dimensionally smaller than the sonic path length. Since eddy size varies directly with z, the sonic
anemometer cannot be used below a certain height. The 10-cm sonic path length reduces the need
for the required fetch length. The zero-offset shift in the w signal caused by temperature changes on
the sonic transducer precludes the use of a CA27 sonic anemometer (SA) for absolute w measure-
ment. For the same reason, the CA27 SA cannot be used for long-flux averaging periods (i.e., 1 hr).
It has been reported that measurements of w′ over normal averaging periods compared favorably
with values reported by Biltoft and Gaynor (1987) using other SA models. The manufacturer also
claimed that the recent model of the CA27 produced less variability in w′ than those previously
reported by Tanner et al. (1985). The CA27 has a calibration of 1 m/s/V with a range of 4 m/s.
The tower and its attached SA are so designed that the tower axis is made vertical by adjusting
the 4 screws protruding from the corners of the concrete foundation. Ensuring that the CA27 SA is
perfectly vertical is accomplished by adjusting the ball-junction so that the air bubble-level on the
top of the CA27 SA is centered.
The temperature measurement device is a 13-m chromel-constantan thermocouple (Model
127, Campbell Scientific, Inc.) with a response of approximately 30 Hz with its signal calibrated to
4C/V. The absolute air temperature is not measured; instead, the measurement is referenced to the
temperature inside the base-mount of the CA27 (Fig. 3). Based on the manufacturer's data, the
reference junction has a 20-min interval time constant, which is expected to be adequate for most
flux-averaging periods.
Care should be taken to prevent damage to the fine-wire thermocouple, which is extremely
fragile. Care also should be taken to avoid any deformation of the silver dish on the acoustical
sensors as this can introduce large effects in the wind signal. It is important to note that the
transducers employed in the CA27 must not be exposed to wet environments.
The algorithms used by the 21X Micrologger to compute statistical measures on-line require the
summation of squares, cross products, and individual values. The covariance of signals x and y is
computed from N measurements of xi and yi (where superscript i indicates instantaneous value) as
Σxi yi Σxi Σyi
x ′ y′ =
-
,
N2
N
and the variance of x′ is given by
Σ xi2 Σxi 2
x′ =
-
2
.
N N
In this investigation, x ′ y′ and x ′2 represent 10-min time averages. For the 10-Hz measurement
rate, 6000 samples were needed to generate a single statistical value. In conjunction with using the
21X Micrologger to compute the 20-min variances of temperature, Tanner and Green (1989)
clearly show the effect of computation error and found the error is proportional to the number of
samples in the averaging period. They reported that as the ratio of the fluctuation to the mean signal
σ′ / x becomes smaller, the error grows more significant. The variation of relative error, i.e., the
x
ratio of calculated σx′c to true standard deviation σ x′ as a function of σ x′ / x , was calculated by
t
superimposing the sinusoidal fluctuations of the known standard deviation upon different mean
values. The results are shown below:
σ xc
σx
′
′
Ratio of
Ratio of
σ x′
x
t
0.0050.01
24
0.010.02
21.3
0.020.03
1.31.1
0.05
<1.05
0.10
1.01
17