Experimentally obtained time coherence has traditionally been interpre
ted as streamwise one-dimensional spatial coherence through Taylor's h
ypothesis. We calculate corrections to the high-wavenumber part of the
coherence to account for the errors caused by the deviation from Tayl
or's hypothesis in high-intensity turbulent flows. The small-scale tur
bulence is assumed to be frozen and convected by a fluctuating convect
ion velocity. Both Lumley's two-term approximation and the Gaussian ap
proximation are used in the calculations. In general, we find that the
coherence for cross-stream separations is significantly overestimated
by the direct use of Taylor's hypothesis, the error increasing with w
avenumber; that for streamwise separations is underestimated. The anal
yses are compared with cross-stream coherence measurements in the atmo
spheric surface layer. Our results indicate that predictions from Luml
ey's approximation yield better agreement with experimental data for c
ross-stream separations than those from the Gaussian model. Our study
suggests that reliable measurement of two-point spatial coherence can
be achieved only for scales not too small compared to the sensor separ
ation.