reduced amplitude for that condition, independent of the 

 reflection coefficient. 



For side beams, the degree of bottom roughness directly 

 affects the amount of energy returned. Smooth bottoms allow the 

 majority of the incident energy to forward scatter away from the 

 source. Rough surfaces, on the other hand, provide inward facing 

 facets which allow increased percentages of energy to be back 

 scattered toward the source. Therefore, while rough surfaces 

 tend to decrease the energy returned in vertically incident 

 waves, they also tend to increase the energy received from 

 nonvertically incident waves. The multi-beam configuration of 

 Sea Beam allows this effect to be incorporated into bottom 

 characterization studies. Figure 2 illustrates a typical 

 backscatter record from Sea Beam. 



INTERPRETIVE METHOD 



Before attempting to categorize the backscatter energy 

 character for seafloor mapping, it is necessary to develop a 

 concise description of the energy envelopes associated with each 

 beam. Each gather produces over 8,000 data points, which at a six 

 second ping rate results in an enormous data set for analysis. 

 Therefore, after the various corrections are applied to the data, 

 a curvilinear model is fitted via least squares to each beam 

 using iterative techniques described in van Heeswijk and Fox 

 (1988). For the center beams, a simple Gaussian function 

 appears to be an adequate model for the energy envelope. For the 

 outer beam, the distribution becomes skewed, requiring a more 

 elaborate function. The Rician probability density function has 

 been proposed as the appropriate general model by Stanton (1984). 



The fitted Gaussian curves describe each center beam with 

 terms relating to the amplitude, dispersion, and total integrated 

 energy. As the previous discussion indicated, these various 

 model parameters are intercorrelated. In order to best discern 

 the potential groupings of pings independent of correlated 

 parameters, the data are rotated into an orthogonal vector space 

 through principal components techniques. Figure 3 illustrates a 

 typical data set from the Mid-Atlantic Ridge projected on its 

 first three principal components. The clustering of the samples 

 into groups is apparent. These principal components scores would 

 next be subjected to a clustering algorithm to automatically 

 select distinct groupings. 



Whether these grouping have geological meaning requires that 

 the resulting groups be spatially coherent. Therefore, the next 

 stage of analysis would be to attempt to map the data back onto 

 the earth and investigate the resulting patterns. If such 

 patterns result, ground truth geological data from bottom 

 photography or submersible observations are required to relate 

 the mathematical properties of the return pulses to the physical 



75 



