>But Those Differences Are Subtle
I was recently asked to comment on the Multivariate ENSO Index (MEI). (Thanks, d.) This post compares the MEI to HADSST2-based NINO3.4 SST anomalies. It also removes the linear effects of ENSO from the Global Temperature Record to show the effects of the differences when performing that type of analysis. And since I’d brought the analysis that far, I thought I’d carry the post a step farther and show the opposing effects of ENSO that exist in global temperature anomaly data.
HADISST-based NINO3.4 SST anomalies also show similar results, though I have not included them in this post. I used HADSST2 data here because I will reference this post in an upcoming one about a paper in press, and that paper uses HADCRUT and HADSST-based NINO3.4 data in its analysis. (The paper attempts to perpetuate a myth I’ve discussed before.)
The Multivariate ENSO Index (MEI) is a calculated dataset that illustrates the timing and magnitude of El Niño and La Niña events. Other ENSO indices use Sea Surface Temperature (SST) Anomalies of the central and eastern Equatorial Pacific or the sea level pressure difference between Tahiti and Darwin, Australia. The MEI, on the other hand, uses additional variables that are part of the coupled ocean-atmosphere ENSO processes. Wolter and Timlin (1998) in “Measuring the strength of ENSO events – how does 1997/98 rank?” note in the abstract, “The Multivariate ENSO Index (MEI) is favoured over conventional indices, since it combines the significant features of all observed surface fields in the Tropical Pacific.” Link to Wolter and Timlin (1998):
The MEI is maintained by Klaus Wolter of NOAA. He explains why he believes the MEI is “better for monitoring ENSO than the SOI or various SST indices” on the NOAA MEI timeseries data webpage. (Scroll down to the FAQs.) He writes, “In brief, the MEI integrates more information than other indices, it reflects the nature of the coupled ocean-atmosphere system better than either component, and it is less vulnerable to occasional data glitches in the monthly update cycles.”
AN OVERVIEW OF THE MEI
The NOAA MEI home page provides a further description of the Multivariate ENSO Index (MEI): “El Niño/Southern Oscillation (ENSO) is the most important coupled ocean-atmosphere phenomenon to cause global climate variability on interannual time scales. Here we attempt to monitor ENSO by basing the Multivariate ENSO Index (MEI) on the six main observed variables over the tropical Pacific. These six variables are: sea-level pressure (P), zonal (U) and meridional (V) components of the surface wind, sea surface temperature (S), surface air temperature (A), and total cloudiness fraction of the sky (C).”
With the exception of the surface wind components, all of the above variables should be self explanatory. NOAA describes the U and V surface wind components in their Transport Winds webpage: “The meridional component of the wind, V, is considered positive when the wind [is] blowing from south to north. A south wind has a positive meridional component while a north wind has a negative meridional component. The zonal component of the wind, U, is considered positive when the wind is blowing from west to east. Thus, a west wind has a positive zonal component and an east wind a negative zonal component.” They continue, “For example, a wind that is blowing from the northeast would have a negative meridional component, V, and a negative zonal component, U. Such a wind would have a direction of 45 degrees.”
COMPARING THE MEI TO AN SST-BASED ENSO INDEX
Figure 1 illustrates the MEI data from January 1950 through July 2010. Like the SST-based ENSO indices, El Niño events are represented by positive values and La Niña events are negative. The 1982/83 El Niño event is shown to peak higher than the 1997/98 event. And the 1997/98 El Niño shows a double peak. The NOAA MEI timeseries data webpage presents the MEI data in bimonthly form.
Note: The MEI data in this post was downloaded from the KNMI Climate Explorer. I’ve used MEI “anomalies” since they will have the same base years as the NINO3.4 SST anomalies and allow for a more direct comparison in the following. The use of MEI anomalies shifts the MEI data slightly, but that shift has no effect on this post.
Figure 2 compares the MEI data to NINO3.4 SST anomalies based on the HADSST2 dataset. Since the MEI is presented bimonthly, the NINO3.4 SST anomalies were smoothed with a lagging 2-month filter for this illustration. That is, for example, the average of January and February SST anomalies are displayed in February. The major variations in both datasets are similar in timing but they differ in magnitude for each event. Note, also, that the MEI data seems to shift upwards around 1976.
If we subtract the NINO3.4 SST anomaly data from MEI, that shift becomes more obvious. Refer to Figure 3. From 1976 to 1980, there is additional rise in the MEI that is not present in the NINO3.4 SST anomalies. There is also some obvious additional variability in the MEI.
Averaging the differences between the MEI and NINO3.4 SST anomalies over the periods before and after 1976, Figure 4, provides an idea of the magnitude of that additional variation in the MEI.
It appears the MEI should account for some of the rise in global temperatures caused by the 1976/77 Pacific Climate Shift.
REMOVING THE LINEAR EFFECTS OF ENSO FROM GLOBAL TEMPERATURES
The common method used by bloggers and climate scientists to remove the effects of ENSO from the global temperature record is to scale the ENSO index data so that the change in the ENSO Index agrees with the resulting change in global temperatures and to lag the ENSO Index data a few months. Then the ENSO Index data is simply subtracted from the global temperature data. It seems to make sense. Unfortunately, it only accounts for the linear effects of ENSO and does not account for the fact that El Niño and La Niña events can warm parts of the global oceans and that these warmings can be cumulative. I’ve discussed this in numerous posts, including “More Detail On The Multiyear Aftereffects Of ENSO – Part 2 – La Nina Events Recharge The Heat Released By El Nino Events AND… …During Major Traditional ENSO Events, Warm Water Is Redistributed Via Ocean Currents,” “More Detail On The Multiyear Aftereffects Of ENSO – Part 3 – East Indian & West Pacific Oceans Can Warm In Response To Both El Nino & La Nina Events”, and with animations of numerous datasets in “La Niña Is Not The Opposite Of El Niño – The Videos.”
Putting that aside, let’s use the method to illustrate two points: the additional portion of the aftereffects of the 1976 Pacific Climate Shift accounted for by the MEI, and the opposing effects of ENSO events.
For those who have never attempted to remove the linear effects of ENSO from the global temperature record, I’ll run through the process. As discussed above, first we need to scale the ENSO index data so that the change in the ENSO Index agrees with the resulting change in global temperatures. Figure 5 illustrates the Hadley Centre’s HADCRUT land plus sea surface temperature anomalies from 1950 to present. It also illustrates scaled NINO3.4 SST anomalies, which are being used as the reference ENSO Index in this example. To scale them, the NINO3.4 SST anomalies are simply multiplied by a factor, and in this example, I used a scaling factor of 0.18. I’ve also shifted the NINO3.4 SST anomalies upwards 0.12 deg C so that they will align with Global Temperature anomaly data during the evolution phase of the 1997/98 El Niño. I’ve used the 1997/98 El Niño event as reference since it is the most significant El Niño event that was unaffected by volcanic aerosols, and its SST anomalies were measured by satellites and in situ buoys.
The NINO3.4 SST anomalies have also been shifted back in time (lagged) two months to account for the delayed response of global temperatures to the change in tropical Pacific SST. In other words, it takes global temperatures a few months to respond fully to the ENSO event. As you can see, the leading edges of the two datasets align well. This makes sense since the central and eastern tropical Pacific (20S-20N, 180 to 80W) represent a major portion of the globe, about 9%. The magnitudes of the two variations from trough to peak are also similar during the evolution phase with the scaling factor I’ve used.
In Figure 6, I’ve started the graph in 1995 to show how well the scaled NINO3.4 SST anomalies and the Global Temperature anomalies align during the ramp up of the 1997/98 El Niño.
The next step is to subtract the NINO3.4 SST anomalies from the Global Temperature anomaly data. The remainder is shown in Figure 7. I’ve highlighted the periods that include the impacts of the explosive eruptions of El Chichon and Mount Pinatubo.
A Stratospheric Aerosols dataset was introduced in a 1993 paper (Stratospheric aerosol optical depth, 1850-1990) by Sato et al. It can be used to account for the impacts of these eruptions (and those of other volcanic eruptions from 1950 to 1999). Estimates of the peak impact on global temperatures from the Mount Pinatubo eruption vary from 0.2 to 0.5 deg C. I’ve scaled the Sato Index data so that it accounts for approximately 0.35 deg C.
Figure 8 illustrates the Hadley Centre’s HADCRUT Global Temperature anomaly data after it has been adjusted for the impacts of volcanic aerosols and the linear effects of ENSO (using the NINO3.4 SST anomalies). Most of the dips and rebounds from the eruptions of the El Chichon and Mount Pinatubo eruptions have been eliminated. I’ve highlighted the apparent step changes caused by the multiyear aftereffects of the 1986/87/88 and 1997/98 El Niño events. Also note the gradual ramp up in temperatures after the 1976 Pacific Climate Shift. There are no other 5-year periods with a gradual rise similar to that. Does this represent the time required for global temperatures to respond to the sudden 1976 upward shift in eastern Pacific Sea Surface Temperatures?
In Figure 9, the average adjusted global temperature anomalies for the periods before and immediately after 1976 are shown. The average adjusted global temperature before 1976 is 0.12 deg lower than it is for period of 1977 to 1986. And for those who are interested, I’ve also illustrate the average temperatures for the two periods after the 1986/87/88 and 1997/98 El Niño events.
We can run through the same process using the Multivariate ENSO Index (MEI) as the ENSO Index. Refer to Figures 10, 11, 12 and 13. The scaling factor used with MEI data was 0.16, as shown in Figure 10.
Figure 14 shows the averages of the adjusted global temperature anomalies for the periods before and immediately after 1976. Recall that the shift was 0.12 deg C, using the NINO3.4 SST anomalies to remove the linear effects of ENSO events. Using the MEI dataset, that shift decreases to 0.05 deg C. This was a very simple comparison. Referring back to Figure 3, the period averages are strongly impacted by how the MEI addresses the 1982/83 El Niño. But it’s a starting point for anyone interested in evaluating this further.
Figure 15 compares the HADCRUT global temperature anomaly datasets after they have been adjusted for ENSO using the MEI and NINO3.4 SST anomalies. Both datasets have also been adjusted for volcanic aerosols using the same Sato Index scaling factors. Another major difference appears to be how the MEI removes the extra El Niño peaks. Both datasets show the ENSO-induced shifts in global temperature anomalies. And both show the multiyear ramp-up from 1976 to 1981, but as illustrated in Figures 4, 9, and 14, the MEI accounts for more of that ramp-up than NINO3.4 SST anomalies.
And now, the second reason for this post.
WHAT CAUSES THE ADDITIONAL VARIATIONS?
If we look again at Figure 15, there are still large year-to-year and multiyear variations in the global temperature anomaly datasets after they’ve been adjusted for ENSO and volcanic aerosols. What causes those additional variations?
Detailed analyses of ENSO, like Trenberth et al (2002) “Evolution of El Nino–Southern Oscillation and global atmospheric surface temperatures”, have shown that parts of the globe warm with a rise in NINO3.4 SST anomalies and others cool. Refer to Figure 16, which is the color version of Figure 8 from Trenberth et al (2002). The correlations with a 0 month lag is shown highlighted in red. Link to Trenberth et al:
Figure 16 (Figure 8 from Trenberth et al 2002)
One would think that after the positively correlated impacts of the ENSO events are removed from the global surface temperature record, as we’ve just done, the remainder would include variations from those areas that are negatively correlated.
This can be shown if we invert either ENSO Index dataset and compare it to what’s left over after the linear effects of ENSO and the volcanic eruptions have been removed from global temperature anomalies. Refer to Figures 17 and 18. Much of the additional yearly and multiyear variability can be explained as warming during La Niña events, and cooling during El Niño events. Note how some of the global responses to the variations in the inverted NINO3.4 SST anomalies are exaggerated while others are suppressed. Why?
In summary, the MEI accounts for more of the 1976 Pacific Climate Shift than the HADSST2-based NINO3.4 SST anomalies. This also holds true for the NINO3.4 SST anomalies based on other SST datasets (HADISST and ERSST.v3b) as shown in Figure 19. Note: The Oceanic NINO Index (ONI) is based on ERSST.v3b data.
Using the simple analysis in this post, the MEI appears to account for more of the aftereffects of the 1976 Pacific Climate Shift (0.07 deg C) than the SST-based ENSO Indices. And there are some additional subtle differences in the MEI data.
And as shown in Figures 17 and 18, when the linear effects of ENSO are removed from Global Temperature anomalies, the remainder logically shows variations that reflect the opposing effects of ENSO.
The post title is The Multivariate ENSO Index (MEI) Captures The Global Temperature Impacts Of ENSO Differently Than SST-Based Indices. It would be up to you as a user of the MEI to determine if the subtle differences mean it’s better.
Regarding the methods used to remove the linear effects of ENSO, Trenberth et al (2002) write in the paper linked above, “Although it is possible to use regression to eliminate the linear portion of the global mean temperature signal associated with ENSO, the processes that contribute regionally to the global mean differ considerably, and the linear approach likely leaves an ENSO residual.”
And as shown in the posts linked earlier, those residuals can be considerable.
The data presented in this post are available through the KNMI Climate Explorer:http://climexp.knmi.nl/selectfield_obs.cgi?someone@somewhere