GFDL - Geophysical Fluid Dynamics Laboratory

55. Tropical tropospheric warming revisited: Part 2

Posted on January 20th, 2015 in Isaac Held's Blog

Vertical profile of temperature trends averaged over 20S-20N in two models.  Solid: trends in a 30-year (1970-2000) realization of the CM2.1 coupled model using estimated forcing agents from 1970-2000.  Dashed: simulation using the atmosphere/land component of CM2.1 with the same forcing agents but running over the sea surface temperatures generated in this realization of the coupled model.

The previous post summarizes the results from a recent paper, Flannaghan et al 2014, that uses atmosphere/land models running over observed sea surface temperatures (SSTs) to look at the consistency between these models and observations of tropical tropospheric temperature trends.  The idea of using this kind of uncoupled model is to try to put aside the issue of SST trends in the tropics and focus more sharply on the vertical structure of the temperature trends. Because models are so consistent in producing a warming trend that is top-heavy in the tropical troposphere, due to the strong tendency to follow a moist adiabatic profile, and because this pattern of change has numerous ramifications for tropical climate more generally, any possibility that this warming profile is wrong takes precedence over other issues in tropical climate change, in my view.  I interpret the results in Flannaghan et al to say that microwave sounding data, at least, does not require us to reject the hypothesis provided by climate models for the vertical profile of the tropical temperature trends.

To follow up on the last post, I would like to discuss, or at least mention, some other issues regarding this setup, in which SSTs are simply prescribed as a boundary condition.  Could there be something fundamentally flawed about this approach?  This may seem like a technical issue, but a large fraction of atmospheric model development takes place in this prescribed SST framework to try to separate biases due to atmospheric model imperfections from those due to the ocean/sea ice model, so it is important to understand its limitations.  I have used this kind of setup in a number of posts to address other issues, for example #2,#10, #11, #32, #34; any limitations to this decoupled framework could affect my own thinking about a variety of climate change issues.

In a coupled model we integrate the atmosphere/land state A and the ocean state O forward in time

\partial A/\partial t = \mathcal{A}(A, O); \,\,\,\, \partial O/\partial t = \mathcal{O}(A, O)

Let’s assume that the atmosphere/land model \mathcal{A} is deterministic and that the SST is the only piece of information about the state of the ocean that the atmosphere feels, so the first equation can be replaced by \partial A/\partial t = \mathcal{A}(A,  SST).  Run the coupled model and store off the SSTs; then run the atmospheric model in isolation, reading in these time-evolving SSTs as needed.  You’ll get the same answer. So what’s the problem?

The problem is that this kind of perfect substitution is not physically relevant since the whole point is to run over observed SSTs and compare to atmospheric observations. The atmospheric model is imperfect, we don’t know the initial conditions precisely, and the atmospheric model is chaotic — any perturbation in the atmosphere due to model imperfections no matter how small or due to differences in initial conditions will grow in time.  Think of a hurricane that develops in the coupled model and imprints a cold wake on the SST.  In a slightly perturbed realization of the atmospheric model no storm is present at that point and the cold wake appears out of thin air.  And a storm develops elsewhere with no corresponding cold SST signature, possibly resulting in a biased storm strength.  Could biases in storm strength get rectified somehow and bias the tropospheric temperature trends?  Or consider the Madden-Julian Oscillation (MJO), an important mode of variability in the tropical Indian and western Pacific oceans with a 30-50 day time scale.  A number of models indicate that coupling dynamically to SSTs affects the amplitude and frequency of the MJO.  If uncoupling the atmosphere from the SSTs alters MJO variability, could this difference in variability be rectified to affect tropospheric temperature trends?

On the other hand, our theories of ENSO variability are typically consistent with a picture in which the atmospheric component of this coupled variability can be understood as the response to the SST anomalies, so in this picture one can regenerate the atmospheric state through ENSO cycles by running an uncoupled atmospheric model over observed SSTs. The difference in the case of ENSO is evidently that there is little atmospheric variability at this low frequency in the tropics in the absence of the SST anomalies.  In the storm and MJO examples intrinsic atmospheric variability imprints itself on the SSTs and the back effect of these SSTs on the atmosphere is then distorted if the coherence of this atmospheric variability and the SSTs is destroyed.    (At least that is my understanding).  See Bretherton and Battisti 2000 for another interesting example, involving the North Atlantic Oscillation.

How do you tell if this kind of thing is important or not?  One way to start is through a  “perfect model” test.  Take the SSTs from a coupled model and prescribe these as the boundary condition for the atmosphere/land component of this same model — perturbing the initial conditions to create another realization of the atmospheric state.  Bruce Wyman at GFDL did this using 30 years (1970-2000) from a single run of the CM2.1 coupled model (this is run2 in the CMIP3 archive) and comparing what we get running the atmosphere/land model over these SSTs.  The uncoupled model turns out to be slightly warmer in the tropical upper troposphere, by 0.15C on average.  Interestingly, this difference is biggest during cold ( La Nina) phases in the model — as seen below, using 12-month running means.  (As pointed out in other posts, this model produces too many super- El Nino events, creating too much variability in tropical temperatures compared to observations.)

The figure at the top of the post compares the vertical profiles of the trends in these two models.  (Like many other models, this coupled model warms too much in the tropics over this time period.) The coupled model warms a bit more aloft than the uncoupled model running over the coupled model’s SSTs, consistent with the difference between the blue and red lines in the time series plot converging over time.  I think this is worth pursuing, to understand these differences better, and to check with other models to see if they are consistent.  But the bottom line is that the differences in trend are small in this model.  Hopefully this smallness is robust, providing confirmation that we can in fact use prescribed-SST models to address this vertical-profile-of-tropical-trends issue. This conclusion could be different for other aspects of climate change.

On a related point, in discussions of surface vs upper tropospheric trends you often see comparisons with the land+ocean surface temperatures rather than SSTs in isolation.  But in the kind of simulations described here, the land is thought of as part of the “atmosphere” — land temperatures are free to change in response to SSTs and forcing agents.  Part of the rationale for running over prescribed SSTs is that the SSTs in large part evolve more slowly than the intrinsic variability in the atmospheric state.  But the land surface has time scales comparable to the atmosphere — decoupling them could distort the diurnal cycle among other things.  So we typically do not try to prescribe the land temperature in this kind of simulation.  One could try to get around this by prescribing the temperature of land layers at depths not affected by the diurnal cycle, for example.  But there is a more important reason to avoid prescribing land temperatures when considering this vertical profile issue.

The moist adiabatic temperature profile depends not only on the temperature of the parcel near the surface that one starts with, but also its water vapor content.  Over the ocean we can hopefully get away with thinking in terms of temperature only because the relative humidity of air near the surface is so strongly constrained as the climate warms (see Post #47).  But over land relative humidity is freer to change.  One can argue that, at least in regions that maintain some convection,  the surface temperature changes will try to maintain consistency with the changes in near-surface relative humidity so that one still ends up at more or less the same temperature as the oceanic parcels when lifted to the upper troposphere. To accomplish this the land has to warm more in regions that dry out with warming.  See Joshi et al 2008 and  Byrne and O’Gorman 2013.  This is a topic that I plan to return to soon in a future post.   Comparing trends in tropical land+ocean near-surface temperatures to trends in upper tropospheric temperatures is confusing without taking the changes in relative humidity into account.

Another issue that come up when thinking about these prescribed SST atmosphere/land models is the extent to which changes in the forcing agents can modify the tropospheric tropical temperature profile even with fixed SSTs — particularly reductions in ozone affecting temperatures near the tropopause and greenhouse gases/aerosols affecting lower tropospheric temperatures over land.  This would be worthwhile returning to in another post as well.

[The views expressed on this blog are in no sense official positions of the Geophysical Fluid Dynamics Laboratory, the National Oceanic and Atmospheric Administration, or the Department of Commerce.]

4 thoughts on “55. Tropical tropospheric warming revisited: Part 2”

  1. Isaac,

    Thanks for this post. You mention that models have a strong tendency to follow moist-adiabatic lapse rates with warming. But the CMIP models produce a wide range of warming profiles in simulations of 21st century climate. For instance, Paul O’Gorman [O’Gorman & Singh (2013)] has found that the ratio of tropical warming between the 300 hPa and 500 hPa levels varies between 1.2 and 1.6 across the ensemble. It seems hard to understand how such a result could arise if the tropospheric mean temperature profile is locked to a moist adiabat, no matter how that moist adiabat is related to near surface temperature/entropy distribution. Perhaps the flatness of tropical tropospheric temperatures is not sufficient to ensure the mean temperature is set solely by the convecting regions?

    Do you think the large range of amplification we found is a feature of coupled simulations only? I am not aware if anyone has looked at mid vs. upper tropospheric warming in the +4 K CMIP simulations, for instance.


    O’Gorman, P. A. & Singh, M. S. (2013) Vertical structure of warming consistent with an upward shift in the middle and upper troposphere Geophys. Res. Lett., 40, 1838-1842

    1. Hi Marty, thanks again for the useful comment. Your upward shift argument makes sense when confined to the upper troposphere, consistent with the FAT hypothesis in particular (post #39), and the analysis of the CMIP5 results is pretty convincing. We were thinking of your paper when we were putting together ours, but because it is not clear how to connect your argument to the surface we did not try to bring it into our discussion. For better or worse, we decided to focus on the relation between SSTs and upper level warming.

  2. Please debunk Tropospheric Hotspot problem..

    Climate change deniers are pointing at the lack of a ‘tropospheric hotspot’. Recently someone sent me this which shows that predictions of warming have all overestimated what was eventually observed.

    Skeptic claim: No Climate Model ever used by The IPCC can replicate the current lack of warming over 16 years. I think that’s a very important Scientific fact that is being ignored. That’s how the IPCC works, it ignores what doesn’t fit leaving only the evidence that does fit.


    1. I have received a few comments like this in recent days. Regarding the hotspot controversy, it is important to me to divide this into two distinct problems. 1) Can models simulate observed tropospheric temperature trends given observed SSTs, and 2) can models simulate the observed SST trends. I think it is important to separate them if we can because the two are often confused and they point to potential problems of very different types in our models. In a series of posts (#20, #54, #55, #61), and in a couple of papers with colleagues discussed in those posts, I have tried to isolate and address the first of these problems. (This particular post addresses the relatively technical issue of whether this separation is itself well-posed.) My view, always subject to revision of course, is that there is no clear discrepancy between models and observations when the models are given observed SSTs over time as their lower boundary condition. (I am referring here to the tropics, where the hotspot controversy is focused.) Not every one agrees with this, but I have done my best and this is my considered opinion at present. Among the key things I and my colleagues have tried to point out in our studies is the importance of thinking about the trends in surface temperature over the oceanic regions where convection is occurring when trying to relate tropical surface and tropospheric trends.

      The second point is really the key. As I put it at the top of post #63, unhelpfully since it is covering all bases, the discrepancy between the observations and forced SST trends in models is “… due to some combination of internal variability, incorrectly simulated climate responses to the changes in forcing agents, and incorrect assumptions about the forcing agents themselves.”. In my view this is essentially the hiatus issue, and the problem is most clearly seen in the tropics. It is unfortunate that this issue is often discussed with the unhelpful focus on whether the observed trend over some period is or is not positive — the key point for me, and I think we agree on this, is the difference between modeled and observed trends. I would personally prefer to see this discussed in terms of the SST data itself and not the tropospheric trends. These are two different things that are not connected as simply as is often assumed, as discussed in the blog posts mentioned above, confounding questions (1) and (2).

      As far as IPCC is concerned, I have allowed this comment here but I am not interested in blanket statements about “IPCC this or that”. There are plenty of other places to discuss this online as you are well aware. I am interested in the science. If you have an issue with the the hotspot or hiatus discussion in Chapter 9 of the WG1 report that’s great — no serious student of this subject could possible agree with everything in the report — but you need to point to the specifics of that discussion that you disagree with if you expect me to address it here. It is also important to me to keep the discussion here focused on the content of specific blog posts. I am discussing question (1) in this post — I am not discussing deficiencies in modeled SST trends over certain time periods. But I do want to return to that important question fairly soon.

Leave a Reply

Your email address will not be published. Required fields are marked *