David Reusch

Earth & Environmental Systems Institute, Penn State

Nonlinear (paleo)climatology: Studying polar climate with self-organizing maps

Room 811 AOSS, Tuesday, April 10, 2007, 2:30 PM

Abstract

Proxy histories from ice cores, tree rings, and other climate indicators are required to extend understanding of climate change to the period before instrumental records. Available observational and instrumental records are too limited spatially and temporally to fully characterize natural variability, particularly in the polar regions. To study these issues from a new perspective, self-organizing maps (SOMs), an analysis tool from the field of artificial neural networks, have recently been used to examine a number of important cryospheric and atmospheric datasets from the polar regions, including Greenland ice cores, Antarctic sea ice, and North Atlantic sea level pressure.

SOMs enable unsupervised classification of large, multivariate/multidimensional data sets (e.g., time series of the atmospheric circulation or sea-ice extent) into a fixed number of distinct generalized patterns or modes representing the probability distribution function of the input data. These patterns are organized spatially as a two-dimensional grid such that distances represent similarity (adjacent patterns will be most similar). When applied to atmospheric data, for example, the analysis yields a nonlinear classification of the continuum of atmospheric conditions. In contrast to principal component analysis, SOMs do not force orthogonality or require subjective rotations to produce interpretable patterns.

The sets of generalized patterns resulting from a SOM-based study concisely capture the spatial and temporal variability in the dataset of interest. For example, the changing position and timing of greatest sea ice extent conditions, month to month and year to year, is readily seen. Similarly, the annual progression of expansion and retreat can be concisely summarized. With respect to a large (58 site) suite of Greenland ice cores, SOM analysis of annual accumulation captures the high spatial diversity seen in climate records from the GIS, including nonlinear gradients in latitude and elevation.