Estimation of Forest Biomass Based on ICESat−2/ATLAS Data in Jingdong

In this study, Jingdong Yi Autonomous County of Yunnan Province was selected as the research area, and space-borne LiDAR ICESAT-2/ATLAS was used as the main information source. On the basis of denoising and classifying ATLAS data, Kriging interpolation based on geostatistics realized the spatial expansion of the index points of ATLAS spot parameters from "point" to "surface". A forest biomass estimation model was established based on 265 biomass survey plots. The results showed as follows: Based on random forest importance ranking, the 5 parameters of ATLAS light spot with strong correlation with forest biomass are as follows: h_max_canopy_abs,h_mean_canopy, ph_ndx_beg, solar_elevation, and solar_azimuth. Variance function analysis was performed on the 5 parameters, and the optimal variation function model is selected according to the determination coefficient and spatial autocorrelation The spatial interpolation effect of the three parameters h_max_canopy_abs, solar_elevation and solar_azimuth was the best using the spherical model , h_mean_canopy and ph_ndx_beg had better effects with the exponential model. Based on random forest regression, a remote sensing estimation model of forest biomass in the study area was established, with modeling accuracy R²=0.7941 and RMSE=23.0047 t/hm², taking the above-ground biomass of 265 plots as explained variables and corresponding 5 parameters as explanatory variables. The model can be used as an estimation model of forest above-ground biomass in the study area. The forest biomass in the study area was estimated based on the verified RF model, and the estimated value was 31269874.76 t. The forest biomass calculated by the 2019 ground survey group in the study area was 26674465.55 t as the reference truth value, with an estimated accuracy of 85.3%,The spatial distribution is basically consistent with the actual calculation results.The results showed that the forest biomass estimation based on ICESAT−2/ATLAS data had a good effect.