Bob Tisdale suggests that the way some folks have combined the PDO and AMO datasets t produce a new curve is wrong, and here is his supporting analysis. – Anthony
Guest post by Bob Tisdale
Including A Discussion Of Its Use In Wyatt el al (2011)
REFER TO THE UPDATE AT THE END OF THE POST
UPDATE 2 (July 15, 2011): This update clarifies that my comments about Wyatt et al (2011) pertained only to an illustration in the poster and not to the paper itself. The illustration in the poster does not appear in the paper. And the update also provides links to the comments by Marcia Wyatt (the lead author of the paper) on the cross post at WattsUpWithThat. This update begins after Figure 6, under the heading of “WYATT ET AL (2011)”.
Graphs that illustrate the sum of the Atlantic Multidecadal Oscillation (AMO) and Pacific Decadal Oscillation (PDO) data, Figure 1, have appeared in blog posts for more than three years. The first example I can find appeared in a pdf document written by Joe D’Aleo of IceCap: US Temperatures and Climate Factors since 1895. It appeared shortly after in the January 25, 2008 post Warming Trend: PDO And Solar Correlate Better Than CO2 at Watts Up With That? and in numerous posts since then. More recently, the AMO+PDO dataset appeared in the May 22, 2011 post Arctic Cycles – AMO+PDO corresponds to Arctic station group and was referred to in the June 2, 2011 post “Earth itself is telling us there’s nothing to worry about in doubled, or even quadrupled, atmospheric CO2″, which was also posted at Jeff Id’s blog The Air Vent as Future Perfect. A description of the “AMO+PDO” dataset and a link to a spreadsheet can be found at the January 25, 2008 at 9:13 pm comment from the first post at WUWT. A variation on the AMO+PDO graph has also recently found its way into a poster for Wyatt et al (2011) paper Atlantic Multidecadal Oscillation and Northern Hemisphere’s climate variability.
The AMO+PDO curve has been compared to a number of surface temperature variables. Unfortunately, the AMO and PDO datasets cannot be summed.
THE AMO IS DETRENDED SST ANOMALY DATA, BUT THE PDO IS NOT
The AMO data through the NOAA Earth System Research Laboratory (ESRL) AMO webpageis detrended North Atlantic Sea Surface Temperature (SST) Anomalies. (For those who would like an explanation of detrending, refer to the discussion of Figure 6 below.) The PDO, on the other hand, is the product of a principal component analysis of detrended North Pacific SST anomalies, north of 20N. Basically, the PDO represents the spatial patterns of the North Pacific SST anomalies that are similar to those created by El Niño and La Niña events. Since the responses of the North Pacific SST anomalies to El Niño and La Niña events are also impacted by Sea Level Pressure, the PDO and El Niño-Southern Oscillation (ENSO) proxies like NINO3.4 SST anomalies can differ at times.
If one were to detrend the SST anomalies of the North Pacific, north of 20N (the same method used to create the AMO data), standardize it, and compare it to the PDO, the two curves (smoothed with an 11-year filter) appear to be inversely related, Figure 2.
In fact, if we invert the PDO data, multiply it by -1, Figure 3, we can see that they are inversely related and that the detrended North Pacific SST anomalies lead the inverted PDO data for much of the time. That inverse relationship indicates that, over decadal time periods, when the PDO is rising, the detrended SST anomalies are falling and vice versa.
In short, the PDO is an abstract form of North Pacific Sea Surface Temperature data that does not represent the Sea Surface Temperature of the North Pacific. For that reason, it cannot be used to determine the impact of the North Pacific SST on Global Temperatures.
THE “AMO+PDO” DATA AND ITS COMPONENTS
There’s another curious thing about the “AMO+PDO” dataset that can be seen if we plot it along with the AMO and the PDO data used to create it. Refer to Figure 4. Notice how the AMO minimum in the early 1900s is much lower than the minimum in the 1970s. It should not be if the North Atlantic SST anomalies have been detrended.
Figure 5 shows the current AMO data from the NOAA/ESRL website smoothed with an 11-year filter. The early 20thCentury minimum should be comparable to the 1970s minimum.
THE AMO DATA USED IN THE “AMO+PDO” SPREADSHEET IS ACTUALLY NORTH ATLANTIC SST DATA
As noted earlier, the NOAA Earth System Research Laboratory (ESRL) calculates the Atlantic Multidecadal Oscillation (AMO) data by detrending North Atlantic SST anomalies. They use the coordinates of 0-70N, 80W-0 for the North Atlantic. The current version of the ESRL AMO data can be found at ESRL : PSD : Download Climate Timeseries: AMO SST. It’s identified on the webpage as the “AMO (Atlantic Multidecadal Oscillation) Index”.
The AMO data used in the “AMO+PDO” spreadsheet and graphs has not been detrended. In other words, it’s “raw” North Atlantic SST anomaly data. The NOAA ESRL/PSD appears to have changed how they present AMO data sometime between 2007/08 and now. They note on the Atlantic Multi-decadal Oscillation portion of their Climate Indices webpage, “this index is newly computed from a new dataset. Please use it and note that it supersedes the old indices. The data is calculated from the Kalplan SST. See the AMO webpagefor more details.” Or the ESRL had two AMO datasets available online back in 2007/08.
The older version of the ESRL/PSD AMO data used in the AMO+PDO dataset (through 2006) is still available online: AMO(unsmoothed): Standard PSD Format. It’s linked to the AMO – NOAA Earth System Research Laboratorywebpage. They list the source as “Calculated from the HadISST1.” That’s wrong. The linked data is based on Kaplan SST data, not HADISST. They also note that the data is “Area averaged SST in the Atlantic north of 0”. There’s no mention of detrending.
A NOTE ABOUT DETRENDING
For those who are unsure what I’ve meant by detrending, refer to Figure 6. It’s a graph borrowed from my post An Introduction To ENSO, AMO, and PDO — Part 2.The trend of the SST anomalies is determined, and the trend values are subtracted from the SST anomaly data, “flattening” the trend.
WYATT ET AL (2011)
UPDATE 2 (July 15, 2011): The following discusses an illustration from the poster for Wyatt et al (2011). In an earlier update that I placed at the end of this post, I had noted that the illustration from the poster does not appear in the paper, but since that update is at the end of the post, many readers may have missed the clarification. With this update, I wanted to reinforce that my comments are not about the Wyatt et al (2011) paper; they are about an illustration from the poster that did not appear in the paper.
I also want to call your attention to the comments by Marcia Wyatt (lead author of Wyatt et al) on the WattsUpWithThat cross post of On The AMO+PDO Dataset. Marcia’s first comment appears at June 8, 2011 at 8:52 pm . And for those interested, I replied to Marcia here: June 8, 2011 at 10:26 pm. Refer also to her additional comments at June 11, 2011 at 9:30 am , and recently on the thread starting at July 6, 2011 at 7:47 am .
Back to the original post.
I was asked to comment on the Wyatt et al (2011) AMO+PDO graph included in their poster, Figure 7. Unfortunately, there’s very little discussion of the graph on the poster and there’s a paywall on the paper, so I have no means of verifying the sources of the data. But…
The same basic problems (the PDO does not represent the SST anomalies of the North Pacific and the PDO is inversely related to the detrended SST anomalies of the North Pacific) apply to the Wyatt et al (2011) AMO+PDO graph.
In addition to that, the note in the poster that “the NH [Northern Hemisphere] surface temperature time series can be nearly perfectly represented as the weighted sum of the AMO and PDO reconstructions” raises a red flag for me. Nearly perfectly? There are significant differences between SST datasets. The Sea Surface Temperature dataset that’s part of the combined surface temperature dataset must be used if a nearly perfect fit is to have any meaning. That is, for example, when referring to the Hadley Centre’s HADCRUT Land+Sea Surface Temperature data, AMO and PDO data based on HADSST2 should be used. Since there’s the paywall on the paper, I can’t confirm if Wyatt et al used the related SST data to create their AMO and PDO data, so the following portion of this discussion (Figures 8 and 9) is for example only.
For AMO data, Wyatt et al refer to Enfield (2001), which used detrended Kaplan SST anomaly data for the North Atlantic. But there are no surface temperature datasets that use Kaplan SST. (This was one of the problems that Tamino had encountered in his AMO post.) ERSST.v3b data is used by NCDC. GISS uses HADISST and Reynolds OI.v2 SST data for their combined surface temperature products. And HADSST2 is used in the Hadley Centre’s HADCRUT. And the differences between Kaplan and the three SST datasets used in Surface Temperature data can be significant, as shown in Figure 8. (For those wondering why the AMO minimums in the 1970s is lower than the early 20thCentury minimums in this graph, I’ve detrended the data starting in 1900.)
If the PDO could be combined with the AMO data (it can’t), using the correct PDO dataset would also be important. Unfortunately, the PDO data was not referenced in the Wyatt et al poster or their guest post at Roger Pielke Sr’s blog. The most-often-used PDO dataset referred to and used in climate studies is the one available through the JISAO website. In fact, Marcia Wyatt’s co-authors Tsonis and Kravtsov referred to the JISAO PDO data in their 2007 paper A new dynamical mechanism for major climate shifts. But, the JISAO PDO data is based on two obsolete SST datasets (UKMO and Reynolds OI.v1) from 1900 to 2001. None of the current surface temperature datasets use UKMO SST or the obsolete Reynolds OI.v1 SST data. And there are again significant differences between the JISAO PDO data and the 1stPrincipal Components of the detrended North Pacific SST data used in the Surface Temperature products, Figure 9.
The other curiosity about the Wyatt et al “AMO+PDO” graph is the weighting: the AMO is multiplied by 0.83 and the PDO by 0.44. The surface area of the North Pacific (20N-65N, 100E-100W) used for the PDO is slightly larger than the North Atlantic (0-70N, 80W-0) surface area used in the AMO. Based on the surface areas, one would expect a weighting of 53% Pacific and 47% Atlantic, or some similarly weighted factors. Again, since the paper is paywalled, I have no idea how Wyatt et al explain the graph, its components or the weighting.
Wyatt et al could replace the PDO data in their AMO+PDO graph with an SST-based ENSO index like NINO3.4 or Cold Tongue Index (CTI) SST anomalies and wind up with similar results. An AMO+ENSO Proxy curve may not fit nearly perfectly with the Northern Hemisphere Temperature anomalies, but that combination should better represent the two indices that impact Northern Hemisphere surface temperature anomalies.
The Pacific Decadal Oscillation is a well-established climate index that’s used for many variables other than surface temperature. Unfortunately, many people mistakenly believe it is calculated the same as (and is therefore comparable to) the Atlantic Multidecadal Oscillation. Do we need a new index to represent the multidecadal variability of North Pacific Sea Surface Temperatures? The amplitude of the multidecadal variations in detrended North Pacific SST anomalies is less than the variations in the North Atlantic, Figure 10. And the frequencies are somewhat different, meaning the two datasets can run in and out of synch.
UPDATE (June 8, 2011)
Here’s a curiosity: I just checked all of the illustrations for Wyatt et al (2011). Figure 2 from the poster, which is Figure 7 in this post, the graph that includes the weighted AMO+PDO dataset, does not appear in the paper. I double checked by having Adobe Acrobat do a word search and the phrases “weighted sum” and “AMO and PDO reconstructions” do not appear in the paper.
And for those interested, Wyatt et al used the ESRL AMO data, the JISAO PDO data, and HADCRUT NH surface temperature data.
The current NOAA/ESRL AMO data is available here:
The PDO data is available through the JISAO PDO website:
All other data is available through the KNMI Climate Explorer: