The weakness of the one-step SST prediction schemes that involve estimating a trend is that this trend is very hard to estimate accurately because of the noise in the data. In the 2007 elicitation, we tried to overcome this limitation by introducing two-step schemes that assume that the MDR SST trend is similar to the Northern Hemisphere temperature trend. The Northern Hemisphere temperature trend is much easier to estimate because of the lower level of noise in the Northern Hemisphere time-series. The two-step method consists of making a statistical prediction of the Northern Hemisphere trend, and then converting that prediction to a prediction of MDR SST. This assumes that the relationship between Northern Hemisphere temperature and MDR SST that held in the past will also hold for the next five years. Whether the two-step method is likely to out-perform the one-step method depends on the degree of similarity between the trends in the Northern Hemisphere temperature and the MDR SST and the relative levels of noise in the two series. In practice, out of sample hindcasts show that the two-step method has worked better in the past (Laepple et al., 2007).
Was this article helpful?