Analyzing the relationship between three-dimensional architectural landscapes and urban carbon emissions using machine learning approaches

Detecting the relationship between urban architectural landscapes and carbon emissions is crucial for achieving China’s carbon-peaking and carbon-neutrality goals. This study aims to investigate the relationship between 3D architectural landscapes and carbon emissions in Qingdao City, based on building 3D information extracted from high-resolution satellite images and carbon emission data from the Center for Global Environmental Research for the year 2020. First, key architectural landscape factors impacting carbon emissions were identified utilizing the Pearson correlation test and Random Forest (RF). A predictive relationship model between architectural landscapes and carbon emissions was built using Support Vector Machine Regression (SVR) and further optimized through the Chaotic Particle Swarm Optimization (PSO) algorithm. The results revealed strong correlations between carbon emissions and factors such as building density, building number, shape, and height. Floor area ratio had the highest impact on carbon emissions, contributing 45.5%, followed by building number, landscape shape index, building coverage ratio, Shannon’s diversity index, and building shape coefficient (BSC). The optimized PSO-SVR model achieved a higher coefficient of determination (R2) in the training dataset (77.32%) and test dataset (76.14%) compared to the SVR model (70% and 64.48%), along with lower mean absolute error (MAE) and mean relative error (MRE). Overall, the PSO-SVR model demonstrated enhanced accuracy in predicting carbon emissions and provided valuable insights for carbon reduction through targeted urban planning and architectural design.