*In this post I show one simple and 2 multiple regression models to assess the role of time, El Nino – La Nina SSTA and volcanic activity (SATO) on UAH global temperature anomaly trends. The 3rd model provides a reasonable approximation of the actual UAH oscillations over the 1979 – Feb, 2011 period. *

**Click Image to Enlarge**

This analysis is similar to previous temperature anomaly regressions (here, here, here) that I have done.

The simple trend line regression shows the overall trend is upward, however, there are several oscillations that the linear trend misses. The yr_frac and Nino34 regression improves on the linear model, however, it undershoots in the early 1980s, overshoots in the 1992-1994 period, periods following significant volcanic activity.

The yr_frac, Nino34 and SATO model improves the fit in the early 1980s and 1992-1994 period and is slightly better in the 1998 and 2010 El Nino periods.

The 3rd model matches the observed 2010 El Nino – La Nina oscillation pretty well so far, indicating that the 2010 – 2011 UAH anomalies are following a predictable pattern.

### Like this:

Like Loading...

*Related*

hi….lucky enough to find your website in searching for sea surface temperature(SST) pattern. Just wondering why we should looking into anomalies pattern rather than the true values of SST. Any explanation?

Cheers

Very nice work. I don’t see an R-squared value, but the model looks very good indeed.

Tamino has done something similar, but used a different index for ENSO… and (surprise!) concluded that the underlying trend is much greater than what you found. It is pretty clear that the lagged Nino 3.4 index is a good predictor of short term trends.

Here is the Tamino post you’re referring to:

http://tamino.wordpress.com/2011/01/20/how-fast-is-earth-warming/

He seems to be using a solar forcing as well, but otherwise the lag times seem about the same as calculated here. It would be interesting to see if MEI vs. Nino3.4 is the primary difference in the resulting trends, and which one has the better fit.

Pretty cool. It would be interesting if you could “project” out ahead to the extent of the Nino lag (5 months?), either without the SATO or assuming no substantial volcanic forcings.

Hi, have you any comments on the presence of autocorrelation in the data – does this carry the risk of a spurious correlation. Can you recommend a text or paper about this?

toby

Excellent question!

I used the gls() function from the nlme package in R. You can read about gls here.