HR: 1340h
AN: NG23D-0108 [Abstracts]
TI: The Analysis of Large Hydro-Climatic Datasets Using Independent Component Analysis
AU: * Sharma, A
EM: a.sharma@unsw.edu.au
AF: School of Civil and Environmental Engineering, University of New South Wales, Sydney, NSW 2052
Australia
AU: Westra, S
EM: seth@civeng.unsw.edu.au
AF: School of Civil and Environmental Engineering, University of New South Wales, Sydney, NSW 2052
Australia
AU: Lall, U
EM: ula2@columbia.edu
AF: Department of Earth and Environmental Engineering, Columbia University
116 Monell Building, Palisades, NY 10964
United States
AB:
Component extraction techniques have been used in studies of a range of hydro-climatic variables both for reducing the
dimensionality of large datasets and to aid in the identification and interpretation of significant modes of climate
variability. The most popular of these techniques is known as principal component analysis (PCA), or the closely related
empirical orthogonal function (EOF) analysis, which uses second-order statistical information to find components that are
uncorrelated, and successively explain the maximum amount of variance of the original dataset. An alternative technique that
has been developed recently is known as independent component analysis (ICA), which is a rotational technique that reduces
higher-order dependencies to find components that are statistically mutually independent - a criterion that is much more
stringent than removing correlation alone. As a result, the resulting independent components (ICs) are presumed to be more
strongly related to the independent causative factors that introduce variability in the data.
The difference between PCA and ICA will be illustrated on two very different datasets. The first dataset is the global sea
surface temperature (SST) anomaly dataset, which has been the subject of numerous previous studies using PCA. In the present
study the ICA results are found to be significantly different from the PCA results, raising an interesting question about how
to interpret the components derived from the two methods. This question is particularly pertinent given that the SST dataset
contains significant internal persistence in both time and space, such that the independent components can not be direct
related to some `independent causative factors', since the influence of these factors (whatever they may be) on SSTs would be
neither linear nor additive.
The second dataset represents the seasonal rainfall time series from over 200 rainfall gauges located throughout the
Australian continent. Once again, the PCA and ICA results were found to be significantly different. Here we make an
assumption that climatic phenomena such as the El Nino Southern Oscillation (ENSO) and the Interdecadal Pacific Oscillation
(IPO) have a quasi-linear influence on Australian rainfall, and examine whether the PCs and ICs are able to highlight the
nature of the influence of these phenomenon. The results of this analysis suggest that ICA has some important benefits in
representing the variability of large climatological datasets such as Australian rainfall in a way that facilitates the
interpretation of the extracted components.
DE: 1816 Estimation and forecasting
DE: 1833 Hydroclimatology
DE: 1854 Precipitation (3354)
DE: 1869 Stochastic hydrology
DE: 4215 Climate and interannual variability (1616, 1635, 3305, 3309, 4513)
SC: Nonlinear Geophysics [NG]
MN: Fall Meeting 2005