Aboveground forest biomass (AFB) has a wide range of applications for investigating ecological functions and services that include: productivity, carbon cycling, biodiversity, and bioenergy feasibility. Federal energy policies have increased the mandated volume of renewable fuel from 9 billion gallons in 2008 to 36 billion gallons by 2022. Forest ecosystems have limits to exploitation and ground-based investigations of biomass at large scales are complicated, and not economically feasible. The goal of this investigation is to understand these limits by developing a modeling approach to predict theoretical biomass potential and examine what environmental factors influence spatial patterns and distributions of biomass within the Missouri River corridor. Information regarding AFB was derived from General Land Office (historic) and Forest Inventory and Analysis (contemporary) data. These datasets are based on plot-level ground inventory and in order to make regional assessments, a statistical methodology was required. We implemented a RandomForest model to map AFB based on a series of physiographic, edaphic, and climatic variables. RandomForest is an ensemble regression tree method that implements stochastic discrimination to determine variable importance and generate robust predictions at fine resolutions (30m2).
Results/Conclusions
All pixel-level biomass values were stratified by ecological subsection where mean values for historic, contemporary, and theoretical categories were compared (144.95, 93.90, and 97.90 Mg ha-1, respectively). Historical-based biomass predictions differed significantly from observed FIA data (X2 = 231.54, df = 19, p-value < 0.001). The RandomForest model successfully predicted observed FIA data (t19 = -1.93, p = 0.06); however, the variation in predicted biomass was significantly different from the observed FIA data (F19 = 7.5, p < .001). Predicted biomass results from RandomForest indicate that the model is capable of making accurate predictions (RMSE = 7.87 Mg ha-1) using landscape-scale descriptors and shows utility for generating importance values for covariates that contributed the most to the regression tree. Information derived from historic biomass values may be used as a baseline measure to determine the status of current forest systems along the corridor and assess the feasibility to reestablish past biomass conditions. The variation between historic and contemporary biomass quantities may be explained by differences in recovery rates from land use changes, which have created regional disparities in potential supplies of woody biomass along the corridor.