Abstract: Based on the four-phase (2002, 2007, 2012, and 2017) continuous forest inventory (CFI) data of Yunnan Province and contemporaneous Landsat imagery, in conjunction with climate and topographic data, this study employed the Random Forest Classifier (RFC) to derive coniferous forest spatial distribution data for the 4 time periods. The fitting performance of 55 models representing 2 categories of ensemble learning techniques specifically Bagging and Boosting was compared. Random Forest Regression (RFR) was identified as the optimal model and subsequently used to invert aboveground biomass (AGB) in coniferous forests across the four periods at the provincial scale. The temporal dynamics and spatial distribution of coniferous forest AGB in Yunnan Province were further analyzed. The results indicate that land use classification using the RFC model for the period 2002–2017 achieved an overall accuracy (OA) ranging from 0.76 to 0.78, with Kappa coefficients between 0.65 and 0.69, demonstrating high classification accuracy. In remote sensing-based modeling and estimation of coniferous forest AGB, decision tree models based on Bagging exhibited greater fitting stability and performance than those based on Boosting. The total AGB of coniferous forests in Yunnan Province was 1.49, 1.33, 1.35 Gt and 1.58 Gt in 2002, 2007, 2012, and 2017, respectively, showing an initial decline followed by an increase. The total AGB in 2017 increased by 6% compared to 2002. From 2002 to 2017, despite declines in certain prefectures, the overall trend of AGB exhibited an increase.