In the ongoing battle to persuade the world that global warming is real and is a problem, advocates are waging a two-pronged attack. And each prong is heavily based on global climate models that have accounted for the largest portion of the research dollars "invested" in this issue.

Not long ago, theirs was but a simple one-pronged attack: Climate models say the world will warm [x] degrees (insert your favorite large number here)—implying that the time for action to combat the forthcoming tragedy had long passed. But so-called global warming "skeptics" (i.e., people who base their arguments on data rather than speculation [e.g., climate models]) cried foul when these self-same models proved incapable of reproducing observed climate variations. Forecasting is always the easy part, it's the verification that's the bane of TV weathercaster and climate modeler alike.

Enter Prong No. 2: fingerprint detection. Though everyone agrees that climate has a lot of inherent variability that serves to screw up comparisons with models, it nevertheless should be possible to detect the impact of human activities (from greenhouse gases)—the so-called human "fingerprint"—on global climate. Take a model, add a slow greenhouse-gas buildup over time, and compare the resulting pattern of temperature change with the observations. If they match up, then presto! The observed changes are caused by greenhouse gases. (Of course, no self-respecting scientist would say "caused," but a journalism major is happy to veer off in that direction.) Most importantly, model forecasts of the future can now be trusted, since they have successfully reproduced past observations.

Quite a story.

This context explains the hype and hoopla surrounding a new study in Science by Scripps Institute of Oceanography scientist Tim Barnett and two colleagues. Barnett used the existing approach for fingerprint detection of air temperatures over land and applied it to the world's oceans using a newly compiled data set that shows changes in water temperatures to a depth of 3,000 meters. Barnett ran a climate model and compared the observed changes since 1995 with the changes in ocean temperatures produced by the model that forced increasing greenhouse gases and sulfate aerosols.

Before we show you the results, here's a very important aside that in reality is more of a main course. No matter how complex a climate model is—no matter how many layers it has, how complex the parameterizations of cloud processes are, or how many soil moisture or sea-ice feedbacks exist—when the model is forced by increasing carbon dioxide, it will warm, and usually in a linear fashion. Sulfates are merely added to lower the future warming rate to a value that's not inherently ludicrous, but rather merely ridiculous. The models can do nothing but produce a warming. They have no choice. More greenhouse gases equals more warming, period.

But for some reason, our real global climate, which apparently hasn't been paying attention to the P. R. from the modeling community, sometimes cools. If this cooling happens for a decade or more (like the surface air temperatures did from end of World War II until the mid-1970s), well, the model is screwed. It simply can't produce a cooling with all those nasty greenhouse gases in the atmosphere. So, for the fingerprinting to work, other things have to be added to the model that can generate a cooling. These can be anything from volcanic expulsions to changes in solar energy to outright cheating (by pre-specifying observed ocean temperatures rather than modeling them, for example, which was the approach of NASA's rocket scientists).

Figure 1 shows model-predicted values of oceanic heat content, averaged from the ocean surface down to 3,000 meters, from 1955 to 2000, compared with the observations. Here, the observed values are highly smoothed (they simply use the mean value for each 10-year period), because otherwise they would not appear to match the model output. Do you think the two lines match? Well, as usual, the model produces a warming ocean, but in every ocean basin except the South Atlantic, the oceans actually cooled between the mid-1970s and mid-1980s. To get around that, error bars are added to the model forecasts (though not to the observations)—showing, according to the authors, "an unexpectedly close correspondence between the observed heat-content change and the average of the same quantity from the five model realizations."

Figure 1. Modeled (shaded region) vs. observed (dotted line) oceanic heat content, averaged from the ocean surface down to 3000 meters, 1995 to 2000.

And now on to the fingerprint detection. In Figure 2, we reproduce the modeled and observed ocean temperatures at depth (down to 2,000 meters) over time. Do they match up? The answer depends of how far away you are from your computer screen right now. If you're looking at this graph with one eye shut from across the room, then you'd better sell your beach house now, because global warming is coming with a vengeance. But step up closer, and let's look at these data a little more carefully.

Figure 2. Modeled and observed ocean temperatures at depth (down to 2,000 meters) over time.

We plotted the observations and model predictions of temperature anomalies at depth at the end of the record (about 1995) for the North Atlantic and North Pacific Oceans (Figure 3). The model essentially produces a huge surface warming that weakens the deeper you go into the abyss. But yet again, nature seems to be operating under a different set of physics. That is one lousy forecast, especially at the surface. If, however, you want the temperature 2000 meters below the surface, where temperatures seem to be unaffected by greenhouse gases, then the model does a fantastic job.

Figure 3. Modeled and observed temperature anomalies at depth (down to 2,000 meters) at the end of the record (about 1995) for the North Atlantic and North Pacific Oceans.

But wait a second. Look again at Figure 1. At the end of the record, the models and observations seem to match perfectly. How is that possible, given Figure 3? To demonstrate, quickly estimate the average of the following numbers:

0.35, 0.30, 0.25, 0.15, 0.10, 0.5, 0.3, 0.1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.

And now the following row:

0.12, 0.13, 0.2, 0.12, 0.12, 0.12, 0.1, 0.07, 0.06., 0.05, 0.05, 0.05, 0.04, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.

If you said the two averages were both pretty darned close to zero, you win a new slide rule!

And that's also essentially what Figure 1 shows. In the top layers of the ocean, where all the temperature variations are taking place, the model does in fact do a lousy job. But since the averages are taken over a layer that extends down to 3000 meters, most of which includes no variation, then the model is now excellent because the important fluctuations are averaged out. (This situation is not unlike the unforgettable brouhaha over Benjamin Santer's claim that he first detected the human-induced greenhouse warming fingerprint in the atmosphere in 1996: The surface and lower atmosphere temperatures don't really match, but he used a statistic that depended heavily on strong cooling of the stratosphere for confirmation).

What we are seeing with the Barnett paper is more of the same. We have claims that a general circulation model can reproduce ocean temperatures when, in reality, it cannot. We have evidence of a human fingerprint in ocean temperature patterns that arises only when the data are substantially smoothed. And we have a press corps that's even more convinced of the certainty of significant human-induced global warming. In fact, however, evidence for the human global warming fingerprint remains elusive.

Reference:

Barnett, T.P., D.W. Pierce, and R. Schnur, 2001. Detection of anthropogenic climate change in the world's oceans, Science, 292, 270–274.