Net primary productivity (NPP) is a key indicator for assessing the quality of forest ecosystems, and its maximum value (NPPmax) reflects the potential for forest productivity. Studying vegetation NPPmax is crucial for the rational utilization of forest resources and sustainable forest management. This study focuses on Larix olgensis A. Henry plantations in Jilin Province as the research subject. Using the national forest inventory data of permanent plots in Jilin Province, NPPmax data extracted by MOD17A3 from 2000 to 2022, and environmental factors (terrain, climate, and soil), this study applied correlation analysis, machine learning algorithms, and Shapley additive explanation (SHAP) methods to analyze the driving factors of NPPmax change. Among the four machine learning algorithms, the random forest model (RF) predicted the NPPmax of L. olgensis plantations with the highest accuracy (R2 = 0.540, RMSE = 11.320 gC·m-2·a-1), followed by the boosted regression tree model (R2= 0.523), artificial neural network (R2= 0.519), and support vector machine model (R2= 0.501). The SHAP analysis based on the RF model revealed that the main factors affecting the change in NPPmax were the annual accumulated temperature above 18℃, total nitrogen, and elevation. The annual accumulated temperature, total nitrogen content and elevation had a threshold effect on NPPmax. The NPPmax significantly varied when the annual accumulated temperature above 18℃ approached 200℃, the soil total nitrogen content reached 2.0 g/kg, and the altitude was approximately 800 m above sea level. By using a machine learning model and SHAP analysis to address the complex nonlinear relationship between NPPmax and the environment, we can better explain the mechanism between NPPmax and environmental predictors. An obvious threshold effect exists between the environmental variables and NPPmax, which provides a basis for evaluating the site quality of L. olgensis plantations.