IMPORTANT COPYRIGHT INFORMATION
The following PDF article was originally published in the Journal of the Air & Waste
Management Association and is fully protected under the copyright laws of the United
States of America. The author of this article alone has been granted permission to copy
and distribute this PDF. Additional uses of the PDF/article by the author(s) or recipients,
including posting it on a Web site, are prohibited without the express consent of the Air
& Waste Management Association.
If you are interested in reusing, redistributing, or posting online all or parts of the
enclosed article, please contact the offices of the Journal of the Air & Waste Management
Association at
Phone: +1-412-232-3444, ext. 6027
E-mail: [email protected]
Web: www.awma.org
You may also contact the Copyright Clearance Center for all permissions related to the
Journal of the Air & Waste Management Association: www.copyright.com.
Copyright © 2006 Air & Waste Management Association
TECHNICAL PAPER
ISSN:1047-3289 J. Air & Waste Manage. Assoc. 59:1370 –1378
DOI:10.3155/1047-3289.59.11.1370
Copyright 2009 Air & Waste Management Association
A Comparison of Lagrangian Model Estimates to Light
Detection and Ranging (LIDAR) Measurements of Dust
Plumes from Field Tilling
Junming Wang
Department of Plant and Environmental Sciences, New Mexico State University, Las Cruces, NM
April L. Hiscox
Department of Environmental Sciences, Louisiana State University, Baton Rouge, LA
David R. Miller and Thomas H. Meyer
Department of Natural Resources Management and Engineering, University of Connecticut,
Storrs, CT
Ted W. Sammis
Department of Plant and Environmental Sciences, New Mexico State University, Las Cruces, NM
ABSTRACT
A Lagrangian particle model has been adapted to examine
human exposures to particulate matter ⱕ 10 ␮m (PM10) in
agricultural settings. This paper reports the performance
of the model in comparison to extensive measurements
by elastic LIDAR (light detection and ranging). For the
first time, the LIDAR measurements allowed spatially distributed and time dynamic measurements to be used to
test the predictions of a field-scale model. The model
outputs, which are three-dimensional concentration distribution maps from an agricultural disking operation,
were compared with the LIDAR-scanned images. The peak
cross-correlation coefficient and the offset distance of the
measured and simulated plumes were used to quantify
both the intensity and location accuracy. The appropriate
time averaging and changes in accuracy with height of
the plume were examined. Inputs of friction velocity,
Monin–Obukhov length, and wind direction (1 sec) were
measured with a three-axis sonic anemometer at a single
point in the field (at 1.5-m height). The Lagrangian model
of Wang et al. predicted the near-field concentrations of
dust plumes emitted from a field disking operation with
an overall accuracy of approximately 0.67 at 3-m height.
Its average offset distance when compared with LIDAR
measurements was approximately 38 m, which was 6% of
IMPLICATIONS
A Lagrangian model has been demonstrated to have the
potential to estimate near-field PM10 dispersion from agricultural disking operations. The major model improvements
over traditional plume models are that it can simulate moving sources and plume meander. Therefore this technique
can be used to provide accurate PM10 dispersions for other
agricultural operations and other moving sources (e.g.,
road dust).
1370 Journal of the Air & Waste Management Association
the average plume moving distance during the simulation
periods. The model is driven by weather measurements,
and its near-field accuracy is highest when input time
averages approach the turbulent flow time scale (3–70
sec). The model accuracy decreases with height because of
smoothing and errors in the input wind field, which is
modeled rather than measured at heights greater than the
measurement anemometer. The wind steadiness parameter (S) can be used to quantify the combined effects of
wind speed and direction on model accuracy.
INTRODUCTION
Particulate matter (PM) is regulated by the U.S. Environmental Protection Agency (EPA) as part of the National
Ambient Air Quality Standards. Eulerian and Lagrangian
models are widely used to simulate transport of PM and
other pollutants.1–15 Eulerian models for estimating scalar
transfer by turbulence have been limited by their inability
to accurately model the dispersion of material from nearfield sources.16 Lagrangian models explicitly consider the
diffusion of material in the near- and far-field.16 Lagrangian models have been used to detail the variability of the
subgrid concentrations in Eulerian grids to examine human exposure to toxins.15 This ability to detail spatial
variability is quite valuable in agricultural settings and
was the reason Wang et al.17 adapted a Lagrangian model
for agriculture dust dispersion.
In agricultural, construction, and other settings
where soil is frequently disturbed, little is known about
the frequency and intensity of aerosol doses received at
short distances away from the disturbance because of the
transient nature of local dust plumes and the difficulties
in making accurate concentration measurements in dynamic plumes.18 Thus, specific field, crop, and weatherrelated best management practices to minimize dust exposure in agriculture have not been defined. The authors
Volume 59 November 2009
Wang, Hiscox, Miller, Meyer, and Sammis
North
Experimental
Field #1
100 m
3D Sonic Anemometer
Experimental
Field #2
Horizontal
scan angle
~ 60°
LIDAR
Figure 1. Experimental farm with field and sensor locations.
have adapted the Lagrangian particle model of Wilson
and Shum6 to estimate the dynamic plumes and aerosol
concentrations of PM ⱕ 10 ␮m (PM10) generated by agricultural field operations.9 The authors hope to use this
model to evaluate management practices across the entire
range of field crops, soils, moisture, weather conditions,
and locations in the United States and elsewhere.
Before this can be done, the model’s accuracy and use
protocols must be established. Hiscox et al.19 used an
aerosol light detection and ranging (LIDAR) to accurately
measure dynamic dust concentrations and calculate nearfield human exposures from a field disking operation. The
purpose of this paper is to compare these direct LIDAR
measurements of dust plume movement and dynamic
concentrations to these model-generated measurements
to quantify the accuracy and precision of the model
predictions.
The wind statistics inputs (e.g., friction velocity,
Monin–Obukhov length, and wind direction) to a Lagrangian model are traditionally time averages of 30 min
to several hours.3,6 The applications presented here are at
the field scale or, more formally, in the “near-field” where
the “persistence” causes the particle trajectories to be
nearly straight lines.20 The authors hypothesize that
short-time (e.g., sec) wind statistics inputs will improve
the model performance in the near-field, when compared
with long-time inputs, by capturing more accurate highfrequency information. There is little literature available
to report Lagrangian model performance using short-time
inputs.
The first objective of this paper is to quantify the
performance accuracy of the Lagrangian model (Wang
et al.17) of dust emissions from an agricultural field operation. Secondly, the response of the modeled plumes to
variations in the wind steadiness is examined. Finally, the
model performance of short- (1 sec) versus long-time (3,
30, and 60 min) inputs was quantified.
Volume 59 November 2009
METHODOLOGY
Model Description
A dynamic Lagrangian, local field-scale model of dust
dispersion from an agricultural disking operation was developed and presented in Wang et al.17 Distribution of
aerosol particles of different sizes that remained airborne
after a few meters was measured in this experiment by
Holmén et al.18 They showed that over 96% of the airborne material generated in this disking operation was
equal to or smaller than 10 ␮m. (Field-scale and near-field
are used interchangeably in this paper and mean that the
simulation domain is ⬍1000 m in the horizontal direction and ⬍100 m in the vertical direction.) The atmospheric inputs for the model are the 1-sec wind statistics:
friction velocity, u* (m 䡠 sec⫺1); Monin–Obukhov length,
L (m), and wind direction (degrees). In each time step, the
model simulates particle flights in three dimensions on
the basis of the wind turbulences calculated from the
wind statistics and a Markov chain algorithm. The concentration (mg 䡠 m⫺3) at a location at time t (sec) is estimated by calculating the mass of the particles in a unit
volume cube at the location at time t (details are in Wang
et al.17).
Field Experiments
The model was used to simulate the dynamic dust dispersion experiments of disking operations in an irrigated
cotton field near Las Cruces, NM, described by Holmén et
al.18 and Hiscox et al.19 The experimental field was approximately 2.8 ha. Figure 1 presents an aerial photo of
the site annotated with the location of the field and
instrumentation. Experimental field 1 was used in the
spring of 2005 to measure dust emissions from field preparation and planting operations. On March 31, 2005, the
field was worked for disking operations. The field was a
mixture of Armijo clay loam and Harkey loam soil types.
The average soil moisture was 28%.
Journal of the Air & Waste Management Association 1371
Wang, Hiscox, Miller, Meyer, and Sammis
Table 1. Averages (3-, 30-, and 60-min) of micrometeorological inputs and wind steadiness (S).
3-min Mean Inputs
Run Number
1
2
3
4
5
6
7
8
9
10
11
12
13
Average
Standard deviation
u* (m 䡠 secⴚ1)
L (m)
Wind
Direction
(ⴗ)
0.34
0.42
0.52
0.27
0.24
0.50
0.26
0.34
0.32
0.42
0.25
0.29
0.35
0.35
0.09
⫺8.80
⫺22.70
⫺79.00
⫺7.60
⫺4.10
⫺40.50
⫺3.10
⫺14.00
⫺10.50
⫺32.00
⫺27.40
⫺73.70
⫺11.00
⫺25.72
24.16
47.90
14.30
34.30
11.80
26.20
14.40
35.00
11.40
8.50
⫺9.00
29.40
38.30
42.70
23.48
15.70
30-min Mean Inputs
S
u* (m 䡠 secⴚ1)
0.96
0.98
0.97
0.99
0.95
0.98
0.99
0.99
0.97
0.97
0.98
0.98
0.96
0.98
0.02
0.37
0.42
0.46
0.45
0.48
0.38
0.30
0.41
0.36
0.35
0.33
0.31
0.46
0.39
0.06
A three-dimensional sonic anemometer (CSAT3,
Campbell Scientific, Inc.) was located at a height of 1.5 m
at the field edge to measure the 20-Hz wind component
velocities and temperature (u, v, w, T). The friction velocity (u*), Monin–Obukhov length (L), and wind direction
were obtained for each sampling pass at each second.
Dust plume size, shape, and movement were measured remotely with the University of Connecticut’s portable backscatter elastic LIDAR at approximately 4-sec intervals for each horizontal-slice scan. The LIDAR
specifications are listed in Hiscox et al.21 The LIDAR is
capable of scanning in horizontal or vertical planes. For
this disking operation, a series of horizontal slices was
designed to scan the entire plume. The lowest elevation of
the scan was a horizontal slice just above the field, and
successive slices were collected at increasing elevations of
approximately 3-m intervals. A full scan consisted of 15
different elevations, completed in approximately 45 sec,
and was repeated 5– 6 times for each pass depending on
the dust’s persistence. The slices from each scan were
combined in LIDAR data analysis software to characterize
the three-dimensional plume for each full scan.19 At the
end of each pass of the disking implement, the tractor
stopped at the end of the field and turned off its engine
until all sampling was completed and the dust plume
generated had moved out of the sampling area; another
pass across the field was then made. This procedure allowed the use of each pass as an independent replicate.
The source strength input (235 mg 䡠 sec⫺1) to the
model was determined from point concentration measurements using the general method in Holmén et al.22
The detailed procedures and calculations to determine the
source strength are described by Wang et al.23
Model Simulations and Comparison
Complete LIDAR, sampler, and micrometeorology data
were available for 13 disking passes. The model was run
four times for each pass using four different time averages
1372 Journal of the Air & Waste Management Association
60-min Mean Inputs
L (m)
Wind
Direction
(ⴗ)
S
u* (m 䡠 secⴚ1)
⫺14.70
⫺19.30
⫺24.00
⫺23.00
⫺35.40
⫺20.90
⫺275.00
⫺22.90
⫺17.20
⫺15.50
⫺105.60
⫺228.60
⫺32.60
⫺64.21
83.64
25.00
28.20
31.50
37.60
20.00
8.80
58.20
32.60
14.50
20.60
42.50
47.40
28.10
30.38
13.09
0.94
0.93
0.92
0.92
0.94
0.95
0.97
0.92
0.94
0.94
0.95
0.95
0.92
0.94
0.02
0.43
0.41
0.42
0.45
0.44
0.38
0.29
0.42
0.36
0.38
0.30
0.27
0.41
0.38
0.06
L (m)
Wind
Direction
(ⴗ)
S
⫺21.90
⫺18.80
⫺20.30
⫺23.80
⫺27.50
⫺20.40
⫺192.70
⫺25.90
⫺17.40
⫺20.90
⫺178.00
⫺301.50
⫺24.40
⫺68.73
89.14
27.20
26.00
26.20
29.00
20.90
15.70
67.90
20.10
16.10
21.90
53.40
60.90
21.30
31.28
16.82
0.92
0.91
0.91
0.91
0.91
0.93
0.96
0.91
0.92
0.92
0.95
0.93
0.92
0.92
0.02
of the input micrometeorology data (1 sec, and 3, 30, and
60 min). Input data for the averaging periods of 3, 30, and
60 min are presented in Table 1. The 1-sec input data are
not shown because of the size of the dataset.
The model concentration outputs were compared
with the LIDAR scans by comparing LIDAR-measured to
model-calculated horizontal arrays (hereafter called
maps) of aerosol concentration at three different heights
above ground (3, 9, and 15 m). At each height, the comparison statistics were the two-dimensional lagged crosscorrelation coefficient, E(x,y), and the offset distances of
the peak coefficients. The coordinates (x and y) of the
coefficient represent the lagged distances in the two horizontal directions in the concentration data matrix, where
x was in the tractor traveling direction and y was normal
to the traveling direction (Cartesian coordinate system)
(Figure 2). The simulated data for each map were shifted
in x and y distances (step length ⫽ 5 m) in the two
directions, then the corresponding correlation coefficient
E(x,y) of the overlapped parts of the simulated and the
measured data was calculated. The two-dimensional
shifted cross-correlation coefficient, E(x,y), is (after Mayor
et al.24)
E 共x,y兲 ⫽
冘
冘 冘
冋 冘 冉冘 冊 册
冋 冘 冉冘 冊 册
N
a i,j b i,j ⫺
a i,j
b i,j
2
N
a i,j 2 ⫺
1/2
2
⫻ N
b i,j 2 ⫺
(1)
a i,j
1/2
b i,j
where a (in the shifted simulated data) and b (in the
measured data) are the overlapped matrices; i, j are the
indices in either matrix (i.e., ai,j and bi,j are the element of
the ith row and jth column in either matrix); and N is the
number of points in either the a or b matrix. Comparison
of these coefficient plots and the offset distances of the
peak coefficient allowed us to quantify how closely
Volume 59 November 2009
Wang, Hiscox, Miller, Meyer, and Sammis
(a)
(b)
(c)
(d)
(e)
(f)
Figure 2. Illustration of cross-correlation coefficient calculation. Overlapped shaded areas show the image matrices of the simulated and
measured concentration data for the calculation of the shifted cross-correlation coefficient of E(x,y). The coordinates (x and y) of the coefficient
represent the lagged distances in the two horizontal directions in the concentration data matrix. Panels a and b show the original measured and
simulated images. Panels c–f show the overlapped area for different x and y values.
matched were the distribution and location of spatial
clustering in the modeled and measured plumes. The peak
correlation coefficient shows how close the particle distribution spatial patterns were between the modeled and the
measured plumes; and the offset distance shows the location difference between the two plumes, measured and
modeled (Figure 3). The best/worst match will be when
that peak correlation coefficient is 1/0 and offset distance
is 0/infinite.
RESULTS
A sample visualization of the simulation versus LIDAR
measurement is plotted in Figure 4. These simulation
output maps are from the 1-sec inputs (data not shown)
and 3-min inputs (shown in Table 1) compared with the
measured LIDAR maps for run no. 6.
Volume 59 November 2009
In this simulation, the cross-correlation coefficients
between the measured and the simulated plumes are plotted in Figure 3. At 3-m height, the peak correlation was
0.82 using 1-sec average input data versus 0.71 using
3-min average input data. At 9 m, the peak coefficients
were 0.74 versus 0.57, and at 15 m, the peak coefficients
were 0.57 versus 0.59 for the 1-sec and 3-min average
input data, respectively. Generally, the accuracy of the
model prediction decreased with an increase in height
and averaging time, as demonstrated by the decreasing
cross-correlation coefficients.
At 3 m, in the simulation of short-time inputs the
offset distances were x ⫽ ⫺15 m (horizontally on the
maps) and y ⫽ 50 m (vertically) (the coordinates for the
highest correlation coefficient) (Figure 5). This means that
the simulated plume at 3 m was 15 m off of the measured
Journal of the Air & Waste Management Association 1373
Wang, Hiscox, Miller, Meyer, and Sammis
(a)
(b)
(c)
(d)
Figure 3. Illustration of the peak correlation coefficient and the offset distance calculations. Panel a shows the measured plume. Graphs b– e
show different peak correlation values and offset distances for different simulated plumes.
toward the x negative direction and 50 m off toward to
the y positive direction, which is a total vector offset
distance of 52.2 m (Figures 4 and 5).
Correlation coefficient maps were calculated at three
heights above ground, 3, 9, and 15 m, and at four different averaging times for the input weather variables, 1 sec
and 3, 30, and 60 min. The summary statistics are presented in Table 2, in which all of the values are 13-run
averages and run-to-run variation. The average correlation coefficients range from 0.67 to 0.48. There is an
average 23% decrease in correlation between the measured and modeled plumes with a height increase of from
3 to 15 m. They also show an average 7% decrease at 3-m
height when averaging time is increased from 1 sec to 60
min. Thus, the model prediction of concentration is
poorer with increasing height and increasing averaging
time.
Table 2 also presents the run-to-run variability of the
correlation coefficients expressed as the coefficient of
variation (CV ⫽ standard deviation/mean). These statistics show a relative increase in average variation of 20%
with each 3-m increase in height, and an increase of 13%
with each increase in averaging time (1 sec to 3 min, 10%;
1374 Journal of the Air & Waste Management Association
3–30 min, 11%; 30 – 60 min, 18%). Thus, the random
error of the model prediction of concentrations increases
with increases in height and averaging time.
Table 2 presents the average absolute offset distance,
which is the horizontal distance between the measured
and modeled plume center. These ranged from 35.1 to
49.4 m. There was no detectable change with time averages, but there was an average 8% decrease (improvement) in the offset distance with each 6-m height increase. The variability of the offset distance again shows
no consistent change with different time averages but
indicates an average 70% increase in the variability with
each 6-m height increase.
DISCUSSION
The predictions of plume dispersion presented above are
an improvement over previous work presented in the
literature. Stein and Wyngaard25 noted that the “inherent
uncertainty” in the dispersion process, because of the
stochastic nature of the turbulence of the atmosphere, is
defined as the ratio of their root mean square difference in
a measurement period and the ensemble average of aerosol concentrations. They noted that previous experiments
Volume 59 November 2009
Wang, Hiscox, Miller, Meyer, and Sammis
Figure 4. The comparison of model simulation of (a, d, g) short-time (1-sec) inputs, (b, e, h) LIDAR observation, and (c, f, i) model simulation
of long-time (3 min) inputs of instantaneous normalized dust concentration (dimensionless) at 3-, 9-, and 15-m heights from a disking operation
at 152 sec after tractor was started. LIDAR and simulated concentration data were divided by their maximum value at each height. The tractor
traveled from right to left. Tractor start point was (246,0). Tractor speed was 1.41 m 䡠 sec⫺1. The simulation time period was 152 sec, u* ⫽ 0.50
m 䡠 sec⫺1, L ⫽ ⫺40.5 m, and wind direction ⫽ 14.4°.
generally showed the values of the inherent uncertainty
in pollutant dispersion from a near-surface source in the
atmosphere boundary layer exceed 50% for an averaging
time on the order of 1 hr. This maximum would be
expected to increase with shorter averaging times. Therefore, compared with previous work the average peak coefficient of 0.67 at a height of 3 m in this study is an
improvement on previous work. In general, the model
concentration errors are due to the model smoothing the
actual LIDAR-measured concentrations. The model individually handles each dust particle, whereas the LIDAR
showed that the dust actually moved in clumps of various
sizes. Further improvement of the model predictions
would have to include the mechanisms that define these
clumps. Also, dynamic wind data at multiple horizontal
locations and multiple heights need to be measured or
modeled to overcome the rapid spatial structure changes
in the wind field.
The ratio of offset distance to the downwind plume
moving distance is the parameter used to measure the
location difference of the modeled and measured plumes.
Its magnitude is due to the accuracy of the plume location,
which is primarily dependent on the variability of the wind
direction and speed. On the basis of the average measured
Volume 59 November 2009
wind speed (4.8 m 䡠 sec⫺1) and the average simulation time
(126 sec), the average plume moving distance was 602 m
(⫽ 4.8 ⫻ 126). The average offset distance was 38 m; therefore, the ratio was only 38/602, or 6%.
The wind speed and direction are never constant over
short averaging times. Therefore plumes are advected and
dispersed in the near-field by nearly instantaneous fluctuations driven by turbulent structures. It is therefore
important to use time averages of the wind measurements
that reflect the time scale that is most influential in dispersing the plumes in the region of interest. The surface
layer time scale (z/u*)26 ranged from 3 to 7 sec at a height
of 1.5 m up to approximately 70 sec at a height of 15 m.
This indicates that the predictable lifetime of wind eddies
was at least an order of magnitude shorter than the 3 min
it took to disk one pass across the field. Thus, as the input
weather data are averaged over greater and greater time
periods, the short-term wind field structure is lost and
only its statistical properties are maintained if the wind
field time series is statistically stationary. The increases in
error of the model predictions of concentration that were
positively correlated with increased time averages are
likely due to the longer time averages smoothing the
Journal of the Air & Waste Management Association 1375
Wang, Hiscox, Miller, Meyer, and Sammis
Figure 5. Sample two-dimensional cross correlation between the LIDAR-measured and the modeled dust concentration (short-time inputs vs.
long-time inputs) at (a– c) 3-, (d–f) 9-, and (g–i) 15-m height. Tractor traveled from right to left. Tractor start point was at (246,0). Tractor speed
was 1.41 m 䡠 sec⫺1. The simulation time period was 152 sec, u* ⫽ 0.50 m 䡠 sec⫺1, L ⫽ ⫺40.5 m, and wind direction ⫽ 14.4°.
short-term fluctuations of wind direction and speed. The
higher concentration prediction errors that were positively correlated with increased height are most likely due
to errors in the model predictions of the wind properties
at the higher elevations because they were only measured
at a single height (1.5 m).
Wind direction fluctuations and associated wind
speed fluctuations are the mechanisms that move the
plume horizontally back and forth in the meandering
process. To classify these fluctuations of the wind direction and speed, a wind steadiness parameter (S) was used
that combines these two. Singer27 proposed that the constancy of the wind, k, defined as the mean vector wind
៮ជ / V,៮ can
velocity divided by the mean scalar wind speed, V
be used for classification purposes. The range of k is from
0 to 1. A value of 1 means the wind direction did not
change over the averaging period. A value of 0 means a
1376 Journal of the Air & Waste Management Association
completely symmetrical wind speed and direction distribution during the averaging period. The steadiness factor,
S, is defined as
S⫽
2
arcsin 共k兲
␲
(2)
The equation transforms the constancy into a linear function. The angular deviations range from 0 to 180 °. If k is
1 then S is 1, and if k is 0 then S is 0. A change of 0.1 in S
represents a deviation in the wind of 18° over the averaging period. The average S for each of the input runs and
averaging time lengths are presented in Table 1.
Similar plots of u* (not shown) demonstrate that
longer time averages smooth the variation in wind turbulence intensity but do not change the overall average.
Volume 59 November 2009
Wang, Hiscox, Miller, Meyer, and Sammis
Table 2. Overall means and run-to-run variation of concentration (peak correlation coefficient) and offset
distance (modeled vs. LIDAR-measured).
Averaging Time of Inputs
Error Measure
Distance offset (m)
CV of distance offset
Peak correlation coefficient (PCC)
CV of PCC
Height (m)
1 sec
3 min
30 min
60 min
3
9
15
3
9
15
3
9
15
3
9
15
48.7
41.9
35.1
0.20
0.47
0.63
0.67
0.52
0.48
0.19
0.24
0.26
0.98
40.5
42.3
42.8
0.25
0.52
0.64
0.66
0.56
0.54
0.24
0.22
0.29
0.98
47.9
49.4
39.1
0.27
0.36
0.61
0.61
0.55
0.50
0.22
0.26
0.36
0.94
49.3
42.7
37
0.20
0.45
0.62
0.62
0.53
0.54
0.25
0.37
0.35
0.93
Wind steadiness
Also, there is little difference between the 30 and 60 min
averages as expected in a fully developed convective
boundary layer.26
The combination of wind speed variations and wind
direction bias in the wind steadiness parameter (S) shows an
overall decrease in wind steadiness with increased averaging
times (Figure 6), which explains the decrease in concentration prediction accuracy with longer time averages.
The run-to-run values of wind direction are plotted in
Figure 7 for the different averaging times and show a wind
veer (clockwise movement) with longer-term averages.
This is interpreted to mean that the longer time series of
wind direction were not stationary and likely were the
cause of the large regular offset distances.
In this study, the wind was very steady (steadiness ⬎
0.9); that is, wind direction and wind speed did not
change much. This resulted in little model performance
difference between the 1-sec and 3-min average input
runs. When wind is not steady (steadiness value is low),
the model performance from 1-sec inputs would have
better performance than that from the long-time average
inputs. It should be noted that the availability of a LIDAR
Figure 6.
inputs.
made nearly instantaneous measurements of concentration in the air possible. This has not been previously
available to test dispersion models at the local scale. Thus
the short-time-average predictions could be tested in this
project.
The wind data measurements were limited to one
point at the field edge. Because of the spatial variations of
the dynamic wind field where specific structures had a
lifetime on the order of seconds, the 1-sec inputs captured
the plume meanders during the 3-min pass on those occasions in which the driving eddies were translated from
the anemometer to the tractor or vice versa. But for much
of the time, this was not the case. Therefore, there was
little difference between the 1-sec and 3-min average input results. This suggests that dynamic wind data at multiple horizontal locations and multiple heights, measured
or modeled, are likely to improve the model performance
in the near-field.
CONCLUSIONS
The Lagrangian model of Wang et al.17 predicted the
near-field concentrations of dust plumes emitted from a
Run-to-run wind steadiness for 3-, 30-, and 60-min
Volume 59 November 2009
Figure 7. Run-to-run wind direction with different time averages.
Journal of the Air & Waste Management Association 1377
Wang, Hiscox, Miller, Meyer, and Sammis
field disking operation with an overall accuracy of approximately 0.67 at a height of 3 m. Its average offset
distance when compared with LIDAR measurements was
approximately 38 m. The model is driven by weather
measurements, and its near-field accuracy is highest when
input time averages approach the turbulent flow time
scale (3–70 sec). The wind steadiness parameter (S) can be
used to quantify the combined effects of wind speed and
direction on model accuracy.
The Lagrangian model presented here is a significant
improvement to the state of the art for simulating nearfield PM10 dispersion from a disking operation. The improvements are primarily due to the inclusion of moving
sources and inputs of weather parameters averaged to
the surface layer time scale (3–70 sec). However, the
model accuracy decreases with height because one-point
wind data measurements at a height of 1.5 m cannot
represent the wind field well at higher heights. The model
was parameterized for the near-field surface layer. Therefore, the model should not be used to simulate regional
(horizontally ⬎ 1000 m and vertically ⬎ 100 m) dust
dispersion.
ACKNOWLEDGMENTS
This project was supported by National Research Initiative
Competitive grants 2007-55112-17849 and 2004-3511214230 from the U.S. Department of Agriculture Cooperative State Research, Education, and Extension Service Air
Quality Program. It was also supported by the University
of Connecticut Storrs Agricultural Experiment Station and
the New Mexico State University Agricultural Experiment
Station. The authors are grateful to the staff at the Agricultural Experiment Station at New Mexico State University for their generous cooperation during the field experiments. The authors thank Dr. Jinfa Zhang at the
Department of Plant and Environment Sciences, New
Mexico State University, for allowing us use his cotton
field for the experiments.
REFERENCES
1. Etyemezian, V.; Ahonen, S.; Nikolic, D.; Gillies, J.; Kuhns, H.; Gillette,
D.; Veranth, J. Deposition and Removal of Fugitive Dust in the Arid
Southwestern United States: Measurements and Model Results; J. Air &
Waste Manage. Assoc. 2004, 54, 1099-1111.
2. Etyemezian, V.; Gillies, J.; Kuhns, H.; Nikolic, D.; Watson, J.; Veranth,
J.; Labban, R.; Seshadri, G.; Gillette, D. Field Testing and Evaluation of
Dust Deposition and Removal Mechanisms; Final Report; Western States
Air Resources Council: Lake Oswego, OR, 2003.
3. Aylor, D.E.; Flesch, T.K. Estimating Spore Release Rates Using a Lagrangian Stochastic Simulation Model; J. Appl. Meteorol. 2001, 40,
1196-1208.
4. Chow, J.C.; Watson, J.G.; Green, M.C.; Lowenthal, D.H.; DuBois,
D.W.; Kohl, S.D.; Egami, R.T.; Gillies, J.A.; Rogers, C.F.; Frazier, C.A.;
Cates, W. Middle- and Neighborhood-Scale Variations of PM10 Source
Contributions in Las Vegas, Nevada; J. Air & Waste Manage. Assoc.
1999, 49, 641-654.
5. Aylor, D.E.; Ferrendino, F.J. Dispersion of Spores Released from an
Elevated Line Source within a Wheat Canopy; Boundary Layer Meteorol.
1989, 46, 251-273.
6. Wilson, J. D.; Shum, W.K.N. A Re-Examination of the Integrated
Horizontal Flux Method for Estimating Volatilization from Circular
Plots; Ag. Forest Meteorol. 1992, 57, 281-295.
7. Pardyjak, E.R.; Speckart, S.O.; Yin , F.; Veranth J.M. Near Source Deposition of Vehicle Generated Fugitive Dust on Vegetation and Buildings: Model Development and Theory; Atmos. Environ. 2008, 42, 64426452.
8. Hagen, L.J.; Van Pelt, S.; Zobeck, T.M.; Retta, A. Dust Deposition near
an Eroding Source Field; Earth Surface Proc. Landforms. 2007, 32, 281289.
1378 Journal of the Air & Waste Management Association
9. Pace. T.G. Methodology to Estimate the Transportable Fraction (TF) of
Fugitive Dust Emissions for Regional and Urban Scale Air Quality Analyses;
U.S. Environmental Protection Agency: Washington, DC, 2005.
10. Chen, H.; Bai, S.; Eisinger, D.; Niemeier, D; Claggett, M. Modeling
Uncertainties and Near-Road PM2.5: a Comparison of CALINE4,
CAL3QHC, and AERMOD; Task Order Numbers 67 and 80; University
of California–Davis (UCD) Caltrans Air Quality Project; UCD: Davis,
CA, 2008; available at http://aqp.engr.ucdavis.edu/Documents/2008Chen-Dispersion-Model-Report-PM-final.pdf (accessed 2009).
11. Gokhale, S.; Raokhade, N. Performance Evaluation of Air Quality
Models for Predicting PM10 and PM2.5 Concentrations at Urban Traffic
Intersection during Winter Period; Sci. Total Environ. 2008, 394, 9-24.
12. Holmes, N. S.; Morawska, L. A Review of Dispersion Modeling and Its
Application to the Dispersion of Particles: an Overview of Different
Dispersion Models; Atmos. Environ. 2006, 40, 5902-5928.
13. Stein, A.F.; Isakov, V.; Godowitch, J.; Draxler, R.R. A Hybrid Modeling
Approach to Resolve Pollutant Concentrations in an Urban Area;
Atmos. Environ. 2007, 41, 9410-9426.
14. Vardoulakis, S.; Valiantis, M.; Milner, J.; Apsimon, H. Operational Air
Pollution Modeling in the UK: Street Canyon Applications and Challenges; Atmos. Environ. 2007, 41, 4622-4637.
15. Vlad, I.; Irwin, J.S.; Ching, J. Using CMAQ for Exposure Modeling and
Characterizing the Subgrid Variability for Exposure Estimates; J. Appl.
Meteorol. Climatol. 2007, 46, 1354-1371.
16. Van den Hurk, B.J.J.M.; Baldocchi, D.D. Random-Walk Models for Simulating Water Vapor Exchange within and above a Soybean Canopy; National Oceanic and Atmospheric Administration Technical Memorandum ERL ARL-185; Air Resources Laboratory: Silver Spring, MD, 1990.
17. Wang, J.; Hiscox, A.L.; Miller, D.R.; Sammis, T.W.; Meyer T.H. A
Dynamic Lagrangian, Field-Scale Model of Dust Dispersion from Agriculture Operations; ASABE Trans. 2008, 51, 1763-1774.
18. Holmén, B.A.; Miller, D.R.; Hiscox, A.L.; Yang, W.; Sammis, T.W.;
Bottoms, R. Near-Source Particulate Emissions and Plume Dynamics
from Agricultural Field Operations; J. Environ. Chem. 2008, 59, 117134.
19. Hiscox, A.L.; Miller, D.R.; Holmén, B.A.; Yang, W.; Wang, J. Near Field
Dust Exposure from Cotton Field Tilling and Harvesting; J. Environ.
Qual. 2008, 37, 551-556.
20. Raupach. M.R. Applying Lagrangian Fluid Mechanics to Infer Scalar
Source Distributions from Concentration Profiles in Plant Canopies;
Ag. Forest Meteorol. 1989, 47, 85-108.
21. Hiscox, A.L.; Nappo, C.J.; Miller, D.R. On the Use of LIDAR Images of
Smoke Plumes to Measure Dispersion Parameters in the Stable Boundary Layer; J. Atmos. Oceanic Technol. 2006, 23, 1150-1154.
22. Holmén, B.A.; James, T.A.; Ashbaugh, L.L.; Flocchini, R.G. LIDARAssisted Measurement of PM10 Emissions from Agricultural Tilling in
California’s San Joaquin Valley. Part II: Emission Factors; Atmos. Environ. 2001, 35, 3265-3277.
23. Wang, J.; Miller, D.R.; Sammis, T.W.; Hiscox, A.L.; Yang, W.; Holmén,
B.A. An Improved Model for Calculating Emission Factors for Agricultural Tilling Operations; J. Environ. Qual., submitted for publication.
24. Mayor, S.D.; Tripoli, G.J.; Eloranta, E.W. Evaluating Large-Eddy Simulations Using Volume Imaging LIDAR Data; Monthly Weather Rev.
2003, 131, 1428-1452.
25. Stein, A.F.; Wyngaard, J.C. Fluid Modeling and the Evaluation of
Inherent Uncertainty; J. Appl. Meteorol. 2001, 40, 1769-1774.
26. Stull, R.B. An Introduction to Boundary Layer Meteorology; Kluwer: Boston, MA, 1988.
27. Singer, I.A. Steadiness of the Wind; J. Appl. Meterol. 1967, 6, 10331038.
About the Authors
Junming Wang is a research scientist and Ted W. Sammis
is a professor with the Department of Plant and Environmental Sciences at New Mexico State University. April L.
Hiscox is an assistant professor in the Department of Environmental Sciences at Louisiana State University. David
R. Miller is a professor emeritus and Thomas H. Meyer is an
associate professor with the Department of Natural Resources Management and Engineering at the University of
Connecticut. Please address correspondence to: Junming
Wang, Department of Plant and Environmental Sciences,
New Mexico State University, MSC 3Q, P.O. Box 30003,
Las Cruces, NM 88003-8003; phone: ⫹1-575-646-3239;
fax: ⫹1-575-646-6041; e-mail: [email protected].
Volume 59 November 2009