A new paper by Deser et al. (2012) (free access) is likely to have repercussions on discussions of local climate change adaptation. I think it caught some people by surprise, even if the results perhaps should not be so surprising. The range of possible local and regional climate outcomes may turn out to be larger than expected for regions such as North America and Europe.
Deser et al. imply that information about the future regional climate is more blurred than previously anticipated because of large-scale atmospheric flow responsible for variations in regional climates. They found that regional temperatures and precipitation for the next 50 years may be less predictable due to the chaotic nature of the large-scale atmospheric flow. This has implications for climate change downscaling and climate change adaptation, and suggests a need to anticipate a wider range of situations in climate risk analyses.
Although it has long been recognised that large-scale circulation regimes affect seasonal, inter-annual climate, and decadal variations, the expectations have been that anthropogenic climate changes will dominate on time scales longer than 50 years. For instance, an influential analysis by Hawking & Sutton (2009) (link to figures) has suggested that internal climate variability account for only about 20% of the variance over the British isles on a 50-year time scale.
I believe Deser et al.‘s results are important and a wake-up-call, because climate change projections used for the studies of climate change impacts have usually been based on a limited amount of regional climate model (RCM) and global climate model (GCM) simulations. Thus, they may not have sufficiently acknowledged the wide range due to internal variability.
Past research projects such as ENSEMBLES (25 runs with a combinations of RCMs/GCMs, thereof 13 independent GCMs simulations – link & final report), NARCCAP (5 independent simulations with GCMs – link), the UKCIP (11-member ensemble of RCMs), and PRUDENCE (included four different simulations with GCMs – assumed to reflect internal variability of climate (Beniston et al., 2007)) may not have — if Deser et al. are correct — accounted for the wide range of outcomes associated with different large-scale atmospheric flow.
The reason is that – separately – they imply a statistical sample (determined by the number of independent GCM simulations) of different large-scale atmospheric flows that may be too small to represent the actual range of possibilities.
Deser et al.‘s analysis was based on simulations carried out with one model (CCSM3), where they carried out nearly identical simulations 40 times which only differed by using slightly different starting points. Simulations carried out with one model, where the set-up is slightly different, is known in the climate science world as an single-model ensemble simulation.
The purpose of such ensemble runs is to explore the sensitivity of the predictions to different descriptions of the atmosphere at the start of the simulations. For instance the location and strengths of low- and high-pressure systems affect the flow of heat and moisture and the subsequent climatic evolution.
The importance of Deser et al.‘s findings can be seen in their figures (one reproduced below), where they highlighted the model simulations giving the smallest and greatest local response.
Figure 1 from Deser et al. (2012): The maps show the average response from the entire ensemble as well as the simulations which gave the warmest and coldest future. The middle panel shows the time evolution of the temperature at a selection of locations and averaged over areas (red curves mark the warmest simulations and blue the coldest). The right panel shows how many simulations in the ensemble that gave temperature trends of different values.
I think there were some surprising aspects in Deser et al.‘s results. Not that I didn’t expect natural multi-annual variations to be unimportant (on shorter time scales, they are very pronounced), but what strikes me is the strong contrast (on a 50-year time scale) between the global mean temperature (lower graph), which was not very sensitive to the regional atmospheric circulation, and the regional temperatures which were strongly influenced.
It has long been recognized that local and regional climate would warm at different rates than the global mean, but not with such large differences as presented by Deser et al. at the time scales of 50 years and for continental scales. Their results imply that while some regions could experience almost zero warming over 50 years, this will be compensated by substantially stronger in other regions (because they also find that the global mean temperatures to be largely insensitive to the different model initial conditions).
These results also imply a surprisingly long persistence of weather regimes in different parts of the world. Usually, one tends to associate these with inter-annual to decadal scales. However, Deser et al observe:
Such intrinsic climate fluctuations occur not only on interannual-to-decadal timescales but also over periods as long as 50 years… even trends over 50 years are subject to considerable uncertainty owing to natural variability.
These findings were in particular important for the winter season at mid-to-high latitudes. Hence, they could be entirely attributed to chaotic dynamics. On the other hand, the two simulations that they highlighted in their study represented extreme cases, and most of the simulations suggested that the future outcome may be somewhere in between.
My interpretation of Deser et al.‘s results is that the range of possible future temperatures gets broader at the same time as the most likely outcome follows a warming curve. This means that the most likely scenario is warming for the future while there still is a small possibility that the temperature for a particular location hardly changes (or even cools) over a 50-year period.
If each simulation is equally likely and the distribution of results given by the ensemble of runs gives an indication of likelihood, then the most likely outcome is described by the value that most of the ensemble members cluster around.
Another way to look at this is that the signal-to-noise ratio doesn’t increase much over time, which makes some of the take-home messages from Brown & Wilby (2012) important to heed. They argue that downscaling for the future involves too many unknowns, but can nevertheless provide tentative input to a more comprehensive risk assessment based on many factors, all of which may not necessarily be related to climate.
The Deser et al. paper also sends a message to people who study regional climate change based on a small selection of climate models (i.e. only from their own country). There is a need to include the maximum amount of reliable information about the local climate, and that would include the range of possibilities and the probabilities associated with these.
It is of course possible that the climate model used by Deser et al. exaggerates the variability in the large-scale atmospheric flow. However, their findings may also potentially explain the observation that led Oldenborgh et al. (2009) to conclude that Europe has warmed faster than projected by the climate models.
p.s. The ENSEMBLES project did involve a “grand ensemble” (climateprediction.net); however, this has so far not been used extensively for describing local and regional climate. The grand ensemble has been more appropriate for describing the global mean state and large-scale phenomena (link to experiments).
References
- C. Deser, R. Knutti, S. Solomon, and A.S. Phillips, "Communication of the role of natural variability in future North American climate", Nature Climate Change, vol. 2, pp. 775-779, 2012. http://dx.doi.org/10.1038/nclimate1562
- E. Hawkins, and R. Sutton, "The Potential to Narrow Uncertainty in Regional Climate Predictions", Bulletin of the American Meteorological Society, vol. 90, pp. 1095-1108, 2009. http://dx.doi.org/10.1175/2009BAMS2607.1
- M. Beniston, D.B. Stephenson, O.B. Christensen, C.A.T. Ferro, C. Frei, S. Goyette, K. Halsnaes, T. Holt, K. Jylhä, B. Koffi, J. Palutikof, R. Schöll, T. Semmler, and K. Woth, "Future extreme events in European climate: an exploration of regional climate model projections", Climatic Change, vol. 81, pp. 71-95, 2007. http://dx.doi.org/10.1007/s10584-006-9226-z
- C. Brown, and R.L. Wilby, "An alternate approach to assessing climate risks", Eos, Transactions American Geophysical Union, vol. 93, pp. 401-402, 2012. http://dx.doi.org/10.1029/2012EO410001
- G.J. van Oldenborgh, S. Drijfhout, A. van Ulden, R. Haarsma, A. Sterl, C. Severijns, W. Hazeleger, and H. Dijkstra, "Western Europe is warming much faster than expected", Climate of the Past, vol. 5, pp. 1-12, 2009. http://dx.doi.org/10.5194/cp-5-1-2009
Hi Tamino and Hank,
Thank you for your replies. Tamino, you are correct that kriging can provide a better method to weight temperatures spatially provided there are sufficient observations with which to estimate the variogram particularly if there are discontinuities in the spatial variance-covariance structure. I do not know how many data points Berkeley had available for Alaska but it has to be many to interpolate temperature for such a large state or it is unlikely to offer much improvement. The interesting data on the UAF GI site are the temperature records for each of the stations. They show that warming trends vary considerably throughout the state with the greatest change happening on the North Slope (not surprising). Some correlate well with the putative “phase shift” of the PDO in 1976 and again in about 2005. The PDO cannot be a cause of long-term temperature change. It only (as I understand the index) indicates changes in how heat is distributed and as such contributes to local variation, which is what Deser et al. addressed. Currently, we seem to be in a negative PDO phase, which likely is contributing to a string of hard winters in SE Alaska despite a long-term warming trend. An interesting side note is that Eaglecrest ski area in Juneau was in serious financial trouble from lack of snow during the early part of last decade. Now they are doing well.
Hank, I agree with you on Wendler’s statement about expecting a linear relation between temperature and CO2. To me, his analysis is naive at best and using r-squared to proportion relative effects is particularly egregious. At the very least, there are much better statistical methods for model selection such as Akaike information criteria (AIC). With respect to your question about who is arguing that warming reduces the need to protect winter range for deer, the answer is many of the U. S. Forest Service planners and foresters working on proposed timber sales within the Tongass National Forest. They are part of an agency that is in serious decline, budget-wise, mission-wise, and staffing-wise. Since WW2, they have been wedded to harvesting timber as their main mission and important funding support. That mission largely is over and the Tongass is one of the last places with timber resources available for harvest at an industrial scale despite the fact that it is heavily subsidized by the taxpayer. Hence, they are doing anything they can to keep that dead horse going.
dave
Thanks for you input Dave. I know many here do not like to hear about the influence of the PDO, but I do, and appreciate your insight. One question, based on the negative turn of the PDO, would you expect Alaskan temperatures to remain at their current levels or decrease (even slightly) in the coming decades?
> expecting a linear relation between temperature and CO2
Interviewed by Ken Rudin on National Public Radio this morning, Ralph Reed gave that as his reason believing there’s not a connection.
http://pubs.acs.org/doi/abs/10.1021/es2044008
Minding Our Methods: How Choice of Time Series, Reference Dates, and Statistical Approach Can Influence the Representation of Temperature Change
“What may seem to be simple or arbitrary choices in these matters could potentially infuse significant bias into the interpretation of data, thereby distorting the representation of climate variability in Alaska and handicapping potential strategies for response. In this paper we demonstrate and emphasize how the use of different time scales, reference dates, and statistical approaches can generate highly disparate results, suggesting that careful use of these tools is critical for correctly interpreting and reporting climatic trends in Alaska and other polar regions.”
and this illustrates the Minding Our Methods abstract:
http://pubs.acs.org/appl/literatum/publisher/achs/journals/content/esthag/2012/esthag.2012.46.issue-14/es2044008/production/images/medium/es-2011-044008_0004.gif
Hi Dan H,
I expect Alaska and my part of SE Alaska to generally warm over the next few decades. I expect a negative PDO to mute that warming somewhat and a positive PDO to enhance it. However, my main concern in the short-term is if ocean processes like the PDO and ENSO interact with general warming to increase the risks of extreme cold and snowy periods and extreme warm periods which will play havoc with ecological communities.
dave
More Intense North Atlantic Tropical Storms Likely in the Future
http://www.sciencedaily.com/releases/2012/11/121130151651.htm
is, I suppose, a ‘stronger regional difference’ but I would have posted this on an Unforced Variations thread if one had been open.
Global Warming is indeed a phenomenon we can’t runaway from. I found motleygreen.com and was able to see that there are acts I can do to save the environment.
> motleygreen
I wonder who owns that site, widely spammed on blogs lately. It offers a 3-minute cancer cure: … motleygreen.com/beta/article-details-cure-cancer-using-your-own-technology-of-emotions-74
Beware or be amused: claims a thinkwrap license; looking at their web page signifies accepting their terms of service; privacy policy is “you consent” to whatever they want. Yeek.