Abstract
Plants sequester carbon in their aboveground components, making aboveground tree biomass a key metric for assessing forest carbon storage. Traditional methods of aboveground biomass (AGB) estimation via Forest Inventory and Analysis (FIA) plots lack sufficient sampling intensity to directly produce accurate estimates at fine granularities. Increasing the sampling intensity with additional FIA plots would be labor and time intensive, particularly for large-scale carbon studies. Utilizing remote sensing (RS) data, such as Airborne Light Detection and Ranging (LiDAR), aerial imagery, and satellite images can significantly enhance the efficiency of forest carbon monitoring efforts. The principal objective of this study is to utilize the random forest (RF) algorithm to build predictive AGB models. We utilized 67 explanatory variables, which were extracted from three RS sources resulting in nine RF models. Each RF model was subjected to variable selection, hyperparameter tuning, and model evaluation. The optimum model considered 28 explanatory variables, with root mean square error (RMSE) of 27.19 Mgha(-1) and R(2 )of 0.41. Combining LiDAR with image metrics increased the accuracy of prediction models, serving as a pivotal tool for large area biomass mapping and carbon related decision making.