Comparison of OMI NO<sub>2</sub> tropospheric columns with an ensemble of global and European regional air quality models
Abstract. We present model results for tropospheric NO2 from 9 regional models and 2 global models that are part of the GEMS-RAQ forecast system, for July 2008 to June 2009 over Europe. These modeled NO2 columns are compared with OMI NO2 satellite retrievals and surface observations from the Dutch Air Quality Network. The participating models apply principally the same emission inventory, but vary in model resolution (0.15 to 0.5°), chemical mechanism, meteorology and transport scheme. For area-averaged columns only a small bias is found when the averaging kernel is neglected in the comparison to OMI NO2 columns. The reason for this is that TM4 a priori profiles have higher NOx concentrations in the free troposphere (where sensitivity to NO2 is high) and higher NOx concentrations in the surface layers (where sensitivity to NO2 is low) than RAQ models, effectively cancelling the effect of applying the averaging kernel. We attribute these low NO2 concentrations in the RAQ models to missing emissions from aircraft and lightning. It is also shown that the NO2 concentrations from the upper part of the troposphere (higher than 500 hPa) contribute up to 20% of the total tropospheric NO2 signal observed by OMI. Compared to the global models the RAQ models show a better correlation to the OMI NO2 observations, which are characterized by high spatial variation due to the short lifetime for NO2. The spread in the modeled tropospheric NO2 column is on average 20–40%. In summer the mean of all models is on average 46% below the OMI observations, whereas in winter the models are more in line with OMI. On the other hand the models on average under-predict surface concentrations in winter by 24% and are more in line with observations in summer. These findings suggest that OMI tropospheric columns in summer over polluted regions are biased high by about 40%. The diurnal cycle and profiles in the regional models are well in line, and the profile shapes correspond well to results from the global models. The analyses against OMI observations have proven to be very useful to initiate model improvements, and to quantify uncertainties in the retrieval product.