A Case Study of Radar Observations and WRF LES Simulations of the Impact ofGround-Based Glaciogenic Seeding on Orographic Clouds and Precipitation. Part I:

Observations and Model Validations

XIA CHU

Department of Atmospheric Science, University of Wyoming, Laramie, Wyoming

LULIN XUE

Research Applications Laboratory, National Center for Atmospheric Research,*

Boulder, Colorado

BART GEERTS

Department of Atmospheric Science, University of Wyoming, Laramie, Wyoming

ROY RASMUSSEN AND DANIEL BREED

Research Applications Laboratory, National Center for Atmospheric Research,

Boulder, Colorado

(Manuscript received 20 December 2013, in final form 18 June 2014)

ABSTRACT

Profiling airborne radar data and accompanying large-eddy-simulation (LES)modeling are used to examine the

impact of ground-based glaciogenic seeding on cloud and precipitation in a shallow stratiform orographic winter

storm. This storm occurred on 18 February 2009 over a mountain inWyoming. The numerical simulations use the

Weather Research and Forecasting (WRF)Model in LESmode with horizontal grid spacings of 300 and 100m in

a domain covering the entire mountain range, and a glaciogenic seeding parameterization coupled with the

Thompson microphysics scheme. A series of non-LES simulations at 900-m resolution, each with different initial/

boundary conditions, is validated against sounding, cloud, and precipitation data. TheLES runs then are driven by

the most representative 900-m non-LES simulation. The 100-m LES results compare reasonably well to the

vertical-plane radar data. Themodeled vertical-motion field reveals a turbulent boundary layer and gravity waves

above this layer, as observed. The storm structure also validates well, but themodel storm thins andweakensmore

rapidly than is observed. Radar reflectivity frequency-by-altitude diagrams suggest a positive seeding effect, but

time- and space-matched model reflectivity diagrams only confirm this in a relative sense, in comparison with the

trend in the control region upwind of seeding generators, and not in an absolute sense. A model sensitivity run

shows that in this case natural storm weakening dwarfs the seeding effect, which does enhance snow mass and

snowfall. Since the kinematic andmicrophysical structure of the storm is simulatedwell, future Part II of this study

will examine how glaciogenic seeding impacts clouds and precipitation processes within the LES.

1. Introduction

The main motivation for advertent weather modifi-

cation remains precipitation augmentation. By 2025, an

estimated 3 billion people will be subject to severe water

shortages (Black and King 2009). Glaciogenic cloud

seeding started in the mid-1940s when scientists at

General Electric Research Laboratory demonstrated

*TheNational Center for Atmospheric Research is sponsored

by the National Science Foundation.

Corresponding author address: Xia Chu, Dept. of Atmospheric

Science, 1000 E. University Ave., University ofWyoming, Laramie,

WY 82071.

E-mail: xchu1@uwyo.edu

2264 JOURNAL OF APPL IED METEOROLOGY AND CL IMATOLOGY VOLUME 53

DOI: 10.1175/JAMC-D-14-0017.1

� 2014 American Meteorological Society

that dry ice and silver iodide (AgI), which have a similar

crystal structure to ice, could initiate ice in a laboratory

supercooled liquid cloud. When dry ice pellets or AgI

nuclei were released from an aircraft into supercooled

stratus clouds, the affected cloud rapidly cleared (Schaefer

1946; Vonnegut 1947), although no in situ or remote

sensing probes were available back then to prove the

interpretation that ice crystals formed and consumed the

available liquid water.

Nevertheless, weather modification programs mush-

roomed following this discovery, and by 1951 cloud

seeding was conducted in about 30 countries. The impact

of cloud seeding typically was assessed by comparing

precipitation in a pre-seeded period or a nearby control

area to that in the treated period or area (e.g., Bruintjes

1999). Such experiments have been hindered by the

overwhelming ‘‘noise’’ of precipitation, that is, the nat-

ural variability at fine temporal and spatial scales

(Garstang et al. 2005). In view of the difficulty to dem-

onstrate efficacy, the National Research Council rec-

ommended in several reports that the evaluation of the

experiments must be carefully designed, conducted, and

analyzed (National Research Council 2003). Aside from

large randomized experiments with a strong statistical

basis, physically based experiments were encouraged to

improve our understanding of the changes in cloud mi-

crophysical and dynamical processes resulting from

cloud seeding. This was the main motivation of the AgI

Seeding Cloud Impact Investigation (ASCII) campaign,

conducted over the mountains of southeast Wyoming

in 2012 (Geerts et al. 2013) in the context of the 2007–14

WyomingWeatherModificationPilot Project (WWMPP),

a randomized dual-mountain experiment (Breed et al.

2014). The present paper analyzes a 2009 pre-ASCII case.

Numerical simulations of cloud seeding have the ad-

vantage that the weather can be repeated with and with-

out seeding; that is, the seeding can be treated as a model

sensitivity experiment. The first dynamic, 2D numerical

simulations of convective clouds originated in the 1960s

(Ogura 1963; Orville 1965). Clouds and precipitation were

bulk parameterized, with originally two and later four

species of water plus water vapor. Bin models have been

developed as well (Young 1974), but since they are com-

putationally expensive, they tend to simplify the dynamics.

Simple cloud models have been adapted to simulate the

impact of glaciogenic (e.g., Simpson andWiggert 1971) and

hygroscopic (e.g., Klazura and Todd 1978) seeding. With

dramatic advances in computational power, the represen-

tation of aerosol and cloud processes in models now en-

ables a more realistic simulation of the cloud seeding

process. Recently Xue et al. (2013a,b) developed an AgI

cloud seeding ice initiation parameterization for use in the

Thompsonmicrophysics scheme (Thompson et al. 2008) in

theWeather Research and Forecasting (WRF; Skamarock

et al. 2008) Model. The Thompson microphysics scheme

has been proven to be a superior scheme for wintertime

precipitation forecasting (Liu et al. 2011; Rasmussen et al.

2011). Xue et al. (2014) show that the dispersal of AgI

nuclei released from the ground over complex terrain

validates well, at least in one dry case.

Observational seeding impact studies have been

plagued by their inability to answer two questions. First,

what is the magnitude of seeding-induced changes in

comparison with natural precipitation variability, which

can be large even in a rather steady storm? Second, are

the observed changes in a nearby control (untreated)

area really representative of natural changes in the tar-

get (treated) area? These are questions that can only be

answered by a realistic numerical simulation. Both will

be addressed in this study.

The objectives of this work are to describe the dynam-

ical and cloud structure of a winter storm observed over

a mountain in Wyoming, to evaluate the ability of WRF

large-eddy simulation (LES) to capture this structure, to

examine changes in low-level reflectivity in this stormboth

upstream and downstream of the AgI generators, and to

evaluate the ability of this WRF LES with AgI cloud

seeding parameterization to reproduce these changes.

This is the first part of a two-part study. Section 2

describes the instruments, case selection, and synoptic

conditions. Section 3 summarizes the AgI seeding pa-

rameterization, themodel setup, and the selection of outer

domain driver dataset. Section 4 examines observed and

modeled storm structure and indicates the AgI seeding

impact, and section 5 examines observed and modeled

changes in radar reflectivity during seeding in target and

control regions. A discussion is provided in section 6, and

conclusions follow in section 7. Part II of this work, which

is not yet complete, will use a sensitivity study to examine

how ground-based AgI seeding impacts cloud micro-

physical and precipitation processes within the LES.

2. Instruments and ambient conditions

The case study presented here took place on 18 Feb-

ruary 2009 over the Medicine Bow Range in southern

Wyoming (Fig. 1). While the AgI generators were sup-

ported by the WWMPP, this case was not one of the

WWMPP randomized, mountain-blind seeding cases. In-

stead, it is one of the seven nonrandomized cases aimed at

examining how glaciogenic seeding affects the reflectivity

profile in orographic snow storms (Geerts et al. 2010).

a. Instruments and flight patterns

The aircraft used in this study is the University of

Wyoming King Air (UWKA), equipped with numerous

OCTOBER 2014 CHU ET AL . 2265

in situ cloud and atmospheric probes, as well as a pro-

filing Doppler millimeter-wave Wyoming Cloud Radar

(WCR) (Geerts et al. 2010). The WCR is the most im-

portant model validation instrument in this study, as it

documents the vertical structure of reflectivity and hy-

drometeor vertical velocity over the depth of the storm

and down to ;30m above the ground level. Because

AgI nuclei are released at ground level, the vertical-

plane reflectivity measurements very close to the terrain

are important. The WCR affords a resolution high

enough to capture a significant fraction of the spectrum

of turbulent eddies. The WCR pulse width during this

flight was 37.5m, with range sampling at 15-m intervals.

WCR profiles are sampled every 4m along the flight

track, but the along-track resolution degrades somewhat

with range as profiles start to overlap because of the

antenna beamwidth (0.58 for the nadir beam and 0.88 forthe zenith beam). At a typical range of 1 km, the WCR

resolution is 10m in horizontal and 40m in vertical,

which in the horizontal is an order of magnitude better

than that of the 100-m WRF LES. The vertical distri-

bution of layers in theWRFLES is such that the average

interval in the lowest 1 km is;50m, that is, comparable

to the radar resolution.

The UWKA flight pattern in the case studied here

(1641–2024 UTC 18 February 2009) is shown in Fig. 1.

TheUWKA repeated this geographically fixed ‘‘ladder’’

(or rather ‘‘lawnmower’’) pattern four times at a con-

stant flight level (;4.27 kmMSL), each time entering on

the downwind side (track 5) and exiting on the upwind

side (track 1), to avoid any possible contamination by

aircraft-produced ice particles on consecutive flight legs.

Each ladder contains five tracks, with track 1 upwind of

the AgI generators, that is, track 1 serves as the control

track. The other four tracks are the target tracks

downwind of the AgI generators. The AgI generators

were activated as soon as the UWKA cleared track 2

during ladder 2, in this case at 1809 UTC. Thus we

refer to the period of the first two ladder patterns as

NOSEED, that is, precipitation was not affected by AgI

FIG. 1. Terrain map with flight pattern and observation sites. The location of this region in

Wyoming is shown as a red box in Fig. 2.

2266 JOURNAL OF APPL IED METEOROLOGY AND CL IMATOLOGY VOLUME 53

seeding. A time window of 30min was allowed for the

AgI nuclei to disperse across the ladder pattern, before

the third ladder was flown, again starting at track 5. The

mean low-level wind speed and direction obtained from

a sounding released at 1800 UTC from Saratoga (Fig. 1)

was such that 30min was sufficient for the AgI nuclei to

disperse to track 5 in this case, so data from the entire

third and fourth ladder can be considered to fall within

the SEED period. All 20 tracks were flown at constant

level (;600m above Medicine Bow Peak) and straight

and level to minimize the impact of the horizontal wind

on the WCR up- and down-antenna radial velocities. In

fact the resulting Doppler velocity is corrected for hor-

izontal wind contamination using flight-level winds,

yielding a good estimate of the hydrometeor vertical

velocity. More details on the operation of the WCR on

this flight can be found in Geerts et al. (2011).

Somemeasurements were made on the ground during

the flight at sites shown in Fig. 1. A dual-frequency

passive microwave radiometer located in Cedar Creek

estimated the liquid water path (LWP). The LWP is

retrieved from two microwave frequencies (23 and

31GHz). The retrieval algorithm uses typical surface

pressure and profiles of temperature and humidity

(Ware et al. 2003). This algorithm compares rather well

with independent measurements (Crewell and Löhnert2003). The radiometer was pointed toward theMedicine

Bow Range at a grazing angle above the terrain to

capture clouds over the mountain rather than over the

upwind valley. The precipitation rate was measured

at six Environmental Technology, Inc. (ETI), gauges,

some closer to the AgI generators than others (Fig. 1).

These are shielded, unheated gauges in which the

snowfall is melted in a glycol solution. Themeasurement

uncertainty of these gauges is discussed in Rasmussen

et al. (2001).

b. Case selection

The case on 18 February 2009 was chosen for an in-

depth analysis for several reasons. First, the WCR echo

top, precipitation rate and radiometer LWP were rather

steady during the 4-h flight period, as will be shown later.

Second, the precipitation was truly orographic and

rather shallow (cloud-top temperature ;2248C), al-

though ice was naturally present at all cloud levels and

along all tracks, according the WCR reflectivity profiles

and 2D particle probes aboard the UWKA. Third, the

temperature at the level of the AgI generators was 298to 2108C, cold enough for extensive activation of AgI

nuclei (DeMott 1995). Fourth, the AgI generators were

(just) in cloud, according to the lifting condensation

level (LCL) derived from the upwind sounding data

and local relative humidity measurements (;90%).

Fifth, supercooled liquid water was present at flight level

during all four ladder patterns, in agreement with the

Cedar Creek radiometer. Sixth, the upstream sounding

indicates that the low-level wind direction (;3048) wassuch that the AgI plumes crossed the four downwind

tracks (Fig. 1). And finally, the upstream Froude number

[Fr5 U/(NH)] was large enough (Fr5 1.4) for AgI nuclei

to be advected into cloud and over the mountain, rather

than around the mountain. Here H is the height of the

mountaintop above the upwind valley, U is wind speed,

andN is theBrunt–Väisälä frequency. TheFroudenumberis calculated using sounding values averaged between thesurface and mountaintop level. For N, a combination of

dry and moist Brunt–Väisälä frequencies is used. Thedry and moist Brunt–Väisälä frequencies were calcu-lated below and above the LCL, respectively.

c. Synoptic condition

The synoptic situation during this case is illustrated in

Fig. 2 using the 12-kmNorthAmericanMesoscale (NAM)

operational model output. The 700-mb (1mb 5 1hPa)

level is chosen as low-level map (Fig. 2a) because it is be-

tween the level of the AgI generators and the mountain-

top. The target area was under rather steady conditions,

far from any front, with a northwesterly wind around

13m s21 at 700mb.A quasi-stationary upper-level trough

(vorticity maximum) stretched fromOregon to Colorado

(Fig. 2b), well south of the target area, resulting in weak

winds aloft in the target area, and weak subsidence at

500mb (not shown). Farther east, this trough connected

to a mobile frontal trough over the Great Plains. The

area of 700-mb cold-air advection associated with this

mobile upper-level trough was well to the east of the

target area (Fig. 2a). The target area experiences no

significant 700-mb temperature advection (Fig. 2b), al-

though the Saratoga sounding (shown below) suggests

weak geostrophic warm-air advection at low levels.

3. Model configuration

a. Approach and physics choices

This study uses WRF version 3.4.1 (Skamarock et al.

2008), with a configuration shown in Table 1. The model

was run in non-LESmode using two-way-nested domains

with horizontal grid spacings of 2700 and 900m (Fig. 3);

the outer domain was driven by the Climate Forecast

SystemReanalysis (CFSR; https://climatedataguide.ucar.

edu/climate-data/climate-forecast-system-reanalysis-cfsr).

Then WRF was run in LES mode using other two-way-

nested domains with horizontal grid spacings of 300 and

100m, driven by (but not coupled with) the 900-mWRF

non-LES run. In other words, there is one-way nesting

OCTOBER 2014 CHU ET AL . 2267

between the non-LES and LES domains. Mirocha et al.

(2014) show that nesting a fine LES domain within

a coarser surrounding LES upon which turbulence

can begin to develop is superior to nesting the fine LES

directly within the mesoscale (non-LES) simulation.

The terrain files had a resolution matching the atmo-

spheric grid for each domain. The simulations on the two

non-LES domains were spun up for 15 h starting at

0000 UTC 18 February 2009. The LES run started at

1500 UTC, with lateral boundary conditions updated

every 30min from the 900-m non-LES model output. A

vertically stretching distribution of 60 eta levels was

adopted for all four domains, as in Xue et al. (2010, 2012,

2013a,b, 2014). Details of the AgI cloud seeding pa-

rameterization can be found in Xue et al. (2013a).

The Mellor–Yamada–Janjic (MYJ) planetary boun-

dary layer (PBL) scheme was used in the non-LES do-

mains based on its good performance in wintertime

orographic clouds (Xue et al. 2014). The LES runs at 300

and 100m assume that the resolved turbulent and

terrain-driven eddies are responsible for the bulk of the

vertical transfer of momentum, heat, water vapor, hy-

drometeors, and AgI nuclei across the PBL. This is an

important assumption, since these eddies are responsible

for AgI nuclei mixing from near-ground generators into

the cloud layer. The LES run at 300m captures essential

dynamics in complex terrain (Chow et al. 2006; Weigel

et al. 2007), while the 100-m LES run resolves the most

energetic convective eddies (Xue et al. 2014). The AgI

nuclei are released in the first layer in the LES, that is, at

FIG. 2. Maps at 1800 UTC 18 Feb 2009 of (a) 700-mb height (thick black contours; 30-m interval), temperature

(colors), and wind barbs (full barb5 10ms21) and (b) 300-mb height (thick black contours; 60-m interval), wind speed

(colors), and wind barbs. The data are from the 12-km NAM; U.S. state boundaries and the Great Salt Lake are

shown as thin black lines. The red box represents the area shown in Fig. 1.

2268 JOURNAL OF APPL IED METEOROLOGY AND CL IMATOLOGY VOLUME 53

;10m above ground level (AGL), matching the height

of the generators. The AgI nuclei flow rate and size

distribution for the generators used in the WWMPP are

described in Super et al. (2010).

Xue et al. (2014) compared the vertical dispersion of

AgI particles in a 100-m LES run to airborne acoustic

ice nucleus counter measurements, for a flight over the

Medicine Bow Range on a day without cloud at and

below flight level. The AgI generators operated in that

case are the same as those used in the present case study.

They found that the modeled AgI particle concentra-

tions below the lowest flight level (starting as low as

280m AGL) were several orders of magnitude higher

than those within flight levels, and that at flight levels the

LES AgI particle concentrations matched the ice nu-

cleus counter measurements, at least within the range of

uncertainty of the instrument.

The 1.5-order subgrid-scale turbulent kinetic en-

ergy (SGS TKE) closure model is used in the LES runs

(Deardorff 1980; Moeng 1984). There is no SGS TKE in

900-m non-LES run that drives the LES runs; in fact, the

initial SGS TKE and the lateral boundary SGS TKE are

zero. Mirocha et al. (2014) have shown that these zero

SGS TKE conditions do not significantly impact the

inner domain of (idealized) LES simulations once to-

pography or wind/temperature perturbations exist in the

FIG. 3. TheWRF domains. Two-way-nested non-LES simulations are run in the 2700- and 900-m-resolution boxes.

Model output from the 900-m simulation is then used to drive two-way-nested LES runs in the 300- and 100-m

domains.

TABLE 1. WRF domains and physics choices.

Configurations/

domains 2700-m non-LES 900-m non-LES 300-m LES 100-m LES

Simulation time 0000 UTC 18 Feb 2009–0000 UTC 19 Feb 2009 1645–2100 UTC 18 Feb 2009

Grid points 360 3 720 210 3 420 540 3 330 810 3 810

Time step (s) 12 4 0.14 0.05

Nesting No Yes No Yes

Vertical levels 60 layers

Radiation Community Atmospheric Model shortwave and longwave schemes

PBL MYJ scheme None

Land surface Noah land surface scheme

Microphysics Thompson scheme with Xue et al. (2013a) AgI cloud seeding parameterization

Turbulence Horizontal Smagorinsky first-order closure 1.5-order TKE closure

OCTOBER 2014 CHU ET AL . 2269

domain. The more complex topography and greater

perturbations of wind and temperature induced by the

heterogeneous land surface types in our LES domains

accelerate the SGS TKE generation, compared to ide-

alized simulations (Xue et al. 2014). SGS TKE is always

a small portion of the total TKE simulated by the model

(Xue et al. 2014). Indeed, a valid LES simulation, by

definition, should resolve the most energetic eddies and

only parameterize the unsolved TKE as the SGS TKE.

Since the SGS TKE is always small compared to total

TKE, the zero SGS TKE boundary condition will likely

not impact the total TKE profile much. We will revisit

this issue in Part II. A very short time step of 1/21 s is

applied in the 100-m domain to ensure that the LES run

remains stable: the abruptly changing topography in-

troduces vigorous upward and downward motions close

to the ground, which may lead to a violation of the

Courant–Friedrichs–Lewy criterion in the vertical.

b. Selection of outer domain driver dataset

The simulations of this case study focus on the 4-h

UWKA flight time period. A simulation over such short

time scale is very sensitive to the initial and lateral

boundary conditions. A slight error in the timing of a

synoptic feature along the boundary of the 2.7-km do-

main (Fig. 3) or an error in the initial conditions can

profoundly impact the simulated weather overMedicine

Bow Mountain during the UWKA flight, via the nested

900-m non-LES that drives the LES 300- and 100-m

runs. Three different drivers for theWRF non-LES runs

were evaluated: the CFSR, the North American Re-

gional Reanalysis (NARR; http://www.emc.ncep.noaa.

gov/mmb/rreanl/), and the analysis data fromReal-Time

Four-DimensionalDataAssimilation system (RT-FDDA;

http://www.rap.ucar.edu/technology/model/rtfdda.php)

fromWWMPP. Of these three, the best dataset to drive

the non-LESWRF runs (and, subsequently, the 300- and

100-m LES runs) was selected based on comparisons

with a radiosonde sounding from Saratoga (Fig. 1),

snowfall at several stations, and radiometer LWP.

The comparisons indicate that the NARR is the least

accurate initialization with excessive low-level stability,

benign orographic clouds, and relatively weak pre-

cipitation. Although the CFSR initialization produces

a less stable PBL than observed over Saratoga (Fig. 4a),

it has the lowest low-level humidity bias, and its LWP

over the Medicine Bow Range validates the best. Also,

it is the only reanalysis that captured the dewpoint de-

pression near 450mb (Fig. 4a). All three drivers un-

derestimate snowfall over the 4 h of theUWKAflight, as

compared with ETI gauge measurements at five sites,

with snowfall rate steadily diminishing during this period.

The CFSR cumulative snowfall is 36% of the measured

amount, while the NARR snowfall is just 19%. Since the

CFSR version validates best, it is used to drive the LES

inner domains.

Both the 100- and 300-m resolution LES simulations

were run with and without AgI seeding, but our analysis

focuses on 100-m resolution LES output. The 100-m

LES model output is collected at 20-s intervals for all

wind components, at 1-min intervals for all hydrome-

teors, and at 5-min intervals for all other variables to

validate the model with UWKA and other data. We

refer to the control run (CTRL) and treated run

(TRTD) as the simulation with and without the AgI

nuclei injection from three point sources at the lowest

grid level. The CTRL run covers the entire event from

1500 to 2100 UTC and the TRTD run is from 1800

to 2100 UTC only, with AgI release from 1809 to

2015 UTC corresponding to the actual seeding period.

We validate the model by means of CTRLmodel output

during the NOSEED period and TRTD model output

during the SEED period, to match the WCR observa-

tions during the NOSEED and SEED periods. We also

examine the simulated AgI seeding impact by examin-

ing the simultaneous difference between TRTD and

CTRL during seed period.

4. Model validation of storm structure

a. Ambient conditions

The model soundings at Saratoga are compared to

observations in Fig. 4. The 100-m LES thermodynamic

profiles are almost the same as those from the parent

900-m non-LES. But the 100-m LES captures the low-

level wind veering better because the terrain is repre-

sented in more detail and because of direct eddy

representation rather than PBL parameterization. The

models overestimate the dewpoint by roughlymore than

2K on average below 600mb, thus the model LCLs re-

main lower than observed (Table 2). The average wind

speed from the surface to the mountain top U exceeds

the 900-m non-LES value by about 2.5m s21 in both the

300- and 100-m LES runs. Because of the higher mean

wind speed and weaker stability, the 300-m LES has

a bulk Froude number about twice the observed value.

Both LES runs also yield a smaller bulk Richardson

number because of weaker static stability and greater

wind shear. The very low bulk Ri values in the LES runs

suggest the possibility of wave breaking, which did occur

in the mountain lee, as discussed below.

b. Vertical transect of storm structure

The UWKA flew an along-wind leg before starting

the first ladder pattern (Fig. 1). The WCR reflectivity

2270 JOURNAL OF APPL IED METEOROLOGY AND CL IMATOLOGY VOLUME 53

(Fig. 5a) indicates that the storm was stratiform, rather

weak (peak reflectivity ; 10 dBZ), shallow, and rapidly

thinning on the downwind side of the mountain. The

WCRhydrometeor vertical velocity (Fig. 5b) is depicted

with a white color band (separating blue from red hues)

centered at 21m s21 to take a first-order hydrometeor

terminal velocity estimate of 1m s21 into account. Ac-

cordingly, the warm colors represent sinking motion (as

subsidence causes warming) and cold colors represent

rising motion. The finescale structures near the surface

on the upwind side of the mountain are shear-induced

PBL turbulent eddies. These have been explored over

the samemountain range in Geerts et al. (2011) andXue

et al. (2014). Turbulence is also seen at the higher levels

on the lee side of the mountain, probably because of

wave breaking. Flow subsides in the lee of the crest. The

sudden updraft at about 1628UTC (8 km from the crest)

and the lifting of the snow echo suggests boundary

layer separation, possible a weak hydraulic jump. The

strong flow acceleration near the crest, locally doubling

the cross-mountain wind speed to .25m s21 just below

flight level, and the rapid deceleration to almost zero

FIG. 4. Skew T–logp showing the WRF sounding (black) and the radiosonde sounding (red) at 1800 UTC over

Saratoga: (a) 900-m WRF non-LES; (b) 100-m WRF LES. A full barb represents 10m s21.

OCTOBER 2014 CHU ET AL . 2271

flow just east of the hydraulic jump (Fig. 5c) are consis-

tent with this interpretation. The shear in the hydraulic

jump results in strong turbulence further downwind

(Fig. 5b). The along-track wind speed (Fig. 5c) is derived

from WCR dual-Doppler synthesis using the nadir and

slant-forward antennas (Damiani and Haimov 2006).

The 100-m-resolution LES reproduces a similar pat-

tern of reflectivity (Fig. 5d), vertical velocity (Fig. 5e),

and wind along the transect (Fig. 5f), although there are

differences in resolution (discussed in section 2a). The

model reflectivity is calculated from hydrometeor con-

centrations and prescribed size distributions assuming

Rayleigh scattering, while scattering of 95-GHz (Wband)

radiation by large particles is in the Mie region, so the

WCR reflectivity values are less than the model values.

The reflectivity pattern is similar, with most snowfall

occurring on the upwind side of the mountain. The

model echo top starts to subside well upwind of the crest,

and the snow layer in the lee is shallower and tapers off

more quickly, without being lofted in a hydraulic jump.

The simulated vertical velocity and isentropes (Fig. 5e)

reveal vertically propagating waves similar to those ob-

served. The two main differences are that the model

does not fully resolve the PBL turbulence in the lowest

0.5–1.0 km (which may explain the thinner snow layer in

the lee), and that the hydraulic jump is further down-

wind (;20 km from the crest) and stronger, with a peak

updraft of 9m s21. The isentropes (Figs. 5e,f) suggest

that there may be multiple minor hydraulic jumps and

breaking waves in the lee at various elevations.

The WRF LES captures the flow acceleration across

the crest into the lee (Fig. 5f), but the peak winds are

about 10m s21 less than observed, and the strong winds

spread further downwind, consistent with the leeward

displacement of the main hydraulic jump. The overall

lee subsidence and associated warming are somewhat

overestimated by the WRF LES, at least at flight level,

as the lee warming is stronger than observed (Fig. 5g).

The model also has a slight warm bias at this level,

consistent with the Saratoga sounding (Fig. 4). Note that

the temperature sensor aboard the UWKA, a reverse

flow thermometer, has been corrected for evaporative

cooling following Wang and Geerts (2009).

c. Flight-level storm structure

We compare the model potential temperature u at

flight level to that observed for ladder 4 in Fig. 6. The

model data were interpolated to a constant height above

mean sea level (MSL) corresponding to the average

flight level for this ladder. The model output is valid at

a time close to the average time of ladder 4. The 100-m

LES model has a warm bias at this level, as mentioned

before. At this time significant warming occurs at this

level in the southeast (lee) side of themountain, as at the

time of the along-wind transect (Fig. 5g). The finescale

pattern of lee warming is quite transient. This warming is

due to lee subsidence (Fig. 7). The location of the main

hydraulic jump is evident as a sudden transition from

warm to much colder air, an;5-K difference over a few

100m. Dramatic local temperature variations in some

areas in the lee suggest wave breaking in progress. Most

of the UWKA ladder pattern is on the cooler upwind

side. Only slight warming is observed on track 5 (see

Fig. 1 for the track labeling) at this time; 45min earlier

(ladder 3) this track witnessedmore significant warming.

More important for the present study is the cool ano-

maly in the SW corner of the ladder pattern, both ob-

served andmodeled. As will be shown later, much liquid

water and snow growth occur in this region.

The model vertical velocities correspond well with the

observed values (Fig. 7), since the vertical velocity is

dominated by terrain-driven gravity waves. Modeled lee

subsidence is obvious, starting just upwind of the crest of

the Medicine Bow Range and other smaller mountains

nearby. Modeled subsidence is slightly stronger than

observed along track 5, at least during this ladder pat-

tern (number 4). The strong model updrafts further

downwind are associated with the hydraulic jump. The

sharp vertical velocity variations in the lee indicate

shear-driven turbulence. The hydraulic jump and asso-

ciated turbulence may have been present at the time of

TABLE 2. Sounding parameters of CSFR-driven 900-m non-LES,

the 300-m LES, and the 100-m LES, compared with the Saratoga

radiosonde (OBS). The wind speed U, the Froude number Fr, the

bulk Richardson number Ri, and the wind shear S represent av-

erages between the surface (10m AGL) and the elevation of

Medicine Bow Peak. For Brunt–Väisälä frequency Nds, we use the

dry (moist) value below (above) the cloud base (defined as the

LCL). The Froude number is calculated as U divided by Nds and

by the height H of the mountaintop above the sounding site in

Saratoga. The Richardson number is N2ds/S

2, where S is the mag-

nitude of the shear vector between the surface (10m) and moun-

taintop level. The quantity T* is the temperature at the LCL. The

observed LCL and Nds calculations start at 100m AGL because of

a radiosonde temperature bias when the radiosonde lacks venti-

lation (i.e., near release time).

Parameter OBS

900-m

non-LES

300-m

LES

100-m

LES

U (m s21) 14.0 14.0 16.4 16.5

Wind direction (8) 304 288 287 288

Nds (1022 s21) 0.64 0.4 0.34 0.47

Fr 1.4 2.4 3.2 2.3

Ri 1.2 0.7 0.18 0.34

LCL (m) 702 469 486 489

T* (8C) 29.1 29.0 29.1 29.1

S (1023 s21) 5.9 4.9 8.2 8.2

Shear direction (8) 222.9 195 217.3 218.6

2272 JOURNAL OF APPL IED METEOROLOGY AND CL IMATOLOGY VOLUME 53

FIG. 5. Analysis of a transect flown along the wind (Fig. 1) from northwest (left) to southeast (right) between 1621:45 and 1635:55 UTC.

The wind blows from left to right. WCR (a) reflectivity and (b) vertical velocity above and below flight level. The black stripe is the WCR

blind zone,;230m deep, centered at flight level. (c) TheWCR dual-Doppler along-track wind speed below flight level only. (d)–(f) As in

(a)–(c), but from the 100-m LES with isentropes contoured in each panel at 2-K interval from 290K. (g) The flight-level potential

temperature comparison.

OCTOBER 2014 CHU ET AL . 2273

ladder 4 downstream of track 5. Much weaker updrafts

are present in the model over the upwind valley and

foothills (Fig. 7), consistent with the cool anomaly. These

are important for snow growth.

The relationship between flight-level observed and

modeled w and u is explored further in Table 3, for

all individual tracks. If the vertical velocity and u fields

were purely turbulent (meaning that the spectrum of

these fields is well behaved in the inertial subrange),

there would be no correlation between collocatedmodel

and point observations: validation would have to be done

in terms of distributions, for example, spectral power

(e.g., Xue et al. 2014). On the other hand, a good cor-

relation can be expected in terrain-forced gravity wave

flow, if the model accurately captures the upstreamwind

profile and stratification, which it does (Fig. 4). The high

correlation coefficients indicate that terrain-forced grav-

ity waves dominate w and u variations along the upwind

tracks especially track 1. The r values listed in Table 3

generally are statistically significant, as there is just a 0.1%

FIG. 6. Potential temperature at 4.25 km MSL according to the 100-m-resolution LES at 1940 UTC (color field) and according to

UWKA measurements (ladder pattern 4, centered at the same time). The black contours are terrain heights from 2.0m to 3.5 km with

a 0.5-km interval.

2274 JOURNAL OF APPL IED METEOROLOGY AND CL IMATOLOGY VOLUME 53

and 10% chance that the correlation coefficient for un-

related variables exceeds 0.164 and 0.082, respectively, for

the number of pairs in this comparison. The lack of cor-

relation for track 5, located in the lee, indicate that non-

linear flow and turbulence dominate there. The model

variances in u and w are larger than observed along track

1, indicating a greater-than-observed gravity wave am-

plitude. The observed and modeled variance is smaller

further downwind over tracks 3–5 as the wave amplitude

is smaller and no wave breaking has occurred yet. The

model’s warm bias is evident in all tracks.

d. Orographic cloud liquid water

Model glaciogenic seeding efficacy is very sensitive to

liquid water content (LWC), which in turn is quite sen-

sitive to orographic flow dynamics and the microphys-

ical parameterization as will be explained in Part II.

Therefore we validate model LWC. We first compare

the slant-path LES LWPwith the radiometer LWP. The

radiometer was aimed toward the mountain in two di-

rections, east (808) and southeast (1358), at a 98 eleva-tion angle. The model output is computed along the

FIG. 7. As in Fig. 6, but for vertical velocity. The airborne gust probe measurements are surrounded by a thick black contour to improve

clarity.

OCTOBER 2014 CHU ET AL . 2275

same slant path, not a vertical integral above Cedar

Creek. But both model and observed values are ex-

pressed as a vertical integral (i.e., equivalent water

depth; units mm). The radiometer data have a 1-min

sampling interval. They were averaged over 5min to

be consistent with model data examined at a 5-min in-

terval (Fig. 8). Even so, we cannot expect the individual

LWP spikes to match because of PBL turbulence and

the sensitivity of liquid water to PBL updrafts above

cloud base. The radiometer LWP decreased about 50%

during SEED (1815–2015UTC) east of Cedar Creek, but

slightly increased to the southeast, consistent with a veer-

ing of the surface wind at Saratoga between 1800 and

1900 UTC. But both 300- and 100-m LES runs show a no

significant trend during SEED, in both radiometer beam

directions, because the model does not reproduce the

observed wind shift.

On average themodel LWP in the two directions is less

than the observed values over the period shown in Fig. 8,

even though the model overestimates low-level relative

humidity (Fig. 4). The radiometer measurements are

somewhat uncertain (Ware et al. 2003).

The supercooled LWC measured at flight level is

less uncertain because there are three independent

LWC measurements. They are the Forward Scattering

Spectrometer Probe (FSSP) bin-integrated LWC, the

Gerber LWC, and the Droplet Measurement Tech-

nologies (DMT)-100 hotwire LWC. The Gerber and

FSSP LWC estimates agreed well on this flight, with

correlation coefficient of 0.82. The averaged values of

FSSP and Gerber probes are used as observation. The

comparison between observed and modeled LWC at

flight level (Fig. 9) shows that liquid water is found

mostly near the southwest part of the ladder pattern,

where upstream ascent (Fig. 7) and lower temperature

(Fig. 6) prevail. The highest LWC values are observed

near the turn connecting tracks 3 and 4, consistent with

modeled LWC. However, the model has no liquid water

over higher terrain (above 3.0-km elevation), inconsistent

with observations. Also, the model orographic cloud

became shallower over time, as discussed below. In fact,

the comparison shown in Fig. 9 is possible only for the

first ladder because the model storm intensity decreased

rapidly afterward.

e. Vertical velocity patterns

An example transect of WCR-measured hydrome-

teor vertical velocity is shown in Fig. 10a, for track 4 on

ladder 2. The wind is mostly out of the page of this

transect, but there is an along-transect wind component

from left to right (counterclockwise of 3098) below

;3.5 km, according to the sounding (Fig. 4). Finescale

structures are evident in the lowest 0.5–1.0 km above

the surface in the WCR transect. Both shear-induced

transient turbulence and terrain-driven eddies can be

seen in this layer. Terrain-driven eddy dipoles can be

seen just to the right of the highest terrain, and across

a hill at the far right (northeast) side in Fig. 10a. They

were present also in the three other passes over track 4

(not shown). A spectral density analysis of WCR vertical

velocity as function of height, as in Geerts et al. (2011),

shows that the finescale vertical motions near the ground

represent the inertial subrange, although punctuated

by terrain-forced frequencies. The highest frequencies

(shortest wavelength) resolvedwithin the inertial subrange

depend on the radar resolution. Smoother, wavelike ver-

tical motions can be seen above the boundary layer

(Fig. 10a). Subsidence prevails, because track 4 is located

close to the crest of theMedicine BowRange (Fig. 1), and

airflow aloft plunges into the lee above the crest, as re-

vealed by the along-wind WCR transect (Fig. 5).

The 100-m LES rendition of this transect is shown in

Fig. 10b. Here vertical air motion is shown, hence the

display color changes at 0m s21, whereas in Fig. 10a the

color changes at 21ms21 to facilitate the comparison,

assuming a first-order guess hydrometeor fall speed of

1m s21. The 100-m LES nicely captures the terrain-

forced eddies, for example, in the same two places

TABLE 3. Validation of flight-level vertical velocity w and potential temperature u for the five tracks shown in Fig. 1. Shown are the

average and standard deviation (std dev) for the 100-m LES simulation and for observations, and the correlation coefficient r between

collocated model on observed values. The variables are computed for each ladder separately, and then averaged.

Track 1 Track 2 Track 3 Track 4 Track 5

w (m s21) Model mean 20.65 20.61 20.30 20.10 0.05

Observed mean 20.56 20.41 20.19 20.01 20.13

Model std dev 1.08 0.57 0.48 0.33 0.46

Observed std dev 0.71 0.59 0.51 0.49 0.52

r 0.68 0.42 0.52 0.27 0.18

u (K) Model mean 296.5 295.6 295.3 295.1 295.1

Observed mean 294.7 294.2 294.1 294.0 294.0

Model std dev 1.22 0.66 0.51 0.41 0.37

Observed std dev 0.57 0.45 0.38 0.35 0.32

r 0.60 0.55 0.12 0.09 0.14

2276 JOURNAL OF APPL IED METEOROLOGY AND CL IMATOLOGY VOLUME 53

mentioned above. The horizontal resolution of the WCR

is an order of magnitude better than the model’s, so the

model does not capture the smaller turbulent eddies seen

by the WCR. Nevertheless, the correspondence is rather

good. The 100-m LES also shows subsident flow aloft,

associated with a windward-tilting gravity wave. This

subsidence is strongest just to the right of the highest ter-

rain, consistent with Fig. 7.

5. Model validation of radar reflectivity changes

a. Reflectivity and precipitation rate

Instantaneous surface snowfall rate is very difficult to

measure (Rasmussen et al. 2012). Numerous studies

have established relationships between radar reflectivity

Z and precipitation rate R. In effect near-surface re-

flectivity is a good surrogate measure of surface snow-

fall rate. WCR reflectivity is dominated by ice crystals

in mixed-phase clouds, rather than the much smaller

cloud droplets, because of the sixth-order dependence

on diameter for Rayleigh scattering. But at W band

(3-mm wavelength) large hydrometeors (larger than

;0.8mm) scatter in the Mie regime. Mie scattering is

complex and highly dependent on crystal shape, orien-

tation, and size distribution, but, in general, the number

concentration becomes more important, and the di-

ameter less important, relative to Rayleigh scattering.

Nevertheless, data from in situ precipitation particle

FIG. 8. LWP trend as observed by the radiometer and as simulated by the 300- and 100-m

WRF LES.

OCTOBER 2014 CHU ET AL . 2277

probes and close-range WCR reflectivity values col-

lected in orographic storms over the Medicine Bow

Range, including the 18 February 2009 case, show

a robust relationship between Z (mm6m23) and

R (mmh21) at W band (Geerts et al. 2010; Pokharel

and Vali 2011), consistent with theory (Matrosov 2007).

The relationship used here is from Pokharel and Vali

(2011):

R5 0:39Z0:58 . (1)

The Mie scattering effect tends to reduce reflectivity

relative to centimeter-wave radars for larger particles.

Thus model reflectivity, computed from rain, snow, and

graupel mixing ratios and parameterized size distribu-

tions (Thompson et al. 2004; 2008) assuming Rayleigh

scattering, cannot be compared directly in value toWCR

reflectivity, but patterns and trends can be compared, and

the radar-based precipitation rate from Eq. (1) can be

compared with model precipitation rate. Thus radar re-

flectivity is the main tool to detect the seeding effect.

FIG. 9. As in Fig. 7, but for liquid water content.

2278 JOURNAL OF APPL IED METEOROLOGY AND CL IMATOLOGY VOLUME 53

b. Observations

The earliest UWKA pass over track 4 (Fig. 11a) shows

a high reflectivity band intersected by the mountain,

suggesting optimal snow growth at this level (;3.5 km

MSL). The temperature at this level (2138C) suggests

growth by vapor deposition. In all transects reflectivity is

elevated on the southwest side, but near the surface on the

northeast side, indicating that the southwest side is mostly

upwind of the crest, and the northeast side in the lee, as

mentioned in section 4e. The near-surface reflectivity is

higher during the last two passes in the snow growth re-

gion to the left of themountaintop. ThreeAgI generators,

located ;11km into the page for these cross sections,

were in operation at this time (SEED). The observed in-

crease in reflectivity downwind of the AgI generators, in

particular the leftmost generator located close to an area

of high observed cloud liquid water (section 4d), can be

due to natural causes rather than AgI seeding.

To examine precipitation changes across the ladder

pattern flown four times on the 18 February 2009 flight,

we examine frequency-by-altitude diagrams (FADs)

(Yuter and Houze 1995) of WCR reflectivity. Specifi-

cally, we contrast target tracks (2–5) against the control

track (track 1) during NOSEED and SEED (Fig. 12).

The reflectivity FADs are constructed as in Geerts et al.

(2010). The vertical axis is expressed as above ground

level. All counts are normalized, so FADs based on

different sample sizes can be compared with each

other. The ‘‘data presence’’ line indicates the fraction of

bins with reflectivity values above a range-dependent

threshold at each height level. The depression in the

data presence line between 1.0 and 1.5 kmAGL is due to

the radar blind zone centered at flight level (Fig. 11).

The highest frequencies in the top four panels of Fig. 12

are near the surface, confirming that the precipitation

is shallow. In all four panels the mean reflectivity in-

creases toward the ground, indicating low-level snow

growth, although there may be some sublimation (or

particle breakup) in the lowest 300m above the surface,

especially along the control leg. There are some appar-

ent differences near the surface between SEED and

NOSEED reflectivity FADs of target and control tracks.

Positive differences in Fig 12c indicate more counts

during SEED. A higher frequency of larger reflectivity

(.8 dBZ) and lower frequency of smaller reflectivity

during SEED is evident in the lowest 1.0 kmAGL of the

difference FAD for the target tracks.

The mean reflectivity in the lowest 1.0 km is about

1 dBZ greater during SEED: the solid line in Fig. 12c is

to the right of the dashed line. This implies that snowfall

was ;14% heavier during SEED, according to Eq. (1).

A control track was flown quasi-simultaneously upwind

of the AgI generators. The difference FAD for the con-

trol track (Fig. 12f) reveals an opposite low-level dipole

compared to the one for the target tracks, that is, a de-

crease in precipitation near the surface. If we assume that

the natural evolution is the same over the control and target

tracks, then we can conclude from these two FAD dif-

ference plots that even though the low-level snowfall rate

FIG. 10. (a) WCR hydrometeor vertical velocity transect for the second pass (1947–1954 UTC) over track 4 (see

Fig. 1 for location). (b) The 100-m LES vertical air motion for the same track at about the same time (1945UTC). No

velocity data exist above the WCR echo top at ;5 km MSL in (a).

OCTOBER 2014 CHU ET AL . 2279

was weakening during SEED, a ‘‘positive’’ seeding effect

is observed, in magnitude larger than the net difference

near the surface seen in Fig. 12c. Above 1km AGL (be-

yond the reach of the surface-based AgI seeding) the

differenceFADs are fairly similar, suggesting that natural

changes between NOSEED and SEED periods indeed

are rather uniform across the ladder pattern. Still, the

assumption of uniform natural evolution at low levels in

the entire domain from track 1 to track 5 over the 3.5-h

period remains unproven and impossible to validate with

the available data.

c. Simulations

The reflectivity transects derived from the 100-m LES

run along track 4 in Fig. 13 can be compared with time-

matched WCR transects (Fig. 11). The model storm

depth and near-surface reflectivity steadily decreased

along track 4, especially from the first to the second pass.

Values as high as 30 dBZ are found up to 4 km MSL on

the left (inflow) side in the earliest transect. The highest

values are near 3.5 km MSL, as observed. The storm

becomes shallower than observed during the last three

passes and weakens, especially on the left side, in dis-

agreement with WCR observations (Fig. 11).

The three AgI generators are switched on in the 100-m

LES at the time this occurred in reality, leading to

a parameterized interaction of AgI nuclei with cloud, as

discussed in section 3. Model reflectivity is not enhanced

near the ground around the downwind projections of the

AgI generators during the SEED period (Figs. 13c,d).

Maybe a small region of elevated reflectivity is present

in Fig. 13c, near the ground downwind of the leftmost

FIG. 11. Transects ofWCR reflectivity at different times along track 4 (a),(b) during NOSEEDand (c),(d) during

SEED. Stars indicate the intersection of the center of the AgI plumes with this transect assuming the sounding-

based wind direction listed in Table 2.

2280 JOURNAL OF APPL IED METEOROLOGY AND CL IMATOLOGY VOLUME 53

generator, which is most likely to have an impact because

of the high LWC there (Fig. 8).

Model reflectivity FADs are plotted in the same man-

ner as WCR reflectivity (Fig. 12), for matching locations

and times, in Fig. 14. The 100-m LES simulated refle-

ctivity data were extracted over a domain bounded by

tracks 2 and 5 (target), and over an ;4-km-wide belt

centered at track 1 (control), both during the NOSEED

and SEED periods. The hydrometeor diversity during

NOSEED (Figs. 14a,d) reasonably matches WCR ob-

servations (Figs. 12a,d) but the LES reflectivity distri-

bution is broader, and there is no evidence of low-level

FIG. 12. FADs of WCR reflectivity using (a) all target track passes (2–5) in ladders 1 and 2; (b) as in (a), but for ladders 3 and 4. The

white line in each panel shows the height-dependent percentage of points with echoes. (c) The difference between the data in (b) and (a),

SEED 2 NOSEED. The dashed and solid lines are the average reflectivity profiles from (a) and (b). (d)–(f) As in (a)–(c), but for the

control track (track 1). The precipitation rate in (c) and (f) is based on the Z–R relationship in Eq. (1).

OCTOBER 2014 CHU ET AL . 2281

sublimation, not even in the control region, consistent

with the model’s lower-than-observed LCL (Table 2).

Not surprisingly, the model precipitation near track 1

tends to be weaker than over the mountain (tracks 2–5),

as observed.

As expected from Fig. 13, the storm depth and intensity

are much less during the SEED period, both in the control

and target regions (Figs. 14b,e). The strong dipole in

Figs. 14c,f reflects this weakening from the SEED to the

NOSEED period. The mean reflectivity vertical profiles

for NOSEED and SEED periods in Figs. 14c,f indicate

the reflectivity decreased more in the control area than in

the target area. In fact the near-surface precipitation rate

decreased by 1.3 orders of magnitude in the control area,

but by just half an order of magnitude in the target area,

according to a typical Z–R relationship (Z 5 200R1.6;

e.g., Rasmussen et al. 2003) (The actual model surface

precipitation rate will be explored in Part II.) This can be

interpreted as a ‘‘differential’’ positive seeding signal, if we

can assume that the natural storm changes are uniform

over the entire domain: the natural storm weakening is

slowed down by seeding in the target (Figs. 14c,f). Indeed,

the treatedminus untreated (TRTD2CTRL) differential

reflectivity FAD shows a positive dipole in the PBL (the

lowest 0.5–1.0km) in the same target region, but, as ex-

pected, no change in the control region (Fig. 15). The

TRTD 2 CTRL comparison is based on identical time

periods (the SEED period), but one simulation with the

AgI generators on, the other without any seeding.

6. Discussion

The difference in near-surface mean reflectivity due

to AgI seeding (TRTD 2 CTRL) is a mere 0.8 dBZ

FIG. 13. As in Fig. 11, but for the 100-mLES computed radar reflectivity along track 4, at times corresponding to

the four WCR reflectivity transects (Fig. 11).

2282 JOURNAL OF APPL IED METEOROLOGY AND CL IMATOLOGY VOLUME 53

(Fig. 15a) corresponding to a 12% precipitation increase

according to the above Z–R relationship. Yet the

near-surface SEED–NOSEED difference in the target

region, a nonsimultaneous comparison, is 28 dBZ

(Fig. 14c), corresponding to a.300% precipitation rate

decrease. In other words, natural precipitation vari-

ability overwhelmed the seeding signal in this case. The

storm intensity was steadier in the period starting 1–3 h

before the start of the aircraft-determined NOSEED

period. We evaluated hypothetical SEED–NOSEED

comparisons of the same duration but with an earlier

start time using the 100-m LES model output, but none

had near-surfacemean reflectivity change below 2.0 dBZ.

In other words, themodel suggests that natural variability

(SEED 2 NOSEED) always overwhelms the seeding

signal (TRTD2CTRL) in this case. TheWCR-observed

FIG. 14. As in Fig. 12, but using the 100-m-resolution WRF LES data at matching locations and for approximately matching times. Thus

(c) and (f) show a nonsimultaneous comparison. The precipitation rate in (c) and (f) is based on a Z–R relationship.

OCTOBER 2014 CHU ET AL . 2283

changes (SEED 2 NOSEED) are rather small in the

target region (Fig. 12), and while the storm weakened

somewhat at upper levels, a positive reflectivity change

was observed in the PBL (Fig. 12c). This low-level re-

flectivity change was only11 dBZ, about the same as the

magnitude of change expected from ground-based AgI

seeding, according to the 100m LES (Fig. 15). That may

be a mere coincidence. The low-level storm weakening

in the control region (Fig. 12f) may suggest that the

seeding impact is actually stronger: if we assume that

natural changes are spatially uniform across the moun-

tain, then one can argue that the seeding impact in the

target area is closer to 13 dBZ, that is, a reflectivity

change of11 dBZ in the target area relative to a22-dBZ

change in the control area. But the 100-m LES indi-

cates that natural differences between the two periods

(SEED 2 NOSEED without any seeding in either pe-

riod) are not spatially uniform (not shown, but very

similar to Figs. 14c,f); in other words, the ‘‘control’’

member is not a good control.

This case study then provides tentative answers to

the two questions posed in the introduction: first, the

seeding-induced changes appear small in comparison

with natural precipitation variability. Second, the ob-

served changes in a nearby control (untreated) area may

not be representative of natural changes in the target

(treated) area. Still, we are hopeful that progress can

be made in detecting an impact of AgI seeding on cloud

and precipitation through a synergy of high-resolution

cloud-resolving LES simulations with a series of com-

plementary in situ and remote measurements. The

model output can be explored to understand the cloud

processes leading to changes in snow growth within

the model world and to confine the parameter space of

effective seeding conditions (as will be done in Part II,

for the 18 February 2009 case). The observations may

yield further insight into physical processes whose rep-

resentation in the model may be inadequate (e.g., vali-

dation of LWC), or into processes not represented at all

in the model (e.g., blowing snow; Geerts et al. 2011).

Cases have to be selected carefully, ensuring relatively

steady atmospheric and storm conditions. Datasets

similar to the one presented in this paper have been

collected for a total of 24 cases between 2008 and 2013

over both Medicine Bow and adjacent Sierra Madre

ranges in Wyoming (Geerts et al. 2010, 2013). Several

other cases, with different ambient and storm condi-

tions, will be examined using a combination of highly

resolved remotely sensed data and simulations.

7. Conclusions

Profiling airborne radar data collected on 18 February

2009 over the Medicine Bow Range in Wyoming are

used to validate numerical simulations of the impact of

ground-based glaciogenic seeding on cloud and pre-

cipitation in a shallow orographic winter storm. The

WRFmodel is used in LES mode, with an inner domain

resolution of 100m, in a domain covering the mountain

range (;80 km). A glaciogenic cloud seeding parame-

terization is coupled with the Thompson microphysics

scheme (Xue et al. 2013a). Modeled ambient and storm

conditions during the flight differed depending on the

dataset used for initial and boundary conditions; the

dataset whose 900-m non-LES model output verifies

FIG. 15. WRF LES reflectivity difference FAD for the simula-

tion with AgI seeding vs that without AgI seeding (i.e., treated 2untreated, a simultaneous comparison): (a) target tracks 2–5 and

(b) control track 1.

2284 JOURNAL OF APPL IED METEOROLOGY AND CL IMATOLOGY VOLUME 53

best in terms of upwind stability and wind profiles, LWP,

and surface precipitation is then used for LES simula-

tions down to 100m.

The 100-m LES model compares relatively well to the

more resolved radar vertical velocity and reflectivity in

vertical transects across the mountain, along and across

the low-level wind direction. The modeled storm verti-

cal motion over the mountain reveals a turbulent

boundary layer, 0.5–1.0 km deep, and gravity waves

above the PBL, both as observed. The storm structure

also validates well, but the model storm dissipates more

rapidly than observed by radar. The WCR reflectivity

FADs show a positive seeding effect in that near-surface

reflectivity increases downwind of the AgI generators,

both in absolute terms and relative to the upwind

change, which is negative; that is, the low-level snowfall

rate decreases in the control region upwind of the AgI

generators. Time- and space-matched model reflectivity

FADs do confirm this positive seeding effect in a relative

sense (relative to the upwind change) but not in an ab-

solute sense. In this case the modeled storm intensity

decrease overwhelms the seeding effect, which can be

isolated in themodel. Thus this case study illustrates that

the verification challenge largely comes down to the

detection of a relatively small signal in the noise of

natural variability, even at small time and space scales,

precisely as stated in Garstang et al. (2005).

Acknowledgments. This work and the ASCII cam-

paign were funded by the National Science Foundation

Grant AGS-1058426, the Wyoming Water Develop-

ment Commission, and the U.S. Geological Survey,

under the auspices of the University of WyomingWater

Research Program. The operation of the AgI genera-

tors, the ETI gauges, and the microwave radiometer was

supported by theWWMPP, which is funded by the State

of Wyoming and managed by Barry Lawrence. We used

the NCAR-supported graphing software NCL.

REFERENCES

Black, M., and J. King, 2009: The Atlas of Water: Mapping the

World’s Most Critical Resource. 2nd ed. University of Cal-

ifornia Press, 128 pp.

Breed, D., R. Rasmussen, B. Lawrence, B. Boe, T. Deshler, and

C. Weeks, 2014: Evaluating winter orographic cloud seeding:

Design of the Wyoming Weather Modification Pilot Project

(WWMPP). J. Appl. Meteor. Climatol., 53, 282–299, doi:10.1175/

JAMC-D-13-0128.1.

Bruintjes, R. T., 1999: A review of cloud seeding experiments to

enhance precipitation and some new prospects. Bull. Amer.

Meteor. Soc., 80, 805–820, doi:10.1175/1520-0477(1999)080,0805:

AROCSE.2.0.CO;2.

Chow, F. K., A. P.Weigel, R. L. Street, M.W. Rotach, andM. Xue,

2006: High-resolution large-eddy simulations of flow in a steep

Alpine valley. Part I: Methodology, verification, and sensi-

tivity experiments. J. Appl. Meteor. Climatol., 45, 63–86,

doi:10.1175/JAM2322.1.

Crewell, S., andU. Lohnert, 2003:Accuracy of cloud liquidwater path

from ground-based microwave radiometry 2. Sensor accuracy

and synergy. Radio Sci., 38, 8042, doi:10.1029/2002RS002634.

Damiani, R., and S. Haimov, 2006: A high-resolution dual-

Doppler technique for fixed multi-antenna airborne radar.

IEEE Trans. Geosci. Remote Sens., 44, 3475–3489, doi:10.1109/

TGRS.2006.881745.

Deardorff, J.W., 1980: Stratocumulus-cappedmixed layers derived

from a three-dimensional model. Bound.-Layer Meteor., 18,

495–527, doi:10.1007/BF00119502.

DeMott, P. J., 1995: Quantitative descriptions of ice formation

mechanisms of silver iodide-type aerosols.Atmos. Res., 38, 63–

99, doi:10.1016/0169-8095(94)00088-U.

Garstang, M., R. Bruintjes, R. Serafin, H. Orville, B. Boe,

W. Cotton, and J. Warburton, 2005: Weather modification:

Finding common ground. Bull. Amer. Meteor. Soc., 86, 647–

655, doi:10.1175/BAMS-86-5-647.

Geerts, B., Q. Miao, Y. Yang, R. Rasmussen, and D. Breed, 2010:

An airborne profiling radar study of the impact of glaciogenic

cloud seeding on snowfall from winter orographic clouds.

J. Atmos. Sci., 67, 3286–3302, doi:10.1175/2010JAS3496.1.

——, ——, and ——, 2011: Boundary layer turbulence and oro-

graphic precipitation growth in cold clouds: Evidence from

profiling airborne radar data. J. Atmos. Sci., 68, 2344–2365,

doi:10.1175/JAS-D-10-05009.1.

——, and Coauthors, 2013: The AgI Seeding Cloud Impact In-

vestigation (ASCII) campaign 2012: Overview and pre-

liminary results. J.Wea.Modif., 45, 24–43. [Available online at

http://www.weathermodification.org/publications/index.php/

JWM/article/view/121/pdf.]

Klazura, G. E., and C. J. Todd, 1978: A model of hygroscopic

seeding in cumulus clouds. J. Appl. Meteor., 17, 1758–1768,

doi:10.1175/1520-0450(1978)017,1758:AMOHSI.2.0.CO;2.

Liu, C., K. Ikeda, G. Thompson, R. Rasmussen, and J. Dudhia,

2011: High-resolution simulations of wintertime precipitation

in the Colorado Headwaters region: Sensitivity to physics

parameterizations. Mon. Wea. Rev., 139, 3533–3553,

doi:10.1175/MWR-D-11-00009.1.

Matrosov, S. Y., 2007: Modeling backscatter properties of snowfall

at millimeter wavelengths. J. Atmos. Sci., 64, 1727–1736,

doi:10.1175/JAS3904.1.

Mirocha, J., B. Kosovic, and G. Kirkil, 2014: Resolved turbulence

characteristics in large-eddy simulations nested within meso-

scale simulations using the Weather Research and Fore-

casting Model. Mon. Wea. Rev., 142, 806–831, doi:10.1175/

MWR-D-13-00064.1.

Moeng, C.-H., 1984: A large-eddy-simulation model for the study of

planetaryboundary-layer turbulence. J.Atmos. Sci., 41,2052–2062,

doi:10.1175/1520-0469(1984)041,2052:ALESMF.2.0.CO;2.

National Research Council, 2003: Critical Issues in Weather Mod-

ification Research. National Academies Press, 144 pp.

Ogura, Y., 1963: The evolution of a moist convective element in a shal-

low, conditionally unstable atmosphere: A numerical calculation.

J. Atmos. Sci., 20, 407–424, doi:10.1175/1520-0469(1963)020,0407:

TEOAMC.2.0.CO;2.

Orville, H. D., 1965: A numerical study of the initiation of cumulus

clouds over mountainous terrain. J. Atmos. Sci., 22, 684–699,

doi:10.1175/1520-0469(1965)022,0684:ANSOTI.2.0.CO;2.

Pokharel, B., and G. Vali, 2011: Evaluation of collocated mea-

surements of radar reflectivity and particle sizes in ice

OCTOBER 2014 CHU ET AL . 2285

clouds. J. Appl. Meteor. Climatol., 50, 2104–2119, doi:10.1175/

JAMC-D-10-05010.1.

Rasmussen, R., and Coauthors, 2001: Weather Support to Deicing

Decision Making (WSDDM): A winter weather nowcasting

system. Bull. Amer. Meteor. Soc., 82, 579–595, doi:10.1175/

1520-0477(2001)082,0579:WSTDDM.2.3.CO;2.

——, M. Dixon, S. Vasiloff, F. Hage, S. Knight, J. Vivekanandan,

and M. Xu, 2003: Snow nowcasting using a real-time corre-

lation of radar reflectivity with snow gauge accumulation.

J. Appl.Meteor., 42, 20–36, doi:10.1175/1520-0450(2003)042,0020:

SNUART.2.0.CO;2.

——, and Coauthors, 2011: High-resolution coupled climate runoff

simulations of seasonal snowfall over Colorado: A process

study of current and warmer climate. J. Climate, 24, 3015–

3048, doi:10.1175/2010JCLI3985.1.

——, and Coauthors, 2012: How well are we measuring snow: The

NOAA/FAA/NCARwinter precipitation test bed.Bull. Amer.

Meteor. Soc., 93, 811–829, doi:10.1175/BAMS-D-11-00052.1.

Schaefer, V. J., 1946: The production of ice crystals in a cloud of

supercooled water droplets. Science, 104, 457–459, doi:10.1126/

science.104.2707.457.

Simpson, J., and V. Wiggert, 1971: 1968 Florida Cumulus

Seeding Experiment: Numerical model results. Mon. Wea.

Rev., 99, 87–118, doi:10.1175/1520-0493(1971)099,0087:

FCSENM.2.3.CO;2.

Skamarock, W. C., and Coauthors, 2008: A description of the Ad-

vanced Research WRF version 3. NCAR Tech. Note NCAR/

TN-4751STR, 113 pp. [Available online at http://www.mmm.

ucar.edu/wrf/users/docs/arw_v3_bw.pdf.]

Super, A. B., B. A. Boe, J. A. Heimbach, J. T. McPartland, and

G. Langer, 2010: Comparison of silver iodide outputs from two

different generators and solutions measured by acoustic ice

nucleus counters. J. Wea. Modif., 42, 49–60.

Thompson, G., R. M. Rasmussen, and K. Manning, 2004: Explicit

forecasts of winter precipitation using an improved bulk

microphysics scheme. Part I: Description and sensitivity

analysis. Mon. Wea. Rev., 132, 519–542, doi:10.1175/

1520-0493(2004)132,0519:EFOWPU.2.0.CO;2.

——, P. R. Field, R. M. Rasmussen, and W. D. Hall, 2008: Explicit

forecasts of winter precipitation using an improved bulk

microphysics scheme. Part II: Implementation of a new

snow parameterization. Mon. Wea. Rev., 136, 5095–5115,

doi:10.1175/2008MWR2387.1.

Vonnegut, B., 1947: The nucleation of ice formation by silver io-

dide. J. Appl. Phys., 18, 593–595, doi:10.1063/1.1697813.

Wang, Y., and B. Geerts, 2009: Estimating the evaporative cooling

bias of an airborne reverse flow thermometer. J. Atmos.

Oceanic Technol., 26, 3–21, doi:10.1175/2008JTECHA1127.1.

Ware, R., F. Solheim, R. Carpenter, J. Gueldner, J. Liljegren,

T. Nehrkorn, and F. Vandenberghe, 2003: A multi-channel

radiometric profiler of temperature, humidity and cloud

liquid. Radio Sci., 38, 8079, doi:10.1029/2002RS002856.

Weigel, A. P., F. K. Chow, and M. W. Rotach, 2007: On the nature

of turbulent kinetic energy in a steep and narrow Alpine

valley. Bound.-Layer Meteor., 123, 177–199, doi:10.1007/

s10546-006-9142-9.

Xue, L., A. Teller, R. M. Rasmussen, I. Geresdi, and Z. Pan, 2010:

Effects of aerosol solubility and regeneration on warm-phase

orographic clouds and precipitation simulated by a detailed

bin microphysical scheme. J. Atmos. Sci., 67, 3336–3354,

doi:10.1175/2010JAS3511.1.

——, ——, ——, ——, ——, and X. Liu, 2012: Effects of aerosol

solubility and regeneration on mixed-phase orographic clouds

and precipitation. J. Atmos. Sci., 69, 1994–2010, doi:10.1175/JAS-D-11-098.1.

——, and Coauthors, 2013a: AgI cloud seeding effects as seen in

WRF simulations. Part I: Model description and idealized 2D

sensitivity tests. J. Appl. Meteor. Climatol., 52, 1433–1457,doi:10.1175/JAMC-D-12-0148.1.

——,S. Tessendorf, E.Nelson,R.Rasmussen,D.Breed, S. Parkinson,

P. Holbrook, and D. Blestrud, 2013b: AgI cloud seeding effects

as seen in WRF simulations. Part II: 3D real case simulations

and sensitivity tests. J. Appl. Meteor. Climatol., 52, 1458–1476,

doi:10.1175/JAMC-D-12-0149.1.

——, X. Chu, R. Rasmussen, D. Breed, B. Boe, and B. Geerts,

2014: The dispersion of silver iodide particles from ground-

based generators over complex terrain. Part II:WRF large-eddy

simulations versus observations. J. Appl. Meteor. Climatol., 53,

1342–1361, doi:10.1175/JAMC-D-13-0241.1.

Young, K. C., 1974: A numerical simulation of winter-

time, orographic precipitation: Part I. Description of

model microphysics and numerical techniques. J. Atmos.

Sci., 31, 1735–1748, doi:10.1175/1520-0469(1974)031,1735:

ANSOWO.2.0.CO;2.

Yuter, S. E., and R. A. Houze Jr., 1995: Three-dimensional

kinematic and microphysical evolution of Florida cumu-

lonimbus. Part II: Frequency distributions of vertical ve-

locity, reflectivity, and differential reflectivity. Mon. Wea.

Rev., 123, 1941–1963, doi:10.1175/1520-0493(1995)123,1941:

TDKAME.2.0.CO;2.

2286 JOURNAL OF APPL IED METEOROLOGY AND CL IMATOLOGY VOLUME 53