7. Prediction of Blgh Fteqnency Roughness 



Raving developed a spectral model of sea-*floor topography based on 

 measurements from surface ship sonar systems, the question remains 

 whether this model can be extrapolated Into spatial scales smaller than 

 those resolved by the sounding system. This question Is particularly 

 Important to underwater acoustic applications, where the acoustic fre- 

 quencies of Interest In a scattering problem do not necessarily corre- 

 spond to the spatial frequencies sampled to generate the model. The 

 concepts of measurement noise levels were Introduced In Chapter 4. In 

 this chapter, the effect of estimation errors on prediction will be 

 examined, sources of measured high frequency bathymetry Introduced, and 

 a simple prediction test presented. 



Source of Error In Spectral Estimates 



The model parameters used to describe the amplitude spectrum of sea- 

 floor topography are derived from regression estimates of spectra from 

 profiles of noisy data collected In a generally non-stationary environ- 

 ment. As such, there is estimation error In the model parameters from 

 several sources, which will necessarily result In prediction errors as 

 the model Is extrapolated to high frequencies. Although many techniques 

 are used to reduce these errors, some level of error will always remain. 

 Unfortunately, due to the variability of data quality, variable track- 

 line spacing, and the presence of some level of non-statlonarlty In a 

 data province. It Is not possible to quantify completely the estimation 



107 



