great care is taken, especially with navigation errors, the measurement error can 

 be reduced to the level of 1 mgal or so. 



In using data of varied accuracy, several alternative procedures are possible. 

 One can discard all the older data which is considered less accurate. Such a 

 procedure would seriously reduce the number of gravity measurements used for the 

 study. A second procedure is to incorporate all data irrespective of accuracy. 

 This procedure was followed by Talwani and LePichon (1959) in constructing averages 

 over 1x1° and 5x5° squares in the Atlantic Ocean. However, the accuracy of the 

 averages is certainly degraded by including data of low accuracy. 



Substantial improvements in obtaining the overall gravity field from the 

 gravity measurements can be made if advantage can be taken of the following two 

 facts: 1. The errors in gravity observations are largely systematic rather 

 than random, and 2. Short wavelength variations in gravity are correlated with 

 topographic variations. 



The errors in gravity measurements are influenced strongly by sea condition, 

 and the ship's heading with respect to sea and swell. Most of the measurements 

 in the western North Atlantic have been made by Graf Askania sea gravimeters . 

 Measurements made with this meter are subject to the cross coupling error (LaCoste 

 and Harrison, 1961) . This error was a principal reason for the inaccuracies of 

 surface ship gravity measurements until about 1965 when cross coupling computers 

 (Talwani et al., 1966) which provide real time cross coupling correction came into 

 regular use. Figure 2 shows how the value and sign of the cross coupling error varies 

 with ship's heading (with respect to sea and swell) and Figure 3 shows how, at a 

 constant heading, the value of the cross coupling error changes slowly as the sea 

 state changes. Although the magnitudes of the errors in some of the examples in 

 these figures are atypically large, having been obtained under rough sea conditions, 

 nevertheless these figures serve to illustrate the manner in which these errors 

 change. In cases when one ship's track is crossed by other ship tracks, recognition 



23-5 



