Dr Remediosopened the discussion of Dr Burrows’s paper: Dr. Burrows has shown results from measurements of three species: ozone, nitrogen dioxide and formaldehyde. They imply that further information separating sources such as biomass burning from biogenic emission and stratosphere–troposphere exchange would be very useful. Measurements of carbon monoxide (CO) from the MOPITT instrument can provide a diagnostic of biomass burning signals, assuming that large enhancements of its concentrations are due to this source alone. In principle, the CO data also provide tracer information, allowing for evolution of CO in biomass plumes, with MOPITT daytime “surface” data over land demonstrating lower-to-mid-troposphere transport and nighttime “surface” measurements over ocean indicating mid-to-upper troposphere transport; the “surface” retrieved level for MOPITT data can shift significantly in altitude sensitivity and so characterisation data or averaging kernels have been used to interpret the data (not shown). In the figure shown for September 2000, we see that nighttime CO data indicate upper troposphere outflow over Madagascar, similar to the transport indicated in the paper albeit for a different year. We also observe a strong pattern of CO sources in western to central Africa (indicated by “daytime − nighttime surface” CO over land which our analyses have shown can be sensitive to near-surface enhancements of CO), raising questions of relations between fire types and emissions of trace gases both as primary and secondary products. A further study to the paper presented would therefore be to combine results from GOME or SCIAMACHY on ENVISAT with analyses of MOPITT data.Retrieved “surface” level carbon monoxide (CO) data from the measurements of pollution in the troposphere (MOPITT) instrument on EOS-Terra. Data are global monthly means for September 2000: (a) daytime measurements; (b) nighttime measurements; (c) difference between daytime and nighttime measurements (J. J. Remedios, N. A. D. Richards, U. Friess)Professor Burrowsreplied: The comment and question are very relevant. For our study, which focuses on the September 1997, unfortunately there are no MOPITT data available as it was launched aboard the NASA Terra satellite in December 1999 and began measurements in 2000. Clearly the assimilation of satellite data from MOPITT and GOME and SCIAMACHY for the period from August 2002 to June 2003 when all are working represents a potentially very interesting study. This was however outside of the scope of this study. This should stimulate more studies using the complete set of data from MOPITT, GOME and SCIAMACHY. MOPITT CO data of a different El Niño year (e.g.2002) can be compared with the results of the transport analysis for September 1997. The MOPITT data from September 2002 indicates an eastward outflow of biomass burning emissions over Madagascar, not very dissimilar to that observed by GOME for NO2.Dr Chipperfieldsaid: The authors say that 1997 is studied due to a ‘strong signal’. What do they mean exactly by this—what quantities are large in this period? This may have implications for the discussions which compared this paper with other GOME periods.Professor Burrowsreplied: The 1997 El Niño Southern Oscillation (ENSO) event was the largest yet recorded. Over Indonesia but also over the Horn of Africa and related regions of East Africa, the El Niño phase of the ENSO results in low rainfall and consequently the amount of fire both natural and anthropogenic in origin was high. This results in large biomass burning/biofuel emissions and the month of September 1997 is a particularly good example of this phenomena. Overall a significant enhancement of tropospheric columns of O3, NO2, and HCHO derived from GOME is observed during September 1997, when compared to the wetter September of 1998. The total column O3 retrieved from GOME agree well with that determined the sonde network, SHADOZ (Thompsonet al.), which started in 1998. Over a significant area of Africa and the western parts of the Indian Ocean the tropospheric O3column during September 1997 exceeded that observed in September 1998 by 25 to 100%. So as a result of the strong ENSO 1997 was an extraordinary year in the tropics and the tropospheric amounts of O3, HCHO and NO2were large.1 A. M. Thompson, J. C. Witte, R. D. McPeters, S. J. Oltmans, F. J. Schmidlin, J. A. Logan, M. Fujiwara, V. W. J. H. Kirchhoff, F. Posny, G. J. R. Coetzee, B. Hoegger, S. Kawakami, T. Ogawa, B. J. Johnson, H. Vömel and G. Lebauw,J. Geophys. Res., 2003,108(D2), 8238, DOI: 1029/2001JD000967.Mr Pfrangcommunicated: I would like to comment on the differences between retrieval and modelling results. Meyer-Arneket al. conclude that uncertainties in the meteorological input data (ERA-40 re-analysis data from ECMWF) are the main reason for inaccuracies in the model representation of transport processes in the upper troposphere without sufficiently taking into account simplifications connected with the transport modelling approach employed in the study.In discussions with Johannes Flemming and Sakari Uppala at ECMWF, a number of questions emerged that would benefit from clarification. First, it remains unclear how vertical exchange processes are treated by the model as we assume that the kinematic trajectories are only isobaric. Convection is highly important in the tropics and may force the trajectories to be located on changing height levels. Secondly, we would be interested to learn how the starting levels for trajectories were chosen and how sensitive is the model with respect to changes in the apparently arbitrarily selected pressure levels. The latter point might be of particular interest when considering uncertainties in injection heights of fires. Thirdly, it is mentioned in the paper that 230 000 trajectories were calculated, but it remains somewhat unclear how these trajectories were chosen and partitioned between the different pressure levels and what effect a different choice and vertical partitioning of trajectories would have had on the results of the model. Fourthly, assuming chemical processes are of importance, why would the maximum trajectory density necessarily coincide with the maximum of the tropospheric O3column as indicated in the discussion on page ten? If the ERA-40 data would remain the main reason for inaccuracies in the transport model it would be very useful to quantify the error of the ERA-40 data itself and the error due to temporal interpolation to enable the ECMWF to analyse and possibly further improve the next generation re-analysis data.To summarise, we are certainly aware of uncertainties in the ERA-40 data, but it appears to be premature to conclude from the results presented in this paper that inaccuracies detected in the representation of transport processes in the model are solely due to the meteorological data provided by the ERA-40 re-analysis (as stated on the bottom of page thirteen).ECMWF highly appreciates your use of ERA-40 re-analysis data, since constructive criticism from independent researchers is one of the important tools for the improvement of our products. We would like to draw your attention to a survey of current and future applications of the ERA-40 re-analysis data which is now available athttp://www.ecmwf.int/research/era/era40survey/. We hope the ERA-40 survey will prove valuable to (i) assess the impact of the ERA-40 re-analysis project, (ii) facilitate exchange of information between ERA-40 users, (iii) compile a review of research work using ERA-40 and (iv) make a case for a more comprehensive and longer re-analysis later this decade, which will benefit from the lessons learned from ERA-40. If you are aware of any project using ERA-40 data not being included in the current version of the ERA-40 survey, we would be most grateful if you contact us at era40survey@ecmwf.int.Professor Burrowscommunicated in reply: We would like to thank the ECWMF for their comments. We used ERA-40 data because we consider it to be suitable and an excellent set of data. However as pointed out on the ERA-40 web site, there appears to excessive tropical oceanic precipitation in the ERA-40. This indicates that in the tropics there are some as yet unresolved issues. Although total ozone has been assimilated into ERA-40 from satellites, this is dominated by the stratospheric ozone burden. In our study the tropospheric columns of O3, NO2and HCHO, retrieved from the measurements of the satellite GOME during the strong ENSO of September 1997 are presented and compared with trajectory calculations, determined using the ERA-40 data with and without chemistry calculated along trajectories. This approach has its limitations.The first question raised concerns the relative importance of convection. Clearly this is of great potential significance in the tropics. If the ERA-40 data set effectively smooth over the convective activity, then trajectories represent some average picture. Convection cannot be explicitly considered in the trajectories because it occurs typically at scale smaller than the grid, used for the calculation of the trajectories. However the winds fields of ERA-40 should represent the larger scale behaviour.To investigate the potential importance of convection for the African outflow into the Indian Ocean, we have used the Lifted Index, LI, and assumed that all negative LI are strongly convected. This is considered to be a worst case. Whilst some redistribution of the trajectories is observed, there is still a significant difference in the pattern observed in GOME columns of O3, NO2and HCHO and that predicted for the outflow from Africa.The second question addresses the sensitivity of the starting level. The transport differs for trajectories released at different altitudes. Above Africa in the region selected for study, the majority of air parcels, released at lower altitudes are transported to the west, whereas those released at higher altitudes (200 hPa above the ground) are transported to the east. The “release” altitude of the trajectories was chosen to be at the top of the planetary boundary layer and at the bottom of the free troposphere, because this altitude is the best representation of the height of the airborne measurements used to initialise the chemistry model.For the case that the fire index is non-zero within a grid box, trajectories are started on a 1° × 1° horizontal grid. At each starting position trajectories are released at 50, 100, 150 and 200 hPa above the ground pressure. The meteorological data are represented on model levels.As shown inFig. 2, the overall distribution of the calculated trajectory shows similar flow out of East Africa at all the pressure altitudes. The chemistry calculated along the trajectories follows the same behaviour.The observations of tropospheric trace columns leaving Africa as determined from the GOME observations in September 1997 appear to be significantly further north of the flow taking trajectories and an estimate of convection into account (Fig. 3). In this case the Lifted Index is used to determine instable air and all such air masses are convected.The trajectory density calculated for the trajectories, leaving biomass burning regions at pressures of 50, 100 and 200 hPa.Trajectory density calculations for trajectories leaving the biomass burning region including an estimate of convection.The combined trajectory density for outflow from Africa is extended northwards but beyond longitudes of 60 °E is further south than the GOME observations. Clearly more data and years need to be carefully analysed to see whether September 1997 is a special case or representative of a trend.The third question addresses the selection of trajectories. As described in the text, all trajectories, which enter the region selected for observation were calculated.The fourth question raised questions whether the ozone formed should follow the trajectory density. Our conclusion is based on the fact that the column amounts of O3, NO2and HCHO as observed from GOME follow similar but not identical paths in the outflow from east Africa. The NO2flow is shifted northwards.In summary we agree that convection is an important process in the tropics. We have investigated what impact convection might have and conclude that it is unlikely to explain all the difference between calculated and observed in the outflow from east Africa for September 1997. Similarly we have argued that it is unlikely that inaccuracies in the retrievals of the three trace gas columns from GOME data are the source of the difference. In the longer term the assimilation of GOME data into ERA-40 may be the best way to resolve the origin of discrepancies between our relatively simple approach and account optimally for convection.Dr Monksaddressed Professor Burrows and Jaeglé:(1) Does the distribution of clouds/smoke plumes and the potential changes with seasonetc. bias the emission estimates? In a sense are you sampling different parts of column NO2?(2) Could you develop a spatial/vertical sampling metric, that could be superimposed on the data?Professor Jaegléreplied:(1) Clouds and smoke plumes do have the potential of affecting the retrieval of NO2columns from GOME. We take into account both effects in our air mass factor (AMF) calculation, but these introduce uncertainties in our tropospheric NO2retrieval: 28% for cloud and 30% from aerosols.1We have rejected scenes where the cloud radiance fraction exceeds that from clear sky, corresponding to cloud cover>40%. As a result of this threshold, no NO2columns are available over regions with large cloud cover during some months. This is can be seen in Fig. 3 of the paper during January, where no NO2columns are available over parts of eastern Europe and east Asia.(2) The spatial and vertical sensitivity of our retrieval of tropospheric NO2columns to clouds and aerosols has been discussed in previous papers by Martinet al.1,21 R. V. Martin, D. J. Jacob, K. Chance, T. P. Kurosu, P. I. Palmer and M. J. Evans,J. Geophys. Res., 2003,108, DOI: 10.029/2003JD003453.2 R. V. Martin, K. Chance, D. J. Jacob, T. P. Kurosu, R. J. D. Spurr, E. Bucsela, J. F. Gleason, P. I. Palmer, I. Bey, A. M. Fiore, Q. B. Li, R. M. Yantosca and R. B. A. Koelemeijer,J. Geophys. Res., 2002,107, DOI: 10.1029/2001JD002622.Professor Burrowsreplied:(1) To a reasonable first approximation in nadir viewing ultraviolet and visible electromagnetic radiation are backscattered from the top of clouds and therefore GOME measurements contain negligible information about trace gases below the cloud top. Cloud top height and cloud cover are routinely evaluated for all GOME data by using the O2A-Band absorption. The cloud fraction reveals the African west coast over the Atlantic are frequently covered with clouds. The cloud top height data indicates that these clouds remain at altitudes of less than 2 km, thus significant information from GOME data is available between 2 km and the tropopause. The region of strong smoke from a biomass burning region is usually small compared to a GOME ground pixel.The use of ground scenes, having <10% cloud cover in our analysis minimises the impact of cloud. Our composites represent cloud free regions between clouds. As we have demonstrated in previous work NO2from convective uplifting and lightning can be observed above clouds. In our comparisons of model and measurement, we are comparing as best we can cloud free regions, so we consider that there is not a bias. The 50% back scattered radiation criterion, used by Jaegleet al. is a different selection criterion. As cloud has a higher reflectance than the surface, then we consider that the criteria for cloud screening are effectively probably very similar.Emissions are expected to be seasonally dependent and are. In our study we have focussed on the large ENSO event of 1997, which resulted in dry conditions in Africa in September 1997.(2) The number of individual data points used for each grid point is known. An intrinsic assumption is that the fire pattern and biogenic emissions were reasonably constant during September 1997. This is of course a simplification but not unreasonable considering the size of the GOME ground scene. An evaluation of GOME data reveals that the instrument was sampling all parts of the examined region equally during September 1997.Dr Chipperfieldasked: The two groups using GOME data employ different thresholds (10%, 50%) for the exclusion of data due to cloud contamination. Why is there this difference?Professor Jaegléresponded: Our AMF calculation to convert slant columns to tropospheric columns, enables quantitative retrieval of partly cloudy scenes. The AMF calculation accounts for cloud scattering using cloud fraction, cloud top pressure, and cloud optical thickness observed by GOME and retrieved with the GOMECAT algorithm. Thus our analysis does not need to be restricted to cloud-free or very low cloud cover areas. We do eliminate regions where clouds contribute to more than 50% of backscattered radiation, which corresponds to cloud cover >40%. At these high cloud covers our retrieval becomes more uncertain.Professor Burrowsresponded: The University of Bremen group has developed trace gas tropospheric column data products for effectively cloud free ground scenes, because clouds block the penetration of solar radiation in the troposphere. Data products can also be generated for cloudy scenes. The UB group considers that cloud coverage of larger than 10% within a ground scene of GOME impacts significantly on the determination of the tropospheric columns. The UB retrievals have been validated by comparison with ground based and aircraft data. The approach by Jaegléet al. to deal with cloudy ground scenes is similar but somewhat different. As clouds typically have a significantly higher spectral reflectance than land or ocean, then the 50% reflectance criteria results in a small effective cloud fraction in a ground scene.Dr Sarkarsaid:(1) Thanks for your presentation. Please could you highlight an alternate method (other than TEM) to resolve the total column amount into tropospheric and stratospheric components, especially to analyse Indonesian Forest fire?(2) In India we have approximately 15 ground-based stations operated by the Indian Space Research Organisation which have state-of-the-art equipment to analyse physical and chemical properties of aerosols and trace gases. Do you plan to compare your results with ground-based measurements?Professor Burrowsreplied:(1) From GOME or any nadir-sounding passive remote sensing instrument operating in the ultraviolet, visible and near IR spectral regions, the derivation of tropospheric trace gas columns depends on factors, which are dependent on either the molecule (e.g.absorption cross section or line strength, and the vertical distribution in the atmosphere) or radiative transfer issues (the relative importance of molecular, particle and surface scattering). The determination of the tropospheric columns of gases, having significant but variable amounts in the stratosphere or above, requires an accurate knowledge of the stratospheric or upper atmospheric amount. Thus as ∼90% of O3is in the stratosphere the retrieval of tropospheric O3represents the most difficult challenge. Fishman and co-workers at NASA Langley1and Hudson and Thompson,2at NASA GSFC/University of Maryland, the University of Bremen, and the University of Heidelberg pioneered the use of the residual approaches to derive tropospheric trace gas columns. In its simplest form this approach uses the difference between two measurements: one in a location having a clean and known tropospheric and stratospheric column, the other with a different location with a different tropospheric column. An effective assumption of longitudinally homogenous stratospheric columns is often made. This works well for the monthly composites in the tropics and sub tropics. As one moves to higher latitude variability in the lower stratosphere has to be explicitly taken into account.A related method to derive tropospheric O3columns from the measurements of a nadir-looking instrument is the convective-cloud-differential (CCD) method.3This assumes that high cloud are close to the tropopause and enables the stratospheric column to be separated from the tropospheric column. In the special case of O3, the variation of penetration depth a function of altitude, coupled with the temperature dependence of the to retrieve the tropospheric ozone (Hoogenet al.4and references therein). At present the radiometric calibration of GOME limits the application of this approach. Neural networks have also been used for the determination of profile information.5These methods effectively combine a variety of different measurements together and the information content and results are therefore of mixed origin.In addition models of the stratospheric or upper atmospheric burden, stratospheric have been combined with satellite measurements. For the stratospheric column of the trace gas is derived from a chemistry and transport model (e.g.ROSE/DLR or SLIMCAT). The total column is derived from satellite measurements. The difference between them represents the tropospheric amount. The CTMs take explicitly into account planetary wave activity and the dynamics of the stratosphere.6For retrieval of tropospheric NO2, an additional option is to use the penetration depth dependence of the spectral bands.7Finally for HCHO as the stratospheric column is small the total column is dominated by the tropospheric column.(2) The validation of data products derived from measurements made by instruments such as GOME or SCIAMACHY on satellite platforms is a critical activity within the entire project. Comparisons with ground based and aircraft borne measurements have been successfully undertaken. The University of Bremen team have been involved in ground based measurements around the world. For the validation of tropospheric columns there has been a focus on using European and American networks at this point in time for logistical reasons. Comparison with ground basedin situmeasurements requires care because the satellite instrumentation yield the tropospheric columns, whereas measurements in the lowermost troposphere do not include the column above the instrument. However in a variety of studies we have been able to show that in polluted situations the boundary layer column dominates the tropospheric column. The use of data from India for the validation of satellite measurements is an important potential future project.1 J. Fishman, C. E. Watson, J. C. Larsen and J. A. Logan,J. Geophys. Res., 1990,95, 3599–3617.2 R. D. Hudson and A. M. Thompson,J. Geophys. Res., 1998,103, 22129–22146.3 J. R. Ziemke, S. Chandra and P. K. Bhartia,J. Geophys. Res., 1998,103, 22115–22127.4 R. Hoogen, V. V. Rozanov and J. P. Burrows,J. Geophys. Res., 1999,104(D7), 8263–8280, DOI: 10.1029/1998JD100093.5 M. D. Müller, A. K. Kaifel, M. Weber, S. Tellmann, J. P. Burrows, D. Loyola,J. Geophys. Res., 2003,108(D16), 4497, DOI: 10.1029/2002JD002784.6 W. Thomas, F. Baier, T. Erbertseder and M. Käster,Tellus, Ser. B, 2003,55, 993–1006.7 A. Richter and J. P. Burrows,Adv. Space Res., 2002,29, 1673–1683.Professor Jaegléresponded:(1) Our method to separate the tropospheric and stratospheric NO2columns is similar to the tropospheric excess method (TEM). We determine the stratospheric component of the column by using GOME observations over clean Pacific regions where NO2is low. We then assume that this stratospheric component is longitudinally invariant and subtract it from the total columns. Finally, we correct for the assumption of zero tropospheric NO2over the Pacific by adding to the columns the modelled estimate for tropospheric NO2columns over the clean Pacific. New observations of NO2columns from SCIAMACHY have both nadir and limb measurements, allowing a better separation of tropospheric columns by nadir–limb subtraction under some circumstances.(2) We found that current biomass burning emission inventories over India and SE Asia are likely to be too high by ∼50%. This is consistent with previous studies based on satellite and aircraft observations of CO.1,2It is clear that direct surface observations of CO, NO2and other compounds emitted by biomass burning over India would be very valuable to further constrain biomass burning emissions in this region.1 P. I. Palmer, D. J. Jacob, D. B. A. Jones, C. L. Heald, R. M. Yantosca, J. A. Logan, G. W. Sachse and D. G. Streets,J. Geophys. Res., 2003,108, DOI: 10.1029/2003JD003397.2 C. L. Heald, D. J. Jacob, D. B. A. Jones, P. I. Palmer, J. A. Logan, D. G. Streets, G. W. Sachse, J. C. Gille, R. N. Hoffman and T. Nehrkorn,J. Geophys. Res., 2004,109, DOI: 10.1029/2004JD005185.Dr Stevensonasked Professor Burrows: Can the use of 10 day trajectories in the tropics be justified? Surely it is very likely that such trajectories will encounter convection (i.e.rapid vertical motion unresolved by the large-scale winds) and cannot be trusted? This seems almost guaranteed given the intimate link between lightning and convective clouds.Similarly, the use of a trajectory model (albeit one with multiple trajectories and detailed chemistry) precludes interactions along the trajectory, such as mixing of various sources, and crucial processes such as wash-out, and variations in photolysis rates as clouds are encountered. All these factors suggest that the model results should be viewed as only a first approximation. Use of a more interactive chemistry–transport model would have been preferable (although would still contain multiple approximations).With these points in mind, I find some of the conclusions of the paper unjustified. For example, the model overestimates NO2columns by a factor of two. This is ascribed to an overestimate of the lightning NOxsource. This may be true, but it could be that lack of convection (and hence lack of delivery of HOxto the upper troposphere) may significantly affect the NO2lifetime and resultant NOxconcentrations.Similarly, the apparent misplacement of the outflow plume over the Indian Ocean is put down to errors in the ERA-40 reanalysis fields. Transporting NO2at the wrong vertical level (due to the lack of convection) may contribute here (see also comments from Evans and Chipperfield).Professor Burrowsreplied: The justification of the use of trajectory models depends on the question being posed. We argue that the large number of 10 day trajectories (about 230 000) used in this study should depict reasonably well the large scale motion of the atmosphere. However the pattern of the GOME observations for September 1997, which have themselves a relatively large ground scene, are significantly shifted compared to the pattern determined from the trajectories coming off the coast of East Africa with or without chemistry.Convection is clearly an important process in the tropics and is probably best treated as a transition probability between two model layers in trajectory models. Similarly mixing of the air in the trajectory clearly does take place. In a simple test the trajectories were lifted up to an altitude of 300 hPa over areas where the Lifted Index indicated instable atmospheric conditions. The transport analysis resulted in an eastward transport to the Indian Ocean at latitudes extending over a wider range than in the absence of convection. However beyond longitudes of about 70 °E the plume turns southward and travels significantly further south than is indicated by the GOME observations.In a related study using the same GOME data and the MATCH-MPI 3D CTM model was used to investigate the pattern of behaviour over several years. This CTM accounts explicity for convection but uses NCEP rather than ERA-40 meteorological data, the general flow pattern off Africa is qualitatively in agreement with the trajectory analyses of this study. Thus a difference also exist here.For HCHO, which is produced primarily by the oxidation of volatile organic compounds from biogenic emissions and biomass/biofuel burning in the lowermost troposphere, the agreement in spatial pattern and magnitude between that predicted for the HCHO behaviour and that observed by GOME is reasonable. This probably indicates that the emission estimates of VOC and the production and loss chemistry of HCHO in the model is reasonable. In contrast the agreement for model and observations for NO2and the O3are poor. We have considered explicitly stratospheric–tropospheric exchange as a possible source of O3but have found this is of negligible significance for the region under study in September 1997. As an excess of NOxis likely to reduce the chain length for the photochemical production of O3, we therefore consider that the high NOxand low O3are connected. In the chemistry model clouds are taken into account by using the ISCCP cloud climatology for September 1997.1Photolysis frequencies are calculated as a function of height taking each cloud type, specifically into account. There is however no feedback mechanism concerning the formation or destruction of clouds. We do however explicitly consider loss on particles of soluble chemical species.In summary we recognise the importance of convection and the limitations of our trajectory model but consider that they are unlikely to explain the large scale differences between the flow pattern, observed off the coast of East Africa in the trace gases retrieved from GOME, and that predicted by the model.1 W. B. Rossow,International Satellite Cloud Climatology Project (ISCCP), Technical Document No. 737, World Meteorological Organization, Geneva, 1996.Dr Chipperfieldasked: By using trajectories the authors’ model will only have the large-scale (resolved) advection. It will miss the sub-grid processes of convection and turbulent mixing. What was the resolution of the analyses used to force the model? Can the authors discuss the possible impact of the neglect of the ‘sub-grid’ transport?Professor Burrowsreplied: The ERA-40-data used in this study have a spatial resolution of 1.5° × 1.5°, the vertical coordinate being ECMWF model levels. In order to minimise the problems arising from turbulent mixing within the boundary layer, all trajectories are released from 50 to 200 hPa above the ground (which corresponds 500 to 2000 m). This altitude range represents the transition region between the boundary layer and the free troposphere. The airborne measurements, which were used as initialisation data for the chemistry modelling were made at this altitude range. We therefore hope to have minimised the impact of turbulent mixing within the boundary layer on for example the spatial distribution of fire emissions.Convective processes lead to a redistribution of air masses as a function of altitude, characterised typically by small scale updraft and larger scale downdraft, which may be regarded as a vertical mixing process. Whilst recognising the potential importance of convection at sub grid scales, we consider that the large number of trajectories calculated represent a reasonable approximation to the mean monthly behaviour of the atmospheric dynamics in September 1997 for a snap shot at 10.30 am. As described above we have used the Lifted Index to assess the impact of convection. There is an impact but consider that convection does not explain the observed deviation of the location of the outflow off East Africa into the Indian Ocean in September 1997. Qualitatively similar results with respect to the position of the outflow have been obtained using the MATCH-MPI 3D model, which explicitly takes convection into account. As noted above this uses NCEP rather than ERA-40 meteorological data.Dr Chipperfieldsaid: Can the authors be more specific about the problems they find with the ECMWF ERA-40 data for their studies?Professor Burrowsreplied: The spatial distribution of trajectories and the amounts of tropospheric NO2, HCHO and O3columns calculated using a chemistry model along the trajectory for the September 1997 have been compared with those retrieved from GOME measurements. The main objective of the study was to investigate the overall budget for the different trace gases. In this context the agreement between modelled and observed HCHO is reasonable and its spatial distribution is relatively well captured for trajectories released over the Savannah and the tropical rain forest. The trajectory analysis suggests that the main outflow into the Indian Ocean is around 20 °S, which indicates a shift compared with GOME observations of O3, NO2and HCHO. As described above the use of Lifted Index to assess convection does extend the region of outflow close to the coast of Africa. In contrast the trajectory analysis of fire emissions from the Indonesian forest fires in September 1997, which were located mainly in the lowermost troposphere, yields a spatial distribution of trace gas columns similar to that derived from GOME measurements well [Ladstätter-Weißenmayeret al., 2005].Dr Sarkarcommented: I suggest you could include the convective term in your trajectory model. Particle-in-cell Lagrangian techniques (as suggested by Dr Stevenson) are quite unrealistic with respect to the uncertainties in the input data.Professor Burrowsresponded: In trajectory (Lagrangian) models convection is arguably best described by a transition probability between different layers. However temperature and humidity fields are only available on the grid representation of the input data. Thus processes at sub grid scale therefore are not readily taken into account. We are of the opinion that the overall larger scale mean monthly behaviour of atmospheric dynamics should be reasonably well captured by the trajectories.Dr Evansasked:(1) Given your chemistry transport model does not consider convection do you think that it is a suitable tool for diagnosing errors within the ECMWF winds from the satellite data?(2) How does your model treat the reactive conversion of N2O5to aerosol?Professor Burrowsreplied:(1) In short the intention was not to diagnose errors in ERA-40, but rather to use this data set, which we consider to be the best available. The analysis performed in this study comprises two steps. First the transport of a large number of trajectories is computed using the reanalysed analysed wind fields from the ERA-40 dataset are applied. Secondly the chemistry boxmodel BRAPHO is run on a large number of these trajectories. It is initialised with volume mixing ratio measured by airborne instruments during the TRACE-A campaign, and a parameterisation for lightning NOx. We argue that for the monthly mean situation, the large number of trajectories should capture the large scale pattern of flow. As noted in answers to similar questions from other participants, we have attempted to estimate the maximum impact of convection on the flow by use of the Lifted Index. Convection does have an impact but does not explain the difference in the large scale outflow from east Africa. Similarly comparisons with the MATCH-MPI 3D model, which uses NCEP data and attempts to take convection explicitly into account, also show differences in outflow from Africa. Whilst recognising the potential importance of convection, we consider that at least for September 1997, the average flow in the middle and upper troposphere as expected from ERA-40 appears to be somewhat shifted in latitude as compared to the retrievals of the trace gas columns having different lifetimes.(2) We recognise the potential importance of the night-time removal of N2O5. The chemistry used in this study is based on the Master Chemical Mechanism.1The MCM compilation uses a parameterisation for the reaction N2O5→ NA + NA where NA denotes liquid nitric acid, which then deposits. NA formed in the model is not converted back to any other nitrogen compound. This MCM parameterisation is therefore an effective reaction with a recommended mean rate coefficient. This approach has the disadvantage that the microphysical processes, such as hydrolysis of N2O5on particles (aerosols or clouds), are not considered explicitly. Similarly it may be weighted towards mid-latitude conditions. We therefore expect that this simple parameterisation will underestimate the conversion in areas with high aerosol loading and overestimate it in regions with low aerosol loading. In the tropics in our region of study, the relatively high temperatures in the lower troposphere and rate of photolysis of tropopsheric ozone are such that the loss of NOxby the three body reaction of OH with NO2is probably more important than the removal of N2O5at night. We therefore do not expect that an improved parameterisation of this reaction will explain the poor description of the amount of NO2. However in future studies an improved description of the microphysics such heterogeneous processes will represent a significant improvement of the model. At this time we attribute the poor description of NO2to be in our parameterisation of NOxfrom lightning, which probably overestimates NOx, in line with early results from recent studies of lightning in the tropics. Further work is clearly needed, utilising the knowledge gained from the latter studies and the recent measurements of NO3and N2O5made in the boundary layer and troposphere.1 S. M. Saunders, M. E. Jenkin and R. G. Derwent,Atmos. Chem. Phys., 2003,3, 161–180.Professor Cohenopened the discussion of Professor Jaeglé’s paper: What are the other consequences of high soil NOxemissions for (a) the N2O budget and (b) for HNO3deposition?Professor Jaegléreplied: Microbial soil processes emit both NOxand N2O. The ratio of NO:N2O emissions is dependent on soil environmental conditions, in particular soil moisture. Combining spatially and temporally varying estimates of this ratio with our top-down estimate for soil NOxemissions could be used to infer N2O soil emissions, providing useful constraints for this very poorly quantified source of N2O. One caveat is that soil NOxemissions can be lost by reactions on the vegetation canopy while N2O emissions are not—this would complicate the analysis. In answer to your second point, high soil NOxemissions imply larger HNO3and nitrate deposition. We plan on examining the consistency of our top-down estimate of soil NOxemissions with deposition observations at rural US National Atmospheric Deposition Program sites. In addition we will examine the effects of increased soil NOxemissions on our simulation of background levels of ozone in the summer.Dr Stevensonasked: Do you think the results you find would differ if you used a different chemistry–transport model?Professor Jaegléreplied: Information from the GEOS-CHEM chemical transport model directly affects our global inventory of NOxemissions at two stages: in the AMF calculation (we need to usea prioriinformation on the vertical distribution of NO2), and in the inversion to derive NOxemissions from tropospheric NO2columns (based on model-calculated NOxlifetimes and NOx/NO2ratios). The shape of the NO2vertical distribution over land is mostly determined by the simulated boundary layer depth, and potential errors in this parameter result in a 15% uncertainty in our AMF calculation. As for NOxchemistry, we estimated a 30% error on this factor based on comparisons to observed NOx/NOyratios. Thus overall, the top-down estimate of surface NOxemissions will be affected by the CTM used, but this influence is relatively small compared to other uncertainties. To illustrate this point, we can compare our results to the study of Müller and Stavrakou1who used an adjoint modelling technique based on the IMAGES CTM to derive optimised NOxand CO emissions inventories from GOME NO2observations for 1997 and surface CO observations. Their results are very similar to ours, in particular they also infer a large source of NOxfrom soils. This give us confidence that this result is independent of the inverse modelling technique and CTM model used and comes directly from the GOME NO2observations.1 J. F. Müller and T. Stavrakou,Atmos. Chem. Phys. Discuss., 2005,4, 7985.Professor Cohencommented: In response to Dr Stevenson’s question about whether the inverse model retrieval of NO2will be strongly dependent on which global model is used, I point out that the lifetime of NOxis only a few hours so large scale meteorological differences in global models cannot affect the inversion. Those differences are important for long lived species such as CO. For NO2systematic difference in OH/between models would matter.Dr Chipperfieldcommented: The author quoted comparisons between the GEOS-CHEM and IMAGES model for NO2and said that the agreement is good. What does this depend on and really test for a short-lived species such as this? Is it known how well the models compare for longer-lived species such as CO?Professor Jaegléreplied: With a short-lived species like NO2, which is mostly confined in the boundary layer, comparisons between global models is a test of both the emissions inventory and of NOxchemistry (NOxlifetime). For a longer-lived species such as CO the comparison is a little more difficult as, in addition to chemistry and emissions, one also needs to consider differences in transport.Dr Evansasked: How model dependent do you think the NO2columns are? Would you come to the same quantitative conclusions with a different model?Professor Jaegléresponded: The short answer is that the tropospheric NO2columns are relatively model insensitive, as I noted in my answer to Dr Monks. However, the tropospheric NO2columns will be affected by the actual retrieval algorithm used and most importantly its assumptions for cloud, aerosol scattering, and surface reflectivity. We have done our best to account for these effects, and assessed their impacts on the retrieved tropospheric NO2columns by detailed error analyses. The higher spatial resolution measurements of NO2from newer instruments such as SCIAMACHY and OMI, combined with more information on clouds, aerosols and surface reflectivity should result in more accurate retrievals and reduced uncertainties.Dr Stevensonsaid: You don’t consider the lightning NOxsource, and suggest it is only a small component of the GOME NO2column. This seems to be at odds with Professor Burrows’s paper, which uses the GOME NO2data to investigate the lightning NOxsource. How does your model deal with lightning? Are you using a different GOME NO2product?Professor Jaegléanswered: Our NOxemission inventory includes a 3.5 TgN year−1lightning source. The contribution of lightning NOxto column NO2over land in our model simulation is generally small. For example we conducted a simulation without lightning and found that over Africa lightning accounts for less than 0.25 × 1015molecules cm−2, much smaller than the observed enhancements in the GOME NO2columns1(∼2–3 × 1015molecules cm−2). This small sensitivity to lightning results from the fact that lightning NOxis preferentially deposited in the upper troposphere, where it will be present mostly as NO because of the low temperatures and the resulting slow rate for NO + O3. Our inversion procedure, where we relate NO2columns to surface NOxemissions, thus does take into account the small enhancement from lightning in the model. So, in other words, we take out the lightning influence using the model—assumed inventory, and examine the land surface sources only. Over the ocean—away from any large land NOxsources-lightning NOxwill stand out more clearly and is likely to account for part of the observed enhancements over the Atlantic presented in the previous paper.The main difference in the tropospheric GOME NO2columns used in our analysis and in the analysis presented in the previous paper lies in the retrieval used, and more specifically in the different treatments of scattering by clouds, aerosols, gases, and surface albedo. These differences have been discussed previously.2Briefly the main effects are that our approach results in lower tropospheric NO2columns over cloudy regions, oceans, and regions with high albedo, while we derive larger columns over biomass burning regions. In addition, we calculate an AMF for each scene using daily information from GEOS-CHEM.1 L. Jaeglé, R. V. Martin, K. Chance, L. Steinberger, T. P. Kurosu, D. J. Jacob, A. I. Modi, V. Yoboué, L. Sigha-Nkamdjou and C. Galy-Lacaux,J. Geophys. Res., 2004,109, DOI: 10.029/2004JD004787.2 R. V. Martin, K. Chance, D. J. Jacob, T. P. Kurosu, R. J. D. Spurr, E. Bucsela, J. F. Gleason, P. I. Palmer, I. Bey, A. M. Fiore, Q. B. Li, R. M. Yantosca and R. B. A. Koelemeijer,J. Geophys. Res., 2002,107, DOI: 10.1029/2001JD002622.Dr Monksasked:(1) Is there a fundamental inconsistency between the presented work, in that the Burrows paper suggests a requirement for less (modelled) NO2(see Table 1 in the paper)vs. the conclusion from Jaegléet al. that there is increased soil emissions of NO2.(2) With respect to the soil moisture/NOxpulse mechanism, is there an overpass sampling bias?Professor Jaegléreplied:(1) We use fundamentally different approaches and assumptions, making a direct comparison difficult. In our paper we use surface emissions combined with a Eulerian chemical transport model to calculate NO2columns and then use this as a framework for our inversion analysis to relate GOME NO2columns to surface NOxemissions. Because of the short lifetime of NOx(a few hours), we assume that NO2columns directly map onto local NOxemissions. The paper presented by Professor Burrows uses a trajectory model, which calculates excess NO2along the trajectory. One way to compare the two methods would be to examine the differences in oura prioriNOxemissions over Africa. During September 2000, our African NOxemissions add up to 0.52 TgN (0.36, 0.04, 0.07, 0.05 from biomass burning, biomass and fossil fuel combustion, soils and lightning). Table 1 of the paper gives the excess NO2emissions for September 1997, which account for 0.05 TgNO2. Not knowing background NO2emissions used in this paper, it is difficult for me to relate the excess NO2emissions to total emissions.(2) This is a good point. One expects rainfall to occur mostly in the afternoon, and thus GOME with its 10:30 am equatorial sampling time would miss the initial pulse from soils. Field experiment have shown that pulsing of soil NOxfollowing rain can last for several days, thus GOME would sample the NO2enhancement 1 or 2 days later.Professor Burrowsreplied:(1) The two studies both use GOME data, but the retrieval procedures for the derivation of trace gas tropospheric column amounts, whilst being similar in concept, are different in detail. For example the air mass factors are calculated using different models and assumptions. Overall the general agreement between the different retrieval approaches appears reasonable in areas of high tropospheric NO2. The focus of the two studies is significantly different. The University of Bremen study has selected the period of intense biomass burning and lightning during September 1997 for a case study. The results indicate that the model emission used for HCHO and its precursors yields a reasonable description of the HCHO behaviour. A reasonable parameterisation of lightning in line with other recent studies results in an overestimate of the column of NO2. The O3column calculated is lower than that observed. The University of Washington study has a different focus. We therefore do not consider that there is a fundamental inconsistency between the two studies.(2) At the University of Bremen study we have not investigated the potential soil humidity pulsing mechanism for the emission of NOx, which has been observed in forested region of California. However this mechanism is likely to be only of minor significance for our region of study in September 1997.Professor Cohenobserved:(1) During INTEX/ICARTT we measured NO2profiles from the NASA DC-8. Over the continental US we frequently observed half the NO2column in the upper troposphere.(2) However, so long as the lightning doesn’t perfectly covary with the model representation of soil NO emissions the model’s inadequacies with respect to lightning will not affect its retrieval of soil NO emissions.Dr Shallcrossopened the discussion of Dr Bloss’s paper; In your comparison between model and measurement you conclude that the variability in OH is driven by variability inj(O1D), does not variability in H2O also play a significant role? In the tropics one would expectj(O1D) to vary rather little and for H2O variability to dominate OH levels; does not this limit the usefulness of your method?Dr Blossresponded: Our method for determining the mean global [OH] is to correct the hemispheric global mean OH values from the GEOS-CHEM model using the observations at Mace Head, Ireland and Cape Grim, Tasmania. This approach factors all the OH production and loss mechanisms included in the GEOS-CHEM model into our final values, including the simulated levels ofj(O1D) and of water vapour—we do not use the relationship between OH andj(O1D) to determine the mean global [OH].We do find a strong correlation between monthly mean model OH and the monthly mean modelledj(O1D) for the mid-latitude Mace Head location (Fig. 6 of the paper), but not for the tropics (Fig. 7). OH production depends also upon the levels of ozone and water, together with temperature and pressure, while OH loss depends upon the levels of a range of co-reactants. The observed correlation at Mace Head indicates that the combined effect of these factors varies withj(O1D); it is perhaps most likely that in a marine location such as Mace Head the loss processes are reasonably constant annually, and we might expect the monthly averaged humidity and temperature to show similar trends toj(O1D), hence the dependence of OH upon these factors is obscured.Dr Brauerssaid: The paper by Blosset al. compares local measurement of OH with a global model. However, the main dependence of OH is caused by the solar radiation. Solar radiation, namely photolysis frequencies likej(O1D) are an external parameter to the chemical model with no feed-back. Therefore, it is suggested to correlate the measured and modelled data toj(O1D) and compare the slope of the regression. The slope—or the OH toj(O1D) ratio—is an indicator for the chemical environment.Dr Blossreplied: We agree that a comparison of the simulated gradient of the OHvs. j(O1D) against the observed (on various timescales) could be a useful diagnostic for model performance, especially in locations where the chemical environment (OH co-reactants, NOx) is constant—however such comparison of modelled and measured OH was not the focus of our paper.Professor Cohencommented: The linearity and strong correlation of OH withj(O1D) is to be expected and uninteresting. What is interesting is the variability of the relationship between OH and its sources.Dr Brauersreplied: Rohrer and co-workers presented an analysis of the OH-j(O1D) correlation for datasets recorded in various environments. Surprisingly, the variability of OH at each location is almost entirely explained by thej(O1D) variation and the precision of the OH measurent, even for a 5 year OH dataset recorded at Hohenpeissenberg. Each location can be characterized by its slope and exponent of the relation between OH andj(O1D). The abstract of their presentation can be viewed athttp://www.igaconference2004.co.nz/abstractPreview.asp?absID=334Therefore, the OH variabilty due to other parameters, sinks or sources, might only be visible from experimental data if the local OH–j(O1D) relationship is thoroughly analysed or when different locations are compared.Professor Berresheimsaid:(1) Following the comment by Dr Brauers, other OH field measurements including our long-term measurements at Hohenpeissenberg have shown that the OH–j(O1D) covariance is highly correlated but with distinctly different slopes for different air mass regimes. Thus the Mace Head data set alone cannot be extrapolated by the model to other regimes and/or latitudes as done in the paper by Blosset al. Error estimates should be given or discussed in the paper in this respect.(2) For the tropics there may still be a strong OH–j(O1D) relation. The authors should test the covariance on a hourly rather than only monthly basis.(3) A detailed justification for choosingj(O1D) as a better parameter of choice to plot against OH rather than usingP(OH) has been given in the previous papers by Ehhalt and Rohrer and in the MINOS paper by Berresheimet al. Blosset al. should cite these papers in that regard, as well as the now citable Abstract by Berresheimet al. on their long-term measurements presented at the EGU 2005 and the IGAC 2004.Dr Blossreplied: The Mace Head OH–j(O1D) relationship was not used to determine the global [OH]. We use the marine boundary layer datasets to scale the OH field calculated by the GEOS-CHEM 3-D chemical–transport model. As GEOS-CHEM incorporates a full global emission inventory and chemical scheme, the model will take account of differing air mass regimes, and hence different relationships between OH, and (for example)j(O1D), at other locations. The OH–j(O1D) relationship observed at Mace Head is included in the paper purely to illustrate the likely reasons behind the success of the model simulation of observed OH levels at this location.We have averaged the observed and modelled data on a monthly timescale to remove day-to-day and synoptic variability from the OH levels—this is necessary as advection data is not available for the specific periods/locations of some of the measurement campaigns used, hence the fine scale variation must be removed. We are thus unable to compare the model and measurements on an hourly timescale—and indeed such a comparison would not be appropriate when comparing measurements of a short-lived species such as OH with values calculated by a model with the temporal and spatial resolution of a global CTM. One of the points we wished to make in the concluding section of the paper is that observational data is sparse in the tropics.Our aim is not to suggest thatj(O1D) was a statistically better diagnostic of OH thanp(OH) but rather that the large variability in OH seen at Mace Head was attributable to a changing radiation field rather than a chemical field. We do not in fact make any mention ofp(OH), the primary rate of production of OH from O3photolysis/O(1D) + H2O. If we were seeking to determine the relationship between OH and a single driving factor,j(O1D) would be the obvious choice as advocated by Ehhalt and Rohrer1—however we must be cautious especially when considering different chemical environments—the data used by Ehhalt and Rohrer were obtained at a “rural relatively unpolluted site” in Germany; recent measurements performed in our group have demonstrated that the OH–jcorrelation can break down in more complex environments, as evidenced by elevated levels of OH observed in wintertime in cities due to alkene ozonolysis.21 D. H. Ehhalt and F. Rohrer,J. Geophys. Res., 2000,105, 3565.2 D. E. Heard, N. Carslaw, L. J. Carpenter, D. J. Creasey, J. R. Hopkins, A. C. Lewis, M. J. Pilling, P. W. Seakins,Geophys. Res. Lett., 2004,31, L18112, DOI: 10.1029/2004GL020544.Professor Planeasked: This paper reports mean OH concentrations for both hemispheres. What use are these numbers? Fig. 9 of the paper shows that the average is heavily weighted to the tropical free troposphere, so the concept of a global average seems a strangely old-fashioned concept.Dr Blossresponded: The concept of a mean global OH concentration is the simplest way to define the atmosphere’s capacity to remove most emitted species. While strictly the value can only be applied to determine the lifetime of a species removed solely by a temperature-invariant bimolecular reaction with OH, and which is uniformly distributed throughout the atmosphere, in practice mean global OH values can be used to estimate the removal rate of most species with lifetimes over 1–2 years (e.g.CH4, many HCFCs) and thus provides a useful tool to assess their likely atmospheric persistence and distribution.More quantitatively, the global mean OH average provides a key metric for assessing the performance of global chemistry-transport models. The approach to determining the global average used in this work, constraining the simulated OH field to point field measurements, differs from most previous determinations which have used measurements of tracer species such as methyl chloroform, and have to take account of their non-uniform distribution throughout the atmosphere. It is thus encouraging that good agreement is obtained between this study and values obtained elsewhere.Dr Coxasked: Are the values given for mean [OH] in the northern and southern hemispheres significantly different from each other? No error limits are given.Dr Blossreplied: The mean [OH] values obtained for the northern and southern hemispheres (0.91 and 1.03 × 106cm−3, respectively) are not significantly different in our current analysis—the uncertainties (derived from the uncertainty in the constraining field measurement data only and neglecting model factors) are 19 and 13%, respectively.Professor Ravishankaraobserved: I would like to note that the global average [OH] derived from methyl chloroform concentrations are “weighted” towards the regions where most of the methyl chloroform is destroyed in the atmosphere. This is because methyl chloroform does not carry information about the regions that do not contribute to its loss. This is in spite of the corrections made by the people who carry out such calculations.In your defense regarding the usefulness of a “global” average number for [OH]: yes, it is useful for calculating the lifetimes of molecules which are primarily lostviaOH reactions in the troposphere and whose lifetimes are in the order of a few years (greater than 2 and less than 20 or so). Further, the rate coefficient for their reaction with OH should have a temperature dependence that is similar to that for OH + CH3CCl3reaction. It so happens that there a quite a few of the CFC-substitutes fall into this category, as does methane. However, this concept should not be pushed beyond this issue because there are questions as to what an average concentration means! We cannot assume the reactivity of OH to be the same irrespective of where it is (i.e., the temperature and pressure of its location)! Therefore, if we are not careful, a global mean OH concentration becomes a meaningless concept.Dr Blossreplied: We agree that there are potentially difficulties in the use of tracer measurements to derive a truly global mean OH value—our method, which employs the global distribution obtained from a CTM, scaled by observations, incorporates contributions from all regions of the atmosphere (although in the current analysis the limited observational dataset may introduce similar limitations). Methyl chloroform data are not used in our analysis so the requirement for the reaction kinetics to be similar to those for reaction of CH3CCl3with OH is removed; however strictly speaking a new constraint, that the reaction kinetics be reasonably invariant with respect to pressure and temperature (or that their atmospheric mass-weighted averages are used), is introduced for use of our current (mass-weighted) mean [OH].Professor Jaegléasked:(1) Does the global model get OH right for the right reasons? In other words, have you compared the model simulations to local observations of H2O, CO,j(O1D), NOx, in addition to looking at OH?(2) Given the high variability of OH driven by its short lifetime, to what extent do surface observations give information on global OH levels?Dr Blossresponded: This is a constant worry for models. GEOS-CHEM has been extensively evaluated against observational datasets globally and we believe that it does sufficiently well to provide a useful platform for our subsequent analysis. However, the raft of negative feedbacks inherent in the atmospheric chemical system make it very easy to calculate the right OH levels for the wrong reasons. A particularly under-assessed area of the model is probably the photolysis rates.We agree that the surface observations do not provide an ideal constraint on OH throughout the rest of the atmosphere. We are in the processes of incorporating aircraft data (and other surface sites) into our analysis which will allow a more complete analysis of the process. The inclusion of more locations/dates will also increase our confidence that the model is correctly simulating the photochemistry which determines [OH].Dr Stevensonasked: You state [OH] in Fig. 1 is ‘global mean annual tropospheric mass-weighted OH’. But most of the studies in Fig. 1 derive OH from methyl chloroform—so surely these studies are weighted by where the MC + OH reaction takes place? They can say very little about OH in the upper troposphere or at the poles. But your global model results will include these regions. Wouldn’t it make more sense to compare MC weighted OH concentrations? Lawrenceet al.1compares different measures of [OH] which may be useful.1 M. G. Lawrence, P. Jöckel and R. von Kuhlmann,Atmos. Chem. Phys., 2001,1, 37–49.Dr Blossreplied: Our purpose in Fig. 1 of the paper was simply to compare the recent determinations of mean global [OH] to our value—the methyl chloroform (MCF) tracer studies are amongst the only available observational determinations of mean global [OH]. The studies referred to in Fig. 1 of the paper which utilise observations of methyl chloroform (MCF) to derive mean global [OH] values1–4incorporate modifications to the inversions used to account for the spatial distribution of MCF, thus mass-weighted values of mean [OH] are obtained; however we accept that this may present a difficulty, and in particular may lead to greater uncertainties over OH levels in the colder regions of the atmosphere which contribute relatively little to MCF removal. Our approach, in which the mean OH is derived from the global model field scaled to direct OH observations, does not suffer from this limitation. We agree that one could weight the OH distribution we obtain from our constrained model according to the distribution of the species one is interested in investigating the removal of, and further weight the removal kinetics to their atmospheric average according to the same distribution, and hence obtain a global mean [OH] specific to an individual species such as MCF or CH4which could be compared to the observed MCF/CH4lifetime.1. R. G. Prinn, R. F. Weiss, B. R. Miller, J. Huang, F. N. Alyea, D. M. Cunnold, P. J. Fraser, D. E. Hartley and P. G. Simmonds,Science, 1995,269, 187.2. M. Krol, P. J. van Leeuwen and J. Lelieveld,J. Geophys. Res., 1998,103, 10697.3. R. Prinn, G. J. Huang, R. F. Weiss, D. M. Cunnold, P. J. Fraser, P. G. Simmonds, A. McCulloch, C. Harth, P. Salameh, S. O. Doherty, R. H. J. Wang, L. Porter and B. R. Miller,Science, 2001,292, 1882.4. S. A. Montzka, C. M. Spivakovsky, J. H. Butler, J. W. Elkins, L. T. Lock and D. J. Mondeel,Science, 2000,288, 500.Dr Chipperfieldcommunicated: Given the large impact ofj(O1D) on OH in many parts of the atmosphere should we be putting more effort into measuring and modelling this quantity? This would remove some uncertainties in looking at the more complicated issue of OH concentration. In contrast to box model studies which may factor out the effect ofj(O1D) from the data to detect other effects, for global models (chemical transport model, CTM and general circulation model, GCM) which calculate chemistry in an unconstrained way it is important to check these controlling processes.Dr Blosscommunicated in response: In general while global atmospheric models have been extensively compared to observations of chemical parameters, much less effort is typically made to evaluate the radiation fields, which drive the photochemistry occurring. Agreement between model and measuredj(O1D) is an essential prerequisite to getting the modelled OH correct (for the right reasons) so greater use of and/or generation ofj(O1D) and other radiation data could provide a useful and relatively inexpensive test of model performance.Professor Heardcommented: The very high correlation between [OH] andj(O1D) calculated in this study for unpolluted environments has prompted Dr Chipperfield to suggest thatj(O1D) levels can be used to infer [OH], and therefore that field measurements of [OH] are less important. In defence of OH measurements some observations concerning the polluted environment are pertinent. During the PUMA field campaigns in 1999 and 2000, the Leeds FAGE conducted measurements of OH concentrations at an urban site in both summer and winter.1–3The results showed that despite more than a tenfold decrease inj(O1D) between summer and winter, the average noontime [OH] only decreased by a factor of two. A rate of production study using a constrained box-model indicated that the surprisingly high concentrations of OH in winter are maintained by the reaction HO2+ NO, with HO2being generated by photoylsis of carbonyls (in particular HCHO) and also by O3+ alkene reactions, as well as direct OH production from O3+ alkene reactions. The reaction O(1D) + H2O was shown to be only a minor source of OH in winter. Thus one must be careful with the generalisation regarding correlations between OH andj(O1D), especially when comparing seasonal variations in polluted regions. Although for the PUMA campaign the ratiosj(O1D)summer/j(O1D)winterand [OH]summer/[OH]winterare very different, there is still a significant correlation between [OH] andj(O1D) in winter, as HO2, the major source of OH, is formed predominantly from the photolysis of carbonyls, the rate of which will be closely correlated withj(O1D). The gradient of [OH]versusj(O1D) will be quite different though for summer and winter in the urban environment, and it would be difficult to infer [OH]a priorifrom any measured value ofj(O1D).1 D. E. Heard, N. Carslaw, L. J. Carpenter, D. J. Creasey, J. R. Hopkins, A. C. Lewis, M. J. Pilling and P. W. Seakins,Geophys. Res. Lett., 2004,31, L18112, DOI: 10.1029/2004GL020544.2 R. M. Harrisonet al.,Sci. Built Environ., 2005, in press.3 K. M. Emmerson, N. Carslaw, L. J. Carpenter, D. E. Heard, J. D. Lee and M. J. Pilling,J. Atmos. Chem., 2004, submitted.Dr Blossreplied: We agree: To define the relationship between [OH]ssandj(O1D)a priori, the levels of O3, H2O, OH co-reactants,Tandphave to be known—these (especially the co-reactant levels) are unlikely to be well constrained other than in certain remote marine locations. This relationship could be useful in certain carefully defined environments where we are confident that the radical chain length is short (i.e.OH is formed overwhelmingly through primary production rather than by reaction of NO + HO2etc.) and that the variability inj(O1D) dominates over variability in (e.g.) H2O and temperature in determining OH production, and over the variability in the OH loss rates. However we are not currently able to define thej(O1D)vs. OH relationshipa priorifor most atmospheric locations.We believe that the only way to confidently ascertain the concentration of this important species is to go out and measure it; however making such measurements is expensive and so we need to evaluate the optimal locations in both time and space. We probably have sufficient data in the extra-tropical marine boundary layer; in the future more observations from tropical marine regions, tropical forests, deserts and mega-cities will be necessary.Dr Chipperfieldcommented: These points are well made. The motivation behind the comment onj(O1D) was not that it is an alternative proxy to measuring OH in clean environments, but rather that given its importance in determining the absolute [OH] then effort should be made to make sure that models can accurately reproduce this controlling factor. This is important for global models (CTMs, CCMs) wherej(O1D) is calculated rather than constrained. For these comparisons observations ofj(O1D) are clearly necessary.Professor Dibbleopened the discussion of Dr Seisel’s paper:(1) In Fig. 6, is the equilibrium constant for multilayer or sub-monolayer conditions? To the extent that Fig. 6 reflects multilayer conditions, the ΔHads values for soot and mineral dust are not directly comparable, since the ΔHadsfor soot refers to submonolayer conditions.(2) Using your desorption rate constants for water molecules from mineral dust, I computed anA-factor of 200 s−1. I cannot understand what that number means. Can you offer an explanation?Dr Seiselreplied:(1) For the maximum coverage observed in this study we estimated an upper limit of 40 monolayers and a lower limit of 1 monolayer. Since these values refer to the maximum coverage we consider the equilibrium constants valid for approximately 1 formal monolayer. The activation energy determined from the desorption rate constants (Fig. 4) for sub-monolayer conditions agree with the adsorption enthalpy derived from the equilibrium constants. Therefore, we consider the adsorption enthalpy for water on soot and mineral dust as comparable.(2) TheA-factor (Fig. 4 of the paper) as well as the adsorption entropy (Fig. 6 of the paper) are determined from the intercept and therefore depend on the absolute value of the surface area. As long as we did not know the absolute surface area physically realisticA-factors or adsorption entropies cannot be calculated.Dr Baltenspergercommented: A word of caution concerning the atmospheric implications: in the atmosphere the soot particles will rapidly be coated by condensing secondary material. Thus, the soot aerosol will be internally mixed and their properties will be determined by this condensed material (which dominates in mass) rather than the soot particle. This is confirmed by our observations at the high Alpine site Jungfraujoch, where the scavenging of soot behaves exactly as the total submicrometre aerosol volume.Dr Seiselreplied: This work and the atmospheric implications drawn are related to the surface properties of the particles under study. In the case wheree.g.a soot particle is internally mixed with other compounds the surface properties of the aerosol will change and therefore our results may not longer be applicable to the “new” type of aerosol.Dr McFigganscommented: To extend the point raised by Dr Baltensperger, not only will soot be associated with organic material but also, given extensive field observations in multiple locations, inorganic components (sulphates, nitratesetc…) are invariably internally mixed with carbonaceous components at only modest distances from emission sources. The atmospheric relevance of the hydrophobic nature of soot is therefore limited to locations sufficiently close to source and still at temperature too high for organic condensation (still in vehicle exhaust, for example).Dr Ammannsaid: In this paper, several fundamental parameters describing adsorption of water to mineral dust and soot are determined. These are then related to macroscopic aspects of hydrophilicity of the related atmospheric aerosols. We have recently determined hygroscopic growth of mineral dust aerosol particles (Arizona Test Dust) using a hygroscopicity tandem DMA.1We have related the very small hygroscopic growth observed on these particles to the presence of water soluble salts, mainly sulphate. While the paper presented by Dr Seisel considered water uptake on the surface in the form of BET type adsorption on an ‘inert’ surface, we suggest that water uptake and possibly also the CCN activities will be mainly controlled by water soluble material.A second point, which relates to this with regard to both fundamental adsorption modes and also the macroscopic hygrophilicity, arises from the way the soot samples were taken. Samples collected on a plate in the hot wake of the flame at probably above 200 °C will not contain many of the organic compounds usually associated with soot. Many of these contain a number of hydrophilic functional groups that have therefore not been considered in the present study, but that might be crucial in the behaviour of soot in the real atmosphere.With regard to the adsorption energetics as derived from the present study, the authors already state that the thermodynamic analysis is affected by the fact that it is not clear to what depth underneath the sample surface water actually diffuses in. In addition, I would like to recall the presence of water soluble material mentioned in my previous comment. The driving force for water uptake would then be the dissolution of the soluble material and not multilayer adsorption. This would then lead to a quite different and very challenging thermodynamic analysis of a multicomponent system. Therefore, it might be better to consider pure and well defined materials for determining fundamental adsorption parameters, and to determine lumped parameters, such as net growth curves or water mass changes for authentic materials.1 A. Vlasenko, S. Sjögren, E. Weingartner, H. W. Gäggler and M. Ammann,Aerosol Sci. Technol., 2005,39, 452–460.Dr Seiselreplied: There are still open questions concerning the detailed interaction of water with theses kind of surfaces. In order to get more insight into the problem you mentioned, it is planned to extent this study to surface which are less complex and are therefore better defined. In addition, we intent to use materials which have different surface coatingse.g.organics or sulfuric acid. With these results we may then be able to parameterize the uptake of water on complex, realistic surfaces like mineral dust or soot and get insight in the influence of “reactive” compounds on the surface.Professor Rudichcommented: Dust particles collected during dust storms in Israel were observed to have organic coatings and sulfates on the surface.1 D. Rosenfeld, Y. Rudich and R. Lahav,Proc. Natl. Acad. Sci. USA, 2001,98, 5975–5980.2 A. H. Falkovich, G. Schkolnik, E. Ganor, Y. Rudich,J. Geophys. Res., 2004,109, DOI: 10.1029/2003JD003950.Dr H. Roscoecommented: When reading this paper, there seemed to be some confusion about mechanisms of ice nucleation, which was not relevant to the main body of this excellent lab study. I recognised the confusion well, as I had been similarly confused a year ago whilst preparing a proposal. Searching the literature, all the papers about ice nucleation were about freezing of ice, never about condensation from vapour to ice.“Theory predicts that homogeneous nucleation of ice from vapor should only occur for extreme supersaturations, never observed in natural conditions” (ref. 1).“A calibration curve was developed to allow us to convertthe freezing temperatures to a saturation ratio required for ice nucleation.” (ref. 2).“... ice forming nuclei (IFN)” (ref. 3).“In particular the effects of solutes and mechanical pressure onthe kinetics of the liquid-to-solid phase transition of supercooled water and aqueous solutions to icehave remained unresolved.” (ref. 4).”Here we provide evidence that at atmospheric pressures the conditions leading to theinitiation of freezing in pure waterare those for which the liquid compressibility and the corresponding density fluctuations reach maxima.” (ref. 5).The classic review asserts that condensation to ice is energetically forbidden except at temperatures much colder than found in the atmosphere. But this is only true of the smallest nucleation particles, we are all familiar with rime on fences, and modern cloud-probes which photograph particles show rimed crystals in the 10 μm range.So my question for the audience is, at what size is direct condensation to ice allowed at, say −35 °C—temperatures routine in the upper troposphere, and above the inversion in Antarctica? If this is as small as 1 or 0.5 μm then we have been ignoring a significant amount of nucleation, particularly on broken wind-blown snow in Antarctica, perhaps also at mid-latitudes.1 W. Szyrmer and I. Zawadzki,Bull. Am. Meteorol. Soc., 1997,78, 209.2 M. E. Wise, R. M. Garland and M. A. Tolbert,J. Geophys. Res., 2004,109(D19203), DOI: 10.1029/2003JD0043133 E. K. Bigg and C. Leck,J. Geophys. Res., 2001,106(D23), 32155.4 T. Koop, B. P. Luo, A. Tsias and Th. Peter,Nature, 2000,406, 611.5 M. B. Baker and M. Baker,Geophys. Res. Lett., 2004,31, L19102, DOI: 10.1029/2004GL020483.Dr Seiselresponded: Heterogeneous ice nucleation, especially the direct condensation to ice (deposition mode) does not only depend on the size of the ice nuclei. In order to be an efficient ice nuclei, the particle has to be insoluble in water. Moreover, the surface of the particle has to exhibit an ice-like structure in order to support the build-up of the ice lattice. Therefore, it is difficult to give a particle size at which direct condensation to ice will occur. In addition, atmospheric aerosol particles are often internally mixed, which may increase its water solubility and consequently lower its nucleation efficiency.Professor Zellneraddressed Dr H. Roscoe: The distinction between water and ice nucleation under sub-monolayer conditions is not sensible. The addition of one or several water molecules to a surface can impossibly determine whether an ice or liquid water layer is being formed, since the terms “ice” and “water” refer to macroscopic ensembles of molecules in certain structural environments. By implication, the comment made by Dr. Roscoe may be atmospherically very relevant but it is not pertinent to the present paper.Dr H. Roscoereplied: My comment was stimulated by statements in the Introduction of the present paper, that seemed to suggest that ice nucleation from vapour was possible in the atmosphere, whereas most texts would assert otherwise. It is clearly pertinent, if only because the Introduction takes up over a page. My point is that the textbook view demands re-examination because at the largest sizes it is certainly possible. Moreover it may also be possible to start the process on small diameter particles because, as Prof. Zellner rightly comments and the paper discusses, sub monolayer deposition cannot have the energy costs of bulk deposition at high curvature, which in a liquid are associated with surface tension forces. This is another reason why ice nucleation from vapour may demand re-examination. Naturally, a process which starts only in sub-monolayer conditions may have difficulty with growth rates, but this is something for modellers to consider.Professor Abbattcommented: Ice formation by deposition mode nucleation is a highly viable ice formation mechanism. In laboratory studies where particles are supported on a temperature controlled teflon substrates (from −10 to −60 °C), we have recently shown that a variety of dust particle types are highly efficient IN, nucleating ice at about 105% relative humidity with respect to ice whereas soot is highly inefficient. It may be that the ice nucleation proceeds initiallyviafor the adsorption of water to the particle surfaces, providing a nice connection to the results of Dr Seisel’s . paper.Dr Coeaddressed Dr H. Roscoe: Riming is the collection of liquid drops by an ice particle and subsequent adhesion by freezing. It is a growth process for ice particles of any size, nucleation of ice is the creation of new ice particles. The latter may occur by freezing of water droplets or by nucleation on a surface such as mineral dust. The former, homogeneous nucleation, is only significant at temperatures below −40 °C. The latter can occur at much warmer temperatures. Ice may grow by vapour diffusion and also by riming.Dr H. Roscoereplied: Dr Coe is quite correct, I made a mistake in my haste to formulate the comment. The process seen in the real atmosphere whereby ice forms on larger scales directly from vapour is the formation of hoar frost, which occurs well above −40 °C. The question then is, at what size of deposition surface does hoar frost demand a large supersaturation? (Size is relevant because surface curvature implies an energy cost when depositing a film, a cost which in a liquid film is associated with surface tension.)Dr McFiggansadded: The distinction between heterogeneous and homogeneous nucleation and riming is well established and treated in the literature. The uncertainty is not in the description but in the nature and abundance of ice forming nuclei (IN) (surfaces providing site for heterogeneous nucleation). Homogeneous nucleation, by definition, is nucleation from the vapour without a pre-existing nucleus. It is “fairly” well-described as a function of temperature and thus supersaturation over ice.Professor Planeasked Dr Seisel: When you calculate the uptake coefficients, do you use the BET or geometric surface areas of the sample? Since theγvalues that you measure are quite large, using the BET surface area would lead to an underestimate of the trueγ, whereas the geometric surface area and produces an upper limit.Dr Seiselreplied: For the calculation of the uptake coefficients the geometric (projected) surface area of the sample has always been used. We did not observe any dependence of the initial uptake coefficients on the mass or height of the sample and interpretated this finding with the absence of diffusion into the bulk.Dr Kingopened the discussion of Professor Rudich’s paper by presenting slides demonstrating the use of Raman microscopy to study the uptake of nitric, acetic and formic acid on calcium carbonate particles. The work confirmed that the atomic nitrogen signal seen in the STM studies was indeed nitrate (owing to the Raman N–O nitrate stretch), and that deliquescence of the products occurred below 30% RH. The reaction went to completion. The reaction of acetic acid with calcium carbonate produced long crystals of calcium acetate (confirmed by Raman spectroscopy) that did not deliquesce. The reaction of formic acid with calcium carbonate produced calcium formate that also did not deliquesce at 30% RH. No evidence of new crystal growth has been observed as in the formate case. Neither of the reactions stopped after a surface passivation of calcium carbonate crystals, but appeared to go to completion.Dr Coxcommented: I would like to report some laboratory measurements pertinent to the atmospheric transformation of CaCO3particles into Ca(NO3)2. Fig. 4 shows the uptake coefficient (γ) for N2O5onto a submicrometre CaCO3aerosol measure in a flow tube experiment, as a function of relative humidity (RH).γincreases with RH approaching 10−2at high RH. Nitric acid produced in the heterogeneous hydrolysis of N2O5is expected to react with CaCO3to form (soluble) Ca(NO3)2with release of CO2. This would provide one efficient route for the transformation implied in the analysis of the atmospheric samples.Uptake coefficient of N2O5onto CaCO3measured in aerosol flow tube coupled to a chemiluminescence NOxanalyser. [N2O5]0∼ 500 ppb. Aerosol produced by atomising a suspension of CaCO3in distilled water; surface areas measured by a DMA. Reactive uptake according to reactions: N2O5+ H2O ⇌ 2HNO3; 2HNO3+ CaCO3⇌ Ca(NO3)2+ CO2+ H2O.Professor Waynesaid: Since N2O5hydrolysis is largely a heterogeneous process itself, at least in the atmosphere, the CaCO3particle probably plays a role in the initial step as well as in the subsequent conversion to nitrate.Dr Coxresponded: That is exactly the process we envisage. Following uptake at the CaCO3particle surface, N2O5molecules are hydrolysed to HNO3. The uptake rate increases with relative humidity, which we believe reflects the increasing amount of surface adsorbed water available to support the hydrolysis reaction. CaNO3is more hygroscopic than CaCO3and the hydrolysis may speed up on partially reacted particles due to increased water content, although this effect would be limited by the known inhibiting effect of high [NO3−] on heterogeneous N2O5hydrolysis.Professor Zellnerasked:(1) Why does the surface reaction product not limit the extent of the bulk reaction? Although it is realized that formation of a gaseous CO2product will help the reaction to run to completion, there may still be a limitation in transport of nitric acid into the bulk of the particles. Such limitation could only be prevented if the primary surface product will deliquesce and therefore ease the transport of nitric acid through a liquid surface layer.(2) The deliquescence of inorganic salts is heavily affected by the presence of organic species. Why do you not see the effect of organics in the evolution of your aerosols?Professor Rudichreplied:(1) The product Ca(NO3)2is shown to deliquesce at relative humidity of about 12%. In addition, we presented evidence that the converted particles are liquid at ambient conditions. The liquid layer conceivably allows the fast transport of the reactants and CO2from/to the atmosphere from the unreacted part and hence allow for the reaction to proceed until completion.(2) The effect raised by Prof. Zellner is observed mostly for homogenously mixed organic/inorganic particles. In dust, adsorbed organics are found on the surface of the dust particle.1Therefore it is expected that such coating could delay the reaction of the bulk by slowing mass transfer. However, if the coating is incomplete, the reaction may still proceed and once the reaction starts, it will proceed. Our experiment cannot probe such process, although preliminary experiments in our lab suggest that a thick coating of humic substances on dust may delay the reaction slightly. We do not expect that the minute amount of possible adsorbed organics may than be enough to show a substantial effect on the deliquescence of Ca(NO3)2, and in that case, ESEM methods will not be able to probe such saddle effects, should they exist.1 A. H. Falkovich, G. Schkolnik, E. Ganor, Y. Rudich,J. Geophys. Res., 2004,109, DOI: 10.1029/2003JD003950.Professor Ravishankarasaid: Could you please comment on why you think that your dust particles also had organics? Do you have other evidence for the presence of organics in or with your dust particles? You say that they are mixed with other air masses with organics. (Just a note—mixing is not easy to deal with when you have trajectory calculations to figure out where the air came from. After all, trajectories assume no mixing!)Professor Rudichresponded: We have not claimed that the dust particles probed in this study had any organics on them. However, in previous field studies we showed evidence and discussed the adsorption of organics (mostly pollutants such as PAH and pesticides) on dust transported from North Africa to the Eastern Mediterranean.1At least in this region, the amount and type of organics spends on the dust trajectory: dust that passed of over the Mediterranean Sea contains less, and different types of organic species than dust transported over land and passes over agricultural and urban areas in the Nile Valley and in Southern Israel.1 A. H. Falkovich, G. Schkolnik, E. Ganor, Y. Rudich,J. Geophys. Res., 2004,109, DOI: 10.1029/2003JD003950.Dr Kingaddressed Professor Zellner: In the Raman microscope study of the reaction between HNO3acid and CaCO3in real time it was notable that the uptake of water began after the surface oxidation of CaCO3and then the reaction accelerated uptaking water and HNO3relatively quickly (reaction over in 5–10 min) until a solution of CaNO3was all that remained. There was no evidence of surface passivation.Dr Allanasked: Given that the processed particles are aqueous, it may be possible that they are collected more efficiently on the impactor substrate through the elimination of bounce. Is it possible that the statistics may be biased towards the processed particles?Professor Rudichreplied: It is possible. We have not quantified this process and hence we cannot comment on its possible effect.Professor Donahueasked: Do these data speak to parameterizations of rainout or washout of aerosol as a function of size?Professor Rudichreplied: The current study does not allow for such parameterization and further studies that specifically address these questions should be planned.Dr Baltenspergercommented: According to Fig. 1, RH was below 100% during rain events. From this, I conclude that your site was below the cloud during the rain event. If you had only below-cloud scavenging I would not expect highly preferential scavenging for processed particles.Professor Rudichresponded: The RH shown is from a meteorological station in the region and not from the site itself. During the rain event, the area was inside the cloud.Professor Herrmannsaid: During trajectory analysis the capability of HYSPLIT to identify precipitation periods could have been applied and might have led to further valuable information.Dr Kingopened the discussion of Professor Ziemann’s paper by presenting slides of the very recently published work by King, Thompson and Ward1demonstrating the use of laser Raman tweezers to hold a mixed droplet of oleic acid and synthetic sea-water at atmospheric pressure for 30 min, oxidize the oleic acid on the droplet with gas-phase ozone, follow the decay of oleic acid and the growth of the products (nonanoic acid and nonanal) with Raman spectroscopy. The growth of the droplet size as the droplet becomes more hydrophilic was monitored. The oxidation of the surface film of oleic acid turned a hydrophobic particle into a hydrophilic particle which then grew in size (data were presented for a particle that grew from 6.5 μm to 8 μm with a corresponding growth of nonanoic acid in the particle. Monitoring mixed droplets of oleic acid and synthetic sea-water in the size range of 1–7 μm demonstrated no size change in the absence of ozone.1 M. D. King, K. C. Thompson and A. D. Ward,J. Am. Chem. Soc., 2004,126, 16710–16711.Professor Ziemannreplied: This is an interesting study demonstrating a powerful new technique. The ability to monitor changes in particle compositionin situusing Raman spectroscopy offers important advantages over more destructive mass spectrometric methods that sample particles into a vacuum system for analysis. This is especially the case for particles containing water or other semivolatile species whose distribution between the gas and particle phases could be disturbed during sampling. This method will be useful for monitoring changes in functional groups as a particle reacts, but identifying individual compounds will be challenging and in most cases probably not possible because of the complex mixtures of products typically formed by organic oxidation.Can you distinguish between different organic acids with your Raman technique?Dr Kingreplied: With care, yes. It was possible to tell the difference between azelaic acid and nonanoic acid for instance.Dr McFiggansasked: Given Dr King’s assertion of surface and bulk products could someone please comment on the phase distribution of the products (gas, surface, bulk, solid, solute) as would be expected in real particles in the moist atmosphere?Professor Ziemannreplied: With regards to the products of oleic acid ozonolysis, one would expect nonanal and nonanoic acid to be present primarily in the gas phase, and the others in the particle phase. Because the particulate products are not very water soluble, they will probably be dissolved in the organic phase, which could be solid, liquid, or waxy, rather than in the aqueous phase. Products with carboxylic acid groups may act as surfactants and exist at the air–water or organic–water interface.Dr Kingadded: The question as I remember it was how to probe the phase distribution of whether of a product was at the surface or in the bulk and the answer is with scattering and reflectance techniques used in colloid science. Such experiments are in progress.Professor Rudichmade a general comment: Nonanal is the product with highest vapour pressure, it has been shown to evaporate from OA particles. In addition to the formed peroxides, people have monitored other high molecular weight products.1–51 T. Moise and Y. Rudich,J. Phys. Chem. A, 2002,106, 6469–6476.2 Y. Katrib, S. T. Martin, Y. Rudich, P. Davidovits, J. T. Jayne and D. R. Worsnop,Atmos. Chem. Phys., 2005,5, 275–291.3 Y. Rudich,Chem. Rev., 2003,103, 5097–5124.4 T. Thornberry and J. Abbatt,Phys. Chem. Chem. Phys., 2004,6, 84–93.5 Y. Katrib, S. T. Martin, H.-M. Hung, Y. Rudich, H. Zhang, J. G. Slowik, P. Davidovits, J. T. Jayne and D. R. Worsnop,J. Phys. Chem. A, 2004,108, 6686–6695.Professor Donahueasked: Oleic acid is typically present in organic aerosol in a small fraction with respect to saturated organic acids. In fact, the mole fraction of double bonds is quite small. What effect will these factors have on ozonolysis in real ambient organic particles?Professor Ziemannresponded: This could affect both the reaction products and kinetics. If the particle matrix is relatively nonpolar because saturated organic acids and other oxygenated compounds comprise only a small fraction of the particle organic mass, then stabilized Criegee intermediates (SCI) are expected to recombine to a significant extent with their aldehyde co-products to form secondary ozonides. If the matrix is relatively polar, however, then SCI become more solvated and the probability for reactions with saturated organic acids to form α-acyloxyalkyl hydroperoxides is enhanced. With respect to reaction kinetics, the presence of saturated organic acids can lead to a more waxy or liquid/solid matrix that reduces the diffusion of O3and oleic acid, thereby slowing down the rate of ozonolysis.Dr G. Smithasked:(1) Could the nonanoic acid product from oleic acid ozonolysis be evaporating during sampling into the vacuum chamber of your instrument?(2) The observed rate of oleic acid reaction depends on particle size. What size or size distribution of particles did you use?(3) Under the assumption that the reaction (O3+ oleic acid) is limited by O3diffusion (the so-called “square root” case), the rate of oleic acid reaction in the OA-DOS particles should be about three timesfasterthan for pure oleic acid. Your observations suggest that the reaction occurs at the surface of the particle instead of in a layer below the surface. Have you considered the analysis of your kinetic data in this regard?Professor Ziemannreplied:(1) I don’t think that nonanoic acid is evaporating from the particles during sampling. Experiments and calculations indicate that compounds that have sufficiently low volatility to be present in particles before they enter our thermal desorption particle beam mass spectrometer will not evaporate before they reach the detection region. Instead, I believe the nonanoic acid is evaporating in the environmental chamber where the particle residence times are on the order of minutes. We know from experience preparing nonanoic acid particle standards that they will completely evaporate within a few seconds.(2) Because the size dependence enters the kinetic expression as a particle volume/surface area ratio, we used a particle radius of 0.2 μm calculated from the ratio of the particle volume and surface area measured for the polydisperse aerosol using a scanning mobility particle sizer.(3) I have not considered other reaction mechanisms because I felt that the rather large scatter in my data would make it difficult to draw conclusions. The major point of the analysis was to demonstrate that the uptake coefficients for the pure oleic acid and oleic acid/DOS mixture were in reasonable agreement with literature data, and that the monocarboxylic acid matrix dramatically reduced the reaction rate.Professor Abbattasked: Do you have any evidence from your work of the chemistry that ozone might undergo with oleic acid substrates at very high ozone exposures,i.e.after the initial reaction of the carbon-carbon double bond is complete? I ask this because we have seen the CCN activity of oleic acid particles increase only after the ozone exposure is raised substantially beyond that needed to give rise to the particle size change attributable to loss of nonanal (cf. ref. 1).1 K. E. Broekhuizen, T. Thornberry, P. P. Kumar and J. P. D. Abbatt,J. Geophys. Res., 2004,109(D24206), DOI: 10.1029/2004JD005298).Professor Ziemannsaid: This is an interesting observation, and I do not have a clear answer. It is known that the rate of reaction of ozone with saturated hydrocarbons is very slow, so it seems unlikely that further oxidation of hydrocarbon chains is occurring. One possibility is that at high concentrations ozone might react with some of the peroxide products, such as secondary ozonides or α-acyloxyalkyl hydroperoxides, leading to decomposition. A major decomposition product would most likely be carboxylic acids, which could enhance CCN activity.Professor Ravishankaraopened a general discussion of the papers by Dr Seisel, Professor Rudich and Professor Ziemann: These are very nice works. It really helps us understand the complexity of the issues regarding the formation and reactivities of aerosols.Could the authors address how the information derived in their studies can be applied to the atmosphere? Also, how could one expand such work to make sure that the derived information is more useful for atmospheric modeling.Professor Ziemannanswered: The most important conclusion from my study is that the particle matrix can have a major effect on the reaction products and kinetics of oleic acid ozonolysis. The primary particle-phase products are organic peroxides, but the specific peroxides formed (e.g., secondary ozonides or α-acyloxyalkyl hydroperoxdes) depends on the polarity of the particle matrix and the concentrations of functional groups such as carboxyl, hydroxyl, and carbonyl that can react with stabilized Criegee intermediates. The phase of the particle affects the rate of ozonolysis by altering the rate of diffusion of ozone within the particle. Incorporating these results into atmospheric models would therefore be difficult, since it would require more knowledge of the organic composition and phases of real atmospheric particles than is currently available. Future studies should investigate the products and kinetics of heterogeneous alkene ozonolysis reactions with different alkene classes and matrices, approaching atmospheric complexity when possible, and seek to obtain more information on atmospheric particle composition and phase properties.Professor Rudichreplied: The specific study presented here shows beyond doubt that processing of Ca-containing mineral dust by HNO3or N2O5(as was shown by Dr Cox) occurs in the atmosphere by mixing of polluted air with natural aerosol particles. It now remains to quantify whether this process has implications regarding the possibility of such dust particles to substantially scavenge atmospheric nitric acid, and what are the resulting implications of the processed dust particles on atmospheric issues such as haze formation and cloud formation. We already demonstrated here that the processed dust participles are liquid at ambient relative humidity. Once these processes are quantified, it should be relatively easy to include them in models.Professor Herrmanncommented: Oleic acid was chosen as a proxy for the numerous fatty acids encountered in tropospheric particles. Detailed laboratory studies are now able to fully elucidate the reaction mechanism of such compounds in tropospheric particles as demonstrated by the study of Professor Ziemann.The question is how the enormous amount of information from such detailed studies might be incorporated into multiphase models. Maybe lumping of compounds will become necessary as it has been applied before in gas phase modelling,e.g.by Bill Stockwell.Professor Ziemannreplied: I agree that there will be a need to lump the heterogeneous chemistry of organic compounds as is done for gas-phase reactions. It may be possible to group reactions according to simple compound classes such as alkenes, alkanes, aromatics, and oxygenates reacting either with OH radicals, NO3radicals, or O3. The most difficult of these reactions to model may be the ozonolysis of alkenes since the products and kinetics are both strongly dependent on the particle matrix, which can be extremely complex and is not known for atmospheric particles. Reactions of compounds with OH radicals, which are the major atmospheric oxidant, may be simpler. Because the reaction is likely to occur very close to the surface, there may be no significant matrix effect on the kinetics. This could allow the initial oxidation kinetics to be modelled rather easily. Furthermore, because the reaction of OH radicals with most organic compounds will lead to the same initial product: an alkyl radical, the chemistry may be similar for many compounds. The challenge will be to determine the relative proportions of OH radical reactions that lead to volatile productsviadecomposition compared to those that add functional groups (e.g., hydroxyl, carbonyl, carboxyl, and nitrooxy) to the carbon chain. The former reactions will alter particle composition by volatilizing organic compounds, whereas the latter reactions will lead to products (and particles) that are more polar, less volatile, and more hygroscopic.Dr Seiselreplied: It will certainly not been possible to incorporate all information from detailed laboratory studies into atmospheric models. However, in order to lump together several compounds or make reasonable approximations, details on the reaction mechanism ase.g.the rate-limiting step have to be known. The goal of such detailed studies is therefore not to fill atmospheric models with an enormous amount of kinetic data. Rather, these detailed studies provide the necessary tool to simplify complex reaction mechanism.Dr McFiggansmade a general comment: It is evident that there is vast complexity at every process level and that, from field studies, vast numbers of individual organic components. It is also evident that the atmosphere averages/integrates the kinetic processes, chemical and thermodynamic properties as probed for various atmospheric implications.I don’t anticipate that a single number analogous to “mean global [OH]” would be a meaningful product to describe any aspect of atmospheric aerosol behaviour. However, for purposes of for example cloud activation, direct optical properties, phase partitioning, health effectsetc, it is necessary to reduce the complexity from the detailed studies of all possible systems. How can we strategically plan a meaningful reduction of the complexity as identified in this meeting towards useable products? If we do not do it as a community, parameterisation of the processes used ine.g.climate models will not be based on the physics and chemistry at a detailed rigorous level.Dr Kingmade a general comment: Atmospheric aerosol contains a high proportional of transition metals which are known to catalyse the decomposition of the ozonides formed from the reactions of ozone with unsaturated compounds such as oleic acid. Should studies of the oxidation of the organic alkenes in the atmosphere include these transition compounds?(N. B. The study of Kinget al.1oxidised oleic acid on synthetic seawater droplets containing many transition metals and a different ratios of products were seen).1 M. D. King, K. C. Thompson and A. D. Ward,J. Am. Chem. Soc., 2004,126, 16710–16711.Professor Ziemannreplied: I think that as knowledge is gained about the oxidation of alkenes in pure form and in simple organic mixtures it will be important to investigate more complex systems. Information is needed on organic mixtures more similar to real atmospheric particles and on the effects of species such as water, sulfuric acid, metals, and inorganic salts as well as UV light on the chemistry.Dr Baltenspergercommented: Of course, we all would like to knowwas die Welt/Im Innersten zusammenhält“what holds the Earth together in its innermost parts” (as Goethe says). However, in order to be able to do that we need to understand the fundamental processes. We just should be modest enough and not over-interpret our data. And we should talk to each other: since chemistry and physics talk to each other in a particle, we scientists should do as well.Dr Tuckcommented:(1) I strongly recommend that techniques, such as neutron analysis, that examine the surface layers of aerosols should be applied to real aerosols as well as idealized laboratory populations.(2) Many medical studies of the lung use surfaces of palmitic acid as a laboratory surrogate for the surface of lung tissue. This is interesting from the point of view of health effects, since it is very striking that the TOF-SIMS analysis of real aerosol surfaces done in Finland contain palmitic acid as one of only a very few prominent peaks.11 H. Tervahattu, J. Juhanoja, V. Vaida, A. F. Tuck, J. V. Niemi, K. Kupiainen, M. Kulmala and H. Vehkamaki,J. Geophys. Res., 2005,110(D6), DOI: 10.1029/2004JD005400, Art. no. D06207