Locating and profiling city street trees using Baidu street view images for carbon storage evaluation

Traditional methods for estimating the carbon storage of street trees involve manual sampling, which incurs substantial human, material, and temporal costs in establishing a city-wide comprehensive inventory of street trees. In this study, we propose a multi-task convolutional neural network called STV-MNet to identify individual- level and city-wide street trees from Baidu street view images (BSVIs). We measured the structural and locational information of the identified trees using cylindrical projection and MonoDepth depth estimation network. STV-MNet achieved a mean intersection over union (mIoU) of 0.733 and a mean average precision of 0.881 at IoU 50 % (mAP50) in individual tree identification, outperforming DeepLab v3+ (mIoU of 0.641) and YOLO v3 (mAP50 of 0.767). Validation with street-measured data demonstrates that our method produces more precise estimations for both tree height and breast diameter, with the root mean square error (RMSE) of 0.09 m and the normalized RMSE of 0.005 m for tree height and the RMSE of 0.01 m and the normalized RMSE of 0.016 m for diameter at breast height (DBH). The location prediction of street trees achieves a minimum error of 0.67 m and an average error of 7.37 m. Using the biomass carbon storage equation, we calculated the carbon storage of individual street trees in Changsha City, Hunan Province, China. The results indicate that the total carbon storage of 333,717 street trees in urban areas of Changsha City is 1.64 × 10⁵ tons, and the annual carbon sequestration capacity across the urban areas is 8014.57 tons. In certain areas, street tree resources have enabled the achievement of carbon neutrality in road transportation. This study presents a novel approach to managing urban street tree carbon storage, leveraging STV-MNet for automatic carbon storage estimates, and demonstrates high practical significance in low-cost and city-wide street tree carbon storage estimation.