Least squares tomographic models of mantle P and S velocity structure
from travel times have shown large-scale variations correlated with su
rface tectonic features, as well as coherent structures in the lowermo
st mantle. The reliability of these global features of velocity models
depends on whether velocity throughout the features can be estimated
well simultaneously: one needs to be able to say with confidence that
a feature involving many voxels is likely to be real. We find a lower
bound on how wide a 95% simultaneous confidence region for least squar
es estimates of mantle P or S velocity perturbations must be, as a fun
ction of position in the mantle. Suppose (perhaps optimistically) that
summary ray travel time errors are independent with mean zero and sta
ndard deviation 0.25 s for direct P phases, and standard deviation 0.5
s for all other phases, regardless of distance. If summary travel tim
e residuals are fitted by least squares to a mantle model parametrized
by 4872 10-degrees by 10-degrees voxels, the half width of a 95% conf
idence region for P velocity (using 889,909 summary P rays, 84,046 sum
mary PP rays, 67,228 summary pP rays, 23,024 summary PcP rays, 11,239
summary PKPab rays, 47,227 summary PKPbc rays, and 141,843 summary PKP
df rays) ranges from +/-0.025 to +/-99.8 km s-1. It is wider than 0.19
3 km s-1 in half of the mantle and wider than 0.593 km s-1 in a quarte
r of the mantle, by volume. Under the same assumptions and using the s
ame parametrization, a 95% confidence region for S velocity based on 1
63,354 summary S rays, 13,781 summary SS rays, 7441 summary ScS rays,
and 12,494 summary sS rays ranges from +/-0.031 to +/-infinity km s-1,
is wider than 0.254 km s-1 in half of the mantle and is wider than 0.
554 km s-1 in a quarter of the mantle. With this ray set, parametrizat
ion, and error model, the 95% confidence region around the least squar
es estimate of P velocity includes the radially symmetric iasp91 model
for 86.6% of the mantle's volume. A 95% confidence region around the
least squares S estimate includes iasp91 for 88.9% of the mantle's vol
ume. On a global scale, the mantle's velocity structure is nearly cons
istent with the iasp91 radially symmetric model. Smaller voxels, more
realistic assumptions about the errors, more accurate modeling of the
physics, and three-dimensional structure outside the mantle would make
the confidence regions still wider.