Using common sense one could predict that a coarse model describing any system will provide bumpier results than a more reliable one that reflects the system in a finer mesh or a more extensive sampling period [1].

Well, that’s what a scientific paper is confirming, as I could understand.

It is cited in the last NATURE edition (vol 531, March 3, 2016, page 10, paywall, but don’t pay, it’s just a short highlight) under the title “Where climate models fall short” See: Geophys. Res. Lett. http://doi.org/bcsr (2016) (abstract, paywall for the full text).

Using what is called a toy-model with different degrees of reliability the authors show that the probability to forecast extreme events is higher with low reliability models than with better ones.

“Climate model ensembles tend to be overconfident in their representation of the climate variability which leads to systematic increase in the attributable risk to an extreme event.”

Figure 3    Variation of Fraction Attributable Risk (FAR) for a one-in-10-year (a) and 50-year (b) event as a function of the standard deviation of the model error (β) which is inversely proportional to the model reliability (see Fig. 2). The variation of FAR is shown for different signal (S, trend) to variability (V, residual variability from the trend) ratios where the dots show the number of different levels of β computed. The green area shows the 5-95 %confidence interval sampled by repeating the predictions 1000 times allowing the model error to vary within the chosen standard deviation (β).

The current possibility to hindcast, to assemble all necessary parameters over a time period to compare model results with reality, is limited to a few decades at best. It follows that such models cannot have a high reliability. Only our great great great children may be able to begin to improve this. Can we cope with our impatience?

The reality is that such frequency increase of hurricanes, tornados or other extreme rain events has not yet been observed in a statistically relevant way. Nevertheless, we constantly have to hear that extreme events are one of the bad things that will become more frequent if we continue to fart, breathe and burn fossil fuels. Shouldn’t we ask how reliable are the models making such projections?

It’s surprising that the editorial staff of NATURE has selected such article for their Research Highlights, as it looks like a stone in the garden of climastrologists.

Other ironic aspect: the use of a simplified toy-model to explain that more complex ones will fail because of their intrinsic unreliability.

[1] Yes, it is still allowed to use common sense without having to provide a database full of citations. Until when?

Merci de partager et de diffuser cet article !