THOMAS L. DELWORTH
NOAA/Geophysical Fluid Dynamics Laboratory
As someone who is mostly familiar with modeling and the efforts that go into it, I am very appreciative of the tremendous effort that goes into constructing observational data sets. The data sets are of critical importance in assessing the ability of models to simulate climate and its variability. I have a few comments and a couple of examples from modeling work to emphasize the importance of the work that Dr. Parker is doing, particularly with regard to the importance of sea ice for low-frequency climate variability.
Examination of low-frequency variability contained in a multi-century integration of a coupled ocean-atmosphere model reveals that there are very long time scales associated with a number of oceanic variables in the polar latitudes of the Southern Hemisphere. In particular, there is a tremendous amount of low-frequency variability in the ice thickness, which then has a substantial impact on the model atmospheric variability. Sea ice, in some regions, can virtually disappear for times as long as a decade, with the result that surface air temperatures are much higher during this period. When you look at interdecadal model variability, features such as this are quite striking. It is critical to increase our observational data base for quantities such as sea ice in order to be able to assess whether such model features exist in the real climate system. Some of the data sets produced by Dr. Parker should help with that.
A key point in his manuscript is the theme of attempting to obtain more information from the available observations through the use of empirically motivated assumptions about the spatial and temporal scales of variations in the observed data. It is therefore critical to keep in mind when utilizing this data the assumptions that go into it. Two key assumptions are the degree of persistence in time and the spatial structure of the data. In general, this represents a powerful technique for supplementing existing data sets. There is, however, a potential for this technique to underestimate variability in data-scarce regions. This potential bias must be kept in mind.
An additional point Dr. Parker made is the critical importance of recovering and digitizing existing observations from data-poor regions. Allow me to comment on the importance of this with an additional modeling example. In characterizing low-frequency variability, a simple technique is to compute at each grid point the serial correlation of data time series. This is a measure of the persistence in the data and thus the inherent time scales. The results of such computations using annual mean sea surface temperature computed in a coupled model shows some very intriguing features. The longest time scales of sea surface temperature anomalies appear to be associated with higher latitudes where deepwater formation is occurring in the model. There are also very long time scales in the circum-Antarctic region. These are features that we would like to investigate in the observations, but the limitations in observational data sets make that difficult. For example, computing serial correlations of observed annual mean sea surface temperature from MOHSST4 produces a map with large data-void regions. In particular, the very intriguing model features in variability occur in regions with insufficient observations to assess their validity. Therefore, the technique that Dr. Parker is trying to use to extract more information from the observations available is really a critical one.
The techniques of data analysis described in the manuscript check for physical consistency between independent data. In particular, comparisons were made between derived and observed wind fields, and inconsistencies between the two were noted and studied. This is an important technique, and is an appropriate method to assess what features are real and what features are not.
Finally, as Dr. Parker mentioned, these data sets are of great utility in conducting simulations of the climate of the twentieth century. It would be beneficial to have other variables available in such a format. For example, land-surface processes might have a role in climate variability. One can envision a similar data set of time-varying land-surface characteristics over the twentieth century.