1) Hourly readings were used for computation, starting with the first 



observation. 



2) The remaining hourly readings were used for computation, starting 

 with the second observation, in order to estimate the variability of hourly 



samplings. 



3) and U) The first and second halves cf the BT data, representing the 

 observations made by the USS REHOBOTH and USS SAN PABLO respectively, were 

 analyzed using the half-hourly readings. 



5>) The whole of the half-hourly BT data was analyzed. In each case 

 60 ilagslwere used. 



One preliminary result of the power spectrum computation is the auto- 

 correlation function. The autocorrelation functions for all five of the 

 series were practically identical and are not shown separately here. The 

 function presented in figure 2 is for the hourly readings commencing with 

 the first observation. It is evident that the major cyclical components are 

 found in the band of periods centered at 2k hours. 



The results of the power spectrum analyses are graphed in figure 3 

 for the hourly readings and in figure h for the half-hourly readings. Since 

 a total of 60 lags was used in all analyses, the number' of degrees of freedom 

 available was about 19 for the first four analyses and 39 for the last analysis. 

 Thus, the accuracy of the analyses is somewhat lower than it would be with a 

 longer series of observations. Confidence limits are not shown in the figures. 



IV. RESULTS OF ANALYSIS 



The spectral analysis derived from the 66° F. isotherm shows that most 

 power is concentrated in the band from 20 to 30 hours. This result was fore- 

 shadowed by the autocorrelation function in figure 2 which is markedly cyclical 

 with peaks at multiples of 22 to 26 hours. The semidiurnal period to be ex- 

 pected from tidal motions is apparently neglibible. Although in figure 3 "the 

 small peak between 10 and 12 hours appears to coincide with the tidal period, 

 the autocorrelation function shows that this peak is probably due to the short 

 length of record. The noise level of figure 3, therefore, must be of the order 

 of 20,000 ft.^sec. In figure lb, there is no outstanding power peak in the fre- 

 quency range in which definition is good. The great differences between the 

 spectra from the two ships, taken at different times, indicate that the noise 

 level for the half-hourly readings is also about 20,000 ft.'-sec. 



It is thus likely that the whole spectrum below 20 hours represents 

 white noise with no important spectral components. For an observational in- 

 terval of one-half hour, the spectrum is best estimated in the range from 1 

 to 10 hours ; for observational periods of one hour, adequate definition in 

 the spectrum exists in the range from 2 to 30 hours. 



