The objective of this paper is to minimize the geoid undulation errors
by focusing on the contribution of the global geopotential model and
regional gravity anomalies, and to estimate the accuracy of the predic
ted gravimetric geoid. The geopotential model's contribution is improv
ed by (a) tailoring it using the regional gravity anomalies and (b) in
troducing a weighting function to the geopotential coefficients. The t
ailoring and the weighting function reduced the difference (1sigma) be
tween the geopotential model and the GPS/levelling-derived geoid undul
ations in British Columbia by about 55% and more than 10%, respectivel
y. Geoid undulations computed in an area of 40-degrees by 120-degrees
by Stokes' integral with different kernel functions are analyzed. The
use of the approximated kernels results in about 25 cm (sigma) and 190
cm (maximum) geoid erroers. As compared with the geoid derived by GPS
/levelling, the gravimetric geoid gives relative differences of about
0.3 to 1.4 ppm in flat areas, and 1 to 2.5 ppm in mountainous areas fo
r distances of 30 to 200 km, while the absolute difference (1sigma) is
about 5 cm and 20 cm, respectively. A optimal Wiener filter is introd
uced for filtering of the gravity anomaly noise, and the performance i
s investigated by numerical examples. The internal accuracy of the gra
vimetric geoid is studied by propagating the errors of the gravity ano
malies and the geopotential coefficients into the geoid undulations. N
umerical computations indicate that the propagated geoid errors can re
asonably reflect the differences between the gravimetric and GPS/level
ling-derived geoid undulations in flat areas, such as Alberta, and is
over optimistic in the Rocky Mountains of British Columbia.