Appendix B.l 



Before generating amplitude spectra from a statistically non-sta- 

 tionary sample space, such as the sea floor, one must Initially define 

 provinces which are relatively homogeneous with respect to the frequency 

 spectrum. Since generating an amplitude spectrum directly by Rsurler 

 transform assumes statlonarlty over the length of the Input series, this 

 "province picking" procedure must be performed by estimating the spec- 

 trum discretely In the spatial domain. Although the present application 

 Is new, the concept of estimating spectra In the time/space domain is 

 not. Blackman and Tukey (1958) referred to such estimates as "pilot 

 spectra" and describe two methods for their calculation. Godfrey (1967) 

 also describes a method, very similar to that detailed here, which Is 

 used for predicting non-stationary time series. This section gives 

 details of the procedure used In this study, presents performance tests 

 of the algorithm, and Include full FDRTRAN-77 software for performing 

 the analysis. 



The initial step in processing is identical to that required for 

 running an FFT. The data must be projected onto a straight-line segment 

 and interpolated to even increments in distance. The straight line is 

 generated by a simple least-squares fit of the navigation track; the 

 best results are obtained when relatively straight line navigation is 

 input. Large deviations in the track greatly degrade results. Next, 

 points are mapped onto the nearest location on the least-squares line, 

 without alteration if they are within a designated "pivot distance," or 

 modified according to the local gradient if beyond this distance. 

 Finally, the data are interpolated at a specified Interval using a one- 



135 



