Search

Rapid urbanization and industrialization in China had led to increased pollutants emission. PM 10 is one of the main components of air pollutants, which significantly impacts human health, environment, and regional or global climate. In this study, a new machine learning deep forest (DF) model was used to construct the aerosol optical depth (AOD) and near-ground PM 10 concentration (AOD-PM 10) model. The DF model combines the advantages of deep neural networks and tree models, which can provide model interpretability. Combined with the Himawari-8 AOD, meteorological, and auxiliary factors, the hourly PM 10 concentration in China (spatial resolution: 0.05 × 0.05°) was obtained. The results show that AOD has the highest contribution to the importance of features in the AOD-PM 10 model, accounting for approximately 13.5%, and the contributions of boundary layer height, temperature, and relative humidity to the importance of features were 11%, 8.6%, and 7%, respectively. A 10-fold cross-validation was used to evaluate the performance of the model. The hourly cross validation results from 09:00 to 16:00 (Beijing time) show that the R2 range was 0.82–0.88, and the root mean square error and absolute mean error were 18.55–23.12 μg/m³ and 11.54–16.82 μg/m³, respectively. The R2 values of daily, monthly, seasonal, and annual average PM 10 estimated by the model were 0.87, 0.91, 0.94, and 0.94, respectively. The areas with high PM 10 concentrations are mainly in northern China, especially in the North China Plain, and the peak value of daily average PM 10 can reach 91 μg/m³; the Intraday variation of PM 10 in southern China ranges from 67 μg/m³ to 72 μg/m³. A large-scale dust weather process was analyzed. Based on the AOD-PM 10 model, the contribution of long-range transport dust to PM 10 in China and Northern China were 25.6% and 38.1%, respectively. The PM 10 measured by the station and estimated by the DF model indicated good consistency. • An interpretable deep forest model was used to estimate the hourly PM 10 in China. • A 10-fold cross validation showed that the R2 of hourly PM 10 was 0.82–0.88. • The R2 values of monthly and seasonal average PM 10 were 0.91 and 0.94, respectively. • AOD contributed highest (∼13.5%) to the importance of features in the model. • The contribution of long-range transport dust to PM 10 in northern China was 38.1%. [ABSTRACT FROM AUTHOR]

Carbon monoxide (CO) has notable effects on the atmospheric environment and human health. This study used Himawari-8 top-of-atmosphere radiation (TOAR) data, meteorological factors, and geographic information, combined with an interpretable deep forest (DF) model, to obtain high-spatiotemporal resolution near-surface CO concentrations (temporal resolution: 1 h; spatial resolution: 0.05°) in China from September 2015 to August 2021 and reveal the distribution and variation in city-level CO hotspots. Based on the feature importance method and the spectral absorption characteristics of pollutants, the optimal combination of Himawari-8 band for the TOAR-CO model was selected. The performance comparison of five machine learning models showed that the DF model was the most performant. The hourly average R2 value (using 10-fold cross-validation) of the TOAR-CO model ranged from 0.72 to 0.78 (10:00 to 15:00, Beijing time); the R2 values of daily, monthly, and annual averages were 0.79, 0.97 and 0.98, respectively. The feature importance of the model varied in the different seasons and regions, with TOAR and meteorological factors had significantly contributions to the model. During the study period, the CO concentration in most areas of China showed a decreasing trend, with an average decline rate of 7.2%·yr−1. Despite CO being a stable, long-lived gas, the TOAR-CO model still captured obvious intraday variation in CO, which reached a peak of 0.89 mg/m3 at 10:00 and dropped to 0.69 mg/m3 at 15:00. The spatiotemporal distribution of estimated CO was basically consistent with satellite and reanalysis data. The TOAR-CO model predictions reflected city-level CO pollution, and the distribution of CO hotspots was closely related to industrial structure and human activities. During the COVID-19 pandemic, CO concentration significantly dropped after the 4th week of control. The CO concentrations in the hotspot cities decreased more significantly, by over 50%. The research results provide some supports for the assessment of CO fine-scale pollution and related research. [Display omitted] • Using an interpretable deep forest model to obtain ground fine-scale CO in China. • The contribution of input features to the model existed spatiotemporal differences. • Despite the stable nature of CO, but the model still captured its diurnal variation of 22.5%. • CO city-level hotspots were mainly distributed in areas with frequent human activities. • During the COVID-19 control period, CO hotspots in China had a greater decline. [ABSTRACT FROM AUTHOR]

Accurately identifying and classifying aerosols is a key to understanding their sources, assessing their chemical and physical changes, and comprehending their feedback mechanisms within climate system. However, existing aerosol classification methods have two limitations: First, ground remote sensing products have limited spatial coverage, and polar orbit satellite data suffer from low temporal resolution. Second, satellite-based aerosol classification methods are affected by cloud interference, resulting in data gaps. To overcome these challenges, this study utilized the extreme tree algorithm (ET) and integrated the FY-4A TOAR dataset with meteorological and geographical information, using a sequential forward selection method for optimal input feature selection. Focusing on the Asian region (70–140°E, 15–55°N), three models were used to generate hourly aerosol classification products: the Clear air-Cloud-Aerosol-Mixed cloud and aerosol (CCAM-ET) model, the Dust-Polluted dust-Other aerosols-Mixed aerosols (DPOM-ET) model under clear sky conditions, and the DPOM-ET model under non-clear sky conditions. The evaluated accuracy of the CCAM-ET model was 85%, demonstrating its effectiveness in identifying cloud and aerosol under various conditions. The CCAM model can achieve an accuracy of 78% when aerosols are located above clouds and 59% when they are below clouds. The DPOM-ET model achieved an evaluation accuracy of 89% for clear sky areas and 87% for non-clear sky areas, respectively, and was superior to traditional methods. Long-term data indicated a significant correlation between the distributions of different types of aerosols and human activity. Polluted dust is common in southwest China during spring and in northern China during winter. Additionally, from 2018 to 2020, dust occurrences decreased on the Indian Peninsula, while non-dust aerosols increased significantly in India and Southeast Asia. • An hourly aerosol classification model was built for non-clear-sky conditions. • The model has an 85% accuracy rate in classifying aerosols and clouds. • The model identifies dust more accurately than traditional methods. • Long-term data revealed a significant link between aerosol types and human activity. [ABSTRACT FROM AUTHOR]