City-level total and sub-category energy intensity estimation using machine learning

Energy intensity is an important indicator for measuring the level of energy utilization efficiency in a region. Due to the limitation of energy intensity statistics, current research mainly focuses on the national and provincial scales, making it challenging to develop detailed energy conservation strategies. This study proposes a city-level total and subcategories (coal, oil, gas) energy intensity estimation method based on multi-source remote sensing data and machine learning models. The performance of the machine learning models is validated using the four-fold cross-validation approach. By comparing the various machine learning models, the deep neural network (DNN) model is used to estimate the city-level total and subcategories (coal, oil, gas) energy intensity from 2005 to 2020. The results indicate that nearly two-thirds of the cities in China have experienced a decreasing trend in energy intensity, and the total and sub-categories (coal, oil, gas) energy intensity tends to be lower in the east and south and higher in the west and north. The proposed method extends energy intensity estimation to the city level and across subcategories, providing a scientific basis for cities to develop targeted energy conservation and emission reduction policies.