Application of Weighted Empirical Orthogonal Function Analysis to ship's datasets
Résumé
Marine ship observations over the vast oceanic regions are crucial to studies of climate variability on timescales from the seasonal to multidecadal. However, any climatic analysis of this historical record is hampered by two difficult problems, namely:
- The systematic instrumental errors which contaminate the ship observations. For example, it is well-known that most of the ship-reports before 1940 contain a large majority of uninsulated bucket Sea Surface Temperature (SST) measurements which are biased low, while the data after the 1940s are mostly injection or insulated bucket SST measurements which are biased high (Bottomley et al., 1990).
- The irregular space-time sampling of the ship-reports. For example, Comprehensive Ocean-Atmosphere Data Set (COADS) summaries provide meteorological variables in the form of monthly means for 2° ´ 2° latitude-by-longitude cells (Woodruff et al., 1987). In such datasets, the number of observations used to compute a particular monthly mean reflects the number of ships that cross the box that month. Thus, for a particular month, one cell’s mean may be computed from hundreds of observations, while others may be based on only a few, and there may be many cells with missing means due to the poor spatial and temporal coverage outside the main shipping lanes.
The latter problem attends almost all climate studies from seasonal to multidecadal timescales, but is particularly relevant to the interannual to multidecadal. The classical solution to cope with this problem is to use some kind of objective analysis. This technique spatially smoothes the oceanic fields by filling the data-void areas with reasonable values which are a linear combination of climatology and anomalies observed in the neighborhood of each grid’s cell. The drawbacks of this solution are: First, the need for a very good climatology which has to be constructed before the analysis. Second, the oceanic fields derived from objective analysis are generally over-smoothed with the undesirable consequence of a degradation in the spatial resolution of the data.
The main objective of this paper is to present a new multivariate statistical method to deal with this last problem. The method may be termed weighted Empirical Orthogonal Function (EOF) analysis or weighted Singular Value Decomposition (SVD) analysis and is a generalization of the traditional EOF analysis, or more precisely, of truncated SVD analysis.
Domaines
Océan, AtmosphèreOrigine | Fichiers produits par l'(les) auteur(s) |
---|
Loading...