Abstract: Above-ground biomass of forests is an important evaluation index of forest productivity, and efficient monitoring of it is of great significance to maintaining global carbon balance and protecting ecosystems. The research is based on the canopy height model data, obtained the single tree canopy width boundary through the watershed segmentation algorithm, and then extracted a total of 23 LiDAR variables within the single tree canopy range,combined with 87 sets of measured data from the Penobscot experimental forest, random forest (RF) and support vector machine (SVM) are used to establish forest above-ground biomass estimation model, and the estimation results of the sample plot models are compared, and the prediction results and their accuracy are discussed. The results show that the random forest model and support vector machine model selected in this study have achieved high accuracy in estimating the above-ground biomass of forests. The random forest model has higher estimation accuracy in estimating forest above-ground biomass based on airborne radar data, stronger model generalization ability, better mapping accuracy, and better suitability.

Key words: forest above-ground biomass, machine learning, random forest, support vector machine, airborne LiDAR