A scalable workflow for urban tree inventory and carbon estimation based on UAV LiDAR–hyperspectral fusion

Urban forests are critical nature-based solutions for climate change mitigation, yet accurately quantifying their carbon sequestration potential at fine scales remains a major challenge. In this study, an integrated framework for individual-tree-level species classification and carbon stock estimation in urban forests is proposed, leveraging UAV-based LiDAR and hyperspectral data fusion. To address the challenges of high feature dimensionality and class imbalance, an enhanced feature selection strategy termed adaptive cross-validation with dynamic correlation constraints (ACV-DCC) was introduced. A total of 14,376 individual trees were automatically segmented using a seed-growing algorithm across three types of urban green spaces on the Yuzhong Campus of Lanzhou University. The ACV-DCC method significantly improved classification performance, increasing the accuracy for 18 tree species to 85.67%. Among the tested classifiers, the RF outperformed the SVM and XGBoost algorithms in terms of the accuracy (85%–86%) and robustness. Its nonlinear modeling capacity also enabled accurate prediction of tree structural attributes, supporting a carbon stock estimation of approximately 1.85 × 106 kg on campus. The results demonstrate the scalability and effectiveness of the proposed framework in small- to medium-scale heterogeneous urban green environments. This reproducible and scalable workflow provides a high-precision technical pathway that can be adapted to heterogeneous urban environments worldwide, thereby contributing to global urban carbon monitoring efforts and sustainable ecosystem planning.