Classification-based estimation of urban green space carbon storage using integrated remote sensing imagery and LiDAR data

Accurate and scalable estimation of urban green-space carbon storage is essential for green infrastructure assessment, yet heterogeneity in vegetation types and community structures complicates modeling. This study developed a fractional vegetation cover (FVC)-stratified framework integrating Sentinel-2 imagery and backpack LiDAR for the central urban area of Nanjing, China. A database comprising 406 field plots, each measuring 20 m × 20 m, was established. LiDAR-derived tree height and diameter at breast height were used to estimate plot-level carbon storage, while NDVI, RVI, and SAVI were evaluated as predictors in FVC-specific regression models. Summer-derived vegetation indices correlated more strongly with carbon storage than winter-derived indices. Based on the combined model evaluation, NDVI was selected for low-coverage plots and SAVI for medium- and high-coverage plots. Compared with unified models, stratification improved predictive R 2 by an average of 49.5%. The optimized models were then applied to map carbon storage density within the vegetation modeling domain, revealing pronounced spatial differences among FVC classes and green-space structures. By integrating two-dimensional satellite imagery with three-dimensional LiDAR data and introducing an FVC-based stratification strategy, the framework provides a scalable pathway from plot-level estimation to urban-scale carbon mapping, supporting green infrastructure evaluation and low-carbon urban planning.