In “Identifying signatures of natural climate variability in time series of global-mean surface temperature: Methodology and Insights”, Thompson et al (2009) remove the effects of three natural variables from the Global Surface Temperature record (January 1900 to March 2009). Those three natural variables are El Nino-Southern Oscillation, stratospheric aerosols emitted by explosive volcanic eruptions, and “variations in the advection of marine air masses over the high latitude continents during winter”, which they condense to “dynamically induced variability” or Tdyn in the paper. Thompson et al use “a series of novel methodologies to identify and filter out of the unsmoothed monthly-mean time series of global-mean land and ocean temperatures the variance associated with ENSO, dynamically-induced atmospheric variability, and volcanic eruptions.”
Thompson et al (2009) Link:
Thompson et al (2009) also provided a link to five of the datasets they used and created while preparing the paper. The webpage is identified as “Data for Thompson, Wallace, Jones, Kennedy”:
This post briefly discusses the data made available by Thompson et al (2009), it illustrates the ENSO and volcanic aerosol residuals that remained in the global temperature anomaly data after the effects of ENSO, volcanic aerosols, and dynamically induced variability were said to be removed, and it illustrates the El Nino-induced step changes that resulted from the significant El Nino events that occurred since 1976.
The post does not discuss the erroneous assumption made by Thompson et al (2009), which is that the relationship between ENSO and global temperature is linear. It is not. The non-linear relationship between ENSO and global temperatures was discussed in the following three posts, which all cover the same subject, fundamentally, though there are differences in the presentation:
1. The Relationship Between ENSO And Global Surface Temperature Is Not Linear
2. Multiple Wrongs Don’t Make A Right, Especially When It Comes To Determining The Impacts Of ENSO
3. Regression Analyses Do Not Capture The Multiyear Aftereffects Of Significant El Nino Events.”
The data furnished by Thompson et al actually reinforces the fact that the global temperature response to El Nino events is not linear.
THOMPSON ET AL (2009) DATA
INITIAL NOTE: The title block of the graphs in this post use the nomenclature from the “Data for Thompson, Wallace, Jones, Kennedy” webpage linked above.
Figure 1 illustrates the residual global temperature time-series data after the effects of ENSO, volcanic aerosols, and dynamically induced variability were removed. It is identified throughout this post as “Tdyn/ENSO/Volcano residual global mean”. ENSO continues to make its presence known in the “Tdyn/ENSO/Volcano residual global mean” data in Figure 1, indicating that Thompson et al failed to remove all of the effects. Note the spike from the 1997/98 El Nino and the dip due to the 2007/08 La Nina. Both are reduced in magnitude, but they are still quite visible. There are other El Nino event residuals in the data, as will be illustrated later.
Figure 2 is a comparative graph of the raw “Tdyn/ENSO/Volcano residual global mean” data and Global Surface Temperature anomalies (listed as “Global mean” on the “Data for Thompson, Wallace, Jones, Kennedy” webpage linked above). Thompson et al (2009) identifies the “Global mean” data as HadCRUT3, which is the Hadley Centre’s combined land surface temperature and SST data.
Smoothing both datasets with 13-month filters, Figure 3, helps to highlight the ENSO residuals left in the “Tdyn/ENSO/Volcano residual global mean” data. It also illustrates how well the methods used by Thompson et al (2009) appear to have removed the effects of volcanic aerosols. Note the differences in the datasets immediately following the 1982 and 1991 volcanic eruptions of El Chichon and Mount Pinatubo.
Thompson et al (2009) uses Cold Tongue Index data [5S-5N, 180-90W] as the base for its “ENSO fit” data, Figure 4. Link to “ENSO fit” data:
NOTE: The methods used by Thompson et al (2009) to create the “ENSO fit” (“Volcano fit” and “Dynamic fit”) datasets will not be discussed in this post. Refer to the paper for further information.
Figure 5 is comparative graph of scaled HADSST Cold Tongue Index data (downloaded through the KNMI Climate Explorer) and the “ENSO fit” data. The model used by Thompson et al (2009) exaggerated the Cold Tongue Index data in some months and suppressed it in others.
A time-series graph of the “Volcano fit” data is presented in Figure 6. Link to “Volcano fit” data:
The data source for volcanic aerosols in Thompson et al (2009) is the Sato Stratospheric aerosol optical depth data available though GISS:
Figure 7 compares the “Volcano fit” data and the inverted and scaled Sato Mean Optical Thickness data. There are minor differences in the month-to-month variations between the source data and the model output.
The “Dynamic fit” dataset, Figure 8, is unique to Thompson et al (2009). As noted above, it accounts for the “variations in the advection of marine air masses over the high latitude continents during winter.” They provide a detailed description of the dataset starting on page 9 of the paper. Link to “Dynamic fit” data:
Figure 9 is a comparative graph of the “ENSO fit”, “Volcano fit”, and “Dynamic fit” datasets for those who are interested in seeing their relative magnitudes. The data have been smoothed with 13-month running average filters.
DETRENDED “Tdyn/ENSO/Volcano residual global mean” DATA
To provide an alternate view of the “Tdyn/ENSO/Volcano residual global mean” data, I’ve excluded the first 11 years of data, a short period of declining temperatures, and divided the remainder into three epochs, Figure 10: January 1912 to December 1943 (period of temperature increase), January 1944 to December 1975 (period of flat to declining temperatures), and January 1976 to March 2009 (period of temperature increase). I then detrended the “Tdyn/ENSO/Volcano residual global mean” data during those periods and compared them to the “ENSO fit” and “Volcano fit” datasets, the two major climate variables, to illustrate how much of the ENSO signal remained after the effects of the three variables were removed.
Figure 11 covers the period of Jan 1976 to March 2009. It compares detrended “Tdyn/ENSO/Volcano residual global mean” data to “ENSO fit” and “Volcano fit” data. There appears to be very little of the 1982 volcanic eruption left in the “Tdyn/ENSO/Volcano residual global mean” data, while the some of the 1991 eruption remains.
In Figure 12, I’ve deleted the “Volcano fit” data, leaving a comparison of detrended “Tdyn/ENSO/Volcano residual global mean” and “ENSO fit” data for the period of January 1976 to March 2009. As illustrated there are very large ENSO residuals remaining in the detrended “Tdyn/ENSO/Volcano residual global mean”. Note how the lag varies with each ENSO event. It is apparent that Thompson et al failed to capture and remove a significant portion of ENSO during this period.
Figure 13 illustrates the detrended Thompson et al (2009) “Tdyn/ENSO/Volcano residual global mean”, the “ENSO fit” and the “Volcano fit” data for January 1944 to December 1975. The major drop in detrended “Tdyn/ENSO/Volcano residual global mean” from 1943 to 1947 represents the discontinuity in the HadCRUT3 data that was first discussed in Thompson et al (2008) “A large discontinuity in the mid-twentieth century in observed global-mean surface temperature,” Nature, 453, 646–650, doi:10.1038/nature06982.Link:
The Hadley Centre appears to be using both papers to justify changes they are making to the HADSST dataset. In the concluding remarks of Thompson et al (2009), they write, “THE SST DATA CORRECTED FOR INSTRUMENT CHANGES IN THE MID 20TH CENTURY ARE EXPECTED TO BECOME AVAILABLE IN 2009, and it will be interesting to see how the corrections affect the time history of global-mean temperatures, particularly in the middle part of the century.” [Emphasis added.]
Figure 14 compares detrended “Tdyn/ENSO/Volcano residual global mean” and “ENSO fit” data from January 1944 to December 1975 to illustrate, again, that there are sizable residual ENSO effects in the dataset. Note that during this period there is little to no lag between the “ENSO fit” and the detrended “Tdyn/ENSO/Volcano residual global mean” data, while there were considerable lags between the two datasets from 1976 to present. Is this a function of the magnitude of the ENSO events, where there are longer lags with larger ENSO events?
Figures 15 and 16 are the comparative graphs of detrended “Tdyn/ENSO/Volcano residual global mean” and “ENSO fit” data from January 1912 to December 1943. Figure 15 includes the “Volcano fit” data; Figure 16 does not. Note how there is little agreement between the multiyear variations of the ENSO data and the detrended “Tdyn/ENSO/Volcano residual global mean” data.
The lack of agreement between the two datasets during this period is likely the result of the uncertainties in the datasets, especially the Cold Tongue Index data. Figures 17 and 18 illustrate the number of SST observations for the Cold Tongue region from 1845 to 1991 and from 1900 to 1950. Note how few observations were made in the early part of the Thompson et al (2009) data compared to more recent numbers. Also note that the number of observations between 1900 and 1950 was influenced by the opening of the Panama Canal in 1914 and by the two World Wars. JISAO link:
Would the model used by Thompson et al (2009) have better determined the relationship between ENSO and global temperature during the last two epochs had they excluded the data before 1943? In other words, did the uncertainties in the Global Surface Temperature and Cold Tongue Index data prior to 1943 skew their model so that it failed to identify the true relationship between the datasets in later years. Figures 19, 20, and 21 are comparative graphs of detrended “Tdyn/ENSO/Volcano residual global mean” data and detrended “Global mean” data, which is the unadjusted global surface temperature data, for the three periods. The model used by Thompson et al (2009) appears to have removed little of the effects of ENSO.
A CLOSER LOOK AT THE RESIDUAL DATA FROM 1976 TO PRESENT REVEALS STEP CHANGES DUE TO SIGNIFICANT EL NINO EVENTS
Like many climate bloggers I have removed the linear effects of ENSO and volcanic aerosols from a number of different TLT and surface temperature datasets. While doing so, I’ve noted a curious effect in the data since 1976 but I’ve been hesitant to post the results because of the possibility of claims that I’d somehow manipulated the data to create the effect. Since the data was created by Thompson et al (2009) there should be no way for others to accuse me of misrepresenting the data. They may not agree with my results or how I segmented the data, but that is something else entirely. For them, in the closing, I’ve provided links to my posts that illustrate El Nino-induced step changes in TLT and SST data. Additionally, I would anticipate that someone will note that El Nino (and La Nina) events are not official ENSO events unless NINO3.4 SST anomalies equal or rise above (or fall below) 0.5 deg C. For that someone, global temperatures do not respond only to variations in eastern equatorial Pacific SST anomalies when the ENSO event is official; they respond to the entire ENSO signal.
Figure 22 is a comparison graph of the “ENSO fit” and “Tdyn/ENSO/Volcano residual global mean” data from 1976 to present. Neither dataset has been smoothed. What struck me was, after the initial warming from 1976 to early in 1982, the majority of the rises in the “Tdyn/ENSO/Volcano residual global mean” data occurred during the significant El Nino events of 1982/83, 1986/87/88 and 1997/98. I’ve highlighted the months when the “ENSO fit” data crosses zero for those El Nino events.
If the “Tdyn/ENSO/Volcano residual global mean” data during those El Nino events is eliminated, Figure 23, something else emerges. Note how there appears to be little to no rise in the “Tdyn/ENSO/Volcano residual global mean” data after the 1997/98 El Nino. The data looks flat. Also note how little the “Tdyn/ENSO/Volcano residual global mean” data rose between the 1986/87/88 and 1997/98 El Nino events. There’s the decline in the data between the 1982/83 and 1986/87/88 El Nino events. Last thing to note, the most substantial rise in “Tdyn/ENSO/Volcano residual global mean” data occurred from 1976 to early in 1982.
UPDATE (September 26, 2009): At the suggestion of blogger H.R. on the version of this post at WattsUpWithThat…
A look at the Thompson et al paper – hi tech wiggle matching and removal of natural variables
…I have revised the following papragraph.
We’ll ignore the anomalous period with the negative trend from April 1984 to August 1986 because it is dominated solely by one La Nina event. If we then compare the trends of periods between significant El Nino events, for those periods longer than ~2 years, that is from January 1976 to March 1982 (0.41 deg C/decade), from August 1988 to May 1997 (0.1 deg C/decade), and from November 1998 to March 2009 (0.01 deg C/decade), the trends of the “Tdyn/ENSO/Volcano residual global mean” data have decreased with time. In other words, it has not been accelerating; it has been decelerating. This can be seen quite clearly after the trends of the “non-significant El Nino periods” are added to the illustration. Refer to Figure 24.
In Figure 25, keying off the trend lines, I’ve listed the rises in the “Tdyn/ENSO/Volcano residual global mean” data that occurred during the significant El Nino events of 1982/83, 1986/87/88, and 1997/98. It appears that most of the rise in “Tdyn/ENSO/Volcano residual global mean” data after early 1982 was the direct result of those El Nino events.
Step changes in TLT anomalies and SST anomalies that resulted from significant El Nino events are discussed in detail in:
RSS MSU TLT Time-Latitude Plots…
Can El Nino Events Explain All of the Global Warming Since 1976? – Part 1
Can El Nino Events Explain All of the Global Warming Since 1976? – Part 2