Abstract: In order to evaluate the feasibility of the remote sensing estimation using spaceborne Lidar ICESat-2 (Ice, Cloud, and land Elevation Satellite-2) data for the forest Aboveground Biomass (AGB) in the mountains region, the Random Forest Regression (RFR) model was conducted by combining the Advanced Terrain Laser Altimeter System (ATLAS) photon point cloud data and 54 sample plots in Shangri-La, a typical mountain area in northwest Yunnan, Southwest China. On the basis of the data denoising and classification, the 50 canopy parameters and 3 topographic factors of 74 873 footprints were extracted. A biomass model was established with 53 parameters as the independent variables after the hyper-parametric optimizing RF, and the biomass data was collected from 54 remote sensing plots to serve as dependent variables. The Root Mean Square Error (RMSE), coefficient of determination (R2) and overall estimation accuracy (P1) were used to evaluate the model accuracy. The results showed that: 1) There was the highest significant correlation of the number canopy photons parameters with the forests aboveground biomass (P<0.01) by a correlation analyzing between the 53 footprint parameters and footprint biomass. The other parameters, landsat percentage canopy, canopy photon rate, slope, number of photons and apparent surface reflectance were significantly correlated with biomass (0.01< P <0.05). As such, these parameter variables were set for the remote sensing model of mountain forest biomass. 2) The traditional RF regression model without the slope presented the R2=0.90, RMSE=11.90 t/hm2, P1=80.06%, the accuracy indexes of the traditional RF model with the slope were R2=0.91, RMSE=11.01 t/hm2, P1=81.30%. Among them, there was a 13.97% contribution rate of terrain slope factor to the model, which was less than the contribution rate of landsat percentage canopy, apparent surface reflectance and number canopy photons to the model. Thus, there was a certain effect of terrain factor in the traditional model on the remote sensing modeling of ICESat-2/ATLAS footprints. 3) There was a much higher accuracy of the RF regression model after optimization by the hyper-parameters, where the R2=0.93, RMSE=10.13 t/hm2, P1=83.31%. The improved model was much more suitable for the estimation of forest aboveground biomass in mountainous terrain, compared with the traditional. The total and average biomasses of footprints were then estimated as 1.32×105 t, and 77.41 t/hm2 respectively. Furthermore, there was a higher fitting accuracy of the above-ground biomass model for the mountain forest using RF after parameter optimization, where the R2 and estimation accuracy were above 0.9, and 82%, respectively. Consequently, the improved model can be feasible for the AGB inversion using the footprint parameters that were extracted by ICESat-2 in mountainous areas. According to the spatial distribution of biomass of ICESat-2 footprints, the footprints with the high biomass were mainly distributed in the northern part of the study area, and there were uneven distribution and large regional differences, which were consistent with the spatial distribution of volume in the study area in 2021. Therefore, the ICESat-2 can be used for the forest aboveground biomass estimation. The findings can also provide the research cases for the remote sensing monitoring of forest biomass at low-high altitude areas.