The scale dependence of velocity variances is studied using data collected
from a grassland site, a heather site, and four forested sites. The depende
nce of velocity variances on averaging time, used to define the fluctuation
quantities, is modeled. The crosswind velocity variance is emphasized, bec
ause it is more difficult to model than the other two components and is cru
cial input for applications such as dispersion modeling. The distinction be
tween turbulence and mesoscale variances is examined in detail. Because mes
oscale and turbulence motions are governed by different physics, meaningful
study of the behavior of velocity variances requires adequate separation o
f turbulence and mesoscale motions from data. For stable conditions, the ho
rizontal velocity variances near the surface exhibit a spectral gap, here c
orresponding to a very slow or nonexistent increase of variance with increa
sing averaging time. This "gap region,'' when it occurs, allows separation
of mesoscale and turbulence motions; however, the averaging times correspon
ding to this gap vary substantially with stability. A choice of typical ave
raging times for defining turbulent perturbations, such as 5 or 10 min, lea
ds to the capture of significant mesoscale motions for very stable conditio
ns and contributes to the disagreement with turbulence similarity theory. F
or unstable motions, the gap region for the horizontal velocity variances s
hrinks or becomes poorly defined, because large convective eddies tend to "
fill in'' the gap between turbulence and mesoscale motions. The formulation
developed here allows turbulence and mesoscale motions to overlap into the
same intermediate timescales. The mesoscale variances are less predictable
, because a wide variety of physical processes contribute to mesoscale moti
ons. Their magnitude and range of timescales vary substantially among sites
. The variation of the behavior of turbulence variances among sites is sign
ificant but substantially less than that for the mesoscale motions.