Climate model tuning

[GregLocock]

http://www.mpimet.mpg.de/fileadmin/staff/klockedaniel/Mauritsen_tuning_6.pdf

"During a development stage global climate models have their
6 properties adjusted or tuned in various ways to best match the known state
7 of the Earth’s climate system. These desired properties are observables, such
8 as the radiation balance at the top of the atmosphere, the global mean tem-
9 perature, sea ice, clouds and wind fields. The tuning is typically performed
10 by adjusting uncertain, or even non-observable, parameters related to pro-
11 cesses not explicitly represented at the model grid resolution."

There's rather a lot in it, I haven't read it through yet. Some of the phrases concerning clouds are 'nicely' phrased.


Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[rconnor]

Mauritsen et al 2012 is a very good paper. It was discussed briefly here already (first linked by beej67 at 16 Apr 14 13:19) but more so in passing.

 

[rb1957]

maybe i'm mis-remembering but i thought rconnor said that the models weren't tuned ?

Quando Omni Flunkus Moritati

 

[rconnor]

You are misremembering. Actually, I used excerpts from this very article to defend my view. See my post at 17 Apr 14 14:41 after beej67 linked this paper (it was in response to you no less).

 

[rb1957]

man! you have either a fantastic memory for your posts, or an excellent filing system.

"The “knobs” are tuned to create a better representation of their specific aspects of the climate system. These adjustments can only be made in the range established by observations or peer-reviewed study." i must read the article to see how this tuning is different to matching the historical record. it'd be interesting reading to see how they address the different combinations of "knob position" that achieve the same match to the observation ...

Quando Omni Flunkus Moritati

 

[rconnor]

Haha, sadly I have neither. But the search function makes up for my shortcomings!

 

[rb1957]

"Tuning aims at balancing the Earth’s energy budget by adjusting (sic) a deficient representation of clouds,"

if they're not tuning to match past climate record, how come ... "models are generally able to reproduce the observed 20th century warming of about 0.7 K, and details such as the years of cooling following the volcanic eruptions, e.g. Krakatau (1883) and Pinatubo (1991), are found in both the observed record and most of the model realizations."

"it is usually to be expected that improved performance arises not because uncertain or non-observable parameters match their intrinsic value – although this would clearly be desirable – rather that compensation among model errors is occurring."

this doesn't fill me with confidence about climate models.



Quando Omni Flunkus Moritati

 

[cranky108]

"models are generally able to reproduce the observed 20th century warming of about 0.7 K, and details such as the years of cooling following the volcanic eruptions, e.g. Krakatau (1883) and Pinatubo (1991), are found in both the observed record and most of the model realizations."

You get lucky sometimes.

 

[zdas04]

If models can duplicate "climate" after a once-a-century event like Krakatau (East of Java) and Pinatubo, it is not luck, it is a very heavy hand on the adjustment knob.

David Simpson, PE
MuleShoe Engineering

Law is the common force organized to act as an obstacle of injustice Frédéric Bastiat

 

[cranky108]

So how do you trust a model where the owner can adjust the knob to fit anything that has happened in the past?

I understand that some events can't fit in the model, but how would one know that the knob is not being adjusted for the results that owner wants in the future case?
Even if you think the model is correct, you should be concerned about the tuning going on. This can leave you with a very wrong result.

Question: How many climate models are being used to predict a result? Assuming some of them are wrong, or are being used wring, we should be able to come up with an avreage result. What is the standard devation of those models? What would be a confedence factor in the average results? And who wrong can they be?

I have to question everything, as a leap of faith should only apply to a religon.

 

[IRstuff]

"If models can duplicate "climate" after a once-a-century event like Krakatau (East of Java) and Pinatubo, it is not luck, it is a very heavy hand on the adjustment knob."

No, that's actually the easiest thing to account for, since it's a singular delta-function event, composed of a sudden introduction of heat, gases, and particulates, which can be inputted into the model, and the model should respond accordingly. This is not that different than modeling shock events in FEA programs. Pinatubo, in particular, should have been well instrumented and measured, as should have been the resulting climate afterwards, since everyone was looking for the perturbations. Pinatubo would then be input into the model, and the resultant climate effects should be relatively easier to compare against actual data.

TTFN
faq731-376

Need help writing a question or understanding a reply? forum1529

Of course I can. I can do anything. I can do absolutely anything. I'm an expert!

 

[rb1957]

that's what i thought too ... that they made a step change in the atmosphere aerosols to coincide with these events. but then it's not unexpected, nor remarkable, that the models match the historical record. indeed, remarking about it suggests that this author doesn't appreciate how the models achieved this outcome.

also, it shows that the models are compared with the historical record, the inference being that a model that agrees with the historical record is "better".

Quando Omni Flunkus Moritati

 

[zdas04]

The IPCC assessment reports are based on something like 111 models(that is the number I see most often, but it is not the only number I see). Averaging models is a common activity in this "science", but not in any other field. So, if the range of predicted temperatures on September 4, 2100 is +2 to +12C (compared to some fixed "normal"), and the average is +9C, and the median is +7C and the most frequent is +5C, what is the temperature going to be? Remember that these folks are saying that a +3C temperature change will melt all polar ice. This is like walking up to 111 people on the street and asking "what will the temperature be at noon on November 4, 2014?" and then averaging the results. Meaningless nonsense.

David Simpson, PE
MuleShoe Engineering

Law is the common force organized to act as an obstacle of injustice Frédéric Bastiat

 

[IRstuff]

" it shows that the models are compared with the historical record, the inference being that a model that agrees with the historical record is "better"."

I'm sure what you're getting at. How is a model that disagrees with historical records a good thing? btw, a copy of the article with graphics is here:

http://www.readcube.com/articles/10.1029/2012MS000154?

The authors simply state "models are generally able to reproduce the observed 20th century warming of about 0.7 K, and details such as the years of cooling following the volcanic eruptions, e.g., Krakatau (1883) and Pinatubo (1991), are found in both the observed record and most of the model realizations."

Which, if they're supposed to match the historical records, requires that the models must account for the pertubations, particularly since both volcanic events are "natural."

TTFN
faq731-376

Need help writing a question or understanding a reply? forum1529

Of course I can. I can do anything. I can do absolutely anything. I'm an expert!

 

[zdas04]

[sorry about the disconnect, I was out of the office for a couple of hours]

When I said "very heavy hand on the controls" I meant that they put in actual data into the model for the days/weeks/months after the eruption. That is pretty heavy handed. Based on those specific interventions, 108 years after Pinetubo they would have had another similar sized event. If it happens in 5 years or 50 years or 150 years then the data for 2099 will be very wrong. I can simulate the event pretty easily, it is the timing that has me gobsmacked.

David Simpson, PE
MuleShoe Engineering

Law is the common force organized to act as an obstacle of injustice Frédéric Bastiat

 

[IRstuff]

"When I said "very heavy hand on the controls" I meant that they put in actual data into the model for the days/weeks/months after the eruption. "

That would be, at most, a one-time event. There's nothing that says it would, or need to, repeat on its own; that would simply be sophomoric, at best. Timing is simple. You create a time series of inputs that lasts the duration of the event and add them to the flow of the simulation. We do this sort of thing in engineering analyses routinely, to simulate discrete events. There's neither magic nor nefariousness involved; it's purely discrete math.

TTFN
faq731-376

Need help writing a question or understanding a reply? forum1529

Of course I can. I can do anything. I can do absolutely anything. I'm an expert!

 

[zdas04]

When I calibrate a model to a specific data sample, I then run it on another data sample to see if the results are reasonable. I do the second run without modifying the model. Otherwise what is the purpose of calibrating?

So for this example, you are trying to match a time series so you either discard the time period where the unusual event occurred from your calibration run before the run or after the run (i.e. discard partial data or partial results). It does no good to "match" history by feeding in the exact answers.

David Simpson, PE
MuleShoe Engineering

Law is the common force organized to act as an obstacle of injustice Frédéric Bastiat

 

[TGS4]

The climate models don't match the post-delta-function events well, even with heavy-handed tuning. They stay cool too long and completely miss the "bounce-back" that the instrument record shows. It appears that the models miss the reduced cloudiness in the immediate aftermath of the volcanic event, which results in additional solar heating. The misses are for a short time (about a year), but the total energy miss is substantial. And they all miss cold, which is quite surprising considering that all of the models have run hot compared to reality for the last 15+ years.


But, since clouds are the least-understood meteorological and climactic process, and therefore cannot be managed by basic-physics in the models but rather by heavy-handed tuning, this miss is not surprising.

 

[TGS4]

The other issue with most of the models (based on a review of CIMP5) with respect to vulcanism is that their temperatures drop BEFORE the volcanic activity - as if they are being fed some kind of annual data rather than monthly or weekly data.

Or maybe they have to have the temperatures drop prior to the activity in order to match the total thermal drop. Heavy-handed knob-fiddling. (All double-entendres intended).

 

[rb1957]

" it shows that the models are compared with the historical record, the inference being that a model that agrees with the historical record is "better"."

i think it shows that models are tuned to the historical record, despite contentions to the opposite.

Quando Omni Flunkus Moritati