Climate Dynamics
Observational, Theoretical and Computational Research on the Climate System
© Springer-Verlag 2011 10.1007/s00382-011-1023-3
Electronic supplementary material
Below is the link to the electronic supplementary material.
PDF (4931 KB)
Footnotes
1 These forecasts are available at: http://www.esrl.noaa.gov/psd/forecasts/sstlim/.
2 The leading EOFs of the two GCMs are compared to the observations in the Supplementary Information.
3 van den Dool (1994) showed that, for the atmospheric flow at 500mb to match over an entire hemisphere, one would need a library of 1030 years to find an analogue within the
observational errors. Making a similar estimate, with optimistic assumptions, for Atlantic SST would require more than 105 years of data to provide a natural analogue. Given 140 years of data, the chance of finding a natural analogue is remote.
4 This choice of artificial damping is motivated by noting that a forecast time series with zero correlation skill, but with the same variance as the true time series will produce an RMS error that is a factor of larger than a climatological forecast, assuming Gaussian
distributions. It would be possible to derive optimal damping factors if the mean RMS error was known before the forecasts are made, but in this case, it is not known.
5 For the LIM method, the estimated EOFs can be extended into the masked regions through regression onto the SSTs, allowing a prediction to subsequently be made for all regions. For the CA method, the estimation of the weights in Eq. 10 does not include the masked regions, but Eq. 11 can use all the data to make a prediction.
