Hourly estimation of black carbon in China based on sparse observation data and stacking model

Abstract

Black carbon (BC) is a highly absorbent aerosol that significantly impacts regional air quality, public health, and global climate change. In this study, we develop a machine learning-based stacking model to estimate hourly BC in China, using observational data from 36 BC sites of the China Atmosphere Watch Network (CAWNET) from 2020 to 2023. The stacking model shows strong robustness and high accuracy nationwide, achieving a Pearson correlation coefficient (R) of 0.79 and a root mean square error (RMSE) of 0.60 μg/m³. Compared to Tracking Air Pollution in China (TAP), the stacking model reduces bias by 33.8% in urban areas and 56.9% in rural areas. According to this model, the urban BC concentration is 48.5% higher than that in rural areas across China. North China Plain, Central China, and the Sichuan Basin continue to be the regions with the highest BC concentrations in China, with average concentrations of 1.73 ± 0.66 μg/m³, 1.61 ± 0.55 μg/m³, and 1.71 ± 0.58 μg/m³, respectively, from 2020 to 2023. During the COVID-19, urban areas experienced a consistent decline in BC concentrations from 2020 to 2022, followed by a rebound in 2023 (rural areas rebound in early 2022). Our research highlights that these high-quality hourly estimated BC concentrations can reveal pollution distribution patterns in both urban and rural areas, as well as during specific time periods, thereby providing crucial support for developing more accurate and effective emission reduction strategies, improving air quality, mitigating climate change, and protecting public health.