agricultural carbon emissions and its timing is crucial for accelerating the development of an agricultural powerhouse and achieving the dual carbon goals at an early date. Based on provincial-level agricultural carbon emission data from China covering 2005 to 2023， this study employed Lasso regression models and principal component analysis to identify key factors influencing agricultural carbon emissions. Multiple machine learning prediction models were constructed， and the optimal model was selected to forecast the peak of agricultural carbon emissions. The results indicate： ① Nationally， agricultural carbon emissions showed a declining trend during the observation period； provincially， 16 provinces exhibited decreasing emissions， with Beijing demonstrating the largest reduction. ② Among five machine learning algorithms， the support vector machine model with Bayesian optimization parameter tuning under Lasso feature selection （LASSO-BO-SVR） performed best. ③ Under the baseline scenario， 23 provinces including Beijing have already achieved agricultural carbon peak， while Heilongjiang and Shaanxi are projected to peak before 2030. Conversely， Guangxi， Yunnan， Qinghai， Ningxia， and Xinjiang face challenges in meeting their peak targets on schedule. Under the policy and low-carbon scenarios， the peak levels for these five provinces declined， though the timing of peak attainment varied. This research provides decision-making references for future agricultural carbon emission reduction pathways across China's provinces.