A Statistically Efficient Mapping Technique for Four-Dimensional Ocean Temperature Data

This paper presents a modified objective mapping technique that takes advantage of the strong vertical correlations in ocean temperature profiles. This technique has been used here successfully to generate a uniformly gridded upper-ocean temperature dataset in the southwest Pacific Ocean region from most of the available bathythermograph casts collected between 0¬¨‚àû‚Äö-50¬¨‚àûS and 140¬¨‚àûE‚Äö-180¬¨‚àû, covering the period from 1955 to 1988. Important advantages of this technique over most previous objective methods are its (i) ability to deal with four-dimensional data (space and time), (ii) improved estimate of the first-guess (polynomial) mean, (iii) preservation of the vertical structure of the ocean temperature data, (iv) computational efficiency, and (v) objective error analysis. The technique combines empirical orthogonal function (EOF) analysis, using the singular value decomposition, and objective mapping. In this application of the method, a digital atlas‚ÄövÑvp of upper-ocean temperatures has been created on a grid 2¬¨‚àû ‚àöv= 2¬¨‚àû, at 5-m depth intervals, and comprises a monthly climatology and two three-monthly time series (January, April, July, and October). The time series include a dataset from 1955 to 1988 to 100-m depth, and a shorter period, deeper dataset from 1973 to 1988 to 450-m depth. Only the first five vertical EOFs are needed to explain about 90% of the total variance in the data and to within the a priori noise estimates. The full four-dimensional temperature field was reconstructed using objective maps of the horizontal coefficients corresponding to each of the significant vertical EOFs. Although the method is statistically suboptimal, the final mapped temperature fields are unbiased and consistent with the a priori noise. In this application, the computing time is reduced by a factor of 36, making the mapping procedure feasible on modern workstations.