United States
Environmental Protection
Agency
Atmospheric Sciences
Research Laboratory
Research Triangle Park NC 27711
Research and Development
EPA/600/S3-85/057 Sept. 1985
&ERA Project Summary
A Field Comparison of In Situ
Meteorological Sensors
J. C. Kaimal, J. E. Gaynor, P. L. Finkelstein, M. E. Graves, andT. J. Lockhart
Measurements of wind speed, wind
direction, and the vertical wind compo-
nent from five conventional in situ me-
teorological systems were compared
with similar measurements from a fast-
response sonic anemometer. The sys-
tems tested were an orthogonal three-
axis propeller anemometer, a light
bivane and cup anemometer, a bivane
propeller anemometer, a light cup and
vane with a vertical propeller, and a
vane-mounted propeller anemometer
with a vertical propeller. Computed ac-
curacy and field precision variables
measured by each system are pre-
sented. The response characteristics of
the sensors tested are discussed.
This Project Summary was devel-
oped by EPA's Atmospheric Sciences
Research Laboratory, Research Triangle
Park, NC, to announce key findings of
the research project that is fully docu-
mented in a separate report of the same
title (see Project Report ordering infor-
mation at back).
Introduction
It has recently become clear through
advances in both theoretical and experi-
mental meteorology, that improve-
ments in modeling the transport and
dispersion of pollutants will require on-
site measurements of the atmosphere.
This requirement has in turn generated
questions about our ability to make
such measurements. To help answer
thesequestionstheU.S. Environmental
Protection Agency sponsored at
NOAA's Boulder Atmospheric Observ-
atory (BAD) an experiment designed to
assess the ability of in situ and remote
sensors to measure the mean and
turbulent properties of the lower at-
mosphere. The tests were carried out
over a 3-week period in September
1982. They were designed and con-
ducted with the goal of gaining a
knowledge of the accuracy, precision,
and general performance characteris-
tics of a variety of meteorological
sensors that are commonly used in
environmental studies. The results
should prove valuable in designing
experiments, understanding data from
field studies, and interpreting the in-
herent limits of accuracy and precision
possible in transport and diffusion
models.
The BAO was chosen as the site for
the experiment because of the availabil-
ity of precise profile and turbulence data
from accurate fast-response sensors on
a 300-m tower, as well as comprehen-
sive data-logging facilities. Two cate-
gories of sensors were tested. One con-
sisted of lightweight in situ sensors of
the type that have been frequently used
in the recent past for boundary layer
studies. The other category consisted of
four commercially available Doppler so-
dars, with the capability to measure
wind speed, wind direction, and vertical
component of turbulence, all at various
heights above the ground. The sodar
comparison has been described in an
earlier report. This report deals only
with the in situ instrument comparison.
Description of the Field Experi-
ment
The sensors were selected to provide
a measure of turbulence, in addition to
mean speed and direction, at 10 m
above the ground. The systems com-
pared in this study (see Table 1) were
selected because they represent types
commonly used in meteorological moni-
toring programs relatedto air pollution.
-------
The selection was also made to include
all usual configurations of instruments
capable of describing three-dimen-
sional flow. The reference standard for
comparison was the BAO three-axis
sonic anemometer.
The instruments were mounted on
10-m towers, erected in a line to the
west of the 300-m BAO tower. Each
tower held a different instrument set.
The installation of the instruments was
intended to simulate the best practice
found in operational field programs and
not the best possible practice if money
were of no concern. Various levels of
quality control checks were used to
achieve the best practical results, in-
cluding daily review of instrument per-
formance. One benefit of the program
was the evaluation of the quality control
procedures, calibration checks, and op-
erational methods.
Data Acquisition, Processing,
and Analysis
The output signals from the anemom-
eters were connected through shielded
cables to their signal conditioners lo-
cated in an instrument trailer at the base
of the tower. The outputs of the signal
conditioners were sampled, digitized,
and processed by the BAO data acquisi-
tion system, along with all the signals
from the standard sensors on the 300-m
tower. The sonic anemometer signals
were processed in exactly the same
manner as all the other sonic anemom-
eter signals from the tower. All the
channels were sampled ten times per
second. The data acquisition computer
calculates in real time the means and
variances of samples of the outputs,
converts them into meteorological
units, and prints out the results at the
end of every 20-min averaging period.
For selected periods the complete time
series (10 samples/s) can be recorded
on magnetic tape. The rest of the time
only 10-s non-overlapping averages are
saved. From each of the 10-s periods a
grab sample (one of these instanta-
neous values) is also saved. These grab
samples are needed especially for com-
puting standard deviations since they
retain the high-frequency information
inherent in the data sample. Thus, we
have 120 grab samples for calculating
each 20-min standard deviation.
Scalar averages and standard devia-
tions were computed for all variables.
For U-V-W and SONIC, the instanta-
neous direction was computed from the
measured horizontal wind components.
For the 20-min periods in which there
were major shifts in wind direction, and
for which the scalar average direction
would be meaningless, the wind direc-
tion values were edited out of the data
set. Data were excluded when the wind
was parallel to the line of the towers to
avoid any shadowing effect. Care was
also taken to ensure that misleading
values were not generated when the
wind was from the north. Northerly
winds cast a shadow on the SONIC w
axis and also caused errors in the vane
direction readings from their discontin-
uities at 0 degree. The entire data set
was also carefully edited to remove spu-
rious values caused by instrument mal-
function, line noise, birds, rain, and the
like.
The intercomparisons of first and sec-
ond moments (means and standard de-
viations) are made against the SONIC
measurements, which are considered
the reference or true values. The statis-
tics of comparison, the bias (b) and
comparability (c), are defined as
(1)
C = Kl
(2)
where
N = number of observations
YJ = ith observation of the test instru-
ment
S, = ith observation of the reference
instrument
The field observations for the experi-
ment extended from 1 to 22 September.
Two periods were selected for record-
ing data at the 10 samples/s rate. The
first period, 0800-1540 MST, recorded
on 9 September, represents typical
convection conditions encountered at
the BAO, while the second period,
Table 1. Instrument Selection Summary
1600-2300 MST recorded on 18 Sep-
tember, represents neutral and stable
conditions. These data are used to
explore details of the sensor response
to the turbulence in the flow.
Operational Maintenance and
Quality Control
A field monitoring program from
which data of known quality are re-
quired should be planned with a quality
assurance effort aimed toward that
goal. The quality assurance plan for this
experiment required that the data-
gathering period be bracketed by cali-
brations and that an independent audit
be conducted during the field program.
The ideal time for an audit is at the be-
ginning of the field measurement pro-
gram so that errors can be corrected be-
fore too many data have been collected.
An independent audit serves two pur-
poses: (1) The expectation that some-
one else will check to be sure things
have been done right inspires the per-
son responsible for the operation of the
system to be sure everything is done
and documented within the planned
methodology, and (2) the auditor gives
another layer of authority to the claim
that data are valid and representative.
His report and the records kept for cali-
bration are an important part of the data
history.
The initial calibration documented the
response of the system elements to arti-
ficial known conditions, such as rate of
rotation for speed shafts and position
change for wind vane shafts. Because
we monitored 20-min-average data on
hard copy continuously, we were able
to detect consistent differences in wind
direction that led us to correct the ori-
entation before the measurement
program began officially. Normally a
monitoring program does not have re-
dundant sensors, so differences are not
there to see. It would be prudent to have
the orientation checked by a different
person than the one who did the orien-
tation initially. This may be the most im-
portant field task and perhaps the most
difficult.
Tower No.
Designation
Sensor Type
Manufacturer
1
2
3
4
5
6
U-V-W
C-BIV
P-BIV
SONIC
C-V-W
P-V-W
Gill UVW propeller
Bivane and cup
Propeller bivane
Sonic anemometer
Cup and vane
Vertical propeller
Propeller vane
Vertical propeller
R. M. Young Co.
Meteorology Research, Inc.
R. M. Young Co.
Applied Technology, Inc.
Climatronics, Inc.
R. M. Young Co.
R. M. Young Co.
R. M. Young Co.
-------
The final calibration marks the termi-
nation of the data-gathering period. It
remains for the data themselves to sup-
port the claim that the instruments per-
formed the same between calibrations
as they did during calibrations. When
collocated instruments are used, the
data may show that the instruments are
performing more accurately than the
calibration suggests. Calibration meth-
ods have their own assumptions and
uncertainties, and it is difficult to know
sometimes whether the error reported
in a calibration comes from the mea-
surement instruments or from the cali-
brator. In field calibrations of meteoro-
logical instruments, it is unusual to have
the luxury of test equipment ten times
as accurate as the instrument, as is usu-
ally required in routine quality control
programs. The best quality control in a
field monitoring program is to have an
experienced meteorologist looking at
the data streams in as near real time as
the budget will allow.
During this experiment initial and
final calibrations were conducted by
one of the authors (Lockhart). The inde-
pendent audit was conducted by Dr.
Fred V. Brock of the National Center for
Atmospheric Research. We believe the
quality of data collected for this study to
be as good as anything one can achieve
under field conditions. Accuracies and
precisions reported are illustrative of
what one might expect with careful at-
tention given to calibration and installa-
tion of the sensors.
Results
Comparison of Wind Measure-
ments
Speed readings from U-V-W and
SONIC are scalar averages of the instan-
taneous resultant speeds from two hori-
zontal velocity measurements. All the
other anemometers measure scalar
speed directly. Bias and comparability
of wind speed for the five anemometer
systems compared are given in Table 2.
The most striking fact here is the
substantial negative bias in U-V-W.
The other sensors show a smaller bias.
The non-cosine response in the pro-
pellers is the most likely reason for this
underestimation. No corrections were
made at any stage in the acquisition
and processing of the data to correct for
this effect. Another point of interest is
the large positive bias in P-BIV. Upon
reviewing the calibration data, we
found that this instrument had a small,
uncorrected error of +0.3 m/s, which
accounts for this positive bias. P-V-W,
which has a propeller similar to P-BIV,
but attached to a vane fixed in the
horizontal plane, had a rather small
bias.
Table 2. Bias and Comparability for
Wind Speed
Instrument b (m/s) c (m/s) N
U-V-W
C-BIV
P-BIV
C-V-W
P-V-W
-0.43
-0.13
0.33
-0.13
-0.16
0.53
0.35
0.48
0.36
0.34
1279
760
760
760
760
b = bias
c = comparability
N = number of observations
The two cup anemometers compared
in this experiment are very different in
size. C-BIV cups were rather large, while
the C-V-W cups were small and light.
However, they showed almost identical
performance in measuring wind speed,
both in terms of bias and comparability.
The latter figure is a measure of the de-
gree of scatter in the data. The results
indicate that the overspeeding problem
attributed to cups is not a problem here.
Comparison of Wind Direction
Measurements
Bias in wind direction measurements
is usually an indication of our inability
to align the sensor properly, since we
have no reason to suspect that the
vanes do not line up with the wind. In
the experiment, great care was taken to
line up the instruments, and they were
checked several times by independent
observers under field conditions. There-
fore, these bias data provide a measure
of the expected absolute accuracy pos-
sible from any wind direction observa-
tion under normal conditions. The bias,
b, and comparability, c, for the mea-
surements of wind direction, 6, and
standard deviation of the wind direc-
tion, cre, are given in Tables 3 and 4.
With the exception of P-BIV and
P-V-W, the observed biases in Table 3
are larger than the sensor resolution.
The comparability figures in the same
table are somewhat disturbing. They in-
dicate that the scatter in the observa-
tions is about 5°. This scatter is a mea-
sure of the confidence one can place in
any one, or a small group, of wind direc-
tion observations. These results should
be of interest, and concern, to those in-
volved in such fields as diffusion model
verification, where a 5° difference in
wind direction can cause a considerable
difference in the prediction of ground
level concentrations.
The values of CTB in Table 4 show, on
average, good agreement of ae with
SONIC. The scatter in the measure-
ments represented by c in Table 4, is
small, about 3°. There is not much dif-
ference between instruments. The light-
est vanes (in C-BIV and C-V-W) did have
the least scatter with respect to the ref-
erence. It is to be expected that all these
instruments will measure a slightly
smaller
-------
Table 5. Bias and Comparability for aw
Instrument b (deg) c (deg) N
u-v-w
C-BIV
P-BIV
C-V-W
P-V-W
-0.08
0.06
0.01
-0.07
-0.07
0.10
0.11
0.06
0.08
0.09
840
809
803
782
780
in the magnitude of b and c for the pro-
pellers in U-V-W, C-V-W, and P-V-W is
reassuring.
Comparison of o^ Values
The wind elevation angle () was
measured directly by the two bivanes
and obtained indirectly from the wind
components measured by the U-V-W
system. These measurements and their
standard deviations, a,,,, were compared
with the 0-4, computed from SONIC.
Agreement is good according to the val-
ues in Table 6.
Table 6. Bias and Comparability for a^,
Instrument b (deg) c (deg) N
U-V-W
C-BIV
P-BIV
-0.30
0.65
-0.81
2.07
1.80
1.85
890
804
806
The slight positive bias of the lighter
bivane is attributed to underdamping.
That the U-V-W instrument agrees so
closely with the bivanes is reassuring in
view of the fact that it is used so fre-
quently for turbulence measurements.
Comparison of Sigma Meters
There are various devices on the mar-
ket that purport to compute an effective
standard deviation from input analog
signals. These have frequently been
used with meteorological equipment,
usually wind vanes, to produce a cre
value for preselected averaging times.
Early sigma meters processed analog
signals, using variations on an R-C cir-
cuit, to estimate the standard deviation
of the signal. With the availability of mi-
croprocessor chips, digital computation
is now widely used. In this experiment
two sigma meters, denoted A (analog)
and D (digital), were tested. Several dif-
ferent input signals were used to see if
any significant differences could be de-
tected; in each case the two sigma
meters saw the same input at the same
time.
Standard deviations estimated by the
meters were compared with o- values
computed by the BAO data-logging sys-
tem from the same input signals. No
comparison is made with the sonic
anemometer in this evaluation.
The results of this comparison
(Table 7) indicate that for 6, 4>, and w
inputs the analog sigma meter signifi-
cantly underestimates the standard de-
viation. Both systems show consider-
ably more scatter for ae than for a$ or
aw, but the analog system shows scatter
almost twice as large. The scatter is ap-
proximately equivalent for a^,; for aw
the analog system performs slightly
better than the digital in terms of both
bias and scatter. Not surprisingly the
performance of both systems deterio-
rates with increasing levels of turbu-
lence. The reasons for this seem clear
for the analog system, but are less so
for the digital. In any event, the trend
toward more digital electronics and on-
site digital data processing and logging
should produce improvements in the
digital meters and in the development
of new algorithms for real-time analy-
ses of meteorological data.
Sensor Response to Wind Fluc-
tuations
An important objective of this study
was to determine how well our sensors
respond to turbulent fluctuations in the
flow. Published data on response
lengths and distance constants enable
us to derive response functions indi-
rectly. The bias and comparability
statistics presented in the foregoing
pages offer additional clues. A direct ap-
proach is to compare spectra for typical
flows from the candidate sensor and a
reference sensor, such as the SONIC.
Toward this end, we had recorded data
from all our sensors at the full 10 sam-
ples/s rate on two days, 9 and 18 Sep-
tember. Times series from these
records were subjected to spectrum
analysis procedures. Some typical plots
of frequency-weighted spectral intensi-
ties as functions of frequency, n, are
presented in Figs. 1-3.
There is, in general, good agreement
between the sensor and the SONIC
spectra at middle and low frequencies.
At the high-frequency end, the SONIC
spectra fall off at a rate consistent with
predictions for the inertial subrange,
which is less steep than the falloff of the
other sensor spectra. The sensor re-
sponse starts to separate from the
SONIC spectrum at approximately the
same wavelength in all three stabilities.
(We assume wavelength \ = U/n, fol-
lowing Taylor's hypothesis.) The verti-
cal arrows represent our best estimate
of this separation point, which corre-
sponds to wavelengths of 4.4 m for C-
BIV, 7.0 m for P-BIV, and 32 m for C-V-
W.
On closer examination, one finds a
tendency in both bivanes to overesti-
mate spectral contributions in the
middle- to low-frequency range in un-
stable and neutral air, and to under-
estimate contributions in the middle- to
high-frequency range in very stable air.
For the vertical propeller, the under-
estimation in stable air is more exten-
sive, almost a factor of 2 across the en-
tire spectral bandwidth. Intermittent
stoppage of the propeller, when the
wind drops below its S response
threshold, can produce such a depres-
sion in spectral levels. (The effect would
be comparable to the effect of adding
zeros to a time series. When time series
are thus expanded, a correction factor is
usually applied to restore the spectrum
to its proper level.)
Transfer functions derived from com-
posite plots of spectra from each sensor
normalized by the SONIC spectra are
presented. For variable x, the transfer
function Tx(\.) is defined as
TX(X) =
[Sx(\)]sensor
[SX(X)]SONIC
(3)
where Sx(\) is the spectral estimate at
wavelength X.
The overestimation in bivanes shows
up very clearly. Underdamping explains
Table 7.
Input Signal
6 (P-BIV)
6 (C-BIV)
<)> (P-BIV)
<|> (C-BIV)
w (U-V-W)
Bias and Comparability of
and the Wind Sensors
Type of v Meter
A
D
A
D
A
D
A
D
A
D
Standard Deviations
b (deg) (m/s)
-3.6
-1.5
-3.8
-1.2
-0.8
0.7
-1.3
0.6
-0.05
0.12
Computed by the Sigma
c (deg) (m/s)
10.3
5.1
10.8
5.5
1.3
1.3
2.1
1.9
0.06
0.14
Meters
N
595
653
345
354
480
479
88
88
157
157
-------
I
to
70'
70°
10-
10-
10'
I I I I
Unstable (U = 3.4 m/s)
SONIC
C-V-W
I
II I
Neutral (U= 6.5 m/s)
I
I
I I T
Stable (U= 2.1 m/s)
v_**
70"4 70"
70"
70° 701
70"
70'2 70"
n(Hz)
10° 701
70"'
70-'
70"
70°
70'
Figure 1. Spectra of w from SONIC and C- V- W for three stability conditions.
or underdamping. The close agreement
the increase near \c, the cutoff fre-
quency, but the continued overestima-
tion at lower frequencies remains a puz-
zle.
Of the two bivanes, C-BIV has the bet-
ter wavelength response, presumably
because of its lighter construction. De-
pendence on the slower cup anemome-
ter for its wind speed information has
had little effect on C-BIV's w response.
(This conclusion is supported by the fact
that the transfer function for w follows 4>
and not the speed.) These favorable re-
sults notwithstanding, we had found
earlier that C-BIV's bias for
-------
ro1
70°
are
measured with reasonable accuracy
(scatter of ±3° in o9, ±2° in o^). The
scatter increases linearly with mag-
nitude (50° in cre, 15° in o^).
3. When transfer functions for w are
compared, a clear difference
emerges between the bivanes and
the propellers. The bivanes tended to
overestimate w but also responded
to wavelengths as short at 4.4 m. The
propellers did not overestimate w,
but neither did they respond well to
wavelengths shorter than 32 m. In-
termittent stoppage of the propeller
was probably responsible for the
drop in spectral levels observed in
the light wind stable case. The re-
sponse to <|> is the same as for w.
4. Sigma meter performance degrades
with increasing turbulence. The dig-
ital meter shows smaller bias and
less scatter than the analog meter in
most cases.
TirU. S. GOVERNMBff PRINTING OfflCE:1985/5»-l 11/20709
-------
J. C. Kaimal and J. E. Gay nor are with National Oceanic and Atmospheric
Administration's (NOAA) Wave Propagation Laboratory, Boulder. CO 80303;
P. L. Finkelstein (also the EPA Project Officer, see below) is with Atmospheric
Sciences Research Laboratory, Research Triangle Park, NC 27711; M. E.
Graves is with Northrop Services, Inc., Research Triangle Park, NC 27709; and
T. J. Lockhart is with Meteorological Standards Institute, Fox Island, WA
98333.
The complete report, entitled "A Field Comparison of In Situ Meteorological
Sensors," (Order No. PB 85-196 988; Cost: $13.00, subject to change) will be
available only from:
National Technical Information Service
5285 Port Royal Road
Springfield. VA22161
Telephone: 703-487-4650
The EPA Project Officer can be contacted at:
Atmospheric Sciences Research Laboratory
U.S. Environmental Protection Agency
Research Triangle Park, NC27711
United States
Environmental Protection
Agency
Center for Environmental Research
Information
Cincinnati OH 45268
Official Business
Penalty for Private Use $300
EPA/600/S3-85/057
Q000329
u s
------- |