Data limitations become extremely severe when one attempts to quantify natural variability on decade-to-century time scales from proxy records. What information can we really derive regarding the patterns of large-scale variability from widely scattered proxy measurements? I think that further effort is required in two areas to maximize and better quantify the information content of the entire climate data base, both instrumental and proxy. First, we need more studies along the lines of data system tests. More specifically, using the more dense data distribution of recent decades, we should determine the loss of information as the distribution of observations is decreased. This will help establish confidence limits for estimates of large-scale averages based on the sparse distribution of observations and proxy series obtained during earlier decades and centuries.
Second, we need to place more emphasis on integration of information from various proxy sources, with the goal of obtaining a better picture of the pattern of global variability during earlier centuries. This is admittedly a formidable task when one is dealing with decade-to-century time scales. I do not believe that the concept of an overall integration of information in decade-to-century variations has yet to take root in the proxy community, but I think it should be strongly encouraged as a long-term goal.
GROISMAN: Two quick comments, one for Gene and one for Henry. Gene, I just wanted to note that the SST variability does not reflect conditions over the Arctic. There is about a 25 percent difference in amplitude between SST and marine air temperature over the open ocean, the latter being higher, so there may be other reasons for the difference you cite. Henry, the problem with tree rings is that there are very few 1000-year-old trees, so in a time series you can't resolve the same mean variance.
DIAZ: I used only air temperature over land, no SSTs. Also, I started in 1891, when we had fairly good coverage. The changes in data distribution don't have much influence on the large-scale averages. As for the tree rings, we did try to compensate by looking at decadal means and interdecadal variability, and I don't think the middle section has much of a problem.
KEELING: I have a figure that is based on the Jones-Wigley temperature record, with a slightly different smoothing. Some of the early variability and the apparent biennial signal are probably the result of sparse data; beginning in 1951 the record is fairly homogeneous. You can see the 1958, 1961, 1963, 1982 El Niños. I'd like to know whether there is some difference in the quantity of data, or in its processing now that satellite data and ground data are being mixed, that has changed data variability in the last decade.
TRENBERTH: Perhaps when you fit the spline you are taking out more of the variance at the end.
GHIL: Henry, I was puzzled by the apparently much higher interdecadal variability in the second of your first two tree-ring figures.
JONES: It's partly due to the indexing procedure that there is more low-frequency variance in the top curve than the bottom one, though some of it could be real. KARL: It's important to remember that there are many different ways to classify variability. I suspect that I could show more or less of it just by defining it differently. On this particular figure, I think we need to be aware that one is for a longer season than the other, and the shorter period of a time series will give you higher variability.
DIAZ: Another point is that getting longer time series from proxy records may not improve our understanding of climate change mechanisms.
GROOTES: I want to comment on the Quelccaya ice-core records you used. As you go down from the surface, the seasonal cycle disappears. It reappears about 1880 and continues significant to about 1480. Apparently when there are low values the record is smoothed so little that the seasonal cycle is preserved. As temperature changes it becomes more difficult to calibrate the isotope changes; you may be looking at post-depositional changes rather than the signal itself. Tree rings and ice cores both require particular attention to understanding exactly what you're recording, especially if you are talking about temperature.
MYSAK: For anyone who's interested, I'd like to mention that Brian Luckman at the University of Western Ontario has tree-ring records from the Canadian Rockies for about the last thousand years. He's showing a nice cycle of 150 years or so.
MCGOWAN: Are you comparing tree rings taken from similar altitudes at those different locations? Also, how does the variance of instrumentally measured mountain temperatures compare with that of lowland temperatures?
DIAZ: Most of the data sets are from moderate elevations. As for the high/low elevation temperature variances, the trends are reduced with respect to surface, but they're still comparable.