In a recent post , we discussed the
factors that may be driving the North Atlantic Oscillation up or
down, as opposite trends have been put forward to explain droughts in the
Mediterranean and the unusual behaviour of hurricane Sandy . What
kind of simulation could provide an answer ?
First of all, some lengthy, but hopefully
interesting background- you may need to read our previous post to better get the context. The North Atlantic Oscillation index
summarizes a picture the winter-to-winter climate variability in the North Atlantic
European sector, ranging from the intensity of the westerly winds, through
temperature in Scandinavia and precipitation in the Mediterranean. A
more positive index in a single month or winter season indicates that westerly winds have been stronger and Mediterranean precipitation
has been less than usual. Basically, the NAO describes the strength
of the Azores high pressure system.
So far, climate simulations so far tend to
show that the NAO should turn more positive due to greenhouse gas
forcing. More recently, a few studies (see here for a popular-scientific story) suggested that the retreat of
sea-ice in the Arctic in autumn would induce anomalous heat fluxed
from the ocean that in turn would push the NAO to a more negative
state. So we have two opposing pulls both originated in the increase
of atmospheric concentrations of greenhouse gases. Let us have a look first at
the NAO index observed since 1820.
Perhaps the best description of
this figure is that it is hard to say that something spectacular is
happening recently. There is an upward trend from 1900 until 1960
followed by a downward trend thereafter. The interannual variations
are very strong and we find extreme positive and negative values
distributed along the whole period. The upward trend encompass a
period of global rising temperatures from 1900 until 1945 and a
period of cooling tin 1960-1975 and warming temperatures from 1975
onwards. So at first sight it is difficult to ascribe the NAO
variations in October to global temperature variations and thus to
diminishing sea-ice. It is also noteworthy that the negative trend in
the last 30 years, which is linked to increasing precipitation in the
Mediterranean is not compatible with the future projections derived
from climate models included in the IPCC 2007 report. This has been
highlighted in a recent paper by one of our students, Armineh
Barkhordarian. Essentially, all models predict decreasing
Mediterranean precipitation in autumn, whereas observartions indicate an increase in the last 30 years.
A view of the NAO index simulated in a simulation with the model ECHAM4-HOPE (ECHO) over the past millennium and in the this century under the most pessimistic scenario of greenhouse emissions neatly illustrates this:
A view of the NAO index simulated in a simulation with the model ECHAM4-HOPE (ECHO) over the past millennium and in the this century under the most pessimistic scenario of greenhouse emissions neatly illustrates this:
We see again very large interannual
variations but a certain tendency for the NAO to be in a more
positive state in periods with stronger external forcing, such as the
Mediaeval Warm Period and the 20th century, and even more
clearly in the 21st century. This behaviour is quite typical of the
models included in the IPCC 2007 report. It was too much work for
this post to calculate the NAO index for all models for just the
month of October, but the winter mean index has been analysed before
by others:
Not all models display a clear trend in
the future, but if a trend is present, it is positive. I have tested
that for a few models that exhibit a trend, the behaviour of the
simulated October NAO index is consistent with the simulated whole-winter
NAO. In particular, the model ECHO-G and the model ECHAM5-OM, both
being sequential versions of the same model ECHAM, display the same
NAO trend.
Finally, let us consider the model
ECHAM6-OM, also known as model MPI-ESM-LR, which is already part of
the suite of models included the forthcoming IPCC 2013 report. This result turned to be quite surprising. The figure below shows the October NAO index
obtained in a scenario simulation driven by the most pessimistic
scenario RCP 8.5, which is roughly comparable to the previous
SRES scenario A2:
We have three simulations with the
MPI-ESM model driven by the same scenario. The difference among the simulations are the
initial conditions in 1st January 2006. This small
ensemble would roughly represent the uncertainty due to internal
climate variability, distinct and independent in simulation, as opposed to the common signal generated by the
external forcing.
The model ECHAM6 shows a quite
different behaviour. The NAO index in October shows almost no trend
until about 2080, and a negative trend thereafter, when the
greenhouse gas forcing becomes really strong It seems thus that the
NAO in this model is less sensitive to the external forcing, and that
the NAO reaction is opposed to the one simulated by ECHAM5 and
ECHAM4. This is not the first case of a flip in the NAO trend when a
new version of a climate model is introduced. In an somewhat older
paper we already noticed that the British model HadCM2 and the German model ECHAM3 model produced
opposite NAO trends when driven by increasing concentrations of
greenhouse gases, with ECHAM3 later agreeing with the subsequent version of the British model HadCM3.
Even more intriguing is now the fact
that the winter NAO index, i.e. the mean over October to March,
and not only picking the October month, does not display any discernible trend in the MPI-ESM simulations:
And now the question for you, interested
reader: what type of simulation should be conducted to disentangle
the influence of the 'direct' greenhouse gas forcing and of the
indirect influence of diminishing Arctic sea-ice cover ?
Posts
13 comments:
Hi eduardo,
nice to have some science here again.
What about the naive approach, prescribing sea ice, similar to Petoukhov and Semenov (2010)?
"And now the question for you, interested reader: what type of simulation should be conducted to disentangle the influence of the 'direct' greenhouse gas forcing and of the indirect influence of diminishing Arctic sea-ice cover ?"
The standard answer should be:
Disentangle the influences.
I suggest a couple of further simulations:
1.) Set arctic sea ice and GHG concentration as constant.
2.) Set GHGs constant and treat a shrinking sea ice scenario as an input parameter.
3. Set sea ice constant and include rising GHSs.
What do you think, Eduardo? Is it possible to perform those runs?
I suspect, you've just waited for these suggestions. ;-)
If the models would show in these simulations contradictory results as given in your post, the answer would be: The models are not useful to make NAO projections.
Andreas
Setting a sea ice constant seems to be a rather bad idea for different reasons, among them ...
http://www.nasa.gov/vision/earth/lookingatearth/quikscat-20071001.html
Here's a another approach to a better understanding of arctic sea ice cover, published by a division of MIT ...
"Discovery of feedback between sea ice and ocean improves Arctic ice extent forecast"
http://web.mit.edu/press/2012/ocean-currents-and-sea-ice.html
http://ecco2.org
V. Lenzer
I am not sure what the right answer would be in this case. Yes, the standard response is 'keep GHG constant and prescribe sea-ice cover (or is it better ice thickness, do the simulation; then prescribe GHG and keep sea-ice extent/thickness constant.
It may work, but let us look again at the long millennial simulation with ECHO-G. This simulation was driven by relatively large variations in solar irradiance and volcanic forcings, in any case smaller than more recent simulation. Can we discern the evolution of global temperatures, say LIA, Late Maunder Minimum, MWP,. in the NAO index ? Actually, not really. We do see a very long-term multi-centennial behaviour of the NAO that roughly corresponds to the sequence MWP-LIA-recent warming. Only in the future, when the model is driven by a very strong GHG forcing, is when the NAO index does something 'unprecedented'. This shows that the level of internal variability is vary large (see the blue line in the millennial simulation) How are we going to detect a signal with a simulation 10 or 20 years long ? would we be able to pick any 20-year period in that simulation and detect that it is different from any other 20 year period.
Semenov and Petoukov conducted six different simulations, each with constant and prescribed lower boundary conditions (sea-ice cover) and and each simulation running over 100 model years.
Thanks Eduardo for revisit this topic and to delve into it a bit deeper. Very much appreciated!
First a few additional thoughts regarding the choice of the October NAO. Autumn sea ice decline may push the October NAO towards the negative state instantanously, but since AO gets pushed negative in lockstep, a more persistent positive feedback is conceivable. It acts to keep the Barents and Kara Sea ice free due to the blocking tendency south of Greenland. As a consequence, the NAO stays negative throughout most of the winter (self-sustaining WACCy-pattern (Warm Arctic Cold Continent) as the continental cold air - once in place - favours restoration of high pressure). Thus the sea ice-NAO correlation is a winter rather than a autumn phenomenon? Not sure.
It is however interesting to note that the October and Winter NAO show little correlation throughout the record which indeed raises the question in how far the October NAO is a meaningful indicator for the sea ice effect. Even some of the models seem to exhibit this behavior as you have nicely pointed out in above. On a more general note, we should keep in mind that the sea ice influence upon the Arctic or NH-circulation is minimal in any case (even if we were to fully understand the involved feedback mechanisms). In my humbled opinion, it is highly unlikely to find a signal in a single month over a mere 7 year period in the NAO index due to sea ice effects. As already said, chances might be higher for the whole winter, despite the fact that other competing players complicate matters a lot (random noise, ENSO, AMOC-strength, solar variability).
Regarding the precipitation trend in the Mediterranean, the small but discernable NAO decline in October seems indeed be linked as you say, but whether this is a "real" model problem or due to well-known simplifications would be hard to tell. For example, without explicit cloud microphysics, you are left with the shifts in the larger scale circulation. Resolve clouds and you may see fewer but heavier precipitation which suddenly matches the observations regardless of the NAO trend in the model. The same is true with the vertical extent of the stratosphere in the model. High top simulations change the sea level pressure over the Mediterranean in the winter considerably (with concomitant precip changes) (e.g. Scaife et al. 2012). We shouldn't blame the models for mismatches at scales they are essentially not designed for. As a modeller, I don't expect the current models to do particularly well on the regional scale (at least not in regions, where convective precipitation plays are vital role). Or to put it differently: I don't have much confidence in regional projections ... if any at all.
And this brings me to your question. In fact, hvw and Andreas have answered the question already. Prescribed sea ice conditions are the way to go. Could become some sort of standard simulation similar to the AMIP runs with prescribed SSTs. We may call it IMIP ;-).
Hi Eduardo,
Just saw your recent comment after writing. The simulation should comprise at least a century. Assume ice-free summer from say 2025 onwards and then slowly extend the duration of the ice free state. I don't know how good the Cryosphere guys are in reproducing the annual cycle, but if they get it right, we may only prescribe the summer low and nudge ice towards this condition. This way, the autumn and winter circulation response in the Arctic is fairly unconstrained. Petoukov and Semenov held 2005 and 2006 SSTs fixed and GHGs were fixed as well. Instead of 100 years which they run their 6 simulations, one would have to run an ensemble of 100 simulations (or more). Whether this is sufficient to overcome the large internal variablity which you mention is difficult to say. The results will tell ...
On top of that, the results would of course be model dependent and may lead to fundamentally different conclusions for different models. In fact, given the spread in the NAO evolution already present (due to a few "outlier" models) one should expect a variety of responses.
Karsten,
it seems to me that we basically agree. Maybe I misunderstood you in the last blogpost 'The perfect spin', but my point there was that it is virtually impossible to attribute any characteristics of Sandy to the diminishing Arctic ice cover, at least if the mechanistic explanation goes through the behaviour of the NAO. If we need simulations over 100 years to identify the signal, the signal itself must be indeed very weak.
Related to this, I have been reading a paper on the summer NAO (Balde et al., The summer North Atlantic Oscillation in CMIP3 models
and related uncertainties in projected summer drying in Europe):
'[39] With regards to the variation of the projected SNAO
trend across models, the existence of a significant upward trend, or lack thereof, does not appear to be related to the skill of the model in reproducing the observed SNAO pattern (cross-reference columns 6 and 8 with 16), so this perfor-
mance cannot be used as a metric to give credence to some projections over others. On the other hand, it is suggestive
that all models with pronounced positive SNAO trends have relatively coarse resolution (2.5 Â 2.75 , or lower, see Table 1), whereas the higher resolution MIROC3.2h and ECHAM4 models (1.1 Â 1.1 ) have non-significant trends.'
It seems that we even do not really understand what is the mechanism by which even a strong GHG forcing may drive the NAO heat flux .
A few months ago I would believe that the predictions for a Mediterranean drying in the future were quite robust, but I am not sure now.
I would suggest that we first agree on a hypothesis, which we want to falsify (or hope to fail falsifying).
Is it:
a) A massive sea ice anomaly of this and that type causes the NAO in the followiong cold season (Oct-February?) to be negative.
or b)
A massive sea ice anomaly of this and that type enhances the probability of the NAO in the following cold season (Oct-February?) to be negative.
In a) and b) the time scale would not play a role - we could do an equilibrium experiment with an
atmospheric GCM alone, with prescribed sea ice and SST conditions.
If however, the time scale plays a role, or a disequilibrium between the oceanic and atmospheric state (we are in a transient phase of the evolving GHG signal), then the hypothesis may be
c) A quickly evolving GHG signal causes global changes as well as sea-ice anomalies, which together push the NAO towards a negative state.
We could assume that the present GCMs (or those of AR4) are not good enough to simulate the sea ice component, so that the de-iceing in the Arctic is too slow. We must then use a coupled GCM, with ocean, atmosphere and sea ice components, impose a fast growth of GHG concentrations PLUS, possibly, a prescribed sea-ice depletion, by nudging the sea ice state to a prescribed configuration (extent, thickness).
Does this makes sense?
I would qualify the null hypothesis. This discussion was prompted by the hypothesis put forward by Scientific American that Hurricane Sandy was modified by anthropogenic climate change. Thus the target season should be October-November, since we do not have hurricanes in January-March.
We would have two potentially factors pulling in opposite directions, but which are not in principle independent:GHG and sea-ice in the Barents-Kara sea. We dont know to what extent the 'direct GHG' forcing is not transmitted via sea-ice, since we dont have simulations with modern climate models with increasing GHG gases but constant sea-ice (maybe some old simulations from the 90's?) , and we dont have results obtained with constant GHG and strongly diminishing sea-ice, since the Petoukhov and Semenov simulations show only results for the winter months. To complicate things further, P&S find a non-linear and seasonally dependent response : the initial reduction of sea-ice has a strong response in December; further reductions have little impact in December but a stronger impact in February:
' As shown in Figure 2, warming and strengthening of U850 take place when SIC reduces from 100% to 80%, particularly pronounced in January and February. Further decrease in SIC from 80% to 60% for February (or from 80% to 40% for January or 60% to 40% for December) is not accompanied
by noticeable changes in SAT and U850. In contrast, sharp cooling and weakening of U850 occur in the model when SIC is reduced to 40% for February (or to 20% for December and January). '
My view would then be that we need two ensembles of coupled transient simulations over 20-years: in one the model runs freely driven by GHG; in the second sea-ice is artificially reduced over the summer months; the null hypothesis would be that the October NAO trends are the same
Edu, ow do you want to do this "in the second sea-ice is artificially reduced over the summer months"? By nudging towards prescribed state of sea ice? Specifying the sea ice may lead to unwanted inconsistencies with the oceanic state. (Nudging may also, but possibly less so.)
Eduardo,
first, I saw my initial sentence terribly mangled due to not checking before sending. It should (obviously) read: "Thanks Eduardo for revisiting this topic and for delving into it a bit deeper."
The "perfect spin" argument was that a negative NAO could be a plausible transition scenario (acting in the opposite direction as the underlying GHG forincg) despite the fact that literally all models don't project anything like that. I'd say that the little dispute was about why the models fail in this regard. We always agreed that it is indeed virtually impossible to attribute any characteristics of Sandy to the diminishing sea ice cover. It's a mere hypothesis, though a plausible one in my opinion. If there is a signal, it is very weak and will likely remain weak.
While the general circulation change with increasing GHG forcing seems fairly well understood and supported by palaeoclimate evidence, the NAO response is less well understood if I interpret the literature correctly. While the Mediterranean drying is a robust signal in most models due to the northward shift of the jet, the inclusion of the entire stratosphere in the models might trigger changes just over Europe. I was quite intrigued by Adam Scaifes talk a few months back in this regard. However, I would be careful with the distinction between drought and precipitation. As already mentioned in my previous comment, more precipitation does not necessarily reduce the risk of drought. In fact, the frequency might already have increased as suggested recently by Hoerling et al. 2012. I still think it is reasonable to assume that this trend continues, especially in the eastern Mediterranean which is less affected from NAO changes anyways. Most would however agree, that regional precipitation projections remain highly uncertain.
Re the length of the simulation period. I would generally favour a longer run (century) over a 20 year period, as it may enable us to find out for how long the potential sea ice effect can overwhelm the GHG driven NAO response (scenario C described by Hans). On the other hand, a 20 year ensemble simulation may provide plenty of information already (which would something similar to scenario B then). If the focus is solely on hurricanes, one would indeed be limited to October and November. Otherwise, winter storms can get quite nasty either. Hence focusing on hurricanes is only part of the story. Several options in any case ...
Hans,
I guess we would have to consider this carefully. If the modelled sea-ice is really far away from observations, maybe nudging would not push the model strongly enough.
Nevertheless, as Karsten said, it seems that such an experiment would be valuable on its own, independently of Sandy.
Btw, I have gotten an email from a colleague from the Uni Hamburg who felt prompted by the Klimazwiebel post to start a short simulation with a simpler model . So it seems that there is some interests in other quarters as well
Hello, devil's advocate speaking.
Apparently the robust knowledge available about determinants of the NAO on all timescales can be appropriately summed up as a big question mark.
Now you can either go the way of the black-box, by prescribing sea-ice and looking at what many years of an ensemble simulation come up with. Or, you actually go deep and try to uncover a piece of the mechanism that drives this oscillation.
In the former case, you better have a large multi-model ensemble, as there seems to be considerable disagreement between simulation codes. Then, why not use what is already there, classify sea-ice state, NAO, SST, and everything else that might play a role in the CMIP3/5 data and do a purely statistical approach? Perhaps because it is likely to get inconclusive results?
The latter case is more ambitious and work has been done in that direction before, I believe. What Hans proposes for case a) and b), simulation of atmosphere only, assumes that NAO doesn't couple back to sea-ice state, on annual and inter-annual time-scale. I believe to remember having read something stating the contrary. In any case you would want to control not only sea-ice and SST, but everything else that has been proposed as having an influence, such as continental snow cover, oceanic convection, thermohaline circulation and whatnot. What I don't understand about "prescribe something"-experiments with AOGCMs is how validity is maintained if the model becomes un-physical. How can you prescribe sea-ice and SST and still have the energy balance close, for example? I guess I am not understanding something fundamental here?
Post a Comment (pop-up window,non-moderated)