A machine learning-based protocol to support visual tree assessment and risk of failure classification on a university campus

Abstract

Tree failure risk assessment involves visually evaluating trees by considering three essential factors: identifying potential targets that may be affected if the tree falls, assessing the potential consequences of the fall, and determining the likelihood of tree failure. This assessment was used to evaluate the safety of trees in a study area at Thammasat University Rangsit Center, Thailand. In two priority-selected areas for tree risk management, 3659 trees representing 139 species were assessed, and to understand the spatial patterns of tree health conditions and risks, the study employed a GIS-based mapping methodology to manage tree inventory and analyze the spatial patterns of tree health conditions and risks. A decision tree protocol based on the chi-squared automatic interaction detector (CHAID) algorithm, which employs machine learning, was used to evaluate the risk of tree failure. Our study successfully identified seven variables that are crucial in assessing the risk of tree failure. According to the findings, the overall accuracy rate of failure risk classification was 87.35 %, and of all the trees evaluated, 280 trees (7.65 % of the total) representing 34 different species were at high risk. It is recommended to start the assessment process by evaluating important variables such as tree cavities, pest infestations, mechanical damage, dead branches, and epicormic growth. Machine learning protocols, integrated with GIS, are shown to be effective, spatially-explicit, decision-support tools for detecting tree failure potential and assessing risk ratings. Application of these tools improves tree risk management practices.