Urban building age prediction using machine learning and its application to carbon sink estimation

Abstract

Building age is significant for urban planning, energy demand estimation and climate change mitigation. This paper utilized building shape metrics, spatial indicators, and explanatory distance predictors of point of interest (POI) data to predict building age, based on 32,826 sample buildings in Qingdao City. The spatial cross-validation (SP-CV) method was used for the evaluation of predictors’ contribution to building age in Random Forest model and the prediction of building age in Support Vector Regression model. The study analyzed the building age characteristics and estimated the carbon sequestration of concrete buildings. The results showed that the distance to 8 adjacent buildings and the POIs, building height and coordinates were significant predictors of building age. The coefficient of determination (R²) exceeded 0.8, and both the mean absolute error (MAE) and root mean square error (RMSE) were less than one year. More than 57 % of the buildings in Qingdao City were over 30 years old. Low-rise buildings and residential buildings have the highest average ages, at 25.7 and 24.7 years, respectively. The average building age increased with distance from the city and district centers. The total carbon sequestration of concrete buildings was estimated at 1.86 million tons in Qingdao City. These findings are critical for urban-scale building energy estimation, urban risk assessment, and climate change mitigation, and they provide valuable insights for urban planning and low-carbon urban renewal.