From Pixels to Carbon Emissions: Decoding the Relationship Between Street View Images and Neighborhood Carbon Emissions

Under the pressing imperative of achieving “dual carbon” goals and advancing urban low-carbon transitions, understanding how neighborhood spatial environments influence carbon emissions has become a critical challenge for enabling refined governance and precise planning in urban carbon reduction. Taking the central urban area of Xining as a case study, this research establishes a high-precision estimation framework by integrating Semantic Segmentation of Street View Images and Point of Interest data. This study employs a Geographically Weighted XGBoost model to capture the spatial non-stationarity of emission drivers, achieving a median R² of 0.819. The results indicate the following: (1) Socioeconomic functional attributes, specifically POI Density and POI Mixture, exert a more dominant influence on carbon emissions than purely visual features. (2) Lane Marking General shows a strong positive correlation by reflecting traffic pressure, Sidewalks exhibit a clear negative correlation by promoting active travel, and Building features display a distinct asymmetric impact, where the driving effect of high density is notably less pronounced than the negative association observed in low-density areas. (3) The development of low-carbon neighborhoods should prioritize optimizing functional mixing and enhancing pedestrian systems to construct resilient and low-carbon urban spaces. This study reveals the non-linear relationship between street visual features and neighborhood carbon emissions, providing an empirical basis and strategic references for neighborhood planning and design oriented toward low-carbon goals, with valuable guidance for practices in urban planning, design, and management.