Thursday, 17 September 2015

Are climate models running hot or observations running cold?

“About thirty years ago there was much talk that geologists ought only to observe and not theorise; and I well remember some one saying that at this rate a man might as well go into a gravel-pit and count the pebbles and describe the colours. How odd it is that anyone should not see that all observation must be for or against some view if it is to be of any service!”
Charles Darwin

“If we had observations of the future, we obviously would trust them more than models, but unfortunately…"
Gavin Schmidt

"What is the use of having developed a science well enough to make predictions if, in the end, all we're willing to do is stand around and wait for them to come true?"
Sherwood Rowland

This is a post in a new series on whether we have underestimated global warming; this installment is inspired by a recent article on climate sensitivity discussed at And Then There's Physics.

The quirky Gavin Schmidt quote naturally wanted to say something similar to Sherwood Rowlands, but contrasted to Darwin I have to agree with Darwin and disagree with Schmidt. Schmidt got the quote from to Knutson & Tuleya (thank you ATTP in the comments).

The point is that you cannot look at data without a model, at least a model in your head. Some people may not be aware of their model, but models and observations always go hand in had. Either without the other is nothing. The naivete so often displayed at WUWT & Co. that you only need to look at the data is completely unscientific, especially when it is in all agony their cherry picked miniature part of the data.

Philosophers of science, please skip this paragraph. You could say that initially, in ancient Greece, philosophers only trusted logic and heavily distrusted the senses. This is natural at this time, if you put a stick in the water it looks bent, but if you feel with your hand it is still straight. In the 17th century British empiricism went to the other extreme and claimed that knowledge mainly comes from sensory experience. However, for science you need both, you cannot make sense of the senses without theory and theory helps you to ask the right questions to nature, without which you could observe whatever you'd like for eternity without making any real scientific progress. How many red Darwinian pebbles are there on Earth? Does that question help science? What do you mean with red pebbles?

In the hypothetical case of observations from the future, we would do the same. We would not prefer the observations, but use both observations and theory to understand what is going on. I am sure Gavin Schmidt would agree; I took his beautiful quote out of context.

Why I am writing this? What is left of "global warming has stopped" or "don't you know warming has paused?" is that models predicted more warming than we see in the observations. Or as a mitigation sceptic would say "the models are running hot". This difference is not big, this year we will probably get a temperature that fits to the mean of the projections, but we also have an El Nino year, thus we would expect the temperature to be on the high side this year, which it is not.

Figure from Cowtan et al. (2015). Caption by Ed Hawkins: Comparison of 84 RCP8.5 simulations against HadCRUT4 observations (black), using either air temperatures (red line and shading) or blended temperatures using the HadCRUT4 method (blue line and shading). The shaded regions represent the 90% range (i.e. from 5-95%) of the model simulations, with the corresponding lines representing the multi-model mean. The upper panel shows anomalies derived from the unmodified RCP8.5 results, the lower shows the results adjusted to include the effect of updated forcings from Schmidt et al. [2014]. Temperature anomalies are relative to 1961-1990.

If there is such a discrepancy, the naive British empiricist might say:
  • "the models are running hot", 
but the other two options are: And every of these three options has an infinity of possibilities. As this series will show, there are many observations that suggest that the station temperature "observations are running cold". This is just one of them. Then one has to weigh the evidence.

If there is any discrepancy a naive falsificationist may say that the theory is wrong. However, discrepancies always exist; most are stupid measurement errors. If a leaf does not fall to the ground, we do not immediately conclude that the theory of gravity is wrong. We start investigating. There is always the hope that a discrepancy can help to understand the problem better. It is from this better understanding that scientists conclude that the old theory was wrong.

Estimates of equilibrium climate sensitivity from the recent IPCC report. The dots indicate the mean estimates, the horizontal lines the confidence intervals. Only studies new to this IPCC report are labelled.

Looking at projections is "only" the last few decades, how does it look for the entire instrumental record? People have estimated the climate sensitivity from the global warming observed until now. The equilibrium climate sensitivity indicates how much warming is expected on the long term for a doubling of the CO2 concentration. The figure to the right shows that several lines of evidence suggest that the equilibrium climate sensitivity is about 3. This value is not only estimated from the climate models, but also from climatological constraints (such as the Earth having escaped from [[snow-ball Earth]]), from the response to volcanoes and from a diverse range of paleo reconstructions of past changes in the climate. And newly Andrew Dessler estimated the climate sensitivity to be 3 based on decadal variability.

The outliers are the "instrumental" estimates. Not only do they scatter a lot and have large confidence intervals; that is to be expected because global warming has only increased the temperature by 1°C up to now. However, these estimates are on average also below 3. This is a reason to critically assess the climate models, climatological constraints and paleo reconstructions, but the most likely resolution would be that the outlier category, the "instrumental" estimates, are not accurate.

The term "instrumental" estimate refers to highly simplified climate models that are tuned to the observed warming. They need additional information on the change in CO2 (quite reliable) and on changes in atmospheric dust particles (so-called aerosols) and their influence on clouds (highly uncertain). The large spread suggests that these methods are not (yet) robust and some of the simplifications also seem to produce biases towards too low sensitivity estimates. That these estimates are on average below 3 is likely mostly due to such problems with the method, but it could also suggest that "the observations are running cold".

In this light, the paper discussed over at And Then There's Physics is interesting. The paper reviews the scientific literature on the relationship between how well climate models simulate a change in the climate for which we have good observations and which is important for the climate sensitivity (water vapour, clouds, tropical thunderstorms and ice) and the climate sensitivity these models have. It argues that:
the collective guidance of this literature [shows] that model error has more likely resulted in ECS underestimation.
Given that these "emergent constraint" studies find that the climate sensitivity from dynamic climate models may well be too low rather than too high, it makes sense to investigate whether the estimates from the "instrumental" category, the highly simplified climate models, are too low. One reason could be because we have underestimated the amount of surface warming.

The top panel (A) shows a measure for the mixing between the lower and middle troposphere (LTMI) over warm tropical oceans. The observed range is between the two vertical dashed lines. Every coloured dot is a climate model. Only the models with a high equilibrium climate sensitivity are able to reproduce the observed lower tropospheric mixing.
The lower panel(B) shows a qualitative summary of the studies in this field. The vertical line is the climate sensitivity averaged over all climate models. For the models that reproduce water vapour well this average is about the same. For the models that reproduce ice (cryosphere), clouds, tropical thunder storms (ITCZ) well the climate sensitivity is higher.

Concluding, climate models and further estimates of the climate sensitivity suggest that we may underestimate the warming of the surface temperature. This is certainly not conclusive, but there are many lines of evidence that climate change is going faster than expected as we will in further posts in this series: Arctic sea ice and snow cover, precipitation, sea level rise predictions, lake and river warming, etc. In combination the [[consilienceof evidence]] suggests at least that "the observations running cold" is something we need to investigate.

Looking at the way station measurements are made there are also several reasons why the raw observations may show too little warming. The station temperature record is rightly seen as a reliable information source, but in the end it is just one piece of evidence and we should consider all of the evidence.

There are so many lines of evidence for underestimating global warming that science historian Naomi Oreskes wondered if climate scientists had a tendency to "err on the side of least drama" (Brysse et al., 2013). Rather than such a bias, all these underestimates of the speed of climate change could also have a common cause: an underestimate of global warming.

I did my best to give a fair view of the scientific literature, but like for most posts in this series this topic goes beyond my expertise (station data). Thus a main reason to write these posts is to get qualified feedback. Please use the comments for this or write to me.

Related information

Gavin Schmidt wrote the same 2 years ago from a modellers perspective: On mismatches between models and observations.

Gavin Schmidt's TED talk: The emergent patterns of climate change and corresponding article.

Climate Scientists Erring on the Side of Least Drama

Why raw temperatures show too little global warming

First post in this series wondering about a cooling bias: Lakes are warming at a surprisingly fast rate


Cowtan, Kevin, Zeke Hausfather, Ed Hawkins, Peter Jacobs, Michael E. Mann, Sonya K. Miller, Byron A. Steinman, Martin B. Stolpe, and Robert G. Way, 2015: Robust comparison of climate models with observations using blended land air and ocean sea surface temperatures. Geophysical Research Letters, 42, 6526–6534, doi: 10.1002/2015GL064888.

Fasullo, John T., Benjamin M. Sanderson and Kevin E. Trenberth, 2015: Recent Progress in Constraining Climate Sensitivity With Model Ensembles. Current Climate Change Reports, first online: 16 August 2015, doi: 10.1007/s40641-015-0021-7.

Schmidt, Gavin A. and Steven Sherwood, 2015: A practical philosophy of complex climate modelling. European Journal for Philosophy of Science, 5, no. 2, 149-169, doi: 10.1007/s13194-014-0102-9.

Brysse, Keynyn, Naomi Oreskes, Jessica O’Reilly and Michael Oppenheimer, 2013: Climate change prediction: Erring on the side of least drama? Global Environmental Change, 23, Issue 1, February 2013, Pages 327–337, doi: 10.1016/j.gloenvcha.2012.10.008.


  1. Your points are well made, but I think you may have slightly mis-interpreted the quote you attribute to Gavin (which he attributes to Knutson & Tuleya). I don't think it is intended as an observations/data are better than models argument, I think it is very simply that if you're trying to predict/project what will happen in the future, you might trust actual observations of the future more than you might trust models of the future. Since the former is not possible, we simply have to use models.

  2. Hi ATTP, I know I was unfair to Gavin. I had hoped I had made that clear by writing: "The quirky Gavin Schmidt quote naturally wanted to say something similar to Sherwood Rowlands, ..."

    Being such a beautiful quote, it was tempting not to take it out of context. ;-)

  3. Have updated the post (not everyone reads the comments) to make clearer that I was unfair and took the quote out of context. It is such a beautiful quote, it was too tempting not to use it. ;-) Have also added the attribution to Knutson & Tuleya. Thanks ATTP, is a nice article.

  4. I was describing to an acquaintance the other day what models were trying to do and why they don't always exactly predict what the climate does in the short term. I used the analogy of a car journey. It went...

    If you asked me how long it would take you to drive from London to Edinburgh, the model in my head would be based on the speed limits and the type of roads—all of which I've driven in the past. I'd add up these journey times and then think 'seven and a quarter hours; and refine my model further by adding a few stops for pees and maybe lunch'. So having mentally done the sum I'd say "around 8 hours in total...", and then add "...depending on the traffic" to cover myself. If I then drew out a graph of the model (journey) I'd distribute the pee and lunch stops where seemed appropriate, and that would be my finished prediction.

    Now if you actually drove the journey you might take your pee stops in different places and have your lunch either early or late. Anyone comparing my prediction with the actual journey as it happened might say "you're running ahead" or "running behind" the prediction, which could at some stages look like I got it wrong. But, low and behold, you arrive at the predicted time... subject to traffic!

    I hope that's useful. Sorry to simplify the debate.

  5. That is a nice example. Simplifying the debate is perfect. And examples outside of climatology helps people attached to their identity as mitigation sceptic to more easily see the reasoning errors.

    (Which probably still does not help much, the most vocal people do not seem to care much about being reasonable. But others might.)

    Love this headline: In a blind test, economists reject the notion of a global warming pause.

  6. Hm, interesting. Do you think there is still room for further warming the present (like Karl & Co. recently did) and/or cooling the past in order to avoid model falsification? It seems like you think there is still further room for this?

    Since the troposphere is expected to warm at least as fast as the surface (but RSS data shows only insignificant warming), it seems to me that further adjustment upwards will totally clash with RSS and the tamperature scam will be bust.

    Best regards, Carl

  7. Carl, yes I would not be surprised if there is a larger trend bias in the observations. The transition to Automatic Weather Stations is most likely to lead to cooling (a smaller radiation error due to the smaller sensor and often forced ventilation). This transition, with modern equipment, also has made it possible to site the stations better; they no longer need to be close to where the observer lives or works. (In the USA this transition started early when the technology was still quite primitive and cables short, that is no longer a problem.)

    Yes, there would be a conflict with the tropospheric temperatures. I do not understand why the people of UAH are so proud that their dataset does not contain the tropic hotspot and thus shows too little warming. We have a lot of evidence that this hotspot does exist and should exist.

    Furthermore, we know that this dataset is buggy, it depends on a large string of satellite of different builds. A nice dataset for a global overview at best, certainly not something you would prioritise when it comes to trends. When UAH still claimed that the troposphere was cooling, before they made the huge adjustments to their dataset, scientists also did not ignore the evidence from so many better quality sources that showed warming (station data, sea surface temperature, a wide range of changes in nature, melting glaciers, stronger precipitation, less snow cover) in favour of this dataset on which only a few people work once in a while.

  8. Hi again, I did not mention UAH, did I?

    The RSS data is firmly in the hands of warmists, but they still have some scientific integrity left, it seems. Probably they will soon feel the pressure from Obamas goons, like Karl did.

    Regards, Carl

  9. Hi again, I did not mention that UAH was the only problem, did I?

    Also the RSS dataset needs to make such huge adjustments. Nothing wrong with that, but then you do need to be able to show these adjustments are right. With just a bunch of people working on this part-time it is hard to attain sufficient quality levels.

    Do you, for example, know of a good validation study? Where people simulated what the satellites would observe based on a global atmospheric temperature dataset. Then applied the retrieval algorithms for tropospheric temperature on that and checked if the temperature trends were the same?

    I do not. I do know such studies for the land surface temperature. In fact, I did one.

    And that would still not protect us from unknown unknowns, which make a large research community even more important. But it would be a start.

  10. All you have done is demonstrate perfect confirmation bias. Models are always dependent on and verified by the data - not the other way around. If you modify the data to confirm your model then that is the most unscientific act imaginable. I really should not have to explain why but just start with the known fact that models are much simplified and very incomplete. The purpose of a model is to simulate as well as possible the real world in the full knowledge that it cannot be anything other than a crude approximation. Model approximations of reality and reality itself are not considered as equally valid in any other field so why should climate science be different?

    The large chunks of missing physics and gross simplifications to the maths in the models may not even be the main problem but rather the input which assumed natural variation was small and declining. Once that assumption is made then any warming perforce must be taken to be man-made. The models merely added a mathematical wrapper to this circular reasoning. If however the models and future obs agree then you can say that the initial assumption was justified. If they don't agree then you must adjust the input to reflect the reality that natural variation has obviously been underestimated. That is how the science must be done and Gavin knows this very well. Whether you are just deluding yourself or you really don't understand this is unclear but if bridges were built your way they would inevitably collapse and kill people.

  11. jgdes, did you read the same post as the one I wrote? I did not write that we should believe the models unconditionally and "modify the data" to make it fit.

    I only argued that the reverse is just as stupid, to ignore the models, to believe the data unconditionally and just "modify" the models.

    This post is a call for more research. We should understand the reasons for the differences. With all the other evidence that there is a cooling bias in the station observations, I would see this as the most likely candidate. But let's wait and see what the evidence will say.

    I see you work on Finite Element Modelling. Thus apparently you do see some use in modelling. In the end you will have to build the design to see whether it really works. That is comparable to the research we should now do.


Comments are welcome, but comments without arguments may be deleted. Please try to remain on topic. (See also moderation page.)

I read every comment before publishing it. Spam comments are useless.

This comment box can be stretched for more space.