Abstract: Forest above ground biomass is a key variable in the study of global carbon cycle and climate change, and accurate estimation of forest biomass is of great significance for carbon emission quantification and environmental monitoring. Airborne LiDAR provides an effective means to estimate forest biomass at regional scales. Based on the forest survey plot data and airborne LiDAR data, the estimation accuracy of different models was analyzed by the combination of Support Vector Machine (SVR), including random forest (RF), K-nearest neighbor algorithm (KNN) and ensemble learning (Stacking). According to the optimal estimation model, the above ground biomass of the forest in the study area was inverted and the spatial distribution map was plotted, and the results showed that: 1) height characteristics contributed the most to the estimation of biomass among all four variables, followed by intensity characteristics and canopy characteristics; 2) The stacking ensemble learning algorithm showed the best prediction performance in the regression fitting process, with R² values exceeding 0.71, and performed well in the estimation accuracy of different forest types; 3) The results of machine learning algorithm modeling and estimating the biomass of three different forest types: coniferous forest, broad-leaved forest and mixed coniferous and broad-leaved forest, showed that the coniferous forest was better than the broad-leaved forest, and the broad-leaved forest was better than the mixed coniferous and broad-leaved forest. 4) Based on the optimal fitting model, the average value of forest above ground biomass in Puwen Forest Farm was 213.3 t·hm⁻², and has a high consistency with the measured biomass of the forest farm (209.7 t·hm⁻²).