Autocorrelation functions of sea surface temperature (SST) and depth of the 20-degrees-C isotherm (D20) are estimated at 95 locations along the tropical Pacific expendable bathythermograph (XBT) ship-of-opportunity tracks, and used to determine statistical parameters required for optimal interpolation. The parameters are variances for signal and noise, and spatial/temporal decorrelation scales. Estimates were made for two periods of time: June 1979 to May 1982 preceding the 1982/83 El Nino, and June 1979 to May 1983 including it. Parameters for the first period indicate smaller scales and weaker signals than the ones for the second. The difference results from El Nino, whose large temperature signal dominates the statistical structure. Understanding the dynamics of the smaller interannual signals and in particular the precursors to El Nino is as important as understanding El Nino. The smaller scales are therefore recommended for optimal interpolation and design of the XBT network. Three summaries of the scales are presented: 1) averages in three degree latitude bands; 2) values in dynamic regions representing the major zonal currents; and 3) median values for 95 estimates in the tropical Pacific 18-degrees-N to 18-degrees-S. The median values are recommended for design of the tropical XBT network. The median scales for the depth of the 20-degrees-C isotherm are: 3-degrees latitude, 15-degrees longitude and 2 months; and the median signal-to-noise (amplitude) ratio is 0.75. Scales for SST are greater than or equal to scales for D20, thus subsurface temperature is the limiting factor in designing the network. Analysis of mapping errors suggests that optimal sampling requires two to three samples per decorrelation scale. Two modes of XBT sampling are recommended for TOGA. Broadscale sampling for horizontal mapping of the temperature field is recommended at a density of two samples per scale, which requires an XBT station every 1.5-degrees latitude and 7.5-degrees longitude, monthly. The broadscale mode should be implemented in as large an area as possible with available shipping. Sampling on repeated transequatorial sections for time series studies is recommended at a density of three samples per scale, which requires 18 sections per year, with stations every 1-degree latitude. The time series mode is useful for more accurate studies of thermal structure and currents, and should be implemented on a few routes in each ocean which transect major thermal features.
