Field observations from 1,193 trees representing four common campus species were scored with expert-weighted preliminary indicators. Species-specific decision trees were combined into a meta-classifier for unlikely, possible or likely failure. Overall accuracy was 79.03% with Cohen’s kappa 0.580. The model was more reliable at the unlikely and likely extremes and underdetected possible failures, so it is a prioritization aid rather than a replacement for advanced diagnostics.
Key findings
- Species-specific models achieved 79.03% accuracy and kappa 0.580. • Defect and failure patterns varied by species. • Ambiguous possible failures were underdetected and need expert follow-up.
Why this matters globally
The protocol could standardize large urban-tree screens and prioritize diagnostic resources while preserving arborist review for uncertain cases.
Thai researcher contribution
A Thammasat University team uses Thai field data to develop an urban safety and ecosystem-management tool.
Limitations to consider
Labels reflect expert visual assessment rather than observed failures; one campus and four species require temporal and external validation.