Data assimilation of surface air pollutants (O3 and NO2) in the regional-scale air quality model AURORA

In the present work, a bias-aware optimal interpolation in conjunction with the Hollingsworth–Lönnberg method to estimate error covariance matrices was applied as data assimilation algorithm in the regional scale air quality model AURORA to assimilate ground level O3 and NO2 concentrations. The study was conducted over the domain Belgium including part of its neighbouring areas with grid resolution of 3 × 3 km2. Data assimilation was carried out for the retrospective simulation in post-processing (offline) mode for a summer and a winter month. Observations were provided by the AIRBASE data archive. Since the air quality model AURORA is presumed to represent background conditions, only the background stations within the domain have been taken into account. The validation of the proposed method was carried out by leaving observations of ten monitoring stations out in one run of the data assimilation process and another ten stations out in the next run and so on (a “leave ten out approach”). The proposed method has been evaluated in both spatial as well temporal domain against various statistical indicators such as correlation coefficient (CORR), root mean square error (RMSE), index of agreement (IOA) and mean fractional bias (MFB). For both the O3 and NO2, the extensive validation results have clearly shown substantial improvement in the data assimilation results over AURORA free run in both the seasons. The results over 70 validation stations show that CORR increased from 0.4 to 0.8 for O3, 0.3 to 0.6 for NO2 while average RMSE reduced from 27.9 to 12.6 for O3 and from 17.4 to 11.0 for NO2 for the month of June. Similar improvements have been observed for the month of Dec as well. Spatial CORR, IOA for monthly means of both the O3 and NO2 concentrations were also increased considerably. The results clearly indicate that the applied bias aware optimal interpolation in conjunction with Hollingsworth–Lönnberg method is a very promising candidate for the statistical correction of regional scale air quality modelling results for the retrospective simulation.