A couple of highlights

Since I haven’t had much chance to write anything recently, I thought I would briefly advertise a couple of papers that may be of interest to my regular readers. One is by Clare Marie Flynn and Thorsten Mauritsen and is [o]n the Climate Sensitivity and Historical Warming Evolution in Recent Coupled Model Ensembles and compares the CMIP5 and CMIP6 models ensembles.

The CMIP6 ensemble suggests a shift towards a higher equilibrium climate sensitivity (ECS), when compared with the CMIP5 ensemble. The Flynn & Mauritsen paper illustrates that this can’t be due to chance, suggesting that the CMIP6 mean ECS is indeed highly unusual. Consistent with the paper I discussed here, they seem to find that the shift in ECS is mostly due to an increase in the shortwave cloud feedback, mostly in the Southern extratropics. Even though there is a shift to a higher ECS, they also find that none of the models with a Transient Climate Response (TCR) above 2.5oC matches the post-1970s warming. This seems broadly consistent with the results from the paper I discussed in this post.

Probability distribution of the greenhouse gas attributable effective climate sensitivity for the periods 1862-2012 (top) and 1955-2012 (bottom). [Credit: Tokarska et al. (2020)]

The other paper I thought I would highlight is Observational constraints on the effective climate sensitivity from the historical period, by Kasia Tokarska and colleagues. They make use of detection and attribution techniques to derive the surface air temperature and ocean warming that can be attributed directly to greenhouse gas increases. They then use this, together with an energy balance model, to infer the effective climate sensitivity (which they refer to as {S}_{his{t}_{GHG}}). As shown in the figure on the right, they find a 5-95% range of 1.3oC – 3.1oC for the period 1862-2012, and 1.7oC – 4.6oC for the period 1955-2012.

For the two time periods, the median values are 2.0oC (1862-2012) and 2.8oC (1955-2012). However, they do highlight that [O]ur estimate of {S}_{his{t}_{GHG}} is lower than the documented ECS of some climate models (e.g. CMIP5 multi-model mean ECS of 3.22oC; Forster et al 2013), including that of some used in the analysis. However, it is well understood that time-dependent feedbacks might render {S}_{his{t}_{GHG}}\,lower than S at equilibrium. This is because lower values for {S}_{his{t}_{GHG}} than S at equilibrium can be explained by the effects of changing strength of the feedbacks at higher levels of warming.

That was all I was really going to say. Both papers are open access, so I’d encourage those who are interested to read them in more detail.

This entry was posted in Climate change, Climate sensitivity, Global warming, Research, Science and tagged , , , , . Bookmark the permalink.

19 Responses to A couple of highlights

  1. Chubbs says:

    A recent discussion paper documents CMIP3, CMIP5 and CMIP6, model performance across a wide range of climate patterns. The extensive output archive, which is still being updated, is also available. Steady improvement in model performance across generations, but with remaining weaknesses, and a range among models. Penultimate quote below:

    “Also relevant for climate feedbacks, Variable Scores for SWCF, LWCF, RH500, and precipitation have increased steadily across the CMIP generations (e.g. Fig. 10), with magnitudes exceeding the uncertainty associated with internal variability. Scores are particularly high for CMIP6 models for which high climate sensitivities have been reported, including CESM2, SAM0-UNICON, GFDL-CM4, CNRM-CM6-1, E3SM, and EC-Earth3-Veg (though exceptions also exist such as in the case of MIRCO6).”


  2. Chubbs,
    That looks interesting, thanks.

  3. JCH says:

    You have a recent paper by Drotos, along with Bjorn Stevens, in which at very high CO2 levels they end up with episodic global cooling events of up to 10k because of low clouds in the eastern and equatorial Pacific, and you have Tapio Schneider’s burn off of low clouds with jungles in the arctic.

    Global variability in radiative-convective equilibrium with a slab ocean under a wide range of CO2 concentrations

  4. izen says:

    While we can appreciate the dotting of i’s and crossing of t’s on the science of AGW, that is not where the problem with this issue resides.

    Consider this output from one of the leading global media outlets, Sky News, perhaps it would not be a viable program in some other Nations, and even in Australia it is framed as an outsider view.
    But the attitude and the subsequent comments show the arena where AGW is disputed. It is not the science…

  5. David B Benson says:

    Wrong thread, but the bigger the ecosystem the harder the fall:

  6. The Very Reverend Jebediah Hypotenuse says:


    Since your posted Sky News video references ‘Electroverse’, which pimps Delingpole articles from Breibart and WUWT, you might find this latest Wiki-scandal amusing.

    Spoiler alert:
    If you create a list of deniers, you’re Hitler. If you destroy a list of deniers, you’re Stalin.

  7. JCH says:

    This little box: episodic coolings of up to 10K versus Eocene highs. No, no important disputes there.

  8. Ben McMillan says:

    The CMIP5 vs. CMIP6 comparison is interesting. There don’t seem to be a really clear subset of high-skill models with narrowly-clustered ECS, so it really looks like there are a broad range of plausible ECS estimates from models. Also the TCS and ECS are not as well correlated as I thought. I guess I was hoping better models and better observations would constrain the range of model-based ECS values some time this decade but doesn’t look like that will happen anytime soon.

  9. jacksmith4tx says:

    “Deep Learning Accelerates Scientific Simulations up to Two Billion Times”

    This is tangentially related to several related fields of science many here follow.
    In a few years we could see some pretty amazing performance improvements by just changing a few lines of code. You can bet we will use these new algorithms to design new vaccines, speed up genetic engineering research and improve energy, climate and weather models.
    Of course this is what they worry about when they talk about the AI singularity too. The rate of change could far out strip our ability to adapt, just like climate change is doing to the whole biosphere.

    “The research team used DENSE to build emulators for 10 different simulation cases from fields such as:
    high-energy-density physics
    fusion energy science
    climate science
    earth science

    Training data was gathered by running each simulator 14,000 times with random input data, except for two simulators that take hundreds or thousands of CPU-hours to run; for these two, less than 1,000 data points were gathered. The team found that the emulator output “generally matches closely” to the simulator output, even for the two cases with limited training data. The emulators ran much faster as well; for simulations that run in “minutes to days,” the emulators can run in “milliseconds to a few seconds.” DENSE also outperformed other non-neural network emulators, such as random forests, or manually-designed neural networks.”

  10. On that deep-learning paper, someone on Twitter brought up an interesting question that the author of the research didn’t directly respond to. This is the question that seems critical: “what are the exact inputs?”

    On Twitter, the author seemed to think it referred to tuned input parameters. But that’s different from a forcing input. The external forcing comes from outside the NN, so can’t be learned from the training data. And because for many systems, the output is a response surface directly resulting from the input, the deep learning system may never capture it. Of course, I may be missing something. Perhaps the entire external forcing is recreated within the NN.

  11. OT, but with the Covid 19 pandemic we have initiated a fascinating experiment to measure effects of sudden decrease in CO2 emissions by our species. There are two impacts that I expect: a fall in the yoy increase of CO2 in atmosphere as measured at MLO. The other impact will be driven by fall in aerosol pollutants from fossil fuel burning. I think we will see an immediate bump up in global temps from the fall in the aerosol pollutants.

    Maybe this gives us something to do while in self-isolation.

  12. Chubbs says:


    Thanks for the Tapio Schneider video and the hiatus-post hiatus SST plot. I would expect climate models to at least get the directional effects of GHG on stratocumulus correct. Per the video, the effect is straightforward and easily parameterized: more outgoing radiation from cloud top = more stable stratocumulus and vice-versa. Further evidence climate models are directionally correct on strat clouds is provided by the recent paper below:

    “We find remarkable agreement between observed and simulated differences in reflected solar and emitted thermal infrared radiation between the post‐hiatus and hiatus periods. Furthermore, a model’s ability to correctly relate Earth’s radiation budget and surface temperature is found to depend upon how well it represents reflected solar radiation changes in regions dominated by low clouds, particularly those over the eastern Pacific ocean.”


  13. JCH says:

    Chubbs – the map I posted above is from the paper you’ve quoted and linked – Loeb.

  14. Chubbs says:

    JCH – haha so it is – preaching to the choir

  15. JCH says:

    The paper that sort of seems at odds, though I’m growing a little more confident it is not, is the Drotos paper, which showed episodic cooling events at high CO2 levels of up to 10K and 0 climate sensitivity.

  16. Dave_Geologist says:

    So the good news if Drotos et al. are right is that warming may saturate at around 5°C. The bad news is that it may oscillate between 0°C and 10°C, on a time frame of a year to a decade. Good luck farming in that world.

    I wonder if something like that was going on in the PETM? Something along those lines might explain the observation in Mediterranean regions of flipping between multi-decade droughts and storms which rolled car-sized boulders hundreds of miles. We know it wasn’t just a matter of localised intense storms hitting only a small region each year – the whole of Spain, for example, went one way or the other at once. It would be like going from a glacial to a very hot interglacial and vice versa in a decade. We know that climate belts shifted hundreds or thousands of miles and some (like the mammoth steppe) appeared and disappeared, but that was with centuries or millennia for ecosystems to adapt.

  17. Chubbs says:

    The Drotos et. al. paper is interesting, but the idealized modeling configuration needs to be considered when extrapolating to the real-world. Per the paper, radiative-convective models are used to provide insight into energy balance, stability and feedback in a simplified system. In particular, this model has a slab ocean and spatially uniform solar heating. There are no continents or rotation. The authors point out that their model doesn’t have the planetary waves caused by differential heating and rotation that play an important role in determining the location of convection. They also mention that adding more realistic conditions, would limit the scale of convective self-aggregation, and change the nature of the response. If the physics studied here was broadly applicable, then there would have been a different eocene.

    To me this paper further highlights the importance of stratocumulus/tropical dynamics as a research objective and is consistent with the Tapeo S result. Stratocumulus require a certain set of conditions and have threshold behavior which introduces stability regimes. The hiatus–>post hiatus transition shows how rapidly the global energy balance can change due to the physics studied here.

  18. jacksmith4tx says:

    Follow up to previous item about speeding up modeling:
    “The new, superparameterized E3SM (SP-E3SM) outperforms the standard E3SM by some metrics, such as in correctly re-creating the daily timing of peak rainfall and in the representation of tropical waves—atmospheric features associated with storms. These results are in line with improvements seen in other superparameterized models.

    SP-E3SM also runs very fast, simulating about 1.2–1.4 years of data per day of computing, compared with roughly 0.2 year of data per day in similar superparameterized models. (SP-E3SM is still slower than the standard E3SM, however, which can produce 5–7 years of data per day.) The researchers achieved this acceleration in large part by restructuring the model’s code to run on DOE’s powerful graphics processing unit computing hardware.”

    Despite the improvements, SP-E3SM suffers from a unique problem known as grid imprinting, which introduces errors into the pattern of rainfall the model simulates. The researchers noted that they are addressing the grid imprinting problem as they refine the new model.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.