| description abstract | Studies of climate variability often rely on high quality sea surface temperature (SST) anomalies. Although the high-resolution National Centers for Environmental Prediction (formerly the National Meteorological Center) optimum interpolation (OI) SST analysis is satisfactory for these studies, the OI resolution cannot be maintained before November 1931 due to the lack of satellite data. Longer periods of SSTs have come from traditional analyses of in situ (ship and buoy) SST observations alone. A new interpolation method is developed using spatial patterns from empirical orthogonal functions (E0Fs)?that is, a principal component analysis?to improve analyses of SST anomalies from 1950 to 1981. The method uses the more accurate OI analyses from 1982 to 1993 to produce the spatial EOFs. The dominant EOF modes (which correspond to the largest variance) are used as basis functions and are fit, in a least squares sense, to the in situ data to determine the time dependence of each mode. A complete field of SST anomalies is then reconstructed from these spatial and temporal modes. The use of EOF basis functions produces an improved in situ SST analysis that more realistically represents sparsely sampled, large-scale structures than traditional analyses. The EOF reconstruction method is developed for the tropical Pacific for the period 1982?92 and compared to the OI. The method is then expanded to the globe and applied to a much longer period, 1950?92. The results show that the reconstructed fields generally have lower rms differences than the traditional in-situ-only analyses relative to the OI. In addition, the reconstructed fields were found to be smoother than the traditional analyses but with enhanced large-scale signals (e.g., ENSO). Regions where traditional analyses are adequate include some parts of the North Atlantic and the North Pacific, where in situ sampling is most dense. Although the shape of SST anomaly patterns can differ greatly between the reconstruction and traditional in situ analysis, area-averaged results from both analyses show similar anomalies. | |
