Assessing the potential of multi-seasonal Sentinel-2 satellite imagery combined with airborne LiDAR for urban tree species identification

Accurate information on urban tree species composition is critical for urban green space ecosystem management. However, achieving large-scale, high-precision species identification in complex metropolitan environments remains challenging. This study assessed the potential of medium-resolution multi-temporal optical imagery combined with airborne LiDAR for tree species classification in large heterogeneous urban areas (> 5000 km²). The results indicate that precise large-scale identification of urban tree species distribution is feasible by integrating multi-seasonal Sentinel-2 imagery with airborne LiDAR data based on a Random Forest hierarchical classification model. The overall classification accuracies for deciduous broadleaf species and evergreen broadleaf species were 63.32% and 76.77%, respectively. Multi-temporal spectra were the primary explanatory variables, with spring bands significantly affecting the classification of deciduous broadleaf species. For evergreen broadleaf species, each season has its own dominant spectral information. Classifications combining data from three seasons outperformed single- or two-season combinations. The incorporation of LiDAR-derived metrics improved the classification results for most species, with accuracy increases of up to 18.75% point for deciduous broadleaf species. Overall, the results demonstrate the effectiveness of combining medium-resolution multi-temporal optical imagery with LiDAR data for urban tree species classification, laying a foundation for quantifying ecosystem services provided by urban trees through remote sensing.