Tuesday, March 22, 2011

Climate smackdown: data versus models?

For decades, skeptics have tried to boil every climate change debate down to this: good, hard data vs. bad, fuzzy models. This caricature lets them attack every model-based climate projection while waiting for data to confirm the reality of human-induced global warming.

Now, here we go again. On February 10, the Wall Street Journal’s professional global warming skeptic, Anne Jolis, trumpeted recent data showing that certain global climate patterns haven’t changed much since 1871. “The weather isn’t getting weirder. The latest research belies the idea that storms are getting more extreme," she wrote, to loud applause from other skeptics. “Another nail in the coffin of anthropogenic global warming,” crowed one.

Less than a week later, the scientific journal Nature presented two papers suggesting that human greenhouse gas emissions have increased the likelihood of heavy rains. One linked the devastating UK floods of the year 2000 to global warming [see it here]. The other identified greenhouse gas emissions as a likely contributor to increases in exceptionally heavy “precipitation events” across the northern hemisphere [see it here]. These results confirm the obvious: a warmer climate leads to more evaporation, and hence more precipitation overall. But they went further, predicting where this extra precipitation would occur and linking it specifically to human emissions.

Both Nature papers relied on climate models, computer simulations of the global atmosphere. When researchers left human greenhouse-gas emissions out of these simulations — simulating the climate as it might be without industrial societies — their models projected fewer heavy precipitation events than observed. When they put them back in, the likelihood of intense precipitation went up.

The skeptic response? “No real data supporting their claims,” one wrote on Andy Revkin's Dot Earth blog. “Just climate models. GIGO [garbage in, garbage out].”

It’s a familiar refrain in the climate change wars. Climate models, goes the tune, are insubstantial fantasies. Tweak their knobs and you can make ‘em say anything. Climate data, on the other hand, are solid, substantial. “Sound science” equals “data, not models.”

But wait — about those data that made Jolis so happy... where exactly did they come from? Here’s a hint: the investigators were awarded over 3 million hours of supercomputer time to do their work. It’s called the 20th-Century Reanalysis Project (20CR, for short). “Reanalysis” is a technique for re-processing past weather data to make a climate dataset.

Here's the paper that got her so jazzed. 20CR began with a comprehensive collection of surface pressure readings covering the period 1871-2008. The project then spent some of those millions of supercomputer hours to pipe those data through a computer forecast system — a simulation model.

That forecast model uses a 3-dimensional grid to represent the atmosphere. The grid mesh contains well over 1 million points, and every one of those points must be assigned a value. Yet the surface pressure readings used as input came from a relative handful of locations — for 1871, the study’s first year, only 62 land stations worldwide. In reanalyses of this type, the vast majority of data  are calculated by the forecast model, not measured by instruments.

So the “data” that had Jolis gloating were in fact largely generated by a computer simulation — the same type of model (though not the same model) used in the Nature studies. According to some skeptics’ own tenets, then, the 20CR data can’t be much more than a scientific fantasy.

True? Of course not. Getting a scientific grip on something as big and complicated as the global atmosphere simply can’t happen without computer modeling. Today, every credible global dataset, without exception, is processed, filtered, corrected, and/or partially generated by computer models. Those who think it’s data versus models — hard evidence vs. squishy algorithms — are living in a long-vanished world, where “science” meant laboratory experiments on highly simplified systems.

So who’s right? Are human greenhouse emissions altering the chances of extreme weather, as the Nature papers suggest? Or does natural climate variability remain unchanged, as 20CR seems to show? Unfortunately, that question can’t yet be answered, because 20CR and the Nature studies addressed different climate patterns that can’t be directly compared. One thing is sure, though: it’s going to take both observations and computer models to find out. Everything we know about the climate — past, present, and future — depends upon our ability to simulate its operation.

The idea that it’s “models bad, data good” just won’t work. We can’t let the skeptics set the terms of the debate. They don’t even understand what the terms mean.


As for the 20CR scientists, they responded to Jolis on Feb. 23. Mild-mannered creatures that they are, they wrote that her opinion "does not accurately reflect our views."
As for the statement that the Twentieth Century Reanalysis Project... shows 'little evidence of an intensifying weather trend': We did not look at weather specifically, but we did analyze three weather and climate-related patterns that drive weather, including the North Atlantic Oscillation. And while it is true that we did not see trends in the strength of these three patterns, severe weather is driven by many other factors.

The lack of a trend in these patterns cannot be used to state that our work shows no trend in weather. Many researchers have found evidence of trends in storminess and extreme temperature and precipitation in other weather data over shorter periods.

Finally, the article notes that the findings are 'contrary to what models predict.' But models project forward, while our analysis looked back at historical observations. We see no conflict between the 100-year-projection of changes in weather extremes resulting from additional carbon dioxide and the fact that our look back at three indicators showed no historical trend.
They fail to point out that their analysis is itself produced by a model.

No comments:

Post a Comment