standard deviation of 0.04. The black line in figure 8 is the 

 fitted straight line and the envelope shows the standard error of 

 estimate. 



Although equation 5 displays a proportionality constant 

 between the vertical and horizontal gradients, the use of this 

 constant is not felt justified at this time because of the necessary 

 averaging of the analog temperature data. We are presently study- 

 ing this interrelationship further, using the digital temperature 

 data, and we hope to be able to substantiate the proportionality 

 constant. However, from the results here it is clear that (1) the 

 vertical-temperature gradient is generally of the order 10 * but 

 can range between 10 _1 and 10 " 3 °C/ft; (2) the horizontal- 

 temperature gradient is generally of the order 10 " 4 but can range 

 between 10 " 3 and 10 " 5 °C/ft; and (3) the two gradients normally 

 differ by two orders of magnitude. 



WAVELENGTHS 



Inspection of the contoured horizontal-temperature-gradient 

 fields reveals a regularity with which the gradient changes sign, 

 strongly suggesting a periodic motion indicative of internal waves 

 or convection cells. Although all information indicates the 

 presence of a broad spectrum of frequencies in the ocean, the 

 simplified interpretation of these data implies that there is a 

 dominant oscillation at an intermediate frequency. 



The power spectrum is a statistical method for gaining 

 spectral information from a data sample of finite length and 

 fluctuating values. The power spectrum U(h) can be shown to be 

 the Fourier transform of the autocorrelation R(\), and is propor- 

 tional to the energy per unit band width. Consequently, any 

 dominant frequency will appear as a peak in the spectrum. 



The power spectrum method of analysis was first applied 

 to NEL Thermistor Chain data of Cruise 4, which took place 

 between San Diego and Hawaii in 1961. Cross sections (data 

 strips) of 8-1/2- and 12-hour durations of depths of an isotherm 

 were subjected to spectral analysis. Figure 9 shows the location 

 of the analyzed data sections (LaFond and Moore, 1964). 19 The 



23 



