INTRODUCTION 



The Bathythermograph (BT) file is one of the more voluminous data files 

 compiled by the National Oceanographic Data Center (NODC). It consists of 

 temperature profile data visually digitized at five meter intervals from ozalid 

 copies of bathythermograph slides (Figure 1). Although tfie size of an ozalid 

 BT slide copy may vary, a representative scale is about two inches to 300 

 meters. Accordingly, a profile is digitized at 0.03 inch intervals on the BT 

 graph scale. This fine resolution is ostensibly recommended for representing 

 subtle temperature gradient irregularities. However, certain undesirable 

 effects accompany such fine scale digitization intervals, namely: (1) the 

 temperature gradient, which is calculated using differences between adjacent 

 data points, is overly sensitive to slight variations caused by reader error and 

 truncation, and is accordingly unreliable; (2) a considerable bulk of computer 

 memory is required to store profile data points (as many as sixty data points 

 for the standard scale BT profile). 



The computer algorithm described in this paper has been developed to 

 reduce an NODC digital BT profile to a select set of regression lines (Figure 2). 

 The objective of the algorithm is to represent a BT profile with economy and 

 precision (negligible departure from the original BT data points). It has the 

 feature that it provides reasonable estimates of temperature gradient between 

 regression points (points of intersection between neighboring regression lines). 

 This algorithm may, with appropriate modifications, be applied to the reduction 

 of any kind of digital profile data set. For example, it may be used to reduce 

 massive STD profile data sets to manageable proportions without compromising 

 their fine scale descriptions of profile gradient. 



PROBLEM DEFINITION 



A. Preserve Features of Profile Gradient 



A BT profile trace may misrepresent not only real temperature values 

 with depth, but real depth and temperature gradient values as well . The 

 accuracy of a mechanical BT profile curve may be further compromised by 

 mechanical and thermal inertia which causes hysteresis in the BT trace between 

 sensor descent and ascent. Perhaps paradoxically, the hysteresis provides con- 

 vincing evidence that the BT trace truthfully describes real irregularities in the 

 temperature profile. Even in cases where hysteresis is pronounced, as a general 

 rule, the humps and bumps registered in a trace during sensor descent match 

 humps and bumps registered during sensor ascent (Figure Kb)). It is a challenge 

 to represent a BT trace digitally without losing the evidence of these subtle 

 humps and bumps. 



