85th AMS Annual Meeting

Thursday, 13 January 2005: 2:45 PM
North Atlantic variability and self-organizing maps: early results
David B. Reusch, Penn State University, University Park, PA; and R. B. Alley
Poster PDF (494.7 kB)
North Atlantic climate variability arises from diverse sources over broad spatial and temporal scales and has been a rich field of study for many years. Despite decades of research and identification of many important properties and behaviors, many important questions remain unanswered. Some of these questions arise from the limitations of the linear tools generally used in climate studies. The nonlinear aspects known to exist in the climate system are thus often only approximated, or sometimes ignored altogether, in the purely linear approach, to the detriment of our understanding. New tools with the ability to handle nonlinear behavior are thus potentially of great value to our study of climate.

The North Atlantic Oscillation (NAO) is an area of particular interest to the study of North Atlantic variability. As the only year-round teleconnection pattern in the northern hemisphere, the NAO has a widespread climatic influence from eastern North America to western Europe (and beyond) and has been widely studied in recent decades. The nonlinearities known to exist in the NAO make it essential to explore nonlinear analysis techniques so as to better understand these aspects.

One such technique is based on self-organizing maps (SOMs). SOMs support analysis of variability in large, multivariate and/or multidimensional datasets. The technique provides a complementary nonlinear alternative to more frequently used but linear tools such as empirical orthogonal function (EOF) analysis. SOMs have a number of advantages including readily handling nonlinear behavior and robust interpolation into areas of the input space not present in the available training input. SOMs also have the benefit of being a completely independent uniformitarian analysis pathway and thus provide independent results for comparison with more traditional techniques.

Preliminary investigations have used 10 years (1980-1989) of winter season (DJF) mean sea level pressure (MSLP) from the NCEP-2 reanalysis to begin exploring the usefulness of SOMs for understanding North Atlantic climate variability, in general, and the NAO, in particular. This work was preparatory to switching to the ECMWF 40-year reanalysis (ERA-40) and adding other meteorological variables to look at other aspects of the problem. We also plan to add climate proxies, such as recent ice cores from Greenland, to explore the relationship between the proxies and the climate of the larger region.

Early results (Figure 1) from analysis of the 10-year period are very encouraging in a number of ways. First, the MSLP patterns extracted by the SOM appear to capture two axes of variability in the NAO: the vertical seesaw of pressure changes in the Icelandic Low and Azores High and the horizontal changes over time in the centers and extent of these two pressure systems. The patterns include both the end members and a suite of intermediate, possibly nonlinear combinations of the two behaviors. The latter highlights a strength of the SOM-based approach versus traditional methods: the ability to handle nonlinear relationships in the data as a matter of course. Second, a stratification of a monthly NAO index (from the Climate Research Unit, University of East Anglia) by high/low values shows that similar extremes in the index can be associated with quite different MSLP patterns from the SOM analysis. While this is not altogether surprising given the linear, two-points-in-space nature of the NAO index, further examination of the patterns, and transitions between them, may reveal new insights about the NAO.

We anticipate further useful insights after switching to the longer ERA-40 dataset, which will make 40+ years of analysis possible. The longer timeframe, with its more complete view of climate variability, should improve the generalization performed by the SOM as it extracts patterns during training, thus improving the robustness of the analysis. The extended dataset will also improve our study of the temporal behavior of the variability through a longer series of extracted patterns.

Given the extensive body of work on the NAO and North Atlantic variability, it remains to be seen how novel our early results are. But even if our conclusions are not new, it is still of interest that an independent analysis path was able to confirm the prior art. And that which is different will add to our understanding of this system.

Supplementary URL: