@article{HU2026, 
author = {Yiwen HU and Zengliang ZANG and Wei DAI and Yi LI and Wei YOU and Lang LIU and Ning LIU and Qun LONG},
title = {Advances and prospects of atmospheric chemistry data assimilation},
year = {2026},
journal = {Journal of National University of Defense Technology},
volume = {48},
number = {2},
pages = {214-227},
keywords = {data assimilation, atmospheric chemistry model, initial condition assimilation, emission source assimilation},
url = {https://www.sciopen.com/article/10.11887/j.issn.1001-2486.25110009},
doi = {10.11887/j.issn.1001-2486.25110009},
abstract = {SignificanceAtmospheric chemical models serve as a crucial tool for accurately forecasting air quality, analyzing pollution causes, and formulating and evaluating emission reduction policies. The forecast results of key atmospheric variables derived from atmospheric chemical models, such as aerosols, clouds, precipitation, radiation, and visibility, provide important support for the atmospheric environmental assurance of military operations including missile guidance, aviation flights, and aerostat station-keeping. However, due to inherent uncertainties in the emission sources, initial conditions, boundary conditions, and physicochemical processes incorporated in atmospheric chemical models, notable discrepancies persist in current simulation and prediction results. Atmospheric chemical DA (data assimilation) is crucial for improving the accuracy of atmospheric chemical predictions by integrating model outputs with multi-source observational data, especially for the optimization of initial conditions and emission sources. Therefore, DA is an important technical method for improving the accuracy of atmospheric chemical predictions and possesses considerable practical and scientific significance for atmospheric pollution prevention and control as well as the atmospheric environmental support of military operations.ProgressAtmospheric chemical data assimilation optimizes the initial field of pollutant concentrations and refines atmospheric pollutant emission sources, thereby enhancing the predictive accuracy of numerical models. Common assimilation methods include OI (optimal interpolation), three-dimensional variational (3D-Var), four-dimensional variational (4D-Var), and EnKF (ensemble Kalman filter). Initial field assimilation improves the accuracy of the initial pollutant concentration field and optimizes the initial conditions of atmospheric chemical models by assimilating observational data on atmospheric pollutant concentrations. For the initial field data assimilation of gaseous pollutants, the direct assimilation of their observational data generally reduces errors in their initial fields and significantly enhances the performance of short-term forecasts. For the assimilation of aerosol initial fields, early studies typically aggregated all aerosol variables into a single state variable for assimilation, then allocated the increment field based on the proportion of each variable in the model. However, this approach fails to account for structural differences in the increment fields of individual chemical components and cannot assimilate observational data specific to each component. With the advancement of observational technologies, the volume of observational data from satellites, radars, and other platforms has increased substantially, which not only provides abundant data support for pollutant initial field assimilation but also poses new challenges to data assimilation techniques. Therefore, the integrated assimilation of multiple variables involving diverse species and particle size bins has become a prevailing development trend. Existing studies have successfully achieved the assimilation of non-conventional observational data such as AOD (aerosol optical depth) and extinction coefficients, demonstrating that the assimilation of initial atmospheric chemical concentration fields plays an indispensable role in improving air quality predictions and elucidating the key physical and chemical processes underlying pollution events.In terms of emission source assimilation inversion, DA leverages observational data on pollutant concentrations to optimize emission inventories compiled via traditional methods, yielding optimized inventories with higher accuracy and faster update rates, which in turn effectively improves the predictive performance of atmospheric chemistry models. EnKF and 4D-Var methods are the most commonly used techniques for atmospheric emission source assimilation. Numerous scholars have conducted research on emission source assimilation systems based on these two methods, and have verified the effectiveness of atmospheric emission source assimilation in enhancing the accuracy of emission sources and the performance of air quality predictions through approaches such as model simulation and forecasting. The research scope of emission source inversion has expanded gradually from pollutants with relatively simple chemical reactions, including CO (carbon monoxide), sulfur dioxide (SO2), and dust, to the emission sources of key atmospheric pollutants involving complex nonlinear chemical reactions, including nitrogen oxides (NOx), VOCs (volatile organic compounds), and fine particulate matter (PM2.5). Moreover, the spatiotemporal resolution of emission source inversion has been continuously improved, with the spatial resolution reaching the ten-kilometer scale and the temporal resolution achieving hourly-scale refined inversion. Existing studies have shown that both EnKF and 4D-Var methods can significantly reduce the uncertainties of emission sources, improve the accuracy of pollutant prediction, and play a crucial supporting role in pollution source apportionment, formulation of emission reduction policies, and evaluation of emission reduction effects.Conclusions and ProspectsAtmospheric chemical DA (data assimilation) is exhibiting a development trend toward refinement, multi-scale expansion, and multi-source synergistic assimilation, with its technical focus gradually shifting to the refined direct assimilation of aerosols and the synergistic inversion of pollution emissions across multiple sectors. The horizontal resolution of atmospheric environmental satellite detectors is continuously improving, and an integrated observation network based on synergistic detection by multi-instrument and multi-platform systems is being progressively constructed. This progress poses increasingly stringent requirements and challenges for atmospheric chemical assimilation research. How to fully integrate high-resolution geospatial remote sensing data into atmospheric chemical DA constitutes a core challenge confronting the field at present, while the in-depth integration of DA with AI (artificial intelligence) algorithms represents a key research direction to break through this technical bottleneck and substantially enhance the accuracy of atmospheric composition analysis and prediction.ML (Machine learning) algorithms possess the capability to extract complex spatiotemporal features from massive datasets, and their computational efficiency following model training is considerably higher than that of traditional numerical models and DA models. Nevertheless, ML lacks clear physical interpretability; its outputs are constrained by the quality of training samples, and it is difficult to apply in regions lacking observational samples (i.e., labeled data). Notably, ML and DA methods not only share mathematical similarities but also exhibit complementary advantages, with theoretical consistency under the Bayesian framework. The integration of these two approaches is expected to overcome the bottlenecks of traditional DA methods in processing multi-source heterogeneous data and addressing low computational efficiency. This integration will further drive the evolution of DA systems toward greater efficiency and intelligence, thereby enabling more accurate analysis and prediction of atmospheric composition.}
}