Improved anthropogenic heat inventory for modeling based on multi-source remote sensing data and machine learning methods

Anthropogenic heat (AH) plays a vital role in the formation of urban heat islands, influencing urban meteorology and air quality. The estimation of anthropogenic heat fluxes (AHF) provides crucial data support for accurate numerical simulations of urban and regional meteorology as well as the atmospheric environment. The following study developed and validated a 500 m high spatial resolution AH inventory across China. Initially, using national and provincial energy consumption and demographic data, the annual total AH emissions from 1985 to 2019 were constructed. These included emissions from industry, transportation, building, and human metabolism, at both national and provincial scales. Subsequently, utilizing the spatial distribution of population and CO emissions, the spatially distributed emissions for total AH and its four components in 2019 were derived, although at a relatively low spatial resolution. Following this, multi-source remote sensing data and machine learning methods were adopted and compared to enhance the inventory resolution. After comparing the linear regression and random forest methods, the BP neural network algorithm was selected to establish an improved AH emission estimation model, resulting in a 500 m resolution AH emission inventory for China. Finally, the high-resolution spatial AH inventory was integrated into the Urban Canopy Model (WRF-UCM) for validation. The results demonstrated that the AH inventory exhibited a reasonable spatiotemporal distribution and component composition. National AHF increased from 0.04 W/m 2 in 1970 to 0.50 W/m 2 in 2020. Among the categories, industry AH emissions were significantly higher than the other three, rising from 0.06 W/m 2 in 1985 to 0.35 W/m 2 in 2019.The average AHF across 19 urban agglomerations in China ranged from 0.28 to 2.60 W/m 2 , with the highest observed in the Yangtze River Delta urban agglomerations. This spatially distributed AH inventory, being closer to the real situation, predominantly improved regional meteorology simulations, reducing the simulation error by nearly half. The root mean square error (RMSE) of temperature for gridded AH showed a minimal change of 0.01 °C, compared to a larger change of 0.09 °C for fixed AH. Additionally, the regional mean simulated PM 2.5 results for gridded AH were − 2.04 μg/m 3 , whereas for fixed AH, these were − 5.81 μg/m 3 . Moreover, because of its extensive spatial coverage, and detailed spatiotemporal characteristics, the new dataset provides robust support for urban climate research, air quality studies, and sustainable development initiatives. • The AH emission inventory with a 500 m spatial resolution for China was developed. • AH emissions from industry rose from 0.06 to 0.35 W/m 2 during 1985–2019. • The AH emissions display a highly heterogeneous spatial distribution. • The improved AH emission inventory helps reducing simulation error by nearly half.