Net ecosystem exchange (NEE) serves as an important indicator for assessing how carbon dioxide (CO2) flows between the land surface and the atmosphere, particularly in the context of climate change. Although in situ instruments can monitor carbon exchanges, the temporal and spatial limitations of these observations necessitate the use of modeling to forecast NEE. This study examined daily NEE variations across four plant functional types (PFTs): deciduous broadleaf forest (DBF), coniferous forest (ENF), mixed forest (MF), and grassland (GRA). Using the random forest (RF) method, a machine learning approach, the study incorporated four environmental variables—solar radiation, air temperature, soil temperature, and relative humidity—as inputs for the model. The measured and predicted values were evaluated using four statistical indices of R², NSE, Bias, and RMSE. The results indicated that DBF had the most accurate modeling performance, whereas GRA had the least. An analysis of the relative significance of environmental variables revealed that solar radiation was the most important and relative humidity the least important across all PFTs. Additionally, soil temperature had a more substantial impact on improving model performance in grasslands compared to air temperature, unlike in forest types. The study also highlighted that uncertainty remains a significant issue during different phenological stages across all PFTs; with the highest uncertainty during the growing season occurring between days 140-220 in forest types and 120-210 in grassland, according to the Julian calendar.