Accurate assessment of expressway roadside vegetation carbon storage is essential for achieving carbon balance and ecological sustainability in the transportation sector. However, this is challenging due to the extensive spatial coverage, marked climatic variation, and pronounced spatial dependence of these vegetation corridors. Focusing on the expressway network of Shanxi Province, this study integrates UAV sample data with Sentinel-2 imagery. A graph convolutional network (GCN) is used to extract road network spatial topology, and estimation models for aboveground vegetation carbon storage are constructed for Shanxi’s northern, central, and southern climatic zones. The performance of XGBoost, Random Forest, and Support Vector Machine models is then compared across these climatic zones, emphasizing the importance of model adaptability and spatial feature integration. The results showed that the incorporation of GCN-derived spatial features significantly enhanced the model’s ability to characterize spatial autocorrelation along linear corridors. Among all models, the XGBoost-GCN model performed best in southern Shanxi, with a validation R² of 0.723. Compared with the overall model, the climate-zoned models improved estimation accuracy by an average of 18.3%, indicating that zoned modeling is more suitable for estimating carbon storage of roadside vegetation under strong climatic gradients. The carbon storage of expressway roadside vegetation in Shanxi Province exhibited an overall spatial pattern of decreasing from south to north. Hydrothermal conditions were identified as the main factors driving its spatial differentiation, while finely managed areas such as interchanges showed relatively high carbon sink potential. These findings indicate that the multi-scale remote sensing approach integrating GCN-derived spatial features with climate-zoned modeling can improve both the accuracy and regional adaptability of carbon storage estimation for roadside vegetation along long-distance expressways, and can provide methodological support for carbon monitoring, carbon sink identification, and differentiated ecological management of transportation infrastructure ecosystems.
Lian et al. (Sun,) studied this question.