Merchantable Biomass and Carbon Estimation of Bottomland Hardwood Oaks in the Arkansas Delta Using Allometric and Random Forest Models

Abstract Bottomland hardwood forests in the Lower Mississippi Alluvial Valley (LMAV) play a critical role in regional carbon dynamics and ecosystem service provision; however, studies to develop biomass estimation models for oak species in this unique ecosystem remain limited. This study developed nonlinear allometric and machine learning Random Forest models calibrated to approximate the Forest Inventory and Analysis (FIA) reported oven-dry bole biomass for merchantable stem sections in seven oak species. The dataset was acquired from the USDA FIA Program on 1,175 individual trees across 34 plots in the Arkansas LMAV between 2015 and 2023. Oak species included in this study were overcup ( Q. lyrata ), cherrybark ( Q. pagoda ), Nuttall ( Q. texana ), Shumard ( Q. shumardii ), water ( Q. nigra ), willow ( Q. phellos ), and post oaks ( Q. stellata ), representing the main and widespread species in this region. Five nonlinear allometric models based on the power model function were developed and evaluated alongside the Random Forest (RF) model. Model results indicated that the RF model achieved the highest predictive accuracy, with a mean absolute percentage error (MAPE) of 12.4% and a relative root mean square error (RRMSE) of 30.9%, outperforming the best-fitted allometric model (Model 2; $$B=a\times {(D{*H}_{m})}^{b}$$ B = a × ( D ∗ H m ) b , which had a MAPE of 17.2% and an RRMSE of 33.3%. External validation using an independent, destructively sampled legacy tree dataset showed lower prediction errors for the RF model (MAPE of 20.7% and RRMSE of 33.8%) and Model 5 ( D 2 H m , MAPE of 22.4% and RRMSE of 34.5%), which suggested that although calibrated using FIA-derived data, these models can reasonably approximate measured biomass when applied to an independent dataset. In addition, analysis of variable importance revealed that the combined height and diameter variable of $${D}^{2}{H}_{m}$$ D 2 H m , and $$D{H}_{m}\text{W}$$ D H m W were the most influential predictors for biomass estimation. Subsequent carbon stock computation showed significant variation across seven oak species (F (6, 1168) = 12.4, p < 0.0001). Overcup oak had the highest mean carbon stock (546.26 kg CO ₂ e), while Nuttall oak exhibited the lowest (267.34 kg CO ₂ e). These findings suggest that the traditional allometric models remain suitable and applicable for biomass estimation, yet the RF model provides a more accurate framework. These models offer forest managers and planners a reliable non-destructive tool for approximating FIA-derived merchantable biomass using commonly available tree attributes, supporting silvicultural planning and extending FIA applications for oak species in the LMAV.