Search

• Annual Q F dataset from different sources at 1 km resolution is built in Shanghai. • Multisource geospatial big data are jointly used in Cubist to disaggregate Q F. • The spatial pattern of Q F I shows hot spot centers over suburbs for the first time. • POIs are the most important variables for all sources of Q F in Cubist models. • The building height data can refine building anthropogenic heat flux mapping. Anthropogenic heat flux (Q F) significantly affects urban surface energy budget. However, refined Q F data are lacking for most cities across the world. To address this deficiency, this study firstly estimated the spatial distribution of the annual average Q F from different sources of industry (Q FI), buildings (Q FB), transportation (Q FT) and human metabolism (Q FM) at district and sub-district level based on multi-source data over Shanghai in 2018. Then, a Cubist machine learning based model was developed to disaggregate Q F into 1 km dataset by fusing points-of-interest (POIs) geospatial big data and multi-source remote sensing data. Results showed that the district-level Q FB contributes most to the total Q F in city center, while Q FI dominates in suburbs. High value areas of 1 km gridded Q FI are mainly distributed outside the central urban area. Q FT presents a divergence pattern from the city center to the outside. The total anthropogenic heat flux is highest in the city center with maximum of 523.03 W m−2. Variable importance analysis revealed that POIs are important for estimations of Q F. Compared with previous studies, Q F values presented here are within the range of exiting results but reveal a more reasonable spatial distribution. This study provides new aspects for spatially refined quantifications of Q F that can supplement future studies on urban heat island and urban climate numerical modeling. [ABSTRACT FROM AUTHOR]