We’ve illustrated and discussed in a number of recent posts how poorly the hindcasts and projections of the coupled climate models used in the Intergovernmental Panel on Climate Change’s 4th Assessment Report (IPCC AR4) compared to instrument-based observations. And this post is yet another way to illustrate that fact. We’ll plot the 17-year and 30-year trends in global and hemispheric Sea Surface Temperature anomalies from January 1900 to August 2011 (the updates of HADISST data used in this post by the Hadley Centre can lag by a few months) and compare them to the model mean of the Hindcasts and Projections of the coupled climate models used in the IPCC AR4. As one would expect, the model mean show little to no multidecadal variability, which is commonly known. Refer to the June 4, 2007 post at Nature’s Climate Feedback: Predictions of climate, written by Kevin Trenberth. But there is evidence that the recent flattening of Global Sea Surface Temperature anomalies and the resulting divergence of them from model projections is a result of multidecadal variations in Sea Surface Temperatures.
WHY 17-YEAR AND 30-YEAR TRENDS?
A recent paper by Santer et al (2011) Separating Signal and Noise in Atmospheric Temperature Change: The Importance of Timescale, state at the conclusion of their abstract that, “Our results show that temperature records of at least 17 years in length are required for identifying human effects on global-mean tropospheric temperature.” Sea surface temperature data is not as noisy as Lower Troposphere temperature anomalies, so we’ll assume that 17 years would be appropriate timescale to present sea surface temperature trends on global and hemispheric bases as well. And 30 years: Wikipedia defines Climate “as the weather averaged over a long period. The standard averaging period is 30 years, but other periods may be used depending on the purpose.”
But we’re using monthly data so the trends are actually for 204- and 360-month periods.
ABOUT THE GRAPHS IN THIS POST
This post does NOT present graphs of sea surface temperature anomalies, with the exception of Figures 2 and 3, which are provided as references. The graphs in this post present 17-year and 30-year linear trends of Sea Surface Temperature anomalies in Deg C per Decade on a monthly basis, and they cover the period of January 1900 to August 2011 for the observation-based Sea Surface data and the period of January 1900 to December 2099 for the model mean hindcasts and projections. Figure 1 is a sample graph of the 360-month (30-year) trends for the observations, and it includes descriptions of a few of the data points. Basically, the first data point represents the linear trend of the Sea Surface Temperature anomalies for the period of January 1900 to December 1929, and the second data point shows the linear trend of the data for the period of February 1900 to January 1930, and so on, until the last data point that covers the most recent 360-month (30-year) period of September 1981 to August 2011.
Note also how the trends vary on a multidecadal basis. The model-mean data do not produce these variations, as you shall see. And you’ll also see why they should, because they are important. Observed trends are dropping, but the model mean trends are not.
I’ve provided the following two comparisons of the “raw” Sea Surface Temperature anomalies and the 360-month (Figure 2) and 204-month (Figure 3) trends as references.
COMPARISONS OF SEA SURFACE TEMPERATURE ANOMALY TRENDS OF CLIMATE MODEL OUTPUTS AND INSTRUMENT-BASED OBSERVATIONS
In each of the following graphs, I’ve included the following notes. The first one reads,
The Models Do Not Produce Multidecadal Variations In Sea Surface Temperature Anomalies Comparable To Those Observed, Because They Are Not Initialized To Do So. This, As It Should Be, Is Also Evident In Trends.
And since those notes in red are the same for Figure 4 through 9, you’ll probably elect to overlook them. The other note on each of the graphs describes the difference between the observed trends for the most recent period and the trends hindcast and projected by the models. And they are significant, so don’t overlook those notes.
There’s no reason for me to repeat what’s discussed in the notes on the graphs, so I’ll present the comparisons of the 360-month and 204-month trends first for Global Sea Surface Temperature anomalies, then for the Northern Hemisphere data, and finally for the Southern Hemisphere Sea Surface Temperature anomaly data. Some of you may find the results surprising.
GLOBAL SEA SURFACE TEMPERATURE COMPARISONS
NORTHERN HEMISPHERE SEA SURFACE TEMPERATURE COMPARISONS
SOUTHERN HEMISPHERE SEA SURFACE TEMPERATURE COMPARISONS
Table 1 shows the observed Global and Hemispheric Sea Surface Temperature anomaly trends, 204-Month (17-Year) and 360-Month (30-Year), for period ending August 2011. Also illustrated are the trends for the Sea Surface Temperature anomalies as hindcast and projected by the model mean of the coupled climate models employed in the IPCC AR4.
Comparing the 204-month and 360-month hindcast and projected Sea Surface Temperature anomaly trends of the coupled climate models used in the IPCC AR4 to the trends of the observed Sea Surface Temperature anomalies is yet another way to show the models have
no shown no skill at replicating and projecting past and present variations in Sea Surface Temperature on multidecadal bases. Why should we believe they have any value as a means of projecting future climate?
Both the HADISST Sea Surface Temperature data and the IPCC AR4 Hindcast/Projection (TOS) data used in this post are available through the KNMI Climate Explorer. The HADISST data is found at the Monthly observations webpage, and the model data is found at the Monthly CMIP3+ scenario runswebpage.
Pingback: An Excellent Multi-Decadal Global Climate Model Hindcast Evaluation By Bob Tisdale | Climate Science: Roger Pielke Sr.
Wow. Really terrific analysis Bob.
Not just in showing that the hindcasts are so far off in the past (when the data was already known) but, more importantly, in demonstrating just how far off the current set of projections is versus the current sea surface temperatures. In addition, I don’t think anyone has provided a good description of the future projections either; so good on you.
Global SSTs over the past 17 years, 0.02C per decade actual versus 0.15C per decade predicted. I think that definitely meets the statistically significant level.
Did you see this?
Thanks, Ibrahim. I had not yet found it.
(Correction. I thought that was a link to a new post. I had read that post of Dr. Spencer’s. Thanks.)
On this sort of thing, I agree with Carrick. A conversion of the data to the frequency domain would tell you a lot more without effectively throwing out a lot of data with what amounts to a low pass filter.
dwrice: If you’re looking for the comment you wrote about this post, you left it on the wrong thread. It’s here, along with my reply:
My reply reads:
dwrice says: “Hi Bob. Just wanted to comment on your Table 1.”
It appears you placed this comment on the wrong thread, dwrice. There is no Table 1 in this post. I assume you’re referring to my post here:
You continued, “I can’t find the link to the HADISST data you cited.”
It’s listed under the heading of SOURCES above, but to save you the time it will take to scroll up there, here’s another link. Just scroll down to SST. HADISST is the dataset at the top of that group:
And you wrote, “I have cross-referenced the global figures with HadSST2 and I find that your published data are very far from what HadSST2 suggests.”
And there are very good reasons for that. HADSST2 is a spatially incomplete dataset for starts. Refer to:
There is a significant amount of data missing in the Southern Hemisphere in HADSST2, and if you’re not aware, the high latitudes of the Southern Hemisphere of the spatially complete, satellite-based SST datasets like HADISST and Reynolds OI.v2 SST data show a significant cooling at those latitudes. An example:
That graph is included in my monthly updates:
Second, HADSST2 has another bias. The Hadley Centre spliced two “incompatible” (for lack of a better word) source datasets together in 1998 and it created an upward shift in the HADSST2 data that does not exist in any of the other SST datasets. The upward shift in the HADSST2 data after 1998 is approximately 0.065 deg C compared to HADISST. That’s a lot of upward bias.
The graph is from this post:
I also discussed and illustrated it with other SST datasets in this post:
That upward shift in the HADSST2 data due to the splicing and the HADSST2 data being spatially incomplete likely account for the differences you found.
You asked, “Maybe HadSST2 is incompatible with HADISST? If so, how?”, and, “Would you care to comment on this?”
Explained and illustrated above.
You concluded, “Some people are watching ”
I’m glad, but if you’re going to double check my work, please use the dataset I used. I almost always include links to my sources under the heading of SOURCES. That way you’re not wasting your time.
Regards – Bob
DeWitt Payne: Do you have a link to where Carrick commented on this post? Also, I attempt to write posts for those who don’t have technical backgrounds, so I prepare them so that anyone can easily duplicate the graphs if they made the effort. Also, most people understand trends, and trends were referred to in Santer el al (2011), which was another reason for preparing this post.
Thanks for that. I left a comment on the other thread which I’ll repeat here.
I still can’t find the specific data set you are referring to. All I get is a link to HADISST SST with questions re time series and Lat/Long, etc. (I have also tried the UK Met Office site but that data is behind a registration wall and listed in integers.)
Is there a direct link to the data you used?
I am surprised that there is such a massive difference between HADISST and HadSST2 based on an upward shift of 0.065 deg C since 1998.
(Met Office also claim that all biases due to lack of sample data are calculated for in HADSST2.)
dwrice says: “I still can’t find the specific data set you are referring to. All I get is a link to HADISST SST with questions re time series and Lat/Long, etc. ” And, “Is there a direct link to the data you used?”
That’s it. Input -90 and 90 for the latitudes and -180 and 180 for the longitudes and click on “Make Time Series.” The next page has three graphs. The top is the raw SST data, the second is the climatology used for anomalies (based on the default base years that you can chnage), and the third are SST anomalies. Above the SST anomaly graph, click on “raw data.” That’s the data I used.
KNMI downloads the data from the respective sources on a monthly basis, around the 25th of each month to keep their site up-to-date.
dwrice or others wanting to duplicate my results, refer also to the following introduction to the KNMI Climate Explorer:
Many thanks. Using this data I see that you are right. Table 1 above is correct as far as HADISST data is concerned. I will make this clear on the blog that sent me here!
I am at a loss to understand why there is such a difference between HadSST2 and HADISST. Both sets appear to be endorsed by the Met Office – so which is the ‘official’ one?
Can you confirm that the IPCC models are projected against HADISST or HADSST2, or neither?
Pingback: Tamino Disparages Bob Tisdale’s Analysis Of Model Predicted Trends In SSTs | Climate Science: Roger Pielke Sr.
Pingback: La temperatura del mar y los modelos climáticos. Muy mala pinta. « PlazaMoyua.com
Bob, you do great work. Thanks.
Typo: You have an extra “no” in this sentence of your closing paragraph.
“Comparing the 204-month and 360-month hindcast and projected Sea Surface Temperature anomaly trends of the coupled climate models used in the IPCC AR4 to the trends of the observed Sea Surface Temperature anomalies is yet another way to show the models have no shown no skill at replicating and projecting past and present variations in Sea Surface Temperature on multidecadal bases.”
Pingback: Tamino Misses The Point And Attempts To Distract His Readers | Bob Tisdale – Climate Observations
Pingback: Shades of Foster Grant | Watts Up With That?
Thank you for excellent work Bob. I have been asking for raw data/graphs for years. This shows the “mis-handling” of the data by the IPCC–again! The Hadley Centre’s “modifications” that miraculously coincide with the IPCC and dramatically differ from their own data is especially troubling. It seems to me that whenever the raw data is produced, it seriously “differs” from those that the IPCC apparently uses after they have been tweaked.
The most galling thing, is that the popular media gobble it up, and when the real data is presented their eyes glaze over.
Please keep up the great work.
Crashex says: “Typo: You have an extra ‘no’ in this sentence of your closing paragraph.”
Thanks. I fixed it with a strike through on the first no.
dwrice says: “Can you confirm that the IPCC models are projected against HADISST or HADSST2, or neither?”
Neither. They are run independently of observational data.
Pingback: More ‘science’ fiction… | pindanpost
Pingback: Satellite-Era SST Anomalies: Models Vs Observations Using Time-Series Graphs And 17-Year Trends | Bob Tisdale – Climate Observations
Doesn’t this puts the lie to Trenberth’s position that the models have not been initialized?
[IPCC AR5 models]
So using the 20th c for tuning is just doing what some people have long
suspected us of doing […] and what the nonpublished diagram from NCAR showing
correlation between aerosol forcing and sensitivity also suggested.
Pingback: On the SkepticalScience Post “Pielke Sr. Misinforms High School Students” | Bob Tisdale – Climate Observations
Pingback: Tisdale schools the website “Skeptical Science” on CO2 obsession | Watts Up With That?
Pingback: Do Climate Models Confirm Or Contradict The Hypothesis of Anthropogenic Global Warming? – Part 1 | Bob Tisdale – Climate Observations
Pingback: Peer reviewed paper shows no warming for at least 20 years. - Page 3 - US Message Board - Political Discussion Forum
Pingback: It Really Should Go Without Saying, BUT… | Bob Tisdale – Climate Observations
This is an excellent approach to plotting trends and clearly indicates the 60 year cycle. We easily see that the mean rate of increase appears to be a little less than 0.06 deg.C/decade. Secondly, the slight decrease in mean rates of increase is consistent with a 900 to 1100 year approximately sinusoidal cycle which has passed a flex and is approaching a maximum perhaps within the next 100 to 200 years. So, yes, the long-term (1000 year) trend is still increasing, though expected to decline within 200 years. And the 60 year superimposed cycle indicates reduced rates of increase for another 20 years or so. There may actually be slight cooling in that period due to the lag in the 360-month trends.
Bob , there’s definitely some mileage in what you are trying to show but you must get beyond “trends”. Sliding trends like this are even worse and can give very different results depending upon the time interval you chose. They are not robust results.
If you want to examine rate of change (and it is a very good idea) do so directly. The difference of each successive month’s, year’s temp (whatever your data is) divided by the interval. You then have dT/dt.
Now smooth with a decent filter like binomial or gaussian that will actually get rid of the higher frequencies and you will have something much more rigorous to blast the models with.
Bit late to the party here, just found a link somewhere that sent me here.
PS if you use a proper filter you will find that the graphs don’t keep shape shifting each time you change the length of the filter. They just get “smoother”, “rounder” as one would expect.
P. Solar says: “Bob , there’s definitely some mileage in what you are trying to show but you must get beyond “trends”. Sliding trends like this are even worse and can give very different results depending upon the time interval you chose.”
I explained in the post why I chose the 17- and 30-year periods. I presented sliding trends because it was used by Hansen et al (2011). It was simply a different way to present the failings of the models.
I saw a presentation that stated the reason for the jump in data in 1998 was that the data before that was estimated based on things like tree rings whereas the figures from that date onwards were “real” measurements. If the same estimating process was continued beyond 1998 you get a consistent data set and the temperature jump disappears.
Does this sound plausible?
Taffekles, the answer to your question is no. The data presented in the post are based on sea surface temperature observations, not tree rings. Your source was wrong.