Satellite observations of tropospheric nitrogen dioxide : retrieval, interpretation, and modelling

The research questions set out in Chapter 1 that guided the investigation in this thesis are repeated here. The answers to these questions contain the most important conclusions of the various chapters and are given below. 1. How can we retrieve accurate information on total and tropospheric NO2 from the backscatter UV-VIS measurements by the Ozone Monitoring Instrument? 2. Can we develop a concept that relates a vertical distribution of a trace gas to a satellite observed column? 3. What is the error buget for tropospheric NO2 retrievals for an operational backscatter instrument like GOME? 4. How can we use GOME tropospheric NO2 observations to constrain estimates of the tropical lightning NOx production? The first research question is topic of chapter 2. One of the ‘Science questions’ posed in the framework of the EOS-AURA mission is: "Is air quality changing?". A contribution to anwering this question may be provided by accurate and precise measurements of tropospheric NO2 columns from the Ozone Monitoring Instrument that become widely available for scientific purposes. To this end, an algorithm for the retrieval of total and tropospheric NO2 has been developed and tested in Chapter 2. This algorithm has been implemented for the operational retrieval of the standard KNMI/NASA OMI NO2 "product". The OMI algorithm builds on the heritage of GOME tropospheric NO2 retrievals, and contains a number of important improvements over previous algorithms. In principle, one of these improvements is the extended size of the spectral fitting window: sensitivity studies indicated that a significant reduction in the slant column uncertainty can be attained for a 405-465 nm window. The derivation of the stratosferische background is improved by accounting for variability along a zonal band. Application of a low-pass filter approach on GOME data shows significant variability in stratospheric NO2 along locations of the same latitude that would lead to otherwise significant systematic errors. Climatological NO2 profiles can be used for troposferic air mass factor calculations. In situations of urban pollution, climatological NO2 profiles simulated with TM3 do not show large variability in their shape. In situations of biomass burning and outflow of NOx-related pollution over comparatively clean areas, the vertical distribution is quite different with NO2 peaking at higher altitudes. We estimate that OMI NO2 columns will be retrieved with a precision of approximately 5% for unpolluted situations, largely due to the uncertainty in the spectral fitting. For situations with NOx levels in the troposphere that are far above background, we expect to measure tropospheric columns with errors up to 60%, largely due to retrieval assumptions on the state of the atmosphere. In Chapter 3, the second research question is adressed. This chapter discussed the strong height-dependent sensitivity of a satellite instrument to a tracer density is discussed in relation to the averaging kernel. This sensitivity was already topic of Chapter 2, but is here discussed in the context of general retrieval theory as developed by Rodgers. It is shown that the averaging kernel provides a direct interpretation of the satellite retrieved column density to users. For intercomparisons with independent data, such as vertical profiles from models or validation measurements, the dependence on a priori assumptions about the profile shape dissapears when the averaging kernel is used. The third Chapter on the retrieval of tropospheric NO2, is Chapter 4. In this Chapter, an extensive error analysis of tropospheric NO2 retrievals is presented in order to answer the third research question. It is shown that GOME tropospheric NO2 retrievals have errors in the 35-60% range, largely determined by air mass factor errors. The most important errors -in order of importance- are errors due to uncertainty in model parameters such as clouds, surface albedo and a priori profile shape. Apart from the error analysis, a number of retrieval improvements has been suggested in Chapter 4. Most relevant is a new method to estimate the stratospheric background from an assimilation approach. This approach has the advantage of accounting for dynamical features in stratospheric NO2, and reduces the otherwise large systematic error in the estimate of stratopheric NO2. A correction for the temperature-dependence of the NO2 cross-section is demonstrated to remove systematic errors on the order of 10%. Finally, we conclude that a correction for the presence of aerosols needs to be accompanied by aerosol corrections in cloud retrieval schemes. Chapter 5 relies on the previous chapters and focuses on the fourth and last research question. In Chapter 5, columns and their error estimates (Chapter 4) are used in an extensive comparison -through the averaging kernel (Chapter 3)- with modelled lightning NO2 columns in order to test lightning parametrisations in TM3 and to impose top-down constraints on the global lightning NOx production. First, it is shown that tropospheric measurements by GOME are sensitive to NO2 produced by lightning. Tropospheric NO2 columns show a rapid increase with the fifth power of the cloud top height for clouds with tops higher than 6.5 km. This estimate of the cloud height-dependence of LNO2 is consistent with the observed power-law relationship of lightning frequencies and cloud top height. Second, a statistical comparison of simulated LNO2 and observed NO2 columns in the tropical region between 40??S and5??N shows that the TM3 model is well capable of reproducing observed patterns of LNO2. This is true for two different lightning parameterisations in TM3. Moreover, the absolute values of modelled and observed (L)NO2 are in good agreement over tropical continents. However, over tropical oceans, the model appears to overestimate the LNO2 contribution to the total tropospheric column. This model bias is likely due to assumptions on the assumed energy ratio (10:1) between cloud-to-ground and intra-cloud lightning, and on the assumed ratio (10:1) between continent-to-ocean convective intensity. For the scheme based on convective precipitation, there are significant regional differences in rainfall-to-lightning ratios that may also lead to the bias over the tropical ocean. From rescaling the modelled LNOx production between 40??S and 5??N, we arrive at a LNOx production estimate of ??1.0 Tg[N] in the 40??S118 Summary, conclusions, and outlook 5??N region. Based on assumptions for rescaling factors in the rest of the world, the global LNOx production in 1997 is estimated to be in the 1.1-6.4 Tg[N] range.