Modeling and spatio-temporal analysis on CO2 emissions in the Guangdong-Hong Kong-Macao greater bay area and surrounding cities based on neural network and autoencoder.

• A method based on Neural Network is proposed for provincial-level CO 2 emissions estimation. • An autoencoder is proposed to downscale the provincial-level CO 2 emissions to city-level. • A complete city-level CO 2 emissions inventory of GBA and surrounding cities is complied. • Spatial dependence and agglomeration characteristics of CO 2 emissions are analyzed. • Spatio-temporal heterogeneity of factors affecting CO 2 emissions is investigated. Significant attention has been given to the issue of CO 2 emissions worldwide, especially for China as the largest emitter. Cities, as the main carriers of China's economic development, account for up to 85 % of CO 2 emissions from energy use in the country. The Guangdong-Hong Kong-Macao Greater Bay Area (GBA) and surrounding cities play important leading roles in low-carbon development. However, compiling city-level CO 2 emissions is challenging due to the limited availability of data. In this paper, a new idea is proposed by using SSA-BP for estimating provincial-level CO 2 emissions and downscaling provincial-level CO 2 emissions to city-level CO 2 emissions by reconstructing missing values of city-level CO 2 emissions with an autoencoder. This study compiles a city-level CO 2 emissions inventory and conducts a spatio-temporal analysis of CO 2 emissions in the GBA and surrounding cities. The results show that CO 2 emissions exhibit a spatial distribution pattern with spatial dependence. Furthermore, according to the analysis of spatio-temporal heterogeneity, GDP influenced CO 2 emissions the most within the five influencing factors, and the influences of GDP, population, and energy intensity were mainly positive, while the influence of patents was mainly negative, and that of trade was both positive and negative. [ABSTRACT FROM AUTHOR]