People familiar with the chaos that is weather will agree that seasonal forecasts based merely on computer simulations are highly speculative and involve much guesswork. The quality of their output leaves little to be desired.
As much as some of these forecasts may be presented with authoritative tones, in the end they all come with a fine print disclaimer concerning certainty, and so they are nothing one can take to the bank.
The average of a series of WAGs
Even the most powerful super-computers using conventional simulations struggle to predict the weather 10 days out, let alone what an entire a winter will be like in terms of temperature and precipitation. The simulations used by the US weather agencies, for example, simply perform a number of data crunching runs — each using different start conditions — to generate a set of scenarios (i.e. wild-ass guesses) and then average them out to create an “ensemble”. Under the bottom line, the ensemble is based on a set of wild ass guesses, and so is correspondingly unreliable.
Woefully uncertain
Not surprisingly, Germany’s flagship daily Frankfurter Allgemeine Zeitung (FAZ) also finds that the ensembles leave very little to be desired and point out that the seasonal forecasts issued by weather agencies such as the NOAA are hugely uncertain, and that clearly the computer simulation methodology is in dire need of improvement.
Even farmers seem to have greater success.
So what can be done to improve forecasting performance?
The FAZ writes that a whole new methodology has been developed by German scientists and that this new methodology promises to vastly improve seasonal forecasts. The FAZ writes that earlier this year a study by the University of Hamburg was published in the Geophysical Research Letters. The press release announced: “Possible for the first time: reliable three-month forecasts for European winters“.
80% winter season accuracy
According to the FAZ, oceanographer Dr. Mikhail Dobrynin of the Hamburg University Institute of Oceanography, Center for Earth System Research and Sustainability and his colleagues have developed a method that works with 80% success rate. Instead of solely relying on a super computer crunching data using a variety of start conditions and then averaging them out, Dobrynin says their method relies also on “teleconnections”, i.e. finding “signals amid the chaos“.
The FAZ reports that for winter forecasts, Dr. Dobrynin has identified four factors: “the snow depth in Siberia, Arctic polar vortex, extent of Arctic sea ice and the Atlantic ocean temperature”, which allow “the path of certain air masses going to Europe to be predicted”. The most critical of these is the snow cover over Siberia in fall.
The NAO index is one key factor in seasonal winter forecasts for Europe. Source: Geophysical Research Letters.
The scientists say that conditions in different geographical regions over the Atlantic “can act as a switch that steers winter weather in Europe” and that focusing on how these interact and behave can allow greatly improved forecasting.
According to the conclusion of the study, the method shows vastly improved performance:
For the real forecast test from 2001 to 2017 the prediction skill of the winter NAO is increased from 0.42 for the full ensemble mean to 0.86 for the subsampled ensemble mean. As a result of a better representation of the winter NAO, the prediction skill for the winter surface temperature, total precipitation, and SLP is improved for considerable parts of the NH.”
NOAA forecasts “hardly better” than guessing
The FAZ writes that “without considering teleconnections, winter in Europe has been virtually unpredictable”.
Dr. Mikhail Dobrynin added that results of forecasts made by the NOAA “were hardly better” than if they had guessed. The new method should thus improve the winter season forecasts considerably. However, he warns that the uncertainty will always exist and we must remain wary of any predictions dealing with seasonal weather.
Interestingly, winter forecasts seem to have nothing to do with CO2 as it is not mentioned in the paper.
Short term forecasts, e.g. weekly monthly or seasonal, do not include CO2 simply because its concentration hardly changes over that sort of time span.
Incidentally, I have been on holiday for the last few days and have just got back. This means that I have some catching up to do. However, in the last few days we have had claims to the effect that: human emissions have little to do with changes in the amount of CO2 in the atmosphere; the sensitivity of temperatures to CO2 is very small; global temperatures have decreased since 1850; a temperature series based on 450 weather stations is somehow more accurate than the BEST one which is based on more than 46,000 time series and so on.
Thanks for reporting this. The method is consistent with that of Dr. Judah Cohen, who forecasts the arctic oscillation for AER. His diagram is:
https://rclutz.files.wordpress.com/2017/09/cohen-schematic2.png?w=1000
He has a fairly detailed description of the mechanism. My synopsis is:
https://rclutz.wordpress.com/2017/10/24/snowing-and-freezing-in-the-arctic/
BTW, in the text of your post, I think you meant “much to be desired”, rather than “little to be desired.”
Do I understand correctly that they will now forecast a NAO index, not weather?
More details in the referenced paper.
I think they must have been watching Joe Bastardi’s Saturday and Daily summaries.
“The quality of their output leaves little to be desired.” Non sequitur
My take is that RANDOM VARIABILITY can never allow an accurate forecast. RANDOM is RANDOM and even with random number generators providing feed backs the results will still be RANDOM. Sometimes the random forecast will be co-incident with the random weather. Good management? Nah -just good luck.
It’s not just that it’s random. It’s also chaotic. Furthermore, climate is the result of multiple random and chaotic systems, none of whose chaotic behavior is quantitatively understood nor whose probability distributions are known. And even if we had all that knowledge and understanding of those phenomena, they would still be totally unpredictable.
As you are obviously aware, Don, but some might not be, knowing the rules isn’t enough. For example, just because you know all the probably distributions of cards in a poker game doesn’t mean you can win. You may be able to limit your losses, but predicting when and by how much you will win is utterly impossible. And that’s only with one known variable, and assuming that no one is cheating. The climate is composed of myriads of processes each with it’s own rules and probablility didstributions, none of which we can predict by itself and certainly not in combination.
This guest post at Judith Curry’s blog grabs the problem by the horns.
https://judithcurry.com/2016/10/16/determinism-and-predictability/
“III. WEATHER AND CLIMATE
If the N-S equations are difficult, weather analysis is much more difficult and climate analysis is infinitely more difficult, the word infinitely being meant literally.”
Can you say Piers Corybn, a history of accurate long term forecasts
It’s the Joe Bastardi method but using oscillations closer to europe
[…] Read more at No Tricks Zone […]
“Dr. Mikhail Dobrynin … and his colleagues have developed a method that works with 80% success rate. … Dobrynin says their method relies also on “teleconnections”, i.e. finding “signals amid the chaos“.”
And IMO that has always been the key — NOT averaging away the apparent noise on a signal, for within that ‘so called’ noise IS the climate signal(s), the chaotic signal(s) that may bend or push the evolutionary elements our climate on its trajectory to the next quasi-stable regime.
Yes it’s known that there are many quasi-cycle events in our climate, and that these events are restrained within strict bounds (Why? — Is another unknown lost in myriad dependencies and feedbacks) but ‘climate science’ is poor at teasing out the most probably evolutionary trajectory, because ‘climate science’ is too busy erasing the chaotic signals as it uses ever more esoteric averaging techniques to render only the most obvious repetitious signals.
What is needed is a method that does not unduly emphasize the quasi-cyclic nature of climatic variations at the expense of the most probable chaotic signal(s). Hopefully Dr. Mikhail Dobrynin and his colleagues have found the beginnings of a method that can be refined into a more generalized technique. IMO we should be gathering such techniques for finding the significant chaotic signal(s) that keeps these oscillations quasi-cyclic and not attempting to find better methods of rendering the cyclic components of the climate oscillations.
[…] https://notrickszone.com/2018/12/18/hamburg-meteorologists-find-new-method-for-better-european-winter… […]