Machine Learning Approaches for Modelling Aboveground Biomass of Dry Evergreen Montane Forest from Essimingor Nature Forest Reserve in Tanzania

AbstractAccurate Aboveground Biomass (AGB) estimation is essential for effective forest management, carbon stock assessment, and climate change mitigation. Traditional allometric models often struggle to capture the nonlinear and complex relationships between forest structural variables and biomass. This study employs Machine Learning (ML) techniques, Random Forest (RF), Support Vector Machine (SVM), and Decision Tree (DT), to develop and evaluate predictive models for AGB estimation in the dry evergreen montane forests of a protected reserve in Tanzania. A dataset comprising Diameter at Breast Height (Dbh), tree height (Ht), Basal Diameter (Bd), slope (Slp), and elevation (Elv) was used to train and validate nine model configurations. Model performance was assessed using R-squared (R2), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). Results show that the SVM model using only Dbh as a predictor achieved the highest accuracy, with R-squared = 1.000, RMSE = 1.419, and MAE = 1.256. Other top-performing models, such as SVM and RF configurations using Dbh and Ht, also showed high predictive accuracy with R-squared values up to 0.999, RMSE as low as 1.589 and MAE as low as 1.279. Overall, the ML models in this study demonstrated superior performance, with R-squared values ranging from 0.957 to 1.000, RMSE from 1.419 to 7.695, and MAE from 1.256 to 5.782. When previously published allometric models were evaluated on the same dataset, their performance was notably lower in most cases, with R-squared values ranging from 0.652 to 1.000, RMSE from 4.217 to 13.423, and MAE from 4.043 to 7.939. These comparisons highlight the enhanced predictive capability and lower error margins of ML-based approaches over traditional models. These findings also highlight the effectiveness of data-driven techniques for AGB estimation in complex forest ecosystems and support their integration into forest monitoring, carbon accounting frameworks, and climate change mitigation strategies.