Forest aboveground biomass estimation based on spaceborne LiDAR combining machine learning model and geostatistical method

Estimation of forest biomass at regional scale based on GEDI spaceborne LiDAR data is of great significance for forest quality assessment and carbon cycle. To solve the problem of discontinuous data of GEDI footprints, this study mapped different echo indexes in the footprints to the surface by inverse distance weighted interpolation method, and verified the influence of different number of footprints on the interpolation results. Random forest algorithm was chosen to estimate the spruce-fir biomass combined with the parameters provided by GEDI and 138 spruce-fir sample plots in Shangri-La. The results show that: (1) By extracting different numbers of GEDI footprints and visualize it, the study revealed that a higher number of footprints correlates with a denser distribution and a more pronounced stripe phenomenon. (2) The prediction accuracy improves as the number of GEDI footprints decreases. The group with the highest R², lowest RMSE and lowest MAE was the footprint extracted every 100 shots, and the footprint extracted every 10 shots had the worst prediction effect. (3) The biomass of spruce-fir inverted by random forest ranged from 51.33 t/hm² to 179.83 t/hm², with an average of 101.98 t/hm². The total value was 3035.29 × 10⁴ t/hm². This study shows that the number and distribution of GEDI footprints will have a certain impact on the interpolation mapping to the surface information and presents a methodological reference for selecting the appropriate number of GEDI footprints to derive various vertical structure parameters of forest ecosystems.