Why Models Run Hot: A Discussion of the IPCC's Overestimation

[Catamount]

It's true that if the BAU scenario did outpace what really happened for carbon dioxide, it wasn't by much. However that scenario had enormously higher methane emissions growth than actually occurred, projecting a near linear growth of emissions where the real world saw what appears to be emissions that went from static yearly inputs to drastic decrease, followed by resumed emissions at only a very modest level compared to the 90s. Far from growing from 350ish MtC/yr and blowing past 400, 500, and more, emissions appear to have tanked by 2000, and only modestly resume in the 2010s if concentration numbers are any indication.

http://www.esrl.noaa.gov/gmd/aggi/aggi.fig2.png

Strangely, I'm having a hard time finding emissions numbers past the late 90s, but nevertheless, concentration change has stalled out pretty strongly, especially compared to the ever-accelerating concentrations rise in FAR's BAU scenario. If emissions had increased the concentration would have started changing faster, not stalled out (obviously).Of course, the more important point is going to be that no matter what, these scenarios are going to be overestimations of net forcings, because they're not going to include the hugely negative trends in the ENSO forcing or recent aerosol hikes. Most of DH's points of criticism are very cogent, and that's one of the biggest. FAR didn't use what we'd consider a full and proper GCM. They just used a simple radiation balance model with a little bit of ocean behavior thrown in which appears to have just been there to moderate the rate of change. That's not even going to account for something like ENSO, let alone represent a correct prediction of the trends of such a forcing that would be necessary to build a proper scenario to test.I believe that may also represent a big criticism of Moncton et al taht DH didn't seem to bring up, but I'm just a layman so I'd be happily corrected there: Moncton et al. appear to basically get their past 20ish-years period by modelling all of climate over that recent period as a response to CO2, deriving the level of climatic response to CO2 by simply subtracting the other major forcings in play outlined by the IPCC and just assuming that anything that remains is CO2->temperature response, and then just plotting that sort of sensitivity of temperature to CO2 over time with the assumption that it's going to explain all behavior over the period they're using for comparison of their model to FAR. Is that going to include the influence of any short-term climatic behavior like ENSO? It doesn't seem so to me. So if that bottoms out, and you get a negative forcing that their simple model doesn't account for, it's going to manifest as reduced sensitivity, and we did have bottomed out short-term natural forcings and increased aerosol production from Asia, and oh hey, look what happened! They got really low sensitivity! Is my admitted lack of expertise here causing me to miss some hidden genius to their methods, or is it really that bad?More to the point, are the models, the modern models, actually overestimating warming? Gavin Schmidt and his co-authors in a recent little piece in Nature Geoscience don't think so. When they took their best shot at including the most up-to-date estimates of the forcings, different than what CMIP5 models are using, most of the disagreement between the temperature and the models disappeared

http://www.blc.arizona.edu/courses/schaffer/182h/Climate/Reconciling Warming Trends.pdf

 

[phyzguy]

More to the point, are the models, the modern models, actually overestimating warming? Gavin Schmidt and his co-authors in a recent little piece in Nature Geoscience don't think so.

I'm having trouble reconciling this quote with the following quotes from the Gavin Schmidt paper:

"Climate models projected stronger warming over the past 15 years than has been seen in observations."
"Why most of the model simulations suggest more warming than has been observed is a second question that deserves further exploration."

It seems clear that the models which were published before the data was available predicted more warming that was observed. At least three peer-reviewed studies are in agreement on this point. Of course, now that the data on the last 15 years is available, it is quite possible to adjust the models so that they "retrodict" the observed warming correctly. Are the models now correct so that they will predict the observed warming in the future? There are a variety of expert opinions on this point, and only time will tell who is right.

 

[Catamount]

I'm having trouble reconciling this quote with the following quotes from the Gavin Schmidt paper:
"Climate models projected stronger warming over the past 15 years than has been seen in observations."
"Why most of the model simulations suggest more warming than has been observed is a second question that deserves further exploration."

I see no contradiction at all. Climate models did project stronger warming than was seen, and Schmidt et al's work to correct that was just some preliminary toying, not even published in a formal paper it looks like, with some small remaining disagreement, so of course there's more work/investigation to be done.

It seems clear that the models which were published before the data was available predicted more warming that was observed. At least three peer-reviewed studies are in agreement on this point. Of course, now that the data on the last 15 years is available, it is quite possible to adjust the models so that they "retrodict" the observed warming correctly. Are the models now correct so that they will predict the observed warming in the future?

Unless I'm reading something wrong, then to be clear, Schmidt et al didn't change the models at all. They changed the inputs. I don't see any place where they changed how the models are actually treating climate. They didn't change the ocean mix layer depth or cloud behavior or some other aspect of how climate works. The problem was that the models were being fed incorrect forcings in the first place, so of course they're going to return incorrect temperatures. If the models are working remotely correctly then you can't get correct results by running the wrong scenario through them in the first place. Once that was corrected, the models, otherwise unadjusted, gave results fairly close to observation. This is indication that the models were always working, not some new way to change and fix them.

As they summarize, "Here we argue that a combination of factors, by coincidence, conspired
to dampen warming trends in the real world after about 1992. CMIP5 model simulations were based on historical estimates of external influences on the climate only to 2000 or 2005, and used scenarios (Representative Concentration Pathways, or RCPs) thereafter. Any recent improvements in these estimates or updates to the present day were not taken into account in these simulations. Specifically, the influence of volcanic eruptions, aerosols in the atmosphere and solar activity all took unexpected turns over the 2000s"

That's not them saying "the models are doing something wrong"; that's them saying "we didn't put the right data into the models".

As for whether models can predict warming, I'm not sure what you mean. Obviously they will never be able to predict the future, but models have been making correct predictions about forcing-temperature response, including multidecal temperature trends, for decades. Even Hansen et al 1988 was doing that fine for awhile, and other works have successfully predicted other aspects of climate as well (eg Hansen et al. 1992 successfully predicting the Pinatubo response). As far as I can tell, they largely just diverge when the real-world forcings diverge from the scenario in the model, and that's not the model's fault. Do models still need work? Sure. Are they completely off-base about how they treat climate? I really don't think so, especially where sensitivity is concerned, because high sensitivity makes sense of an awful lot of climatic observations, and every attempt at constraint, regardless of method and observation, returns a fairly similar value, whether it's ECS from the Last Glacial Maximum (and even outliers there, like Schmittner et al. 2011, still don't diverge hugely), or via water vapor feedback measurements from Pinatubo, or any of half a dozen other radically different methods, all basically returning the same range of numbers (~2.5-3C for most likely values).

The long and short of it is that as best as I can tell Moncton et al's basic contention about sensitivity just doesn't make any sense, and hillariously bad methods aside, is essentially entirely irreconcilable with basically every observation about climate ever made.

 

[PeterDonis]

The problem was that the models were being fed incorrect forcings in the first place

But if we're trying to use the models to make predictions, we don't know the forcings in advance.

Also, one of the items they corrected was ENSO, which is not an external forcing; it's an internal mode of variability. If they can't correctly predict ENSO's behavior, then that's a significant limitation in the models themselves, not in the data they're being fed.

 

[Catamount]

But if we're trying to use the models to make predictions, we don't know the forcings in advance.

Also, one of the items they corrected was ENSO, which is not an external forcing; it's an internal mode of variability. If they can't correctly predict ENSO's behavior, then that's a significant limitation in the models themselves, not in the data they're being fed.

In short, you're saying computer models will never (at least within in foreseeable future timeframe) be able to literally predict the future. I think that was always understood.

Computer models by their very nature can only offer scenarios, giving some idea of how climate might behave under certain conditions. That's a very powerful tool both for learning why climate behaved certain ways in the past for scientific curiosity, and offering risk assessments or useful geoeningeering data going forward. That they'll never be crystal balls is not, as far as I'm concerned, a weakness, because neither is anything else in science.Yes, the inability to explain what ENSO is going to do is a concern, because that is perhaps within the purview of what models are supposed to do. I'm not sure ENSO is ever really going to matter on multidecadal timescales, and if it doesn't it's neither here nor there. Is that really relevant to a discussion of Moncton et al (2015)? I don't think it is, because unless there's a relationship between CO2 change and ENSO behavior, it's not going to change climate sensitivity on any timescale (ie neither TCR nor ECS). At most, it's just one in a big list of source of semi-random fluctuation in the climate system, and mostly just on scales of a few years as far as I know. That climate is going to fluctuate in ways we can't predict for a long time, and that the best we can do is figure out how we might be biasing long term trends as it does, seems to go without saying.

 

[PeterDonis]

I'm not sure ENSO is ever really going to matter on multidecadal timescales

I'm not so sure. If the fraction of time that ENSO spends in the La Nina mode as opposed to the El Nino mode can change on those timescales, then that's a potentially significant change in the climate, even if each individual cycle only takes a few years to run.

 

[PeterDonis]

That they'll never be crystal balls is not, as far as I'm concerned, a weakness, because neither is anything else in science.

There are no perfect crystal balls anywhere in science, true. But the degree to which science's crystal balls approach perfection varies greatly between scientific fields. And what really matters is the accuracy of prediction that is possible relative to the accuracy that is needed to make it sensible to invest huge amounts of resources. In some fields, we have that level of accuracy--in astronomy, for example. If astronomers were to say that they predict that a certain asteroid will hit the Earth in, say, 2037 unless we do something, that prediction would be worth betting large amounts of resources on. But climate science is not one of those fields. That doesn't mean it isn't a valid science; it just means it hasn't reached that level of accuracy.

 

[Catamount]

I think we're largely on the same page here.

I agree that ENSO could definitely count for a lot more than we realize. On the rare occasions I had an opportunity to take a climatology course here or there, the first person I studied under actually lamented that we knew far, far too little about just about all teleconnections, largely because the focus for research just hasn't been there until very recently. He liked to characterize the state of the research by the IPCC Assessment Reports, and I suppose that is their purpose, and I recall him noting that in summarizing the science TAR barely mentioned teleconnections, and that it wasn't until the later research that made it into AR4 that there seemed to be a general scientific consensus that they were actually something really important.

It doesn't of course mean that ENSO is anything important, but yeah it definitely could be. Who knows, maybe external forcings even do influence the time spent in positive or negative phase or the magnitude of one or both.I do also agree that climatology doesn't come anywhere near the accuracy of some other fields of science. I'm loathe to call that a criticism, because it's generally a very hard science to get good data for, but you're right that that has significant implications when weighing climatology's predictions in any kind of risk assessment. Now it should be noted I think that, as Mike Mann is fond of saying "uncertainty cuts both ways", ie things could well be worse than climatologists understand them to be rather than better, but generally speaking I agree that we have to be careful in discussing potential actions to take not to get overzealous given the uncertainties.

Of course, one of the big advantages of dealing with climate vs an asteroid is that if we're smart, we may have opportunity to have our cake and eat it: deal with potential climatic change while doing things we ought to be doing anyways. There needn't be steep net costs to any endeavor outside of their utility in mitigating climatic changes. Unfortunately I fear that might be getting too far outside the purview of this discussion. I'm new on this forum, so I'm sure you know better where that line is drawn than I.

 

[PeterDonis]

it should be noted that, as Maike Mann is fond of saying "uncertainty cuts both ways", ie things could well be worse than climatologists understand them to be rather than better

This is true, but it doesn't make the climatologists' ability to predict what will happen any more accurate, so scientifically speaking it's irrelevant. It could be relevant in a discussion about judgment calls outside the purview of science, but that's a separate issue; see below.

if we're smart, we may have opportunity to have our cake and eat it: deal with potential climatic change while doing things we ought to be doing anyways.

I agree; but as you say, this is getting into the area where PF discussions are generally not supposed to go, since we're now starting to talk about matters of judgment in an area where science's ability to tell us what will happen is limited, rather than the science itself.

 

[johnbbahm]

I was looking at the GISS data set, and noticed something that may be part of the reasons the
Models are running hot.
http://data.giss.nasa.gov/gistemp/tabledata_v3/SH.Ts+dSST.txt
http://data.giss.nasa.gov/gistemp/tabledata_v3/NH.Ts+dSST.txt
The higher anomaly temperatures seem to occur in the colder months of the hemisphere.
temperatures averages , are unlikely to be from higher highs during the
"winter" months, the low temperatures must not be going as low, and increasing the average.
I was trying to think why CO2 would only preform it's quantum greenhouse function during the night.
I was thinking of a population inversion, from sunlight excited nitrogen, vibrationally exciting CO2.
During the daytime, there is likely an endless supply of excited nitrogen, continuously pumping
any available CO2.
The 14 um ground emission finds no absorbers, because they are all busy, in the sunlight hours.
At night the CO2 does absorb and re-emit, but the effect is like a thin layer of clouds.
This would be enough to mess the models up a bit.
It would also be fairly easy to test.
A daytime blue sky spectrum, should not have much of CO2 primary lines at 9.6 and 10.6 um.
If those lines are there, they did not come from being pumped at 14 um.

 

[Bystander]

The higher anomaly temperatures seem to occur in the colder months of the hemisphere.

"Seem?" There's not really anything in the table that reaches out and grabs attention one way or the other.

temperatures averages , are unlikely to be from higher highs during the
"winter" months,

Do you have any basis for such an assertion?

the low temperatures must not be going as low, and increasing the average.

Are you aware of the various methods used for measurements of daily highs and lows and how they've changed over the past century?

 

[johnbbahm]

"Seem?" There's not really anything in the table that reaches out and grabs attention one way or the other.

Do you have any basis for such an assertion?

Are you aware of the various methods used for measurements of daily highs and lows and how they've changed over the past century?

In the GISS data set, 3 months out of 12 NH and 12 SH caused the majority of the increase.
Those months were March, April, and December of the Northern Hemisphere.
The zone maps further broke it down to show the highest average zones were between 44 and 90 degrees north.
What is the likelihood of places say north of a line about Columbus, OH reporting record breaking annual highs,
in March, April, and December?

I am aware the GISS talks about this http://data.giss.nasa.gov/gistemp/abs_temp.html
I think their answer should raise questions about consistency of sampling.
"Q. What do we mean by daily mean SAT ?
A. Again, there is no universally accepted correct answer. Should we note the temperature every 6 hours and report the mean, should we do it every 2 hours, hourly, have a machine record it every second, or simply take the average of the highest and lowest temperature of the day ? On some days the various methods may lead to drastically different results."

 

[D H]

In the GISS data set, 3 months out of 12 NH and 12 SH caused the majority of the increase.

The data do not warrant you saying that. You cannot look at one year. All it takes is a prolonged warm spell in December to completely mess up the statistics for that particular December, or a prolonged cool spell in March.

Climate, and hence climate change, is what happens over longer periods of time. When you look at (for example) the years 2001 to 2014, you'll see

• That temperature rise is more evenly distributed across the calendar than it is for anyone year.
• That there's a lot more year-to-year variability from late fall to early spring than there is from late spring to early fall. Summertime is easy to predict in much of the world. It's going to be hot. Fall, winter, and spring? That's when weather forecasters don't do so well at forecasting.
• On an even longer scale, you'll see that the shift from winter to summer is getting earlier in the year, and the shift from summer to winter is getting later in the year.

 

[D H]

Actually 2014 being the hottest year was directly influenced by those 3 hot months in the NH.

Even if you only look at 2014, what you wrote a couple of posts back ("In the GISS data set, 3 months out of 12 NH and 12 SH caused the majority of the increase.") is not true.

Picking the northern hemisphere data from the three months that exhibited the greatest warming in the northern hemisphere and the southern hemisphere data from the three months that exhibited the greatest warming in the southern hemisphere is naturally going to result in more than 1/4 of the warming (which is what you'd get if the warming was uniform across the calendar). That's to be expected; you've cherry-picked the most extreme data. It does not account for the majority ("over half") of the increase. It's more like 1/3.

 

[johnbbahm]

Even if you only look at 2014, what you wrote a couple of posts back ("In the GISS data set, 3 months out of 12 NH and 12 SH caused the majority of the increase.") is not true.

Picking the northern hemisphere data from the three months that exhibited the greatest warming in the northern hemisphere and the southern hemisphere data from the three months that exhibited the greatest warming in the southern hemisphere is naturally going to result in more than 1/4 of the warming (which is what you'd get if the warming was uniform across the calendar). That's to be expected; you've cherry-picked the most extreme data. It does not account for the majority ("over half") of the increase. It's more like 1/3.

Actually, I just look again, they averaged NH and SH month by month, if just one month March NH 2014 dropped to normal,
I used 80, it brings the average down to 66, and there are still two more abnormally high months.
If I substitute in an abnormally high 85 for each of those months the average for the year drops to 65.
So it is true that those 3 months in the NH threw off the average.

 

[D H]

Actually, I just look again, they averaged NH and SH month by month, if just one month March NH 2014 dropped to normal,
I used 80, it brings the average down to 66, and there are still two more abnormally high months.
If I substitute in an abnormally high 85 for each of those months the average for the year drops to 65.
So it is true that those 3 months in the NH threw off the average.

You are cherry-picking. The same applies to *any* year, including cooler ones. In any given year, there will always be months with extreme temperatures because weather is a bit random. Even with your cherry-picking, your claim that three months in 2014 were responsible for the majority of the warming during 2014 not true.

I don't think you understand the point of this thread. This thread is not about the models that are used to analyze recorded temperatures "run hot" (which is what you appear to be arguing). It is about whether the models that are used to predict temperatures several decades into the future "run hot."

 

[D H]

Thread closed for moderation.