Abstract:With an increasing number of air quality monitoring stations installed around the Chinese mainland, high-resolution aerosol observations become available, allowing improvements in air pollution monitoring and aerosol forecasting. However, the multi scales (especially small-scale) information included in high-resolution aerosol observations could not be effectively utilized by the traditional three-dimensional variational method (3DVAR). This study attempted to extend the traditional 3DVAR to a multi-scale 3DVAR with two iteration steps, two-scale-3DVAR (TS-3DVAR), to improve the effectiveness of assimilating high-resolution observations. In TS-3DVAR, the large-scale and small-scale components of observation information were decomposed from the original high-resolution observations using a Gaussian smoothing method and then assimilated using the corresponding large-scale or small-scale background error covariances which were derived from the partitioned background error samples. The data assimilation (DA) analysis field generated by TS-3DVAR is more accurate than 3DVAR in reproducing the field’s multi-scale characteristics, which could thus be used as the initial chemical field of the air quality model to improve aerosol forecasting. Particulate matter with an aerodynamic diameter of less than 2.5 μm (PM 2.5 ) and 10.0 μm (PM 10 ) from the surface air quality monitoring stations from November 01 to November 30, 2018 at 00:00 were assimilated daily to verify the effects of TS-3DVAR and 3DVAR on the aerosol analysis and forecast accuracy. The results showed that TS-3DVAR better constrained both large-scale and small-scale, especially the spatial wavelengths in a range of 54–216 km and those above 351 km. The average power spectra of the TS-3DVAR assimilation increment in the two wavelength ranges were 71.70% and 35.33% higher than those of 3DVAR. As a result, the TS-3DVAR was more effective than 3DVAR in improving the accuracy of the initial chemical field, and thereby the forecasting capability for PM 2.5 . In the initial chemical field, the 30-day average correlation coefficient (Corr) of PM 2.5 of TS-3DVAR was 0.052 (6.12%) higher than that of 3DVAR, and the root mean square error (RMSE) of TS-3DVAR was 3.446 μg m −3 (16.4%) lower than that of 3DVAR. For the forecasting capability for PM 2.5 mass concentration, the 30-day average Corr of TS-3DVAR during the 0–24 hour forecast period was 0.025 (5.08%) higher than that of 3DVAR, and the average RMSE was 2.027 μg m −3 (4.85%) lower. The positive effect of TS-3DVAR on the improvement of forecasting capability can last for more than 24 h.

3DVar sectoral emission inversion based on source apportionment and machine learning

Study on the 3dvar Emission Inversion Method Combined with Machine Learning in CMAQ

Improvement of PM2.5 Forecast in China by Ground-Based Multi-Pollutant Emission Source Inversion in 2022

Development of four-dimensional variational assimilation system based on the GRAPES–CUACE adjoint model (GRAPES–CUACE-4D-Var V1.0) and its application in emission inversion

Sector‐Based Top‐Down Estimates of NOx, SO2, and CO Emissions in East Asia

Combining physical mechanisms and deep learning models for hourly surface ozone retrieval in China

Optimal estimation of initial concentrations and emission sources with 4D-Var for air pollution prediction in a 2D transport model

Multi-scale three-dimensional variational data assimilation for high-resolution aerosol observations: Methodology and application

The optimization of SO 2 emissions by the 4DVAR and EnKF methods and its application in WRF-Chem

Remote Sensing Fine Estimation Model of PM2.5 Concentration Based on Improved Long Short-Term Memory Network: A Case Study on Beijing–Tianjin–Hebei Urban Agglomeration in China

Ozone Concentration Estimation and Meteorological Impact Quantification in the Beijing‐Tianjin‐Hebei Region Based on Machine Learning Models

Supplementary Material to "tracking Daily NOx Emissions from an Urban Agglomeration Based on TROPOMI NO2 and a Local Ensemble Transform Kalman Filter"

Investigating the Changes in Air Pollutant Emissions over the Beijing-Tianjin-Hebei Region in February from 2014 to 2019 through an Inverse Emission Method

Inverse modeling of Chinese NOx emissions using deep learning: Integrating in situ observations with a satellite-based chemical reanalysis

Remotely sensed and surface measurement- derived mass-conserving inversion of daily NOx emissions and inferred combustion technologies in energy-rich northern China

Lidar data assimilation method based on CRTM and WRF-Chem models and its application in PM2.5 forecasts in Beijing

An Inversion Framework for Optimizing Non-Methane VOC Emissions Using Remote Sensing and Airborne Observations in Northeast Asia During the KORUS-AQ Field Campaign

Guiding the Data Learning Process with Physical Model in Air Pollution Inference.

Understanding the spatial and seasonal variation of the ground-level ozone in Southeast China with an interpretable machine learning and multi-source remote sensing

Four-dimensional Variational Assimilation for SO2 Emission and Its Application Around the COVID-19 Lockdown in the Spring 2020 over China

A Novel Short-Term PM2.5 Forecasting Approach Using Secondary Decomposition and a Hybrid Deep Learning Model