Generic above-ground biomass estimator for urban forests using machine learning

Abstract

ABSTRACT Beyond urban trees’ aesthetic roles in urban landscapes, urban trees have significant environmental and ecological values such as carbon sequestration. In this study, machine learning (ML) namely Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Support Vector Machine Regression (SVR) were used to develop generic AGB estimators for urban trees using the diameter at breast height, total height, and dry wood density of 1051 individual urban trees. The results from the ML were compared with the outputs from a generic allometric equation that was developed using a destructive method. The results showed that the ML represents a good alternative to the traditional destructive method with R2 above 0.9 during training, and R2 above 0.8 during testing. The RF and XGBoost performed better than SVR in the prediction of AGB. However, overall, the AGB predicted using ML was more accurate than the AGB estimated with a generic allometric equation. The generic AGB estimator improves urban forest management by providing an accurate AGB which can support decision-making and can be used for planning, carbon accounting, and monitoring as well as tree species selection and maintenance.