number of available profiles, the NODC archival XBT file was searched for suitable tempera- 

 ture profiles. These profiles were then converted to equivalent sound speed profiles using 

 annual mean salinity profiles from the Nansen cast data at each location and were inter- 

 polated for sound speed at standard depths. Even though the XBT provides less absolute 

 temperature and resultant sound speed accuracy than the Nansen cast, blending the two 

 profile sources is justified in this analysis because there is stronger emphasis on the statistical 

 distribution of positive and non-positive gradients than on absolute sound speed values. The 

 XBT data contributed significantly to the sample size for the region East of Singapore, but 

 were not used in the North Sea where ample Nansen cast stations were available. Some 

 additional profile editing was performed to remove redundant sampling in a few instances 

 where repeated observations were made at the same location and time. 



An initial attempt was made to classify or "pigeonhole" profiles by vertical shape or 

 type. This process is impractical because it requires the selection criteria to be redefined for 

 each data set and because there are few guidehnes for acoustic significance of the various 

 sound speed structures. The overall sound speed gradient computed from the surface to 

 specified depths is of basic acoustic significance in defining the expected mode of propaga- 

 tion, therefore, a statistical approach to gradient distributions was apphed to the data. 

 Surface energy exchange (i.e., heat and wind) and horizontal and vertical advection can 

 affect the vertical sound speed gradient. The presence of the sea floor influences advection, 

 and, therefore, can also influence the gradient. To avoid biasing the statistics, shallow 

 profiles that covered less than 80 percent of the water column were eliminated in all in- 

 stances except in the region East of Singapore because of the limited number of observa- 

 tions. Deep profiles would be expected to display any bottom-related effects. 



Two methods of statistical tabulation were used for the vertical sound speed 

 gradients. Examples of each are provided in the individual site analyses in the following 

 section of this Appendix. Gradients are computed in units of meters/second per meter. The 

 first approach produced a tabular listing by season of the number of positive and non- 

 positive gradients observed in the data set from the surface to each successive standard 

 depth. It also provided statistics on gradient strengths. These tables provided information 

 on whether the overall gradients were positive or non-positive and some measure of 

 correlation between gradient and depth of observations. However, because the identity 

 of individual profiles is lost in this statistical summary, it is not an adequate guide to the 

 selection of representative profiles for acoustic model inputs. 



The second statistical approach preserves the identity of each profile while providing 

 a measure of the distribution of positive and non-positive sound speed gradients. Many 

 profiles in complex shallow water environments may contain both positive and negative 

 gradient layers at different depths in the water column. This is especially noticeable during 

 the spring and fall transition seasons. To classify a single profile as either a positive or 

 non-positive gradient profile requires a definition based on the expected effect of the 

 overall shape on the primary mode of acoustic propagation. The definition also must 

 remain relatively simple if it is to be applicable to all shallow water environments. The 

 basic definition chosen for this analysis will classify a profile as positive if the depth of 

 the absolute maximum observed sound speed value is greater than 10 m, and non-positive 

 if the absolute maximum is observed within 10 m of the surface. This allows a rather large 

 number of possible profile shapes to be grouped together in either category. However, as 

 a first order profile type separator for an assumed near surface sound source, this sorting 

 approach works well. 



70 



