UPDATE: Corrected Figures 2 through 6.
UPDATE 2: Corrected the color-coding in the title block of Figure 4. (Thanks to blogger cassandraclub for noticing it.)
UPDATE 3: I removed the word Anomalies from Figures 2, 3 and 4. And I’ve added two graphs using BOM data for Australia at the end of the post.
In the wake of the heat wave in Australia last summer, I had promised Jo Nova a post about Australia land surface air temperatures. That email exchange took place a couple of months ago. I began work on it a few days ago, the graphs were done, but I hadn’t written the text. Much to my amazement, Anthony Watts published a post about the press release for the Lewis and Karoly (2013) paper Anthropogenic contributions to Australia’s record summer temperatures of 2013 (paywalled). Anthony’s post was titled Claim: Humans play role in Australia’s “angry” hot summer. Lewis and Karoly (2013) were blaming human-induced global warming for the heat wave, but the data I had downloaded indicated Australia summertime temperatures in 2013 weren’t remarkable and the models showed no skill at being able to simulate Australia land surface temperatures.
Please keep in mind that I did not prepare my post about Lewis and Karoly (2013) but the post does shed some light on the paper. Please read Anthony’s post and the abstract of the paper linked above.
For the land surface temperature dataset, I used NOAA’s GHCN-CAMS. It is available for download on a gridded basis through the KNMI Climate Explorer. It has the best spatial coverage of the surface temperature datasets that are regularly updated, because it relies on other surface temperature data in addition to the GHCN data. And it’s also available in absolute form, where other datasets are presented as anomalies. Unfortunately, it has a higher warming trend than the GHCN-only datasets.
I used the coordinates of 45S-10S, 110E-155E for Australia. Figure 1 is a time-series graph of the Australia land surface temperature anomalies from January 1948 (the start of the dataset) to present (May 2013). There is very obvious shift in the data around 1977—possibly a lagged aftereffect of the 1976 Pacific Climate Shift—so I started my comparison in 1979, which is a common start year for surface temperature data presentations.
I had originally looked at the months of January to March, but those commenting on the thread at the WUWT post were also defining the Australian summer as November to January and December to February. So I threw together a couple of additional graphs. One other note: I typically use RCP6.0 for the scenario in my CMIP5-based (IPCC AR5) model-data comparisons, because it’s similar to the A1B scenario, which was the one used most often in CMIP4-based studies. But Lewis and Karoly (2013) went all out and used RCP8.5, so I changed model scenarios for this post. I did not, however, make any other effort to make this post agree with Lewis and Karoly (2013). They picked 9 CMIP5-based models for their study and I used all the 39 models with their 81 ensemble members.
SUMMERTIME MODEL-DATA COMPARISON
The following three graphs compare the 3-month average Australia land surface temperatures (not anomalies), based on the GCHN-CAMS data and the multi-model ensemble mean of the RCP8.5-based models stored in the CMIP5 archive. Figure 2 uses November to January, Figure 3 is for December to February, and Figure 4 includes January to March. As illustrated, no matter which 3-month periods you look at, there wasn’t anything unusual about the land surface temperature for the 2013 season. The other thing that really stands out is the fact that, based on the linear trends, summertime surface temperatures haven’t warmed since 1979. The linear trends are basically flat. On other hand, the models show that summertime land surface temperatures should have warmed at a rate of about 0.22 to 0.236 deg C per decade. Oops, they missed yet again.
MONTHLY MODEL-DATA COMPARISON
That’s not to say that Australia land surface temperatures haven’t warmed since 1979. The monthly data shows that Australia land surface temperatures warmed at a rate of about 0.07 deg C per decade. However, the models show that if greenhouse gases were responsible for the warming, Australia land surface temperature anomalies should have warmed at a rate that’s more than 3 times faster. The modelers still overshot the mark by a sizeable amount.
And as a reference, I’ve replaced the observations-based data with CRUTEM4 in Figure 6, to confirm that the GHCN-CAMS data does show a little extra warming.
Note: The trends in Figures 5 and 6 are based on the monthly data and model outputs, not on the smoothed versions.
STANDARD BLURB ABOUT THE USE OF THE MODEL MEAN
We’ve published numerous posts that include model-data comparisons. If history repeats itself, proponents of manmade global warming will complain in comments that I’ve only presented the model mean in the above graphs and not the full ensemble. In an effort to suppress their need to complain once again, I’ve borrowed parts of the discussion from the post Blog Memo to John Hockenberry Regarding PBS Report “Climate of Doubt”.
The model mean provides the best representation of the manmade greenhouse gas-driven scenario—not the individual model runs, which contain noise created by the models. For this, I’ll provide two references:
The first is a comment made by Gavin Schmidt (climatologist and climate modeler at the NASA Goddard Institute for Space Studies—GISS). He is one of the contributors to the website RealClimate. The following quotes are from the thread of the RealClimate post Decadal predictions. At comment 49, dated 30 Sep 2009 at 6:18 AM, a blogger posed this question:
If a single simulation is not a good predictor of reality how can the average of many simulations, each of which is a poor predictor of reality, be a better predictor, or indeed claim to have any residual of reality?
Gavin Schmidt replied with a general discussion of models:
Any single realisation can be thought of as being made up of two components – a forced signal and a random realisation of the internal variability (‘noise’). By definition the random component will uncorrelated across different realisations and when you average together many examples you get the forced component (i.e. the ensemble mean).
To paraphrase Gavin Schmidt, we’re not interested in the random component (noise) inherent in the individual simulations; we’re interested in the forced component, which represents the modeler’s best guess of the effects of manmade greenhouse gases on the variable being simulated.
The quote by Gavin Schmidt is supported by a similar statement from the National Center for Atmospheric Research (NCAR). I’ve quoted the following in numerous blog posts and in my recently published ebook. Sometime over the past few months, NCAR elected to remove that educational webpage from its website. Luckily the Wayback Machine has a copy. NCAR wrote on that FAQ webpage that had been part of an introductory discussion about climate models (my boldface):
Averaging over a multi-member ensemble of model climate runs gives a measure of the average model response to the forcings imposed on the model. Unless you are interested in a particular ensemble member where the initial conditions make a difference in your work, averaging of several ensemble members will give you best representation of a scenario.
In summary, we are definitely not interested in the models’ internally created noise, and we are not interested in the results of individual responses of ensemble members to initial conditions. So, in the graphs, we exclude the visual noise of the individual ensemble members and present only the model mean, because the model mean is the best representation of how the models are programmed and tuned to respond to manmade greenhouse gases.
We can add Australia land surface temperatures to the list of variables the CMIP5 climate models show no skill at simulating. The others include:
And we recently illustrated and discussed in the post Meehl et al (2013) Are Also Looking for Trenberth’s Missing Heat that the climate models used by Meehl et al (2013) show no evidence that they are capable of simulating how warm water is transported from the tropics to the mid-latitudes at the surface of the Pacific Ocean, so why should we believe they can simulate warm water being transported to depths below 700 meters without warming the waters above 700 meters?
That list is growing quite large.
UPDATE 3: Nick Stokes was correct to point out that I’ve presented a reanalysis with GHCN-CAMS and not data. I can’t complain that Balmaseda et al (2013) are presenting a reanalysis, not data, while looking for Trenberth’s missing heat (see here), and then present a reanalysis without showing the difference between the reanalysis and data.
If we assume that the BOM data is correct, then compared to the BOM land surface air temperature data for Australia, it appears the GHCN-CAMS reanalysis has a cooling bias during summer months. Regardless, as shown in Figures 7 and 8, the RCP8.5-based CMIP5 climate models overestimate the summertime warming rate by 2.4 times and overestimate the monthly warming rate by 3.2 times. Again, the models show no skill at being able to simulate Australia land surface air temperatures.
(The trends in Figure 8 are based on the monthly data and model outputs, not the smoothed values.)