USGCRP logo & link to home

Updated 12 October, 2003

US National Assessment of
the Potential Consequences
of Climate Variability and Change
Climate Indices: ENSO Caveats



Although HADCM2 seems to show better ENSO signatures than CGCM1 in terms of several sea surface temperature (SST) indices ( NINO ), these signatures are the result of variance in the tropical oceans that is too high, both compared to observations and every other GCM, including HADCM3.  As part of the Coupled Model Intercomparison Project (CMIP), power spectra of globally and annually averaged surface air temperatures from several CMIP models (unforced control simulations), including HADCM2 ( spectra ), show that the ENSO-time period variances (< 10 years) in HADCM2 are too high compared to observational variance at those time periods.  Notice also that this error has been corrected in HADCM3.

Mark Stevens from CCCma has calculated several correlation coefficients to highlight tropical variances in several models.  All of these experiments are unforced control simulations, which can be used to understand the model variability.  The upper left panel in each figure is the observations.
Therefore one must be wary of using any of the models to predict trends in ENSO variability.  While the HADCM2 shows the correct ENSO amplitude when using the NINO indices, the ENSO pattern is completely wrong.  The CGCM1, on the otherhand, while simulating a better ENSO pattern, does not get a strong enough ENSO amplitude.   Even if a model gets the correct ENSO signal for the correct reason, it still may not produce the correct teleconnections, which are what determine the ENSO response in the U.S.



Figure 1 (above):  The correlation between the global annual mean and the annual mean at each grid point shows that
the variance in the tropical Pacific Ocean is too large in HADCM2, while CGCM1 is closer to reality.  HADCM2 shows larger Pacific variance than all the other models.  Any SST index anywhere in the tropical Pacific would therefore show significant variance in HADCM2.  (The white areas are the areas where there is not enough observed data as these analyses are part of a study involving observational data).


Figure 2 (above):  The correlation between the NINO3 annual mean and the annual mean at each grid point shows that all the models reproduce the NINO3 ENSO signature.  As Legler and O'Brien have shown ( NINO ) the amplitude of the NINO3 signature is too weak in CGCM1, but the strength of the NINO3 signature in HADCM2 is due to an overprediction of variance in the entire tropical Pacific, rather than a specific ENSO signature. 


Figure 3 (above):  The standard deviation of the annual mean surface temperature anomalies confirms that the tropical variance in CGCM1 is less than that in HADCM2.

US CCSP  logo & link to home USGCRP logo & link to home
US Climate Change Science Program / US Global Change Research Program, Suite 250, 1717 Pennsylvania Ave, NW, Washington, DC 20006. Tel: +1 202 223 6262. Fax: +1 202 223 3065. Email: Web: Webmaster: