We perform inversions of gravity data (geopotential model EGM96) and seismi
c tomography model (SlBRLBM) for the scaling factor (zeta), which relates r
elative density anomalies to relative S-wave velocity anomalies. The gravit
y data and tomographic model are anti-correlated below continents down to a
depth of z = 200 km. This anti-correlation is not present below oceans. Ex
cept for smoothness, which is controlled by a damping factor, no a priori i
nformation is added to the inversion. Data are filtered between degrees l =
11 and l = 16 of the spherical harmonic expansion. This spectral window is
well suited for the study of intermediate-size (2000-4000 km) anomalies in
the uppermost mantle. Calculations are made separately for sub-continental
and sub-oceanic mantle. The sub-continental and sub-oceanic scaling factor
s are significantly different at depths shallower than 260 km. In both case
s, the magnitude of zeta is around 0.05. The sub-continental scaling factor
has a positive root down to,z = 220 km, whereas the sub-oceanic scaling fa
ctor yields positive values down to z = 140 km only. At depth shallower tha
n 350 km, models of zeta do not depend on the damping factor or the viscosi
ty model. At depths greater than 350 km, the resolution of zeta (z) decreas
es significantly and low degrees (l = 2-1) add information from large-scale
anomalies and from the lower mantle. As a result, the shape and values of
zeta for l = 2-16 and l = 11-16 are significantly different at depths great
er than 350 km. A possible explanation of the discrepancies between the sub
-continental and sub-oceanic scaling factor is that intermediate-scale anom
alies are more important in the continental uppermost mantle than in the oc
eanic uppermost mantle. (C) 2001 Elsevier Science B.V. All rights reserved.