Dear COBRA Collaborators, At long last, we have data from the Scripps flasks to share with you. I would first like to apologize for the extreme delay in getting these out. There were many factors involved - measuring O2 is certainly not easy and we're still learning about potential sources of bias, the analytical system had not been built at the time of the flights and unforeseen difficulties in making it work and in developing a data processing system arose, the diagnosing of apparent sampling problems has been difficult, and our decision to pursue CMDL and INSTAAR measurements on these same flasks added further complexity and necessary testing - but the folks in Ralph's lab worked very hard on all of these issues and the only unnecessary delays resulted from my own inability to free up enough time to see the process along. I am now committed to working (with John Miller) on finishing the analysis of these data and further methodological testing in preparation for the Brazil campaign in a timely manner. We filled 300 Scripps flasks during the 2000 campaign. These were analyzed in a purgeing (constant pressure) mode at Scripps for CO2, O2/N2, and Ar/N2. This means of analysis can be critical for avoiding fractionation of O2/N2, but results in a small amount of 'push' gas entering the flask. The flasks were then 'sniffed' at Scripps for CO2 and a dilution value was calculated from the purge, sniff, and push-gas CO2 values. The flasks were then analyzed at CMDL and INSTAAR and these values were back-corrected for the dilution by the push-gas at Scripps. Only the dilution number transfered between labs - the values for the push gas were assigned by CMDL and INSTAAR analyses of prepared flasks. Of the 300 sample flasks, 11 have been flagged for problems during sampling, and 18 have been flagged for problems during analysis at Scripps, leaving 271 'good' flasks of which a small number have been flagged for problems in analysis at CMDL and/or INSTAAR. The data is now available in several forms by anonymous ftp to ftp.cmdl.noaa.gov in directory pub/john/cobra2000. The file cobraflask.ps is a multipage postscript file created by John Miller that shows plots of all of the results broken down by flight segments. The file sioflaskdatav3.xls is an Excel spreadsheet with all of the data from the Scripps flasks, a description of flags and headers, and some example plots. The file sioflaskdatav3.csv is a comma deliminated version of the data, and sioflaghead.txt is a description of flags and headers. The file sioflaskreadme.txt is a copy of this email. The file cobra_scripps_v_cmdl.xls is a comparison between Scripps and CMDL flask data compiled by John Miller. Because the flasks were not collected at exactly the same time, he interpolated between CMDL flasks to obtain values to compare to the Scripps flasks and then filtered for good CO2 agreement to reduce natural variability. After the campaign, we discovered a small scratch in the rotor of the mutiport valve in the sampling unit. It is clear from the flask pressures measured at Scripps and later testing that this scratch allowed sample air to continue to enter a flask from the high-pressure side of the valve for approximately 60 seconds after isolation, totaling about 10% of the flask volume. This does not appear to have caused significant 'smearing' of the sample collection time, but it did affect the O2/N2 and Ar/N2 ratios (see below). We have calculated values from the most recent Harvard data for comparison, using an optimal weighting function consisting of 90% from an exponential with a decay time of 26 seconds preceding the time of flask isolation (approx. equal to flask volume of 1.5L divided by flowrate of 3L/min) and an additional 10% spread over the 52 seconds after the time of flask isolation, and include these values in the dataset. We have also conducted a series of tests comparing CMDL flasks, Scripps flasks, and the method of back-calculating CMDL and INSTAAR values using Scripps dilution numbers. What follows is a description of known offsets and artifacts by gas, followed by a discussion of directions for further scientific analysis. CO2 ----- Not all of the flasks had a representative amount of overlapping Harvard CO2 data (due to the calibration cycle) for comparison. The variable PCTHUCO2 is the percentage overlap, with a maximum of 50 due to the 2 second frequency of the Harvard data. For the 177 flasks that had greater than 90% overlap (PCTHUCO2>45), the SIO-HU difference is 0.27 ppm +/- 0.43 (1sigma). Dropping 3 outliers it is 0.30 +/- 0.33 ppm. For comparison, the remaining CMDL-HU offset is 0.22 ppm +/- 0.42 ppm (below 8 km). There is a systemmatic offset between the CMDL suitcase flasks and the CMDL network flasks of around -0.1 ppm. Thus, SIO-HU is fairly consistent with what you would expect from CMDL(network)-HU. There is also an offset between CO2 measured at Scripps and that measured on the same flasks at CMDL. Scripps CO2 (CO2P) minus the back-calculated CMDL CO2 (cco2b), Scripps post-analysis sniff CO2 (CO2S) minus the initial CMDL CO2 (cco2), and the direct flask comparison tests SIO-CMDL, all give an offset of approximately -0.15 ppm. This is consistent with previously seen effects of viton under suction and we recommend using the Scripps (CO2P) values, which are measured at constant pressure, rather than the CMDL (cco2b) values for these flasks. CO ---- CO also has a well-known but much larger viton effect. This can easily be seen in the postscript figures. The SIO-HU CO difference (removing 3 outliers) is 29 +/- 13 ppb. The SIO-CMDL difference calculated by John is 31 +/- 20 ppb. We knew going in that the Scripps flasks would not produce valuable CO data and we don't recommend using these. CH4 ----- The back-calculated CH4 data appear to be good. The SIO-CMDL CH4 difference calculated by John is 3.6 +/- 7.5 ppb and the error on the test flasks run through the analysis and back-calculation was 0.7 +/- 0.6 ppb. SF6 ----- The back-calculated SF6 values look good on average, with a SIO-CMDL difference of 0.00 +/- 0.08 ppt, however it does look like during some flight segments there were persistent systemmatic offsets of up to 0.06. These offsets are somewhat significant relative to the relatively low observed variability (large industrial signals were 0.2-0.6 ppt). H2 --- Hydrogen appears to suffer from what may be a viton effect, with a SIO-CMDL difference of 21 +/- 10 ppb. N2O ------ The N2O offsets do not seem systemmatic, but do have quite a bit of variability, with a SIO-CMDL difference of 0.02 +/- 0.59 ppb. C13 ----- The Scripps C13 values by themselves look meaningful. However, there is a very consistent offset with the CMDL values of -0.11 +/- 0.04 per mil. Based on laboratory tests, John has identified a robust +0.055 bias in the CMDL suitcase flasks relative to CMDL network flasks, that is also roughly consistent with offsets seen in comparison to network flasks during the WLEF overflights. The dilution back-calculation test flasks had a systemmatic error of -0.06 +/- 0.004. The direct CMDL vs. Scripps flask tests we did (4 flasks of each) gave an average difference of 0.01 per mil. Thus it seems that roughly half of the offset is a CMDL suitcase problem and the other half is somehow a problem in the dilution/back-calculation process but not a problem with the Scripps flasks themselves. We are currently working to resolve these issues further. O18 ----- We do not have anything reliable to compare the O18 values to, as the CMDL flasks have a known water bias. However it is encouraging that plotting Scripps O18 vs. water does not show the same dependence as seen with the CMDL flasks. Ar/N2 -------- For the 81 flasks collected during the 1999 flights, the standard deviation of flask pressure was 11 torr and of Ar/N2 was 7 per meg. This variance in argon is very close to the precision Ralph and Michael Bender get on their network flasks, which was encouraging. A strong indication that something was going on with the 2000 flasks was that the flask pressure and argon values had sds of 31 torr and 17 per meg respectively. Because this increased variance was a result of the scratched rotor problem described above, we can only use the Ar/N2 data as a means to assess and potentially correct for biases in the O2/N2 data (see below). O2/N2 and APO -------------------- We would expect APO (O2+1.1CO2) to be much less variable over the continent and thus is easier to look for biases in APO than in O2/N2. Unfortunately, APO correlates with both flask pressure and with Ar/N2. The correlation with Ar/N2 is likely due to mass-dependent fractionation of the air that entered the flask through the scratch and the correlation with pressure could either be due to surface adsorption effects or fractionation due to pressurized air leaking back out of the flask through the o-ring. For the whole dataset, the slopes are -0.10 permeg/torr (r^2=0.09) and 0.33 permeg/permeg (r^2=0.26). APO should be even more constant at high altitude, and for 61 flasks collected above 5 km the correlation coefficients increase to 0.22 and 0.46. Somewhat surprisingly, Ar/N2 and flask pressure are not correlated at all (r^2=0.00 above 5 km). A combined fit to 36 flasks collected above 7 km gives slopes of -.15 permeg/torr and 0.395 permeg/permeg and an r^2 of 0.81. Perhaps coincidentally, the O2/Ar relationship of 0.395 is exactly what I calculate for fractionation due to Knudsen diffusion; (1-(28/32)^.5)/(1-(28/40)^.5)=0.395. We have calculated 'corrected' values using the combined fit above 7 km and show these as O2N2X and APOX in the database. The corrections have a standard deviation of 7 per meg, which gives a rough estimate of the magnitude of the biases. Because these biases are relatively small compared to the observed gradients of up to 100 per meg, we still hope to be able to use the O2/N2 and APO data to investigate O2:CO2 ratios and air origins. Scientific Directions and Future Tests -------------------------------------------- While most of our efforts to this point have been directed towards sorting out all of the issues described above, we now look forward to delving more deeply into what the data show (and we invite you do so too). For example, we will be interested in characterizing the relationships (O2:CO2, 13C:CO2, etc.) associated with various processes. The photosynthetic uptake signals are very strong and can be seen on many of the plots in cobraflasks.ps. A number of respiration signals can also be identified in some of the early morning flights (e.g. 824a). Biomass burning signals are also evident on a number of flights (e.g. 728a, 809c). And finally, industrial emission signals show up quite strongly on several flights (e.g. 804a2). An interesting comparison can be seen between signals from cities such as Denver and Boston and the signal on 808a from Hayes, KS (the O2:CO2 and 13C:CO2 relationships on this profile were much more consistent with local biomass burning). We will also be looking at airmass origins, comparing the flask data to back-trajectories to see for example if we can separate Northern Pacific from Gulf air. 823c1 is an example of an APO profile that suggests the lower level air came from off the Gulf, and we saw similar signals at WLEF during the 1999 flights. In terms of preparation for the Brazil campaigns, we have a number of sensitivity tests planned to resolve the C13 offsets, to verify normal operations with a new valve rotor in place, and to investigate the possibility of not purgeing the flasks at Scripps, as well as some modifications to the unit including screens on all of the lines to prevent foreign material from getting into the valve and improved diagnostics to prevent the situation that occurred in 2000. Thank you all very much for considerable patience in getting these data. Any and all comments (positive or negative) are certainly welcome. Britt and John -- Britton B. Stephens National Center for Atmospheric Research Atmospheric Technology Division P.O. Box 3000, 1850 Table Mesa Drive Boulder, CO 80307-3000 Phone: (303) 497-1018 Fax: (303) 497-1092