Historical forest aboveground biomass carbon storage (HFCS) dataset

The primary challenge in reconstructing historical forest aboveground biomass carbon storage lies in the difficulty of acquiring forest growth status indicators spanning over a century, irrespective of canopy height, forest age, NDVI, LAI, or SIF. In this study, we utilized historical forest cover percentage datasets from 1900 to 2020 as a surrogate indicator for modelling century-long forest carbon stock estimation. Subsequently, we selected 15 categories of predictive factors, including vegetation, meteorologic, geophysical, soil, and spatial distance factors . To do this, we used the forest aboveground biomass (AGB) dataset from 2019 as the base map. We divided the dataset into a 70% training set and a 30% test set for model training and evaluation. By continuously adjusting the velocity and position of particles using the Particle Swarm Optimization algorithm, we obtained the optimal RF model with optimal model parameters as follows: a number of decision trees of 195, a maximum depth of 10 for each decision tree, a number of features of 12 to consider when looking for the best split, minimum number of samples of 4 required to split an internal node, and a minimum number of samples of 3 required to be at a leaf node.