Tisdale on model initialization in wake of the leaked IPCC draft

Brian H says:
November 15, 2011 at 4:06 am
John B, et al;
The model runs are not “initialized” on real world conditions at all. They’re just tweaked to see what effect fiddling the forcings has on their simplified assumptions. It’s not for nothing that the IPCC says in the fine print that they create ‘scenarios’, not projections. Even though they then go on to treat them as projections.
————————
The IPCC publishes projections based on scenarios. e.g. if CO2 emissions level off (scenario), we would see something like this (projection).

This is not knew news, here’s Kevin Trenberth in his Nature Blog June 4 2007:
“In fact there are no predictions by IPCC at all. And there never have been. The IPCC instead proffers “what if” projections of future climate that correspond to certain emissions scenarios. There are a number of assumptions that go into these emissions scenarios. They are intended to cover a range of possible self consistent “story lines” that then provide decision makers with information about which paths might be more desirable. But they do not consider many things like the recovery of the ozone layer, for instance, or observed trends in forcing agents. There is no estimate, even probabilistically, as to the likelihood of any emissions scenario and no best guess.
Even if there were, the projections are based on model results that provide differences of the future climate relative to that today. None of the models used by IPCC are initialized to the observed state and none of the climate states in the models correspond even remotely to the current observed climate. In particular, the state of the oceans, sea ice, and soil moisture has no relationship to the observed state at any recent time in any of the IPCC models. There is neither an El Niño sequence nor any Pacific Decadal Oscillation that replicates the recent past; yet these are critical modes of variability that affect Pacific rim countries and beyond. The Atlantic Multidecadal Oscillation, that may depend on the thermohaline circulation and thus ocean currents in the Atlantic, is not set up to match today’s state, but it is a critical component of the Atlantic hurricanes and it undoubtedly affects forecasts for the next decade from Brazil to Europe. Moreover, the starting climate state in several of the models may depart significantly from the real climate owing to model errors. I postulate that regional climate change is impossible to deal with properly unless the models are initialized.”
John B my understanding is that the models can only hindcast the climate if sulphates are fed into them. In fact Trenberth himself says that el Nino and the PDO are critical modes of variability that affec the PAC rim countries. How does that square with them evening out over a decade?

Yes, they should be initialized with a natural variability component.
Just call it a temporary ocean circulation energy transfer (additional absorbed for periods and then additional energy released in other periods).
What they will find, however, is that their greenhouse forcing formulae do not work. The GHG impact is much lower in that scenario – less than half in fact.
So, I’m sure some of the climate modelers have played around with some cycles (regular or not) but they soon abandon the effort because then the rest of their math / the GHG forced climate model does not work.

M.A.Vukcevic says:
November 15, 2011 at 5:10 am
——————
I think the AMO and the ENSO are the most important cycles. There are a few others that could be used. But not everyone agrees that the AMO could be like a “forcing” – operating independently and causing its own climate variability.
If some want to use the ocean in general rather than these specific components, then it doesn’t matter and they still end up with a climate model that does not work.

Will somebody please explain to this retired engineer why a linear extrapolation of a curve based on ill defined chaotic sources (especially out to a hundred years) should have any predictive value? It seems to me to be a naive and extremely unscientific approach.

Latitude says:
November 15, 2011 at 4:50 am
,,,
And I say that “climate change signals” are way too small to be modeled…
,,,
Which makes climate models fun games……but worth less than the paper they are printed on
——————
That doesn’t follow. As an analogy, we can model the tides pretty accurately, but if you stand on a beach and watch the waves crashing in, you might be tempted to say “the tide signal is too small to be modelled”. So it is with climate. And the analogy goes further: we can’t model the details of the beach, the turbulence in the waves, the wind on a particular day, etc., but we can still produce tide tables. The waves may be hugely important if you are swimming on that beach, but they do not affect the longer term trend of the tide.

Solomon Green says:
November 15, 2011 at 5:34 am
Comparing the Mauna Loa CO2 chart from 1975 to the present day it appears that the only
“known physics and various forcings” required to replicate the gradient in the IPCC AR4 sea temperature anomalies model is CO2.
—————-
But we don’t only look at 1975 to the present day. Even Bob’s post goes back to 1910 and, short term variations aside, the models look pretty good. And if CO2 is the major forcing that determines how well they depict reality, that should be telling us something about the reality of the effect of CO2.
Unless, of course, all them lib’rul scientists is lyin’ 🙂

The real problem with the GCMs is that they don’t “predict” the past record, they can’t explain the present, and they generally aren’t very good at extrapolating into the future.
Ultimately, they are nonlinear curve fitting routines, with parametric components in them that can match any curve you like for at least some small chunk of it, especially when that curve is predominantly monotonic.
This is the fundamental problem with the whole debate. We have at least moderately accurate/believable temperature data from e.g. ice cores and other proxies that extends back over hundreds of thousands to millions of years. This record clearly indicates that we are in a five-million year long “ice age” interrupted by interglacials that seem to be order of or less than ten thousand years long, the Holocene being one of the longest, warmest interglacials in the last million years.
We don’t know why the world was warmer 5 million years ago — MUCH warmer than it is today, in all probability. We don’t know why it started to cool. We don’t know why it cooled to the point where extensive glaciation became the rule rather than the exception, with CO_2 levels that dropped to “dangerously low”, almost too low to support atmospheric plant life at the worst/coldest parts of it. We don’t know why it periodically emerges from glaciation for a few thousand years of warmth. We don’t know why it stops and returns to being cold again. Within the interglacials and/or glacial periods, we don’t know why the temperatures vary as they do, up and down by several degrees C on a timescale of millennia.
What we do know is that the best predictor of temperature variation on a timescale of centuries is the state not of the atmosphere but of the Sun, in particular the magnetic state of the sun. The evidence for this is, IMO, overwhelming. I don’t particularly care if the physics of this isn’t well understood — that’s a challenge, not a problem with the conclusion. The analysis of the data itself permits no other conclusion to be made. Sure, other factors (Milankovitch cycles, orbital eccentricity changes in response to orbital resonances, axial precession) seem to come into play with the longer time scales but THEY aren’t predictive either.
I am unaware of any model that can predict the the PAST, and this is the biggest problem of all. AGW is predicated on the assumption that we know what the temperature “should” be by some means other than numerology and reading tea leaves. And we don’t. It’s a non-Markovian problem laced with chaotic dynamics on an entire spectrum of timescales from seconds to epochs. Most of the local (decadal) dynamics is determined by a mix of the solar cycle, various effectively chaotic limit cycles (e.g. the PDO or AO) on TOP of the strong correlation with magnetic state of the sun, with additional near-random modulation by aerosols and greenhouse gases and cloud cover and with various unknown positive and negative feedbacks.
The problem, in other words, is “difficult”. It is complex, in the precise sense of the term. Locally, one can ALWAYS fit a segment of the temperature curve over decades to centuries as long as it has a monotonic trend. In the case of the 20th century, there are at least two completely distinct fits that can be made to work “well enough” given average monotonic warming. One is “CO_2 is dominant, previous thermal history and solar state and so on perturbations”. One fits the general upward trend to the general upward trend of the CO_2, then finds parameters to scale the correlations between solar state and temperture so that they match the perturbations.
The other is the opposite — explain all or nearly all of the temperature trend on the basis of solar state and dynamical history, and treat the CO_2 forcing as a small component that contributes to a general monotonic trend that would be there anyway even if there was no anthropogenic CO_2 at all. After all, only somebody very, very stupid would think that the mean temperature would be CONSTANT in the absence of humans. The entire climactic record stands ready to refute any such silliness. Nearly all of that record consists of century-long stretches of monotonic warming and cooling, with occasional abrupt changes bespeaking the existence of multiple poincare attractors in the general phase space and chaotic transitions between them.
Both of these can work, but the fundamental problem with the former is that it DOES NOT WORK outside of the 20th century. It is utterly incapable of explaining not JUST the MWP and LIA, but all of the OTHER significant variations in average global temperature visible in the long term proxy based temperature reconstructions. It will not work if indeed global temperatures drop — quite possibly drop precipitously — over the next thirty years as the sun’s state returns to levels not seen since the 1800s if not the 1600s.
One simple fact ignored by the AGW crowd is this. The last time the sun was as active as it was in the 20th century was 9000 years ago, at the very start of the holocene. Indeed, that earlier burst of solar activity (a double maximum that spanned some 300 years with an interval in between peaks) may have been the event that really gave the Holocene a robust start and finally ended the previous glacial period after the Younger Dryas. More or less monotonic warming over the 20th century has to be interpreted in light of that fact — the assumption of “everything BUT CO_2 levels is either the same or unimportant in determining climate, so CO_2 is the explanation” is simply not justified when something else is so spectacularly not the same as a 9000-year grand maximum in the sun across the very interval where the greatest warming was observed.
So, requiring GCMs to fit all of the data is good, but it isn’t enough. Not only all of the 20th century data, but ALL of the data. Until they can fit the data over the last 2000 years (say) or the last 10,000 years, their models are pointless, the moral equivalent of saying “look, a monotonic trend, I can fit that” with any function that (for some range of parameters) exhibits a monotonic trend. They have to be able to predict the UPS and the DOWNS, the long term variations. And the only way one can possibly do this is by including the sun — not just insolation but its actual state — as one of the fit parameters.
Once this is done — and we can, any of us, perform such a fit in our own minds, just from looking at a graph — the entire argument inverts. Suddenly, because solar state is the BEST predictor, the ONLY predictor capable of getting the right gross features of the T(t) curve, everything ELSE becomes the perturbations, the moderators of a general trend slaved to the sun. Suddenly the quality (e.g. R) of the curves goes way up. Suddenly CO_2 is a modulator, sure, but a relatively unimportant one, tenths of a degree C and part of that moderated by negative feedback from clouds, almost lost in the chaotic noise.
rgb
(Could somebody in Big Oil now send me a check, please? I keep hearing about how anybody who is a “denier” is in the pay of Big Oil, but I can never seem to find BO and send them a bill…)

OK, scratch that, I think I get it.

Bob Tisdale says:
November 15, 2011 at 7:01 am
John B says: “No, they haven’t done that at all. The models do not work by latching on to trends. They work by simulating the effects of known physics and various forcings on a simplified, gridded model of the atmosphere and oceans. If it happens to generate something close to linear, so be it, but they do not in any way assume a linear trend.”
John B, where did I write that the models “assume a linear trend”? Are you twisting what I wrote? Read what I wrote again.
You continued, “The hindcast shows that the models replicate past climate pretty well over scales of decades…”
The data illustrated in Animation 1 very clearly contradicts this statement. You are wasting your time…and the time of those who have come to discuss this.
————–
Bob, you said “latched on to a trend”, as if that is how the models work. You then go on to linearly extrapolate short periods of the model projections, as if that has some meaning.
Figure 1, the one that shows the largest portion of data, is the one that shows most clearly how well the models have worked. Your animation, extrapolating linear trends to arbitrary parts of the data, is at best misleading.