Aerosol observation data assimilation is key to atmospheric environmental prediction. The original CMA chemistry–weather (CMA-CW) 3DVar data assimilation system uses simply fine and coarse particulate matter (PM_2.5 and PM_2.5–10) as control variables, with only concentrations of PM_2.5 and PM₁₀ being assimilated. In this study, a new 3DVar assimilation framework is developed, which considers seven aerosol species (black carbon, organic carbon, soil-dust, sea salt, sulfate, nitrate, and ammonium, instead of PM_2.5 and PM_2.5–10) in refined size segments as control variables, aiming to direct assimilate more aerosol related variables such as aerosol optical depth and lidar extinction and to provide accurate chemical initial fields for the CMA-CW model. Moreover, in the new assimilation framework, a novel equal-proportion distribution and compensation for background error increment is proposed and implemented, which produces physically more reasonable background error and thereby more reliable atmospheric–chemical analysis increments. The new assimilation framework is validated by idealized and real-case assimilation experiments, with reasonable performance. Based on the CMA-CW coupled model with the new assimilation framework and a rapid update cycle configuration, five-day CMA-CW simulation experiments for a widespread heavy fog–haze event in winter 2016 are carried out. The results demonstrate that assimilation of the surface aerosol observation data significantly improves the short-time forecast of atmospheric pollutants, due to refined and more precise information on aerosol compositions brought by the new assimilation framework. Meanwhile, the surface aerosol data assimilation also makes a positive contribution to visibility forecasting, significantly improving the visibility forecast in the heavy pollution areas.

Artificial intelligence (AI) has emerged as a promising alternative to traditional numerical weather prediction (NWP) models. However, the mismatch between training and operational input data limits its practical application in real-time forecasting. This study overcomes this challenge by fine-tuning the FuXi model using the Global/Regional Assimilation and Prediction Enhanced System (GRAPES) reanalysis data, resulting in an adapted version termed FuXi-GRAPES (FuXi-G). FuXi-G was evaluated over the full calendar year 2021. Compared to its baseline configuration, FuXi-G demonstrates marked improvements in forecast skill, particularly in the Southern Hemisphere and the tropics, achieving temporal and spatial accuracy that rivals or slightly surpasses that of NCEP operational forecasts. The FuXi-G model also exhibited a notable reduction in error propagation, outperforming NCEP forecasts by Day 5 and alleviating seasonal forecast biases in the Southern Hemisphere. A case study of Typhoon Khanun underscored the model’s enhanced ability to predict high-impact weather events, including a critical directional shift of the typhoon associated with the breakdown of the subtropical high. These results suggest that fine-tuning AI models can substantially improve forecasting accuracy while avoiding the computational burden of retraining on new datasets, providing a scalable approach for implementing AI models in operational weather forecasting.

For numerical weather prediction (NWP), data assimilation (DA) combines short-term forecasts and various atmospheric observations to achieve optimal initial conditions, based on which subsequent forecasts are launched. With the rapid advancements in numerical models and observing systems, DA has been significantly evolved. Modern methods now can account for uncertainties of state variables across various spatiotemporal scales, incorporate multiscale observation error statistics, and enforce dynamical constrains and model balances. Meanwhile, observations from various platforms, such as ground-based, aircraft, and satellite, have been assimilated. These include data from polar-orbiting and geostationary satellites, radar-derived radial winds and reflectivity, Global Navigation Satellite System (GNSS) radio occultations, etc. To further utilize the advanced observing systems and DA techniques for high-impact weather predictions, target observation strategies have been developed to identify areas where additional observations can yield the greatest predict improvements. Based on the advancements of DA theories and methods, China’s operational systems have made significant progress, establishing advanced operational DA systems. Over the past decade, the forecast skill of 5-day global weather prediction has improved by approximately 15%. The article reviews a century of development in DA, and discusses future directions, including the advanced DA methods, operational frameworks, integration of novel observations, and the synergy between DA and artificial intelligence.

The radiance data of satellite microwave radiometers play a more and more important role in the assimilation of numerical prediction systems due to the all-sky advantage of satellite observations and the all-weather availability of microwave soundings. As an important category of passive microwave radiometer, the application potential of microwave imager in numerical weather forecasting needs further verification and more sufficient exploration. Focusing on the Advanced Microwave Scanning Radiometer 2 (AMSR2) carried on the Global Change Observation Mission–Water (GCOM-W), the thinning scheme within 200 km radius is applied; a quality control scheme including nine issues of checks is developed to screen those contaminations including the sun-glint phenomenon and radio-frequency interferences which can disturb the low-frequency channels; a bias correction scheme on the basis of conventional predictors is employed to effectively reduce systematic deviations of instrument; the difficulty related to observation error evaluation is overcome by posteriori verification. Ten channels of GCOM-W AMSR2 are directly assimilated into the Global/Regional Assimilation and Prediction System–Global Forecast System (CMA_GFS, i.e., the GRAPES_GFS) version 3.0 by the four-dimensional variational (4DVar) assimilation method independently developed by the Numerical Weather Prediction Center of China Meteorological Administrator. A pair of batch experiments for one month indicates that, with the GCOM-W AMSR2 assimilation, the depiction of humidity analysis field is improved, the medium forecasting skills of various precipitation are also improved, and the prediction score card shows obvious positive impacts on the southern hemisphere and the equatorial region. Thus, direct assimilation of the GCOM-W AMSR2 into the CMA_GFS 4DVar is confirmed to be useful for improving the amount of observations in poor-data regions, and it can take advantage of the water vapor sensitivity to promote the skills of humidity analysis and precipitation prediction.