correlation should also be a cosine function. Hence, one might conjecture 
that the oscillatory portion represented by the autocorrelation for the 
second day is indeed due to a periodic component in the time series. Simi- 
larly, the time in which the autocorrelation decreases to some small 
number is a measure of the correlation time for the phenomena beyond which 
the time series is uncorrelated. However, a study of the time series does 
not appear to support these conjectures. In no case can one point out the 
differences in the time series which would account for the differences in 
their autocorrelations. It is not a function which one can compute mentally. 
Suffice it to say that it is useful as an intermediate step in the computa- 
tion of the power spectrum, except for certain exceptional cases. 
The power spectra corresponding to the autocorrelations previously 
discussed are given in Fig. 8. The very interesting result is obtained 
that the power spectra all have the same form within the error associated 
with the results. There are slight differences but these reflect large 
differences in the autocorrelation. For example, at 0 frequency the auto- 
correlation which decreased slowest has the highest value, exceeding by an 
order of magnitude the results of the other two. Again, the autocorrelation 
with one oscillatory cycle shows peaks in its spectrum. These peaks, however, 
are within the errors to be associated with the computation (shown as the 
error bar) and hence must be ignored. The overwhelming similarity is in the 
form of the power spectra, for, excluding the data obtained at frequencies 
less than 1 cycle per hour as reflecting errors due to the finite sample 
time, and cxcluding the data at the highest frequency as perhaps being in- 
Sake by the response time of the thermometers, the power spectra follow 
the f 
at the same buoy and at the same time; it applies also to data obtained at 
law as indicated. This is true for data obtained at various depths 
the same depth at different buoys; and the data shown is data obtained at the 
same depth, at the same buoy, and on successive days. 
Power spectra representing the extremes of those calculated were 
plotted with those of the bottomed thermistors at Bermda ‘=? (see Fig. 9). 
The difference in slope between the thermistor data and the thermometer data 
13 
