One day man will be able to predict the weather far into the future. Maybe. On his blog Cliff Mass notes that weather forecasts have improved in the short run but get lousy fast about two weeks out. As he explains, decent data plus fast computers project forward tolerably well as long as the current values of the weather system can be accurately measured. But as Judith Curry just reminded us in reviewing a paper “Escape from Model-Land”, you can’t predict the behaviour of complex systems that are sensitively dependent on initial conditions (SDIC) because of the famous “Butterfly effect” that explains why, to quote Mass, “small errors in the initial description of the atmosphere and deficiencies in our models inevitably lead to growing errors, and by 2 weeks such errors swamp the forecast.” Mass remains “hopeful that eventually forecast skill beyond two weeks will improve.” If not, we sure won’t find ourselves modeling climate effectively. Which means we sure aren’t now.
Mass is not addressing climate alarmism in this post. But he could be, since his two examples of failed forecasts are the “official NOAA Climate Prediction Forecast for October” on Sept. 19 and another NOAA prediction on Oct. 4. They are not random, instead both said temperatures would be well above average (who saw that coming, especially from NOAA?) and then in fact it was cold. And if taken seriously his analysis has polemical implications for the climate debate with its projections out to 2100 to tenths of a degree using computers that couldn’t predict Christmas on Dec. 25 unless told in advance when it was. (As Rud Istvan put it bluntly about a new and even scarier sea level rise model, “Models produce data only in alarmist climate circles.”)
For instance Mass draws attention to the fact that “our models still have key deficiencies (such as poor description of thunderstorms)”. And it’s not a trivial point. The crucial insight of “chaos theory”, which was very trendy until its implications for climate alarmism began to be understood, is that complex nonlinear systems are impossible to model precisely because doing so requires knowing the current state of the system to a precision below the limits of measurement. The flapping of a butterfly’s wings can create errors that propagate up to the scale of a hurricane on the other side of the world a week later, and we can never know where all the butterflies are. The grand assumption in climate studies is that the intractable complexities of climate simply average out over long time scales making long term prediction possible. Yet back in 2001 the IPCC itself admitted otherwise:
In climate research and modeling, we should recognize that we are dealing with a coupled non-linear chaotic system, and therefore that the long term prediction of future climate states is not possible. (p. 774)
Is all hope lost? Maybe not. As Mass writes, beyond two weeks, “The forecasts are not much better than simply using the average conditions (or climatology).” In keeping with that let us propose that if we want to know what climate is going to do, we should look at the average conditions, a.k.a. what it has typically done. Doing so tells us the climate will fluctuate, that recent changes are trivial compared to those we have seen even over the past 12,000 years of the Holocene interglacial, let alone on longer time-scales like 5 million, 50 million or 500 million years, and that CO2 levels have very little to do with global temperature.
It may not be the forecast you were hoping for. But if the weatherman says rain tomorrow you take an umbrella, right? So if they say uncertainty next decade, you try to be resilient… and skeptical of prophets of doom claiming absurd degrees of precision.