RASMUSSON: There's much to be said for working only with real data. I hope you will also make available fields that are real data only.
PARKER: Yes, MOHSST5 is available, and we'd like to improve it by digitizing all the data possible.
KUSHNIR: If you have a good approximation to the covariance matrix structure, you can use that with more confidence than Laplacian methods to fill up missing regions.
PARKER: That would be similar to the eigenvector technique, which I think would be a good one here. We just haven't got that far yet. Also, I think getting more real data is important; for example, our pre-1953 sea ice is all from an old climatology, so the sea-ice anomalies are the same from year to year.
TRENBERTH: The use of covariances in either space or time depends very much on the nature of your signal. For instance, when you put in greenhouse warming a global-scale structure is superimposed on your fields, and larger spatial and temporal correlations are implied than you might otherwise get.
PARKER: Indeed. What would really be nice would be to have designators categorizing the quality of the data for each of the months you're presenting—for instance, with the Air Force's cloud-cover analysis, designating which are real data and which climatology.
KARL: David, when you said you tried to make the frequency of observations consistent with that of the nineteenth century, did you actually reduce the number of observations in a grid box to match the earlier ones?
PARKER: No, we had to use the box values, so we were just testing for errors resulting from the interpolation scheme. Selecting observations would have been a major computational exercise. But we are aware of that problem.
KEELING: When you are adding the COADS data to your own data set, what do you do when you have both? Can you identify the data sources and tell which are original observations?
PARKER: For the moment we are inserting the COADS SST only if the box is missing from our data set. We hope that with Scott Woodruffs help we will be able to blend the two sets, removing duplicates, by 1994. Ultimately we'd like to have everything that can be digitized.
KEELING: How hard would it be, then, to average your data by anything other than month?
PARKER: In our scheme the data are in five-day periods, or pentads, so that you could do half-months or three-month periods. We also have data sets of the bucket corrections we used so that anyone who wants to remove them from the data as they stand can do so.
JONES: We have now digitized the spring and summer months of that Danish sea-ice chart series that runs from the turn of the century to about 1960. I believe John Walsh is working on the winter ones.
WALSH: We're aggregating the narrative reports into a data set for winter. But I'd like to add that no one has yet taken advantage of the spatial character of these reports the way David has with SSTs. There is also a set of reports going back to the turn of the century from an Antarctic ice station, and some sort of spatial extrapolation procedure used on both of these might give us some measure of interannual variation in certain limited regions.
GHIL: In addition to data origin identifiers and such, I would really like to have error bars with your data.
PARKER: Well, there's a first approximation already there—the root-mean-square error fields—though they might be a lower limit, considering Tom's comment about the number of observations per grid box.
GROISMAN: I wanted to mention the dangers of using the German and Russian climatology of sea ice for the 1920s to 1940s, since sea ice was retreating considerably during this period. I hope that some day you will be able to use some Russian arctic observation data now stored in the Arctic Institute-—not digitized, of course.
LEVITUS: I'm happy to say that we at NOAA are making arrangements with the appropriate officials right now to do just that.