at the U.S. National Climatic Data Center. Support for David R. Easterling was provided under Department of Energy (DOE) Interagency Agreement DE-A105-90ER60952 and NOAA's Climate and Global Change Program.
DAVID A. ROBINSON
I think Dr. Groisman's presentation has been an excellent example of the advantages and pitfalls of dealing with observational data and of putting data together on a variety of temporal and spatial scales. And I can assure you that the paper contains some tremendous information that, unfortunately, Dr. Groisman had time only to gloss over. I would like to break the study down into a couple of segments to foster further discussion.
The first section I shall simply entitle ''obtaining numbers". This includes the problems Dr. Groisman presented that relate to measurement techniques, whether for obtaining liquid rainfall or, more important, getting adequate snowfall measurements. Next, there are the problems in developing an observational network that is homogeneous through time—which, as Dr. Groisman found, was impossible to do for the entire North American continent. He could look only at the southern reaches of Canada and over the United States for the entire century. When it came to looking into the Arctic, he could go back only about 40 years.
Third is the paper's serious criticism, which may have passed people by too quickly, that it is difficult to believe anything that comes out of the primary stations in the United States. The figure that shows the changes at these stations (relocation, gauge changes, wind shields) really substantiates that. So we essentially have different observational networks that vary through time.
Last, there is the whole idea of metadata. We must have enough information on the history of a station or network that we can go back and reconstruct a homogeneous, standardized time series over decade-and century-long periods. While North American metadata is certainly better than African, there are gaps and lapses in the station histories. For precipitation, the impact of station moves is less significant than for temperature. But the gauges that measure precipitation, the wind shields, the 8-inch gauge versus the tipping bucket, and so forth are as important as any problems in measuring temperature in the United States.
It occurs to me to wonder how Dr. Groisman would rank these concerns in terms of importance for obtaining homogeneous precipitation records. To me, the most significant problem is perhaps the switch to the tipping bucket with the introduction of automated surface observing systems at primary U.S. observing stations over the next few years. Some time down the road when we are looking for decadal variations in precipitation, we may have to examine the metadata for the 1990s very carefully to come up with a homogeneous time series.
Another question I have is how deeply inhomogeneities interfere with obtaining true signals. You have to do so much with the data to be able to compare it at different times at an individual station, or among stations or countries. How deeply does this limit our ability to recognize climate change? Tom Karl et al. stated in their 1991 Science paper that it would take several decades to recognize a change in precipitation over central North America that might be an anthropogenic signal. I wonder how much these inhomogeneities in the different gauges and the different national recording procedures would extend or contribute to that lag, particularly when it comes to a noisy record like precipitation.
When it comes to point-to-area transformations, the question is where to draw the line. Dr. Groisman has done a lot of extrapolation, particularly in the western reaches of North America, in order to take into account elevational factors or orographic factors for precipitation. In some cases factors as great as a 70 percent increase were applied to the precipitation results of station networks. At some point one must decide whether to stay with point data, or at least data from a similar elevational area, or to stretch and extrapolate the data in order to obtain a more spatially complete and, in terms of absolute values, a more accurate record. Dr. Groisman did make it clear that the extrapolation does not interfere with any type of trend analysis; my question relates only to the absolute values of precipitation and snowfall generated.
Now for evaluating the numbers. Let us begin with the climatologies of snowfall and precipitation. It was interesting that at a given point in time when there was a larger mean, the standard deviations increased. When it comes to climate change, it is important to know whether we see a change in the mean or a change in the variability of climate.
Next Dr. Groisman looked at time series: the 40-year variety, the 100-year variety, precipitation, snowfall, Alaska, Canada, and the lower 48. There are so many interesting things in these time series-for instance, the major increase in snowfall in the Alaskan time series in the 1980s. Is