Impure Speculation

Recent computer simulations at HadCRU have led to media announcements that global warming will remain “levelled off” until 2009, at which point it will take off like a bat out of hell. The folks at HadCRU know a lot more about the climate system than I do, and their computer models are probably very good. But I admit, I’m skeptical about their forecast.

It’s not the long-term aspect of the forecast I’m suspicious of. The ability of computer models to make successful long-term forecasts has already been demonstrated: James Hansen’s team at NASA GISS forecast decade-scale increases in global temperature that have already come to pass. It’s the short-term aspect that troubles me. Long-term, climate follows known, and very basic, laws of physics: put more energy in, the temperature goes up. But short-term changes involve the details, not in the “big picture” but in the “small picture,” and often the devil is in the details. Also, there are known factors which are unpredictable, and would easily throw any short-term prediction out of whack. If there’s a strong, sustained el Nino in, say, 2008, global temperature won’t “remain levelled off,” 2008 will break the record. If there’s a massive volcanic eruption in 2008, global temperature will take a nose-dive as the planet cools for a couple of years.

So I read the actual paper containing the new predictions. It turns out that the press reports are considerably overblown (surprise!); they give the unmistakeable impression that the HadCRU team has made definitive predictions of the future progress of global warming over the next decade or so, as though we now know with confidence how global average temperature will evolve up to 2014. If you read the actual paper you’ll find that is simply not so. The researchers do claim that their new model, DePreSys (”Decadal Climate Prediction System”) performs demonstrably better at short-term prediction than previous models (which they simulate using a system they call NoAssim), and in fact it does. Comparison between hindcasts from the models (using past data to “predict” what follows in the historical record) clearly demonstrates significant improvement in the predictive ability of the DePreSys model. But it does not demonstrate, nor do the authors claim, the ability to make ironclad predictions.

The principal difference in the DePreSys model is that it includes more detailed data of, and better models the evolution of, heat content (H) in the upper 113 m of the ocean. They also argue persuasively that this is a driver of internal variability in the climate system. Furthermore, they intialize the state of ocean heat content using actual measured data, hence “assimilating” this data, which is why they refer to the previous model as “NoAssim.” As the authors say:

Because the internal variability of the atmosphere is essentially unpredictable beyond a couple of weeks (26), and the external forcing in DePreSys and NoAssim is identical, differences in predictive skill are very likely to be caused by differences in the initialization and evolution of the ocean. During 600 years of the HadCM3, control integration T_s is highly correlated (correlation R = 0.89) with global annual mean ocean heat content in the upper 113 m (H). Furthermore, the correlation is higher when H leads T_s by 1 year (R = 0.56) than when T_s leads H by 1 year (R = 0.32), providing strong evidence that variations in H can force T_s. We also find that H is predicted with significantly more skill by DePreSys than by NoAssim (Fig. 1B), and we conclude that the improvement of DePreSys over NoAssim in predicting T_s on interannual-to-decadal time scales results mainly from initializing upper ocean heat content.

What is their prediction for global average temperature up to 2014? It is shown in their figure 4:

This graph illustrates that nobody’s model anticipated the cooling impact of the Mt. Pinatubo explosion in 1992. The models also underestimated the warming effect of the 1998 el Nino. But in general, the DePreSys model matches observed data better than the NoAssim model, and it does indeed forecast that global surface temperature T_s will remain levelled off until 2009, rising swiftly after that. It also predicts that by 2014, T_s will increase by 0.3 deg.C over the temperature in 2004.

It seems to me that the computer models get the “big picture” right, and hence their long-term forecasts are reliable. Over the next decade, we can expect that the all-time hottest year record will indeed be broken, perhaps several times. But then, we don’t really need computer models to make this forecast, basic physics does the job for us. But despite our ability reliably to predict the long-term big picture, I don’t see that we’re yet able to predict the short-term “little picture.” I’ll go on record as saying that I fully expect a new all-time record year in the next decade, but I wouldn’t bet the farm — or a cup of coffee, for that matter — on which year.

I repeat that the folks at HadCRU know more about the subject than I do, so I certainly could be wrong, and they certainly could be right. And I’ll repeat that the impression given by press reports that the new forecast from HadCRU is anywhere near definitive, does not reflect what the authors state in their paper.

Just to have a little fun, and JUST AS AN EXERCISE, I’ll make my own detailed forecast of the next decade or so. NOTE: THIS IS JUST AN EXERCISE FOR FUN, I DON’T ACTUALLY BELIEVE MY OWN FORECAST.

I have often subjected climate time series to detailed analysis, and there’s a “period” in temperature time series that has come up several times for recent data; it shows, for example, in global average temperature during the modern global warming era. This period is about 3.65 years, and represents reasonably well the “fast fluctuations” of recent global temperature. Statistically, this period is “on the edge” — it could be just the signature of red noise, or it could be a genuine periodic phenomenon whose statistical strength hasn’t yet risen above the background noise. If we look at just the data since 1980, there’s also a roughly 7.75-year period which is most likely red noise. But since I’m playing fast and loose with predictions, I’ll include them both in my forecast, in an attempt to simulate the fast fluctuations of global temperature.

There is also one pattern which is statistically significant, namely an upward trend of 0.017 deg.C/yr, so I’ll include that too. Having identified the patterns to include, it’s straightforward to extrapolate them to the future (Note: I’ve based my analysis on GISS rather than HadCRU data, so the “zero point” of temperature anomaly is different from the HadCru forecast):

According to this forecast, we can expect another peak in the 3.65-year cycle around 2010, so that just might be when the next record-breaking warmth is observed. The next common peak of the 3.65 and 7.75-year cycles won’t come until 2013, so that’s when we would expect a definite new all-time record-breaking warm year for the globe.

A few comments about forecasting are in order. In order to extrapolate patterns into the future (as I have done) reliably, one must establish that the patterns truly indicate not only what the data have done, but the physical process underlying the time evolution. This doesn’t mean we have to understand, or explain, the physical mechanism at work, but it does mean we must establish the statistical significance, and persistence, of the pattern beyond doubt. In fact, extrapolating patterns which represent faithfully what the data did but not what they will do, is one of the most common pitfalls of prediction.

For the cycles I’ve included in this forecast, this is simply not the case. Therefore I reiterate, that the “forecast” I’ve graphed here is just a toy — a game I’m playing with numbers. It is beyond doubt that the steady warming is statistically significant and persistent, and does have a physical basis. It is possible, but frankly not very likely, that the 3.65-year cycle has a physical basis, and that even the 7.75-year cycle does. The likelihood that all components of the forecast are both physically real and will persist is, to be candid, not very good.

It’s interesting that the next peak in the 3.65-year cycle is coming in 2010, not long after the time the HadCRU model predicts global temperature to take off. I’m skeptical of their ability to make predictions over the short term, but I’m even more skeptical of the correctness of my own “prediction.”

Will it continue to get warmer over the long haul? Yes. Will temperatures “remain levelled off” through 2009, then increase rapidly? Maybe. Will we see another peak in 2010, and yet another higher peak in 2013? Possibly. Only time will tell.