Generated using version 3.2 of the official AMS LATEX template
The dispersion of silver iodide particles from ground-based1
generators over complex terrain. Part 2: WRF Large-Eddy2
simulations vs. Observations3
Lulin Xue ∗
National Center for Atmospheric Research, Boulder, Colorado
4
Xia Chu
University of Wyoming, Laramie, Wyoming
Roy Rasmussen
National Center for Atmospheric Research, Boulder, Colorado
Daniel Breed
National Center for Atmospheric Research, Boulder, Colorado
Bruce Boe
Weather Modification, Inc., Fargo, North Dakota
Bart Geerts
University of Wyoming, Laramie, Wyoming
5
∗Corresponding author address: Lulin Xue, current affiliation: National Center for Atmospheric Research,
Boulder, CO 80301
E-mail: [email protected]
1
ABSTRACT6
A numerical modeling study has been conducted to explore the ability of the WRF-based7
Large-Eddy Simulation (LES) with 100 m grid spacing to reproduce AgI particle dispersion8
by comparing the model results with measurements made on Feb. 16, 2011 over the Medicine9
Bow Mountains in Wyoming. The recently developed AgI cloud seeding parameterization10
(Xue et al. 2013b,a) was applied in this study to simulate AgI release from ground-based11
generators. Qualitative and quantitative comparisons between the LES results, observed12
soundings, and airborne/ground-based observed AgI concentrations were conducted. Anal-13
yses of TKE features within the planetary boundary layer (PBL) and comparisons between14
the 100 m LES simulation and simulations with 500 m grid spacing were performed as well.15
The results showed that: 1) Despite the moist bias close to the ground and above 4 km AGL,16
the LES simulation with 100 m grid spacing captured the essential environmental conditions17
except for a slightly more stable simulated PBL compared to the observed soundings. 2)18
Wind shear is the dominant TKE production mechanism in wintertime PBL over complex19
terrain and generates a PBL with about 1000 m depth. The terrain-induced turbulent eddies20
are primarily responsible for the vertical dispersion of AgI particles. 3) The LES-simulated21
AgI plumes were shallow and narrow, in agreement with observations. The LES simulation22
overestimated AgI concentrations close to the ground, which is consistent with the higher23
static stability in the model than observed. 4) Non-LES simulations using PBL schemes had24
difficulty capturing the shear-dominant turbulent PBL structure over complex terrain in25
wintertime. Therefore, LES simulations of wintertime orographic clouds with a grid spacing26
close to 500 m or finer are recommended.27
1. Introduction28
Inadequate or uncertain targeting of seedable clouds from silver iodide (AgI) ground-29
based generators has been a complex and a long-standing problem in winter orographic30
1
cloud seeding programs. The efficacy of the ground-based seeding depends significantly31
on the effective dispersion of the seeding agent in orographic clouds over complex terrain.32
To address how AgI particles released from ground-based generators disperse over complex33
terrain within the Wyoming Weather Modification Pilot Program (WWMPP), which is an34
outcome-focused randomized program (Breed et al. 2011, 2013), a focused field experiment35
was conducted between 9 February and 1 March 2011. Part 1 of this study (Boe et al.36
2013) describes airborne measurements of AgI-based ice nuclei (IN) plumes from ground-37
based generators collected by Weather Modification Inc. (WMI) Piper Cheyenne II research38
aircraft equipped with an updated NCAR acoustic IN counter (Langer et al. 1967; Langer39
1973; Heimbach et al. 1977; Langer et al. 1978; Heimbach et al. 2008; Super et al. 2010). The40
airborne data were collected over the Wyoming Medicine Bow and Sierra Madre mountain41
ranges on nine different days during the field experiment period.42
Previous observational studies have documented the dispersion of ground-released AgI43
plumes over mountainous target regions. An airborne experiment conducted by Super (1974)44
studied the dispersion of an AgI plume over the Bridger Range in Montana using the original45
version of the NCAR acoustic IN counter. The plume width was ∼ 28 degrees and was46
mostly confined to the lowest 500 m above the ridge line. A similar experiment conducted47
by Holroyd et al. (1988) over the Grand Mesa of Colorado showed that the median spread of48
AgI plumes was ∼ 15 degrees and that the median plume height above the crest exceeded49
500 m. They also pointed out that the dispersion efficiency was higher during cloudy days50
than clear days. Measurements of microphysical changes induced by seeding such as high51
concentrations of small ice crystals also indicated that AgI plumes have relatively narrow52
spreads and remain close to the ground (Super and Heimbach 1988; Super and Boe 1988;53
Huggins 2007). More recently, Geerts et al. (2010, 2011) showed, by means of reflectivity data54
from a profiling radar, that the impact of ground-based orographic cloud seeding was confined55
to the planetary boundary layer (PBL), about 1 km deep. Most of the aforementioned studies56
focused on AgI dispersion from a single ground-based generator while the field experiment57
2
within the WWMPP tried to assess the features of AgI plumes from both single and multiple58
ground-based generators, better representing the real WWMPP seeding experiments.59
Besides physical measurements of AgI plumes from ground-based generators by airborne60
and ground-based instruments, numerical models were used to investigate plume disper-61
sion. In the air quality modeling community, Lagrangian particle trajectories and dispersion62
models are commonly used to simulate pollutant transport and the dispersion of hazardous63
materials. Commonly used models include: the Hybrid Single Particle Lagrangian Inte-64
grated Trajectory (HYSPLIT) model (Draxler and Rolph 2003), the Second-order Closure65
Integrated Puff (SCIPUFF) model (Sykes and Gabruk 1997), and the FLEXTRA and FLEX-66
PART models (Stohl et al. 1995, 2005). These models are usually run in an “offline” mode,67
which means that they are driven by meteorological reanalysis data or meteorological con-68
ditions generated by numerical forecast models. However, for certain types of applications69
such as aerosol-cloud-precipitation interactions and glaciogenic seeding effects on orographic70
clouds, “online” calculations of particle transport and dispersion are needed. The reason71
is that the transport and dispersion of such passive particles will impact the microphysics72
of clouds and precipitation, which also influences the flow dynamics through microphysics-73
dynamics feedbacks. These interactions are not represented in “offline” models.74
Bruintjes et al. (1995) was probably the first study in the weather modification com-75
munity that compared online-calculated gaseous tracer concentrations (SF6), using a three-76
dimensional model, with the airborne in-situ observations. This pioneering work demon-77
strated the capability of numerical models to qualitatively and quantitatively capture the78
dispersion of AgI over complex terrain. The interaction between the airflow and the topog-79
raphy was identified as the dominant factor in determining the dispersion and transport of80
tracer materials. Part of the discrepancy between the simulated and observed results can be81
explained by the relatively large grid spacing (2 km) used in Bruintjes et al. (1995). Schicker82
and Seibert (2009) found that a grid spacing of 2.4 km did not reasonably simulate flows83
over complex terrain while 0.8 km did reproduce characteristic flow features. Weigel et al.84
3
(2007) performed Large-Eddy Simulations (LES) using a grid spacing of 350 m to investi-85
gate turbulent kinetic energy (TKE) evolution in a steep and narrow Alpine valley. They86
found that wind shear was the dominant production mechanism for TKE over such complex87
terrain. LES simulations at 150 m grid spacing showed that the stable stratification limited88
turbulent stress to the lowest few hundred meters near the surface of the mountain (Chow89
et al. 2006).90
It is promising that high resolution LES simulations can reproduce flow features over91
complex terrain reasonably well. Moreover, high resolution cloud-resolving LES simulations92
capture the interactions between turbulent eddies and cloud microphysics over complex ter-93
rain. There are some observational evidences that PBL turbulence is important in snow94
growth in cold clouds hugging mountains (Geerts et al. 2011). This study does not examine95
cloud microphysical processes (the case study in question is a dry event), but it serves as a96
prerequisite study for follow-up observational and numerical work (currently in progress) into97
the impact of glaciogenic seeding on microphysical processes in orographic clouds. Specif-98
ically, this study examines the capability of LES at 100 m grid spacing using the Weather99
Research and Forecast (WRF) model to capture AgI dispersion over mountainous topogra-100
phy for a case on Feb. 16, 2011 described in Part 1 (Boe et al. 2013). The model results101
were evaluated using airborne and ground-based observations. Comparisons of model results102
between simulations with 500 m grid spacing and that of LES with 100 m grid spacing have103
also been conducted to provide guidance for future simulations. The observational exper-104
iment on Feb. 16 and the numerical experiment are described in section 2 and section 3105
respectively. The model-observation and model-model comparisons are presented in section106
4 followed by discussions in section 5. The main conclusions are summarized in section 6.107
4
2. Description of the observational experiment108
Nine flights were conducted over the Medicine Bow and Sierra Madre mountains in109
Wyoming between Feb. 9 and Mar. 1, 2011. Operations required that (a) the region110
be devoid of low-level orographic clouds and snowfall in order to allow the aircraft to fly111
under visual meteorological conditions close to the terrain, and (b) substantial flow over the112
mountain. The latter was ascertained by requiring that the Froude number Fr > 1. On one113
of the nine flights, on Feb. 16, 2011 between 2240 UTC and 0216 UTC on Feb. 17, 2011,114
the IN counter registered high IN concentrations on several flight legs, with some well above115
mountain top level (Boe et al. 2013). Hereafter, all times are in UTC. The IN counter at116
the surface site also recorded the passage of AgI plumes (Boe et al. 2013). Given the rich117
observational data, this case was chosen for a LES simulation.118
The synoptic conditions during the observational period are illustrated in Fig. 1 using the119
North American Regional Reanalysis (NARR) data. These data have a spatial resolution of120
32 km and are available every 3 hours. The 700 hPa temperature (color shaded) was chosen121
because it roughly corresponds with the average mountain elevation above sea level. Mod-122
erately strong southwesterly flow persisted over a broad area ahead of a slowly progressing123
trough over California. Local cooling over the Medicine Bow Mountains (within the green124
circle) at about 2 K over 6 hours was due to cold-air advection and diurnal surface cooling.125
Wind speed and direction were fairly steady during this 6 hour period. The Medicine Bow126
Mountains was devoid of low-level clouds, and there was altostratus overhead (Boe et al.127
2013).128
For this case, radiosondes were released from Saratoga, a town upwind of the Medicine129
Bow Mountains (see Figs. 2(a) and (b)), on Feb. 16 at 2200 and on Feb. 17 at 0100.130
The observed soundings along with those from model simulations are shown in Fig. 3.131
The sounding parameters averaged between the surface and the peak of the Medicine Bow132
Mountains are listed in the left column of Table 1. Based on the Saratoga soundings, the133
winds (speed and direction) did not change much and the atmospheric stability decreased134
5
slightly during a three-hour period. These sounding parameters are broadly consistent with135
the synoptic analyses.136
Five ground-based generators (red triangles in Fig. 2) were operated between 2148, 02-16137
and 0115, 02-17. The aircraft flew 8 legs transecting the plumes (see Fig. 2(c) and Fig. 6 for138
the flight pattern) and took measurements from 2319, 02-16 to 0035, 02-17 corresponding139
to local late afternoon1. The flight legs were upwind of the steep mountain crest around140
Medicine Bow Peak to avoid boundary-layer separation and vertical transport in the lee141
of this sharp crest, something commonly observed over this mountain range (French et al.142
2013). An IN counter, with a larger cloud chamber than the airborne version, was operated143
at the surface site (mountain meadow cabin #9 as shown in Fig. 2) to detect AgI plumes144
in downwind regions. The generator operational times and flight times are listed in Table 2.145
More details about this case can be found in Boe et al. (2013).146
3. Description of the numerical experiment147
The WRF model was run on two nested grids with grid spacings of 2500 m and 500 m148
respectively driven by the North America Regional Reanalysis (NARR) data in a non-LES149
mode initially. Hereafter, these two domains are referred to “coarse-resolution” or “non-150
LES” domains. Since a practical technique that communicates information between the151
outter domain in a non-LES mode and the nested domain in a LES mode (two-way nesting)152
is generally unavailable in WRF, a one-way nesting procedure was applied to drive the 100153
m LES simulation. The simulations on the two non-LES grids were spun up for 21 hours154
from 0000 to 2100 on Feb. 16, 2011. They were then run from 2100, 02-16 to 0300, 02-17155
with an output interval of 20 minutes. Subsequently, these outputs were processed to drive156
the LES simulation over the same period with lateral boundary conditions being updated157
every 20 minutes.158
1Local solar noon is at 1919 UTC, and sunset is at 0040 UTC on this day.
6
The topographies of the domains with different grid spacings are shown in Fig. 2. The159
2500 m and 500 m domains consist of 320 by 220 grid points while there are 800 by 800 grid160
points for the LES domain. The vertically stretching coordinate as applied in Xue et al.161
(2010, 2012, 2013b,a) was adopted for all three domains. The vertical grid spacing is less162
than 200 m in the lowest 2000 m above ground level (AGL), which makes the grid aspect163
ratio of the LES domain less than 2 in regions experiencing the strongest AgI dispersion.164
For the LES domain, high-resolution elevation data from USGS 2 were used. For the two165
coarse-resolution domains, regular USGS 30-second data were used.166
In this study, the AgI cloud seeding parameterization documented in Xue et al. (2013b,a)167
was applied to simulate the release of AgI particles from ground-based generators. The size168
distribution of AgI particles from the generators is assumed to follow a lognormal distribution169
with a mean diameter of 0.05 µm and a geometric standard deviation of 2. The mean size170
of the AgI particle is slightly larger than that was specified in Xue et al. (2013b,a) due171
to slightly different ingredients of the AgI solution. The vertical mixing of AgI particles172
by the Mellor-Yamada-Janjic (MYJ) and the Yonsei University (YSU) PBL schemes was173
explicitly simulated in the coarse-resolution domains. No PBL scheme was specified for174
the LES simulation since we assumed that most of the terrainr-induced eddies responsible175
for AgI vertical mixing would be appropriately resolved with a 100 m grid spacing. For176
the LES simulation, a 1.5-order Turbulent Kinetic Energy (TKE) closure model was chosen177
(Deardorff 1980; Moeng 1984). A very short time step of 1/15 s was applied to ensure that178
the LES simulation remained stable3. The detailed configurations of the WRF model for all179
domains are listed in Table 3.180
21 arc-second data (about 30 m resolution) over the area of interest was downloaded from the National
Map Viewer and Download Platform for the LES simulation.3The abruptly changing topography introduces vigorous upward and downward motions close to the
ground which easily violates the vertical CFL criterion when a longer time step was applied.
7
4. Results181
a. Appropriateness of the LES simulation182
Before any LES results can be compared with the observations, appropriateness of the183
LES simulation must be assessed. The first general and important question about such a184
simulation is how much spin-up time is needed. In this study, the coarse-resolution domains185
were spun up for 21 hours before the generators were turned on. However, the limited186
computational resources prevented us from performing a long LES spin-up simulation. Here,187
we investigate the spin-up time issue by analyzing the kinetic energy power spectrum as188
described in Errico (1985) and Skamarock (2004).189
The instantaneous 1D kinetic energy spectra at 15 min intervals averaged over the entire190
horizontal domain (excluding 10 grid points near boundary in each direction) between the191
surface and 2000 m AGL are plotted in Fig. 4 from 0 to 120 minutes. The height of 2000 m192
AGL was chosen because most of the terrain-induced turbulent eddies are only active below193
this level and the vertical resolution of the model is also appropriate for a LES simulation194
below this altitude, as shown later in this study. The blue line in each panel represents the195
k−5/3 slope, which indicates the inertial subrange of the kinetic energy spectrum. The LES196
domain covers 80 km by 80 km, which means that the largest scale in the energy spectrum197
is still within the mesoscale range; therefore the spectrum follows the inertial slope at most198
scales. The deviation of the spectrum from the inertial subrange at scales less than 1 km199
(> 10−3 m−1 in 1/wavelength space) is due to the numerical dissipation of the integration200
scheme in the model (Skamarock 2004) and to the vertical averaging. The effective resolution201
of this LES simulation is about 1 km (where the deviation of the spectrum begins) which is202
10 times the grid spacing of 100 m. This effective resolution-to-grid spacing ratio is slightly203
greater than that for mesoscale NWP model (typically arond 7, Skamarock 2004).204
Apparently, the initial wind field that was interpolated from the 500 m results did not205
produce the correct spectrum. Fifteen minutes into the LES simulation, the wind field was206
8
adjusting to the underlying complex terrain. The turbulence at small scales injected energy207
into the wind field at large scales, causing an “overshoot” of the spectrum with respect to208
the inertial slope. Such adjustments and overshooting still existed after 30 minutes into209
the simulation. At 45 minutes, the wind field reached a balanced state and produced a210
spectrum following the inertial subrange at scales greater than 1 km. The spectrum remained211
basically unchanged after 45 minutes. Such a steady state spectrum will not be achieved if212
the synoptic conditions change significantly during this period. Based on the evolution of213
the model kinetic energy power spectrum, we conclude that the LES simulation needs about214
45 minutes to spin up in this specific case. Since the AgI particles were released from the215
generators after 48 minutes in the LES simulation, the dispersion of these particles should216
be properly simulated by the balanced flow field.217
Another important aspect for LES simulations to be verified is whether the subgrid-scale218
TKE is small compared to the total or resolved-scale TKE. LES by definition requires the219
model to resolve eddies at the most energetic scales. The energy at unresolved scales must be220
modeled or parameterized. The profiles of total TKE, subgrid TKE and half of the vertical221
velocity variance (vertical momentum flux) are illustrated in Fig. 5 for both the upwind222
and downwind regions. The upwind region is defined as the 360 by 360 grid points in the223
south-west corner of the domain while the downwind region is the 440 by 440 grid points224
in the north-east corner4. The profiles from 0 to 2000 m AGL represent values of these225
terms temporally averaged between 2145, 02-16 and 0115, 02-17 (operational time of the226
generators) and spatially averaged within the upwind and downwind regions.227
For both the upwind and downwind regions, the subgrid TKE is a small portion of228
the total TKE at each level. The subgrid-scale TKE to total TKE ratio is around 10%229
to 20% in the lowest 200 m and becomes less than 10% above 200 m. The higher ratio230
close to ground agrees with previous LES studies of a buoyancy-driven PBL (Deardorff231
1980; Moeng 1984). Since the atmosphere was stable and windy, the flow in the downwind232
4The peak of the Medicine Bow Mountains is roughly at grid point (360,360).
9
region was approximately between a resonance state and a boundary layer separation regime233
(1 < Fr < 1.7, Table 1) (Stull 1988). The facts that no terrain-induced gravity waves or234
mountain wave breaking existed in this case and that the interactions between the flow235
and terrain were weaker in the downwind region resulted in a weaker TKE profile in the236
downwind region than in the upwind region. The profiles of vertical velocity variances (Fig.237
5) highlighted the fact that the majority of the TKE resided in the horizontal dimension238
because the wind shear was the dominant TKE production mechanism in this case (see239
detailed discussions in section 5). These profiles also indicated a PBL height of about240
1000 m AGL in this case, which agrees with the observations of Geerts et al. (2010, 2011).241
Such a PBL height suggests that the vertical grid spacing used in the LES simulation was242
appropriate.243
b. Sounding comparisons244
After the appropriateness of the LES simulation is justified, model simulated and observed245
meteorological conditions are compared in this section. Since the NARR data were used to246
drive the simulations, comparisons of synoptic fields between the model and the NARR data247
are inappropriate. We then compare the sounding information at Saratoga in this section.248
The observed and simulated soundings at 2200, 02-16 and 0100, 02-17 are plotted in Fig.249
3. The observations were taken at Saratoga (black curves); simulated soundings from the 500250
m non-LES simulation were also located at Saratoga (red curves); and simulated soundings251
from the 100 m LES simulation were located about 6 km to the east of Saratoga (the star252
symbol in Fig. 2(c)) (blue curves). Both simulations reproduced the observed temperature253
structure above 700 hPa reasonably well. However, they both failed to simulate the moisture254
structure above 4 km AGL (∼ 600hPa). The NARR data apparently diagnosed more mois-255
ture in the free atmosphere than Saratoga sounding indicated, which forced the simulations256
to generate a more humid atmosphere. Close to the ground, both simulations overestimated257
the dewpoint temperature and underestimated the temperature. This is mainly due to the258
10
incorrect soil moisture initialization in the NARR data5. At both sounding times, the 500259
m simulation slightly outperformed the 100 m LES simulation in terms of temperature and260
dewpoint temperature between the surface and 700 hPa. The 500 m simulation also cap-261
tured more realistic low-level wind directions than the 100 m LES simulation. However, such262
differences might be the result of the slightly different sounding location used in the 100 m263
LES simulation.264
Sounding parameters were calculated between 60 m AGL and the peak of the Medicine265
Bow Mountains (Table 1). Observations below 60 m were not included due to artifacts266
introduced during the sounding launch. The model data below 60 m were removed as well267
due to the unrealistic values caused by the erroneous soil moisture initialization. Similar268
to what was found in Fig. 3, both simulations generated realistic wind direction between269
the surface and the mountain peak. The 500 m run simulated slightly better wind speed270
than the 100 m LES simulation. But the 100 m LES simulation captured wind shear and271
shear direction changing trend better than the 500 m simulation6. Both runs simulated a272
more stable atmosphere than the soundings with the 100 m LES simulation being slightly273
more stable than the 500 m simulation. As a result, both simulations under-predicted Fr274
and over-predicted Ri. Nevertheless, both simulations showed the same downward trend in275
stability as in the observations. Due to the moist nature of the NARR-driven simulations,276
the simulated lifted condensation level (LCL) heights were lower than observed. The LCL277
height was especially low in the 100 m LES simulation at 0100, 02-17. Despite the moist278
bias aloft and close to the ground simulated by the model, both the 500 m non-LES and279
the 100 m LES simulations captured the general features of the flow in the PBL where the280
dispersion of AgI particles occurred as will be revealed in the following sections.281
5Saratoga, as a small urban area which is not resolved by the NARR data, has higher surface temperature
and lower soil water content than its surroundings.6Wind shear turned counter clockwise in both observations and 100 m LES simulation while it turned
clockwise in 500 m simulation.
11
c. AgI concentration comparisons282
As discussed in Part 1 (Boe et al. 2013), the IN counter has a time-distributed delay283
averaging 80 seconds from the ingestion of atmospheric IN to the detection of them. Thus,284
the measurements in the form of IN count rate (s−1) accounting for the 80 s delay are plotted285
along the flight tracks in Fig. 6. Figure 6(a) reveals that the horizontal spread of the AgI286
plumes was not significant, with the majority of the plumes evident in the immidiate down-287
wind vicinity of the generators. However, since the flight legs were vertically orientated, the288
horizontal spread of the plumes at low levels may have been wider. The temporal smoothing289
of IN measurements (and thus spatial dispersion because of the aircraft movements) was290
not accounted for in Fig. 6. Thus, in reality, the IN plumes may be more narrow than291
suggested in Fig. 6(a) (Boe et al. 2013). Figure 6(b) illustrates the vertical structure of the292
observed AgI plumes. The vertical coordinate is referenced to mean sea level (MSL). The293
highest count rates (> 10 s−1) were observed in the lowest leg and the count rates gener-294
ally decreased with increasing altitude. The features of observed AgI plumes in this case295
agreed with many previous observational studies (Super 1974; Holroyd et al. 1988; Super296
and Heimbach 1988; Super and Boe 1988).297
Figure 7 shows 3D depictions of LES-simulated AgI plumes and wind fields at 2230 on298
02-16, 0000 on 02-17 and 0130 on 02-17. The wind field showed little change during the 3-299
hour period, which is consistent with the analyses presented in previous sections. Spatially,300
the high-level wind showed constant wind speed and almost uniform wind direction while301
the low-level wind was forced by the complex topography and presented higher variability302
than the wind field aloft. The mid-level wind was more variable than the high-level wind303
but less variable than the low-level wind.304
In Fig. 7, the visible blue plumes indicate AgI particles with number concentration305
greater than 100 L−1, which roughly corresponds to one IN count per second in the ob-306
servations (approximately the lowest concentration measurable with the IN counter in its307
configuration used in this experiment, see Part 1 for details). The AgI plumes remained308
12
narrow during this period. The plumes from generators MP, BR and BCH (see Fig. 2 for309
locations) merged into one large plume at 0000 on 02-17, but the plumes from generators310
FCO and RR2 stayed separate throughout the simulation. The general morphology of these311
plumes in the horizontal compared qualitatively well with the observations (see Fig. 6(a)).312
The vertical extent of the simulated plumes are also illustrated in Figs. 7(b1)-(b3). Note313
that the plume from RR2 (the southeastern generator) had more vertical dispersion than314
the others during this period. Most of the time, the plumes draped over the mountain with315
high concentrations of AgI particles close to the ground.316
To better compare the vertical structure of the AgI plumes, the AgI concentrations are317
plotted along two cross sections (as indicated by two black dashed lines in Fig. 2(c)) in Fig.318
8 for the same times as shown in Fig. 7. The two cross sections, NW to SE orientated and319
about 6 km apart, were chosen to cover the flight legs 3 to 8. Also plotted are the TKE isoline320
of 1 m2s−2 (red lines) and cloud water mixing ratio (black lines). Similar to Fig. 6(b1), the321
plume from RR2 (the southernmost plume) reached a higher altitude than the others at322
02-16 22:30 (Figs. 8(a1) and (b1)). The plumes from FCO (the center plume) and RR2 had323
AgI concentrations greater than 105 m−3 or 100 L−1 (colors that are warmer than yellow in324
the color bar) above the maximum observation altitude of 3800 m MSL (indicated by black325
dashed lines) at this time (Figs. 8(a1) and (b1)). The envelope of the plumes’ upper edges326
is co-located with the active turbulence region (red outlines), which implies that turbulent327
mixing is the primary mechanism for AgI vertical dispersion. Since the instantaneous plume328
morphology is an integrated reflection of the turbulent eddy history, occasional mismatches329
between the plumes’ upper edges and the instantaneous TKE isoline are reasonable.330
Figures 8(a2) and (b2) show that at 0000 on 02-17, the turbulence was not as active331
in the plume locations as 1.5 hours ago. Correspondingly, the plumes stayed close to the332
ground. Only a small portion of the plume from the FCO generator reached 3800 m MSL333
with a AgI concentration greater than 10 L−1 (see yellow areas in Fig. 8(b2)). This match334
between low turbulent activity and weak AgI vertical extent confirms that turbulent eddies335
13
are primarily responsible for the vertical dispersion of AgI particles. Also at this time, clouds336
started to form above 4000 m MSL, consistent with the sounding analyses in section 4b. At337
the later time (0130 on 02-17), the turbulence became strong again over the plume regions.338
As a result, the AgI plumes reached higher than 1.5 hours earlier. The clouds grew into a339
deck locating just above the plumes. Visual observations showed that there was a cloud deck340
close to this altitude when the airborne observations were taken (see Fig. 6 in Boe et al.341
(2013)).342
It is noticed from Fig. 8 that the turbulence was more active over the southern part of343
the domain. Indeed, the southern portions of these cross sections are part of the upwind344
region and the northern parts are associated with the downwind region due to the southerly345
component of the prevailing low level wind. Based on the analyses showed in section 4a,346
the TKE is stronger in upwind region than in downwind region (see Fig. 5). The isoline347
of TKE=1 m2 s−2 is almost exactly the cut off value for downwind TKE (see Fig. 5).348
Therefore, less TKE active region in the northern part of the cross sections showed up in349
Fig. 8. The TKE budget terms in this case will be analyzed in detail in section 5.350
The qualitative comparisons between the observations and LES-simulated AgI dispersion351
presented so far indicate that the horizontal spread of the simulated AgI plumes was similar to352
what was observed and the simulated plumes extended to the highest observational altitude353
with comparable AgI concentrations. Quantitative comparisons are needed to further assess354
the usefulness of the LES simulation. Therefore, we plot the contour-frequency by altitude355
diagrams (CFAD) of airborne measurements (all the data from flight legs 1 to 8) and LES356
results in Fig. 9. In each panel, the X-axis represents the IN count rate in s−1 (the AgI357
count rate converted from the concentration for model results), the Y-axis is the altitude (m358
AGL), and the frequency (ratio of AgI-altitude data points over total number of smaples) is359
color shaded. The observations numbered 2918 along the flight legs, and are plotted in the360
CFAD format in Fig. 9(a). To make the comparisons as consistent as possible, we identified361
the 2918 grid points corresponding to the locations of all the measurements in space. Since362
14
the LES results were output every 5 mins (see Table 3), it is impossible to match the exact363
timing of each measurement. Therefore, the average value of each grid point between 2315,364
02-16 and 0035, 02-17 (the observational period as listed in Table 2) was plotted in Fig.365
9(b). Notice that the observed AgI concentrations as a whole is a subset of the entire AgI366
concentration space (a continuous temporal-spatial manifold), the observations should be367
bounded by the maximum possible values in the entire concentration space. We thus plotted368
the maximum AgI count rate from each of the 2918 grid points between 2315, 02-16 and369
0035, 02-17 in Fig. 9(c).370
Figure 9 shows that low values of both observed and simulated AgI count rate (< 2 s−1)371
dominated the appearance frequency at all altitudes (blue to green or even red colors). The372
CFAD of the average simulated AgI count rate resembled the observed CFAD below about373
400 m AGL, but it deviated in the AgI count rate comparison between 400 and 1000 m374
AGL. In this layer, the LES-averaged CFAD overestimated the frequency of small values375
(low AgI concentration) and underestimated the large values (high AgI concentration). As376
discussed previously, the vertical dispersion of AgI particles was largely determined by the377
turbulent eddies which are highly intermittent in space and time. The LES-averaged CFAD378
(over 1 hour and 20 mins) significantly smoothed out the intermittency and underestimated379
AgI concentration than the observed CFAD. In contrast, the maximum LES-simulated AgI380
concentrations resulting from the strongest turbulent eddy during the measurement period381
should encompass the observed values. Figure 9(c) shows that the maximum LES-simulated382
CFAD overestimated the frequencies of large values at almost all altitudes. Quantitatively,383
the observations lie between the two LES-simulated CFADs. Considering the uncertainties384
associated with the observations and the LES simulation, the CFAD comparisons show good385
agreements between the observed and LES-simulated AgI concentrations.386
Measurements of IN concentrations at the surface site (mountain meadow cabin #9)387
were also collected in the downwind region of the airborne measurements in this case (see388
Fig. 2(c) for the location of this site). Figure 10 shows time series of 15-min averaged AgI389
15
concentration between 2200, 02-16 and 0300, 02-17 for observations at the surface site (panel390
(a)), results of the LES simulation at the closest grid point to the site (panel (b)) and results391
of the LES simulation averaged over the 9 by 9 closest grid points to the site (panel (c)).392
The model values were taken from the first layer which is about 7 m AGL at the surface393
site.394
The LES simulation generally over-predicted AgI concentrations at the surface site but395
remained within one order of magnitude. The over-predicted AgI concentrations maybe396
caused by 1) the inefficient vertical dispersion by the turbulent eddies, and 2) errors in the397
simulated wind direction. Relative to point 1), the LES simulation showed more stable398
conditions than observed (see Table 1). Therefore, the simulated eddies might be weaker399
than the actual conditions, particularly at the surface site (downwind region), which would400
leads to higher concentrations of AgI close to the ground. As for point 2), the LES-simulated401
wind direction was more southerly than the observed wind in the first half of the simulation.402
Thus, more AgI particles were advected from the RR2 generator to the surface site in the403
model. Although the model overestimated the AgI concentrations at the surface site, it404
captured the trends quite well. A weak plume of AgI particles was observed to arrive at the405
site at 2245 on 02-16 and last for about 45 minutes. The LES simulation showed that the406
plume arrived about 15 minutes earlier than observed and had elevated concentrations for407
about an hour. Observations also showed that a second strong plume swept over the surface408
site from 0045, 02-17 to 0200, 02-17 while the model predicted a second wave of elevated409
concentrations between 0130 and 0200 on 02-17.410
d. LES and non-LES comparisons411
The 100 m LES simulation has been validated using the airborne and ground-based412
measurements of AgI number concentration in the previous section. The LES simulation413
with 100 m grid spacing was shown to reasonably capture the AgI dispersion characteristics414
over the Medicine Bow Mountains in the Feb. 16, 2011 case. If we assume the LES simulation415
16
to be a reasonable representation of the reality, we can use the high resolution LES results416
to validate other simulations with lower resolutions. Here, we provide an example of such417
validations using simulated CFADs of AgI concentration and profiles of total AgI number.418
To conduct such validations, three extra simulations over the domain with 500 m grid419
spacing were performed. Each run was a one-way nesting simulation driven by the outputs420
of 2500 m simulation. The output frequency was set to 5 minutes to match that of the421
100 m LES simulation. One of the runs was the LES simulation, one a non-LES simulation422
using the MYJ PBL scheme and one a non-LES simulation using the YSU PBL scheme.423
The CFADs of AgI concentration over the 100 m LES domain (for 500 m simulations, it is424
a subset of the domain as indicated by the black box in Fig. 2(b)) during the AgI release425
period (2145 on 02-16 to 0115 on 02-17) are plotted in Fig. 11 for the 100 m LES simulation426
and the other three 500 m simulations. The X-axis is model simulated AgI concentration427
from 10−3 to 109 m−3 in a logarithmic scale. The Y-axis and color shaded areas have the428
same meanings as the Fig. 9. The data sample included all the model data below 1800429
m AGL over the indicated domain and from all output records. The large data population430
resulted in much smoother CFADs compared to those in Fig. 9.431
The CFAD of the 500 m LES simulation looks very similar to that of the 100 m LES432
simulation. Both CFADs showed high frequencies of data that are confined in concentra-433
tion between 104 and 107 m−3 and in altitude between 0 and 300 m AGL. For the lower434
concentration range, the 500 m LES simulation also predicted similar pattern to the 100435
m LES simulation (blue belt from 10−3 to 103 m−3 and from 300 to 1000 m AGL). Such436
features were also captured by the 500 m MYJ simulation. But it totally missed the high437
concentration features as in both LES simulations. Basically, the 500 m MYJ simulation438
predicted a much shallower mixed layer than LES simulations. Similarly, the 500 m YSU439
simulation also simulated a very shallow mixed layer in which very high concentration of440
AgI stayed close to the ground. Some local maxima close to 103 m−3 at about 200 m AGL441
were also predicted by the 500 m YSU simulation.442
17
The different properties of AgI vertical dispersion simulated by these simulations can be443
found in the profiles of total AgI number as well (Fig. 12). As showed in Fig. 12, the 500 m444
LES simulation predicted a similar profile to the 100 m LES simulation when the generators445
were turned off. The AgI number is greatest in the 100 m LES simulation between 150446
and 1700 m AGL. The 500 m MYJ simulation reproduced the trend of the LES simulations447
but under predicted the number more than an order of magnitude between 200 and 1500 m448
AGL. The 500 m YSU simulation completely missed the main features produced by the LES449
simulations and simulated a very inactive PBL with mininum vertical dispersion. Both LES450
simulations predicted fewer AgI number close to ground compared to non-LES simulations,451
which indicates the low-level vertical dispersion is stronger in LES mode.452
Based on the CFAD and AgI number profile comparisons, the LES simulation with453
500 m grid spacing mostly reproduced the “real” AgI dispersion characteristics. Non-LES454
simulations using PBL schemes had difficulty capturing the shear-dominant turbulent PBL455
structure over complex terrain in wintertime. More analyses on flow features, turbulent456
properties and surface fluxes need be done in the future to better understand why PBL457
schemes failed in this case. In this study, we tentatively suggest that LES simulations458
should be performed for wintertime orographic clouds with a grid spacing close to 500 m459
or finer. In a coarse grid, the MYJ PBL scheme should be used to simulate the glaciogenic460
seeding of wintertime orographic clouds.461
5. Discussions462
The wintertime PBL over complex terrain is not consistent with the traditional buoyancy-463
driven mixed layer concept. This statement can be confirmed by analyzing the profiles of464
TKE budget terms in the TKE tendency equation7. A symbolic form of the TKE tendency465
equation can be written as:466
7The complete equation can be found in Stull (1988).
18
467
∂TKE
∂t= BUOY + SHEA + ADV T + TUTR + PRAD + DISS (1)
where ∂TKE∂t
is the TKE tendency or storage term (Stull 1988), BUOY is the buoyant produc-468
tion term, SHEA is the shear or mechanical production term, ADVT is the TKE advection469
term, TUTR is the turbulent transport term, PRAD is the pressure adjustment of corre-470
lation term, and DISS is the dissipation term. Figure 13 illustrates the profiles from 0 to471
2000 m AGL of the BUOY, SHEA, ADVT, TUTR and PRAD terms over the upwind region472
(same as in Fig. 5) during the AgI release time (2145 on 02-16 to 0115 on 02-17). The473
storage term and dissipation term are ignored to make the plot more readable.474
Since this experiment was carried out in late afternoon on an overcast winter day, the475
buoyancy played an insignificant role in TKE production. Actually, it worked as a destruction476
term since the atmosphere was stably stratified8. Wind shear, on the other hand, dominated477
the TKE production between 100 and 600 m AGL. Such a deep shear layer consistently478
generating turbulent eddies does not exist in a traditional convective PBL over flat terrain.479
The breakdown of the shear term indicates that the negative values below 100 m AGL are480
associated with the persistent positive vertical kinematic eddy fluxes (u′w′ > 0 and v′w′ > 0)481
even if the shears are positive (du/dz > 0 and dv/dz > 0). Such features are suspected to482
relate to the interactions between the fine-scale terrain properties and the low level wind483
field. Further work should be done to decode this phenomenon. Both buoyancy-driven484
and shear-driven turbulent eddies are anisotropic. Buoyant turbulence is more vertically485
oriented while shear-induced turbulence is strongest in the horizontal. The shear dominant486
TKE production showed in Fig. 13 explains the small fraction of the vertical component to487
the total TKE in the PBL as shown in Fig. 5.488
The turbulent advection, transport and pressure adjustment terms showed similar ver-489
tical patterns with the destruction of turbulence at low levels and the production at high490
8Under such stable conditions, an air parcel displaced vertically by turbulent eddies will experience a
buoyancy force pushing it back to its original place (Stull 1988).
19
levels. However, the interception of the curve and the zero line increases from 400 m AGL491
for advection, to about 600 m AGL for transport and to about 900 m AGL for pressure ad-492
justment term. All these three terms might work as production or destruction terms locally.493
But integrated over the horizontal and the vertical, they effectively become zero. These494
terms only adjust or redistribute turbulent energy generated by buoyancy and shear. Figure495
13 shows that these mechanisms moved the turbulent energy generated in the wind shear496
layer to higher levels. The pressure adjustment term was responsible for the nonnegligible497
TKE above 1000 m AGL as shown in Fig. 5.498
The discussions on the TKE budgets indicate that the wintertime PBL over complex499
terrain is very different from the summer time convective PBL over flat terrain. In this case,500
the wind shear induced by the rough terrain existing in a relatively deep layer dominates501
the TKE production that is horizontally orientated. Such a shear-driven turbulent layer is502
not only responsible for AgI dispersion but also believed to enhance orographic precipitation503
by increasing riming efficiency as proposed in Houze and Medina (2005). In this case, the504
shear-driven turbulent layer is entirely within the PBL, as evident from the TKE profile, in505
contrast to the elevated shear zone in Houze and Medina (2005). Clearly, PBL turbulence506
is ubiquitous over complex terrain (where winter storms are usually accompanied by strong507
winds), thus orographic precipitation may rather generally be enhanced by PBL turbulence,508
as suggested by Geerts et al. (2011). It is worth mentioning that the longwave radiation509
cooling of the cloud deck could invigorate the underlying turbulence, which enhances the510
scalar dispersion and snow growth. Further discussion of this topic is the subject of a future511
paper.512
AgI contamination over target areas from the previous seeding operations is a practical513
problem for many randomized wintertime orographic cloud seeding programs. Usually, the514
mean wind speed and the target scale are used to calculate the average AgI dissipation515
time. However, the interactions between the topography and the low level wind have been516
shown to be very complex in this study. The actual AgI dissipation time is believed to be517
20
different than the simple calculated value. This problem is hard to address by the physical518
experiments performed by WMI but it is relatively easy to answer by the LES simulation.519
Based on this LES simulation, it took about 1.5 hours for AgI plumes to transport and520
dissipate away the target region after seeding ceased. This time is significantly longer than521
the simple calculated time of 50 mins.522
6. Conclusions523
A numerical modeling study has been conducted to explore the ability of the WRF-based524
LES model to reproduce AgI particle dispersion by comparing the model results with mea-525
surements made on Feb. 16, 2011 over the Medicine Bow Mountains in Wyoming. The526
recently developed AgI cloud seeding parameterization (Xue et al. 2013b,a) was applied in527
this study to simulate AgI release from ground-based generators. Qualitative and quan-528
titative comparisons between LES results and observed soundings, airborne/ground-based529
observations were conducted. Analyses on TKE features within the PBL and comparisons530
between 100 m LES simulation and simulations with 500 m grid spacing have been performed531
as well. The main conclusions of this study are:532
1 Despite the moist bias close to the ground and above 4 km AGL, the LES simulation533
with 100 m grid spacing captured the essential environmental conditions and simulated534
a slightly more stable PBL compared to the observed soundings at Saratoga.535
2 Wind shear is the dominant TKE production mechanism in wintertime PBL over536
complex terrain and generates a PBL with about 1000 m depth. The terrain-induced537
turbulent eddies are responsible for the vertical dispersion of AgI particles.538
3 The LES simulated AgI plumes were shallow and narrow, in agreement with inherently539
limited observations. The LES simulation overestimated AgI concentrations close to540
the ground due to more stable simulated condition than the real atmosphere.541
21
4 Non-LES simulations using PBL schemes had difficulty capturing the shear-dominant542
turbulent PBL structure over complex terrain in wintertime. LES simulations should543
be performed for wintertime orographic clouds with a grid spacing close to 500 m or544
finer.545
The LES simulation was shown to reasonably simulate the AgI dispersion over the com-546
plex terrain in this study. Using the AgI cloud seeding parameterization, more LES simula-547
tions will be performed in the near future to investigate the glaciogenic cloud seeding effect548
of the wintertime orographic clouds.549
Acknowledgments.550
This study was partly supported by the NCAR Advanced Study Program and the551
Wyoming Weather Modification Pilot Program. X. Chu and B. Geerts are greatful for the552
support of the AgI Seeding Cloud Impact Investigation project (NSF AGS-1058426). All553
rights to the underlying data collected and/or generated with funding from the Wyoming554
Water Development Office (WWDO) from which this report was created, remain with the555
WWDO. This report does not constitute the opinions of the State of Wyoming, the Wyoming556
Water Development Commission or the Wyoming Water Development Office.557
22
558
REFERENCES559
Boe, B. A., J. A. Heimbach, T. W. Krauss, L. Xue, X. Chu, and J. T. McPartland, 2013:560
The dispersion of silver iodide particles from ground-based generators over complex terrain.561
Part 1: Observations by acoustic ice nucleus counters. J. Appl. Meteor. Climatol., 52, to562
be submitted.563
Breed, D., M. Pocernich, R. Rasmussen, B. Boe, and B. Lawrence, 2011: Evaluation of the564
Wyoming winter orographic cloud seeding program: Design of the randomized seeding565
experiment. 18th Conf. on Planned and Inadvertent Weather Modification, AMS, 23–27566
January 2011, Seattle, WA.567
Breed, D., R. Rasmussen, B. Lawrence, B. Boe, T. Deshler, and C. Weeks, 2013: Evaluating568
winter orographic cloud seeding: Design of the Wyoming Weather Modification Pilot569
Project (WWMPP). J. Appl. Meteor. Climatol., 52, submitted.570
Bruintjes, R. T., T. L. Clark, and W. D. Hall, 1995: The Dispersion of Tracer Plumes in571
Mountainous Regions in Central Arizona: Comparisons between Observations and Mod-572
eling Results. J. Appl. Meteor., 34, 971 – 988.573
Chow, F. K., A. P. Weigel, R. L. Street, M. W. Rotach, and M. Xue, 2006: High-Resolution574
Large-Eddy Simulations of Flow in a Steep Alpine Valley. Part I: Methodology, Verifica-575
tion, and Sensitivity Experiments. J. Appl. Meteor. Climatol., 45, 87 – 107.576
Deardorff, J. W., 1980: Stratocumulus-capped mixed layers derived from a three-dimensional577
model. Bound.-Layer Meteor., 18, 495 – 527.578
Draxler, R. R. and G. D. Rolph, 2003: HYSPLIT (HYbrid Single-Particle Lagrangian Inte-579
grated Trajectory) model access via NOAA ARL READY. NOAA Air Resources Labora-580
tory, Silver Spring (2003), NOAA Air Resources Laboratory, Silver Spring (2003).581
23
Errico, R. M., 1985: Spectra Computed from a Limited Area Grid. Mon. Wea. Rev., 113,582
1554 – 1562.583
French, J. R., S. Haimov, V. Grubisic, S. Serafin, L. Strauss, and L. Oolman, 2013: Wave-584
induced boundary-layer separation in the lee of the Medicine Bow Mountains - Part I.585
Observations. J. Atmos. Sci., submitted.586
Geerts, B., Q. Miao, and Y. Yang, 2011: Boundary Layer Turbulence and Orographic Pre-587
cipitation Growth in Cold Clouds: Evidence from Profiling Airborne Radar Data. J. At-588
mos. Sci., 68, 2344 – 2365.589
Geerts, B., Q. Miao, Y. Yang, R. Rasmussen, and D. Breed, 2010: An Airborne Profil-590
ing Radar Study of the Impact of Glaciogenic Cloud Seeding on Snowfall from Winter591
Orographic Clouds. J. Atmos. Sci., 67, 3286 – 3302.592
Heimbach, J. A., A. B. Super, B. A. Boe, G. Langer, and J. T. McPartland, 2008: Compar-593
ison of two acoustic ice nucleus counters. J. Wea. Modif., 40, 54 – 63.594
Heimbach, J. A., A. B. Super, and J. T. McPartland, 1977: A suggested technique for the595
analysis of airborne continuous ice nucleus data. J. Appl. Meteor., 16, 255 – 261.596
Holroyd, E. W., M. J. T., and A. B. Super, 1988: Observations of Silver Iodide Plumes over597
the Grand Mesa of Colorado. J. Appl. Meteo., 27, 1125 – 1144.598
Houze, R. A. and S. Medina, 2005: Turbulence as a Mechanism for Orographic Precipitation599
Enhancement. J. Atmos. Sci., 62, 3599 – 3623.600
Huggins, A. W., 2007: Another wintertime cloud seeding case study with strong evidence of601
seeding effects. J. Wea. Modif., 39, 9 – 36.602
Langer, G., 1973: Evaluation of NCAR Ice Nucleus Counter. Part I: Basic Operation.603
J. Appl. Meteor., 12, 1000 – 1011.604
24
Langer, G., G. Cooper, C. T. Nagamoto, and J. Rosinski, 1978: Ice Nucleation Mechanisms605
of Submicron Monodispersed Silver Iodide, 1,5-Dihydroxynaphthalene and Phloroglucinol606
Aerosol Particles. J. Appl. Meteor., 17, 1039 – 1048.607
Langer, G., J. Rosinski, and C. P. Edwards, 1967: A Continuous Ice Nucleus Counter and608
its Application to Tracking in the Troposphere. J. Appl. Meteor., 6, 114 – 125.609
Moeng, C.-H., 1984: A Large-Eddy-Simulation Model for the Study of Planetary Boundary-610
Layer Turbulence. J. Atmos. Sci., 41, 2052 – 2062.611
Schicker, I. and P. Seibert, 2009: Simulation of the meteorological conditions during a winter612
smog episode in the Inn Valley. Meteorol. Atmos. Phys., 103, 211 – 222.613
Skamarock, W. C., 2004: Evaluating Mesoscale NWP Models Using Kinetic Energy Spectra.614
Mon. Wea. Rev., 132, 3019 – 3032.615
Stohl, A., C. Forster, F. A., P. Seibert, and G. Wotawa, 2005: Technical Note : The La-616
grangian particle dispersion model FLEXPART version 6.2. Atmos. Chem. Phys., 5, 2461617
– 2474.618
Stohl, A., G. Wotawa, P. Seibert, and H. Kromp-Kolb, 1995: Interpolation errors in wind619
fields as a function of spatial and temporal resolution and their impact on different types620
of kinematic trajectories. J. Appl. Meteo., 34, 2149 – 2165.621
Stull, R. B., 1988: An Introduction to Boundary Layer Meteorology. Springer, 666 pp.622
Super, A. B., 1974: Silver Iodide Plume Characteristics Over the Bridger Mountain Range,623
Montana. J. Appl. Meteo., 13, 62 – 70.624
Super, A. B. and B. A. Boe, 1988: Microphysical effects of wintertime cloud seeding with625
silver iodide over the Rocky Mountains. Part III: Observations over the Grand Mesa,626
Colorado. J. Appl. Meteo., 27, 1166 – 1182.627
25
Super, A. B., B. A. Boe, J. A. Heimbach, J. T. McPartland, and G. Langer, 2010: Com-628
parisons of silver iodide outputs from two different generators and solutions measured by629
acoustic ice nuclei counters. J. Wea. Modif., 42, 49 – 60.630
Super, A. B. and J. A. Heimbach, 1988: Microphysical effects of wintertime cloud seeding631
with silver iodide over the Rocky Mountains. Part II: Observations over the Bridger Range,632
Montana. J. Appl. Meteo., 27, 1153 – 1165.633
Sykes, R. I. and R. S. Gabruk, 1997: A second-order closure model for the effect of averaging634
time on turbulent plume dispersion. J. Appl. Meteo., 36, 165 – 184.635
Weigel, A. P., F. K. Chow, and M. W. Rotach, 2007: On the nature of turbulent kinetic636
energy in a steep and narrow Apline valley. Bound.-Layer Meteor., 123, 177 – 199.637
Xue, L., A. Teller, R. M. Rasmussen, I. Geresdi, and Z. Pan, 2010: Effects of Aerosol Solu-638
bility and Regeneration on Warm-phase Orographic Clouds and Precipitation Simulated639
by a Detailed Bin Microphysical Scheme. J. Atmos. Sci., 67, 3336 – 3354.640
Xue, L., A. Teller, R. M. Rasmussen, I. Geresdi, Z. Pan, and X. Liu, 2012: Effects of641
Aerosol Solubility and Regeneration on Mixed-phase Orographic Clouds and Precipitation.642
J. Atmos. Sci., 69, 1994 – 2010.643
Xue, L., S. Tessendorf, E. Nelson, R. Rasmussen, D. Breed, S. Parkinson, P. Holbrook, and644
D. Blestrud, 2013a: AgI cloud seeding effects as seen in WRF simulations. Part II: 3D645
real case simulations and sensitivity tests. J. Appl. Meteor. Climatol., 52, 1458 – 1476.646
Xue, L., et al., 2013b: AgI cloud seeding effects as seen in WRF simulations. Part I: Model647
description and idealized 2D sensitivity tests. J. Appl. Meteor. Climatol., 52, 1433 – 1457.648
26
List of Tables649
1 Summary of the sounding parameters. 28650
2 Generator and flight operational time. 29651
3 Model configurations. 30652
27
Table 1. Summary of the sounding parameters.
Parameters1\Cases2 Observations 500 m non-LES 100 m LES2200 UTC on 2011-02-16
U (ms−1) 22 19 17U dir (o) 227 225 221S (10−2 s−1) 1.25 0.70 1.21S dir (o) 250 265 272N (10−2 s−1) 0.77 1.01 1.10Fr (-) 1.87 1.22 0.97Ri (-) 0.38 2.07 0.83HLCL (m AGL) 2280 1730 1327
0100 UTC on 2011-02-17U (ms−1) 23 21 19U dir (o) 219 222 220S (10−2 s−1) 1.35 1.20 1.51S dir (o) 242 286 265N (10−2 s−1) 0.64 0.77 1.00Fr (-) 2.32 1.78 1.23Ri (-) 0.23 0.41 0.44HLCL (m AGL) 1642 1320 9481 All parameters were calculated between 60 m AGL and the peak
of the Medicine Bow Mountains. U is the mean wind speed; Udir is the direction of the mean wind; S is the bulk wind shear be-tween 60 m AGL and the peak of the Medicine Bow Mountains; Sdir is the direction of the wind shear; N is the Brunt-Vaisala fre-quency (average of the dry and the moist values); Fr = U/(NH)is the local bulk Froude number, where H is the height of be-tween 60 m AGL and the peak of the Medicine Bow Mountains;Ri = N2/S2 is the bulk Richardson number; and HLCL is theheight of the lifted condensation level (LCL) above the ground.
2 For the observations and the 500 m non-LES simulation, the pa-rameters were calculated over Saratoga (see Figs. 2(a) and (b)).For the 100 m LES simulation, the parameters were calculatedover the starred location shown in Fig. 2(c).
28
Table 2. Generator and flight operational time.
Generators Operational timeMullison Park (MP) 2148, 02-16 to 0104, 02-17Barrett Ridge (BR) 2149, 02-16 to 0106, 02-17French Creek Overlook (FCO) 2150, 02-16 to 0111, 02-17Rob Roy 2 (RR2) 2151, 02-16 to 0113, 02-17Beaver Creek Hills (BCH) 2152, 02-16 to 0115, 02-17Flight leg Observational timeLeg 1 2319, 02-16 to 2323, 02-16Leg 2 2326, 02-16 to 2333, 02-16Leg 3 2335, 02-16 to 2342, 02-16Leg 4 2345, 02-16 to 2353, 02-16Leg 5 2355, 02-16 to 0004, 02-17Leg 6 0007, 02-17 to 0014, 02-17Leg 7 0017, 02-17 to 0025, 02-17Leg 8 0027, 02-17 to 0035, 02-17
29
Table 3. Model configurations.
Configurations\Domains 2500 m non-LES 500 m non-LES 100 m LESSimulation time 0000, 02-16 to 0300, 02-17 2100, 02-16 to 0300, 02-17Time step 15 s 5 s 1/15 sOutput frequency 1 h 20 mins 5 minsRadiation CAM short wave and long wave schemesPBL Mellor-Yamada-Janjic (MYJ) scheme N/ASurface Noah land surface schemeMicrophysics Thompson scheme with AgI cloud seeding parameterizationTurbulence Horizontal Smagorinsky first order closure 1.5-order TKE closure
30
List of Figures653
1 Maps of NARR 700 hPa height (thick black contours, 30 m interval), temper-654
ature in Celsius (color shaded), wind barbs (full barbs equals 5 m s−1) and655
300 hPa height (thick white contours, 60 m interval). Green circles represent656
target area. (a) valid 2100, 2011-02-16, (b) valid 0000, 2011-02-17, (c) valid657
0300, 2011-02-17. 34658
2 Topography of (a) 2500 m grid spacing domain, (b) 500 m grid spacing domain659
and (c) 100 m grid spacing LES domain. Red triangles indicate the locations of660
five generators operated on Feb. 16, 2011. The black square symbol represents661
the surface site – mountain meadow cabin #9. The black cross in (a) and (b)662
indicates the town of Saratoga where the radiosondes were released. The large663
black boxes in (a) and (b) represent the domains in (b) and (c) respectively. In664
(c), the abbreviated name of each generator (listed in Table 2) is listed below665
the corresponding red triangle symbol. The flight pattern is illustrated by the666
blue line. The black dashed lines show the locations of cross-sections depicted667
in Fig. 8. The star symbol close to the western border of (c) represents the668
location of the LES-simulated (blue) soundings shown in Fig. 3. 35669
3 Soundings from observations (black), 500 m non-LES simulation (red) at670
Saratoga and from 100 m LES simulation (blue) at starred location in Fig.671
2(c) for (a) 2200, 2011-02-16 and (b) 0100, 2011-02-17. Temperature (oC) is672
the solid line and dewpoint temperature (oC) is the dashed line. Wind barbs673
are in m s−1 (full barb equals 5 m s−1). 36674
4 Average kinetic energy power spectra (energy density in m3 s−2) over the LES675
domain below 2000 m AGL. Each panel illustrates the spectrum at a moment676
from 0 to 120 minutes in 15 min invervals. The blue line indicates the k−5/3677
slope. 37678
31
5 Average profiles of the total TKE (solid lines), the subgrid-scale TKE (dashed679
lines) and half of the vertical velocity variance (dotted lines) over the upwind680
region (red) and the downwind region (green) between 02-16 21:45 and 02-17681
01:15. All terms have units of m2 s−2 38682
6 The plan view (a) and the cross-section viewed from west (b) of the in-situ683
measurements of AgI IN along the flight legs. The IN count rate (s−1) from684
the NCAR acoustic IN counter is color coded. In (a), the terrain is contoured685
every 500 m from 2000 to 3500 m MSL. Generators and the surface site are686
indicated by red triangles and the black square respectively. In (b), the flight687
legs are labeled from 1 to 8. 39688
7 3D depictions of the topography, AgI number concentration (greater than689
100 L−1 for visible plumes) and wind vectors (∼ 2800 m – yellow, ∼ 3600 m –690
blue and ∼ 4400 m – purple MSL). Panels (a) are bird’s-eye view perspective691
from the south, and panels (b) are side views from the southeast. Three692
snapshots are shown at times (1) 2230 on 02-16, (2) 0000 on 02-17 and (3)693
0130 on 02-17. 40694
8 Cross sections of AgI concentration (in logarithmic scale and color shaded in695
log(m−3)), isoline of TKE=1 m2 s−2 (red outlines) and cloud water mixing696
ratio (black contours in intervals of 0.1 g kg−1). Panels (a) are the western697
cross section and panels (b) are eastern cross section in Fig. 2(c). Valid at698
(1) 2230 on 02-16, (2) 0000 on 02-17 and (3) 0130 on 02-17. The black dashed699
lines indicate the highest level at which the in-situ measurements were taken700
(around 3800 m MSL). 41701
9 CFADs of (a) observed IN count rate (s−1), (b) average LES-simulated AgI702
count rate (s−1) between 2315, 02-16 and 0035, 02-17, and (c) maximum LES-703
simulated AgI count rate (s−1) between 2315, 02-16 and 0035, 02-17. All data704
are along the flight legs 1 to 8. Frequencies (0 – 0.03) are color coded. 42705
32
10 Time series of AgI concentration (L−1) at the surface site (mountain meadow706
cabin #9) for (a) observations, (b) 100 m LES results at the closest grid point707
and (c) 100 m LES results averaged over the 9 × 9 closest grid points. 43708
11 CFADs of AgI concentration over the 100 m LES domain for the (a) 100 m709
LES simulation, (b) 500 m LES simulation, (c) 500 m MYJ simulation and710
(d) 500 m YSU simulation, between 2145, 02-16 and 0115, 02-17. Frequencies711
(0 – 0.0012) are color coded. 44712
12 Profiles of AgI total number over the 100 m LES domain for the 100 m LES713
simulation (black), the 500 m LES simulation (blue), the 500 m MYJ simula-714
tion (green) and the 500 m YSU simulation (red) at 0115 on 02-17. Only the715
data between 0 and 2000 m AGL are shown. 45716
13 Average profiles of TKE budget terms (10−4 m2 s−3) over the upwind region717
between 2145, 02-16 and 0115, 02-17. The red line represents the buoyancy718
term, the black is the shear term, the green indicates the advection term,719
the blue line is the turbulent transport term and the orange line shows the720
pressure correlation term. 46721
33
2820
2850
2880
2910
2940
2970
3000
3030
3060
3090
8640
8700
8760
882088808940
9000
9060
9120
9180
92409300
9360
9420
2820
2850
2880
2910
2940
2970
3000
30303060
8640
8700
87608820
88808940
9000
9060
9120
91809240
9300
9360
9420
28202820
2850
2880
2910
2910
2940
2970
3000 3030
3060
8640
8700
876088208880
8940
9000
9060
9120
91809240
9300
9360
9420
a)
b)
c)
-10 -8 -6 -4 -2 0 2 4 6 8
2011
-02-1
6 2
1:0
02011
-02-1
7 0
0:0
02011
-02-1
7 0
3:0
0
Fig. 1. Maps of NARR 700 hPa height (thick black contours, 30 m interval), temperaturein Celsius (color shaded), wind barbs (full barbs equals 5 m s−1) and 300 hPa height (thickwhite contours, 60 m interval). Green circles represent target area. (a) valid 2100, 2011-02-16, (b) valid 0000, 2011-02-17, (c) valid 0300, 2011-02-17.
34
a)
b)
c)
MP
BR
BCHFCO
RR2
Fig. 2. Topography of (a) 2500 m grid spacing domain, (b) 500 m grid spacing domain and(c) 100 m grid spacing LES domain. Red triangles indicate the locations of five generatorsoperated on Feb. 16, 2011. The black square symbol represents the surface site – mountainmeadow cabin #9. The black cross in (a) and (b) indicates the town of Saratoga where theradiosondes were released. The large black boxes in (a) and (b) represent the domains in(b) and (c) respectively. In (c), the abbreviated name of each generator (listed in Table 2) islisted below the corresponding red triangle symbol. The flight pattern is illustrated by theblue line. The black dashed lines show the locations of cross-sections depicted in Fig. 8. Thestar symbol close to the western border of (c) represents the location of the LES-simulated(blue) soundings shown in Fig. 3.
35
Fig. 3. Soundings from observations (black), 500 m non-LES simulation (red) at Saratogaand from 100 m LES simulation (blue) at starred location in Fig. 2(c) for (a) 2200, 2011-02-16 and (b) 0100, 2011-02-17. Temperature (oC) is the solid line and dewpoint temperature(oC) is the dashed line. Wind barbs are in m s−1 (full barb equals 5 m s−1).
36
0 min 15 mins 30 mins
45 mins 60 mins 75 mins
90 mins 105 mins 120 mins
E(k
) (m
-3 s
-2)
E(k
) (m
-3 s
-2)
E(k
) (m
-3 s
-2)
1/wavelength (m-1) 1/wavelength (m-1) 1/wavelength (m-1)
Fig. 4. Average kinetic energy power spectra (energy density in m3 s−2) over the LESdomain below 2000 m AGL. Each panel illustrates the spectrum at a moment from 0 to 120minutes in 15 min invervals. The blue line indicates the k−5/3 slope.
37
0
500
1000
1500
2000
0 0.5 1 1.5 2 2.5 3
He
igh
t (m
AG
L)
TKE (m2 s-2)
Total TKE upwind
Subgrid TKE upwind
1/2 W'W' upwind
Total TKE downwind
Subgrid TKE downwind
1/2 W'W' downwind
Fig. 5. Average profiles of the total TKE (solid lines), the subgrid-scale TKE (dashedlines) and half of the vertical velocity variance (dotted lines) over the upwind region (red)and the downwind region (green) between 02-16 21:45 and 02-17 01:15. All terms have unitsof m2 s−2
38
Longitude
La
titu
de
a)
Latitude
He
igh
t (m
AM
SL
)
b)
1
2
34
5 6
7
8
0 1 2 3 4 5 6 7 8 9 10 11
IN count rate (s-1)
Fig. 6. The plan view (a) and the cross-section viewed from west (b) of the in-situ mea-surements of AgI IN along the flight legs. The IN count rate (s−1) from the NCAR acousticIN counter is color coded. In (a), the terrain is contoured every 500 m from 2000 to 3500m MSL. Generators and the surface site are indicated by red triangles and the black squarerespectively. In (b), the flight legs are labeled from 1 to 8.
39
a1)
a2)
a3)
b1)
b2)
b3)
2011-0
2-1
7 0
0:0
02011-0
2-1
6 2
2:3
02011-0
2-1
7 0
1:3
0
Fig. 7. 3D depictions of the topography, AgI number concentration (greater than 100 L−1
for visible plumes) and wind vectors (∼ 2800 m – yellow, ∼ 3600 m – blue and ∼ 4400 m– purple MSL). Panels (a) are bird’s-eye view perspective from the south, and panels (b) areside views from the southeast. Three snapshots are shown at times (1) 2230 on 02-16, (2)0000 on 02-17 and (3) 0130 on 02-17.
40
a1)
a2)
a3)
b1)
b2)
b3)
Latitude Latitude
AgI conc. in log scale [log(m-3)]
2011
-02-1
7 0
0:0
02011
-02-1
6 2
2:3
02011
-02-1
7 0
1:3
0
Western cross section Eastern cross section
Fig. 8. Cross sections of AgI concentration (in logarithmic scale and color shaded inlog(m−3)), isoline of TKE=1 m2 s−2 (red outlines) and cloud water mixing ratio (blackcontours in intervals of 0.1 g kg−1). Panels (a) are the western cross section and panels (b)are eastern cross section in Fig. 2(c). Valid at (1) 2230 on 02-16, (2) 0000 on 02-17 and(3) 0130 on 02-17. The black dashed lines indicate the highest level at which the in-situmeasurements were taken (around 3800 m MSL).
41
AgI count rate (s-1)
a) b)
c)
AgI count rate (s-1)IN count rate (s-1)
Fig. 9. CFADs of (a) observed IN count rate (s−1), (b) average LES-simulated AgI countrate (s−1) between 2315, 02-16 and 0035, 02-17, and (c) maximum LES-simulated AgI countrate (s−1) between 2315, 02-16 and 0035, 02-17. All data are along the flight legs 1 to 8.Frequencies (0 – 0.03) are color coded.
42
0
50
100
150
200
250
300
350
400
22h 23h 00h 01h 02h 03h
0
50
100
150
200
250
300
350
400
22h 23h 00h 01h 02h 03h
0
50
100
150
200
250
300
350
400
22h 23h 00h 01h 02h 03h
a) b) c)
Fig. 10. Time series of AgI concentration (L−1) at the surface site (mountain meadow cabin#9) for (a) observations, (b) 100 m LES results at the closest grid point and (c) 100 m LESresults averaged over the 9 × 9 closest grid points.
43
AgI concentration (m-3) AgI concentration (m-3)
b) 500 m LES
c) 500 m MYJ d) 500 m YSU
a) 100 m LES
Fig. 11. CFADs of AgI concentration over the 100 m LES domain for the (a) 100 m LESsimulation, (b) 500 m LES simulation, (c) 500 m MYJ simulation and (d) 500 m YSUsimulation, between 2145, 02-16 and 0115, 02-17. Frequencies (0 – 0.0012) are color coded.
44
0
500
1000
1500
2000
1.00E+05 1.00E+08 1.00E+11 1.00E+14 1.00E+17 1.00E+20
He
igh
t (m
AG
L)
Total AgI number
100 m LES
500 m LES
500 m MYJ
500 m YSU
Fig. 12. Profiles of AgI total number over the 100 m LES domain for the 100 m LESsimulation (black), the 500 m LES simulation (blue), the 500 m MYJ simulation (green) andthe 500 m YSU simulation (red) at 0115 on 02-17. Only the data between 0 and 2000 mAGL are shown.
45
0
500
1000
1500
2000
-50 -40 -30 -20 -10 0 10 20 30 40 50
He
igh
t (m
AG
L)
TKE Budgets (10-4 m2 s-3)
Buoyancy
Shear
Advec!on
Transport
Pressure
Fig. 13. Average profiles of TKE budget terms (10−4 m2 s−3) over the upwind regionbetween 2145, 02-16 and 0115, 02-17. The red line represents the buoyancy term, the blackis the shear term, the green indicates the advection term, the blue line is the turbulenttransport term and the orange line shows the pressure correlation term.
46