interpolated values determined from this survey is shown in Figure 10. It is 

 apparent that only a slight decrease in error was achieved by use of the increased 

 sampling rate. In order to more effectively reduce this error, the FFT was 

 computed for that anomalous portion of the model field contained within 10' 

 and 70' latitude and 0' and 60' longitude. Contours of the normalized amplitude 

 spectrum for this part of the model field are shown in Figure 11 . With an estimate 

 of 0.25 cycles per data interval as the frequency at which the spectrum for this 

 smaller area flattens out, a track spacing and sampling interval of 2 nm was 

 selected. Although the lack of a definite trend in the high frequency components 

 indicates that the track direction is not a particularly significant factor for this 

 model field, the distinct trending of the low-frequency components indicates 

 that some control of this factor is required. The model 3 simulated survey, based 

 on the amplitude spectrum in Figure 11, is shown in Figure 12. Figure 13 is a 

 contour of the differences between the spline interpolated data values obtained 

 from the model-3 survey and the true values from the model field. The histogram 

 of these errors, as shown in Figure 14, indicates that the model-3 survey produced 

 a significant decrease in the overall errors in the interpolated data values. 



In order to determine whether the model-3 errors were caused by the 

 cubic spline interpolation algorithm or by the simulated survey design, a two- 

 dimensional, high-pass filter was applied to the one-minute grid values of the 

 model field. This digital filter, which was designed by Fuller (1967) as a 

 general high-pass filter, has the two-dimensional amplitude response shown in 

 Figure 15. The weight function for this filter is symmetric, thus, no phase 

 shifting is involved in the operation. The amplitude response is such that those 

 frequencies which were not adequately sampled with the model-3 survey are 

 retained while the lower frequencies are removed. The contoured result of 

 this filtering operation is shown in Figure 16. By comparing this result with the 

 residual contours in Figure 13, it is apparent that at least some of the errors in 

 the interpolated values from the model-3 survey are generated by locally 

 inadequate sampling. 



28 



