When sparse sea-surface temperature (SST) data are gridded for climatological purposes using minimum curvature splines, the resulting contour plots are often marked by localized high-amplitude anomalies that are not true features of the climatological field. Two traditional approaches to solving this problem are removing the data that cause the anomalies, or smoothing the data prior to final gridding and contouring. One form of smoothing that can be done employs a filter that predicts regional trends in the data and enhances local trends possibly correlated with auxillary information. For example there is a general tendency for oceanographic temperature profiles to be well-correlated within localized regions of similar bathymetry. By choosing a filter that takes such tendencies for localized spatial coherence of data into account, trends that are obscured due to the sparsity of data may be enhanced and more easily than in the contoured grid. Such considerations led to the development of a two-part smoothing filter that uses simple cross-validation and least squares to fit data, in this case SST data, to a regional surface that depends on all the data in the target area, and to localized surfaces delineated by bathymetric range. The regional surface is fit to the SST observations. The errors in the regional fit are then fit to individual surfaces defined over localized areas as delineated by a common range of bathymetry. The data are then interpolated to the nearest grid cell center only if the cell is supported by data. The filter was applied to SST data from the Yellow Sea and the east coast of the United States. The results showed a large improvement in the climatological representation over most of the respective area, but was not able to resolve stable current patterns without additional special processing.
展开▼
