The atmospheric model calculates the winds that blow on the ocean model's surface. Rather than use a full atmospheric general circulation model for this task, we use a statistical atmospheric model. There are two reasons we do this. The first is pragmatic: the statistical model runs 100 times faster than a full general circulation model. This lets us do many more experiments than we would otherwise be able to do in a given amount of time. The second is philosophical: when we construct the statistical model, we go through many steps intended to filter out noise that we believe only degrades the predictive performance of the model. There is no easy or self-consistent way to remove such noise from a general circulation model.
The statistical model is made from observations, as follows. First we decompose historically observed variability in sea surface temperature (SST) into a set of patterns called empirical orthogonal functions (EOFs). We then decompose the historically observed wind field in the same fashion. Lastly, we see, from the historical record, how likely it is that each wind pattern will be associated with each SST pattern. We save all this information in a file. When we go to use the statistical atmospheric model, we simply decompose the model's SST field into the same patterns as historically observed, then apply the wind patterns that are associated with those SST patterns to the ocean model's surface.
Last modified: 25 June 1997
Copyright © 2000 David W. Pierce. All rights reserved.