Oz Kira

Urban vegetation plays a crucial role in regulating microclimates, mitigating the urban heat island effect, improving air quality, and contributing to climate resilience—functions increasingly vital amid accelerating climate change. Despite this importance, mapping vegetation in urban environments remains challenging, particularly at the species level, due to structural complexity, mixed land use, and spectral similarity among vegetation types. In this study, we developed a hierarchical multi-temporal framework using high-resolution (3 m) PlanetScope imagery with time-series analysis to classify urban vegetation and dominant tree species by capturing their phenological trajectories. PlanetScope imagery from 2021 to 2023 was first averaged into monthly composites. These monthly composites were subsequently organized into annual multi-temporal feature stacks (Y.C.), seasonal multi-temporal feature stacks (S.C.), individual seasonal datasets, and monthly datasets. Spectral bands were combined with five vegetation indices (NDVI, EVI, GNDVI, MSAVI, and NIRv) to represent seasonal variations in vegetation characteristics. A Random Forest classifier was applied at the pixel level. The Y.C. yielded the highest performance for broad vegetation classification (Overall Accuracy = 94.33%, Kappa = 0.92), followed by S.C., while individual seasons and months generally produced lower accuracies. Grass and Trees benefited from increased temporal information, whereas Shrubs remained more difficult to classify. Tree-species classification showed a different temporal pattern, with winter yielding the highest accuracy (OA = 89.33%). At the species level, F-scores ranged from 0.66 to 0.90, with species exhibiting distinct seasonal characteristics showing greater separability. These contrasting results demonstrate that the optimal temporal representation depends on classification level, with extended time-series information benefiting broad vegetation classification and seasonal phenological contrast enhancing tree-species discrimination. Our findings underscore the utility of phenology-aware, high-resolution remote sensing for advancing urban vegetation monitoring, green-infrastructure design, and climate-adaptive urban planning.