Please note that this article, which is not highlighted on the
weekly PNAS Tipsheet, is provided as a nonfinal proof.
Nonfinal proofs are accepted versions that may not incorporate
authors’ corrections to the proofs and late additions or
changes regarding conflicts of interest, funding sources, or
author affiliations. PNAS publishes daily online through a
rolling publication process. Due to the large volume of articles
we publish, the journal does not provide updates to the media
for articles that are not highlighted on the weekly PNAS
Tipsheet. Reporters interested in covering this article, which is
not highlighted on the weekly PNAS Tipsheet, should check
directly with the authors of the articles for any such late
additions, which may be included in the published articles.

 Quantifying snowfall from orographic cloud seeding
Katja Friedricha,1, Kyoko Ikedab, Sarah A. Tessendorfb, Jeffrey R. Frenchc, Robert M. Rauberd, Bart Geertsc, Lulin Xueb,
Roy M. Rasmussenb, Derek R. Blestrude, Melvin L. Kunkele, Nicholas Dawsone, and Shaun Parkinsone
a
Department of Atmospheric and Oceanic Sciences, University of Colorado Boulder, Boulder, CO 80309; bResearch Applications Laboratory, National Center
for Atmospheric Research, Boulder, CO 80307; cDepartment of Atmospheric Science, University of Wyoming, Laramie, WY 82071; dDepartment of
Atmospheric Sciences, University of Illinois at Urbana–Champaign, Urbana, IL 61820; and eCloud Seeding Group, Idaho Power Company, Boise, ID 83702

Climate change and population growth have increased demand
for water in arid regions. For over half a century, cloud seeding has
been evaluated as a technology to increase water supply; statistical
approaches have compared seeded to nonseeded events through
precipitation gauge analyses. Here, a physically based approach to
quantify snowfall from cloud seeding in mountain cloud systems is
presented. Areas of precipitation unambiguously attributed to cloud
seeding are isolated from natural precipitation (<1 mm h−1). Spatial
and temporal evolution of precipitation generated by cloud seeding
is then quantified using radar observations and snow gauge measurements. This study uses the approach of combining radar technology and precipitation gauge measurements to quantify the
spatial and temporal evolution of snowfall generated from glaciogenic
cloud seeding of winter mountain cloud systems and its spatial
and temporal evolution. The results represent a critical step toward
quantifying cloud seeding impact. For the cases presented, precipitation
gauges measured increases between 0.05 and 0.3 mm as precipitation generated by cloud seeding passed over the instruments. The
total amount of water generated by cloud seeding ranged from
1.2 × 105 m3 (100 ac ft) for 20 min of cloud seeding, 2.4 × 105 m3
(196 ac ft) for 86 min of seeding to 3.4 x 105 m3 (275 ac ft) for 24 min
of cloud seeding.

 

clouds precipitation
gauge observations

  cloud seeding   radar observations  

W

intertime mountain cloud systems have been seeded to
increase snowpack and water supplies since the discovery
(1) that ice crystals could be generated by injecting silver iodide
(AgI) into a cloud of supercooled liquid (2). During this process,
termed glaciogenic seeding, AgI aerosols are introduced into a
supercooled liquid cloud to nucleate ice crystals, which then
achieve sizes and fall velocities sufficient to fall to the surface.
To date, studies quantifying the impact of glaciogenic seeding
of orographic clouds have employed statistical comparisons between seeded and nonseeded events and times as well as target
and control areas (2). Physical process studies attempt to track or
identify seeding plumes and associate enhancements in snowfall
to quantifying seeding impacts. Reported enhancements in snowfall
due to cloud seeding range between 0.5 and 2 mm h−1 (3–6), and
enhancements in reflectivity echoes unambiguously attributed to
cloud seeding have been reported in a handful of studies (3, 7).
All of these studies are often challenged by the difficulty in distinguishing between natural and seeded precipitation, nonrepeatability of a controlled experiment in nature, and the detection
of a relatively small signal in weather systems exhibiting large natural
variability (2). While most studies using a statistical approach
remain inconclusive about the amount of precipitation generated
by cloud seeding (2), a few overcome the challenges and show
statistically significant increases in snowfall (8) or use a computerintensive ensemble technique to evaluate cloud seeding through
the use of thousands of model simulations (9). What makes this
study unique is that 1) the temporal evolution of a seeded cloud
is documented from the time of AgI injection to the time of
snowfall on the ground, over the full width of the seeded cloud
parcel, and 2) the seeding-induced snowfall is isolated unambiguously from natural precipitation. Instead of traditional
www.pnas.org/cgi/doi/10.1073/pnas.1917204117

statistical comparisons, we introduce here a physically based approach by which we isolate areas of precipitation that were unambiguously generated by cloud seeding and quantify the amount
of precipitation in these areas using precipitation gauge measurements and ground-based radar analyses. This approach is applied
to radar echoes that were attributed to seeding at a time of no or
light (<1 mm h−1) natural precipitation for three cloud seeding
events. This study combines radar and gauge analyses in order
to quantify the spatial and temporal evolution of snowfall from
cloud seeding.
For the three cases presented, unambiguous evidence was
provided in two studies (10, 11) that glaciogenic seeding from an
aircraft led to the production of precipitation that eventually fell
to the surface. However, the amount of precipitation produced
for the three cases was not quantified in these previous studies,
which would be a fundamental step toward investigating cloud
seeding efficacy. Cloud seeding efficacy depends on the type of
cloud seeding operations and the temporal and spatial evolution
of a variety of atmospheric variables that control microphysical
and dynamical processes in clouds and precipitation. Here, we
quantify snowfall accumulation from glaciogenic cloud seeding
and its spatial and temporal evolution based on radar observations and snow gauge measurements. The data were obtained
during 3 d in 2017 (19 January, 20 January, and 31 January),
when orographic clouds were seeded with AgI released from
aircraft in the Payette basin of Idaho during the Seeded and
Significance
Cloud seeding to increase winter snowpack in mountains has
traditionally been evaluated using precipitation gauges and
target/control statistics leading mostly to inconclusive results.
Here, an approach employing radar and gauges is used to
quantify snowfall by first isolating radar returns that are unambiguously the result of cloud seeding in regions with light or
no natural precipitation and then quantifying the seedinginduced precipitation at the ground. The spatiotemporal evolution of snowfall from cloud seeding is quantified. Although
this study focuses only on three cases, the results are a fundamental step toward understanding cloud seeding efficacy
that, for over half a century, has been an unanswered question
for water managers wishing to utilize the technology for water
resource management.
Author contributions: K.F., S.A.T., J.R.F., R.M. Rauber, B.G., L.X., R. M. Rasmussen, D.R.B.,
M.L.K., N.D., and S.P. designed research; K.F., K.I., S.A.T., J.R.F., R.M. Rauber, B.G.,
L.X., R. M. Rasmussen, D.R.B., M.L.K., N.D., and S.P. performed research; K.F., K.I.,
and S.A.T. contributed new reagents/analytic tools; K.F. and K.I. analyzed data; and
K.F., S.A.T., J.R.F., R.M. Rauber, B.G., L.X., R. M. Rasmussen, D.R.B., M.L.K., and S.P. wrote
the paper.
The authors declare no competing interest.
This article is a PNAS Direct Submission.
Published under the PNAS license.
1

To whom correspondence may be addressed. Email: katja.friedrich@colorado.edu.

This article contains supporting information online at https://www.pnas.org/lookup/suppl/
doi:10.1073/pnas.1917204117/-/DCSupplemental.

PNAS Latest Articles   1 of 6

EARTH, ATMOSPHERIC,
AND PLANETARY SCIENCES

Edited by John H. Seinfeld, California Institute of Technology, Pasadena, CA, and approved January 30, 2020 (received for review October 2, 2019)

 Natural Orographic Wintertime Clouds: The Idaho Experiment
(SNOWIE) (11). On all 3 d, both burn-in-place and ejectable
flares were used to release AgI from a seeding aircraft. The
aircraft flew repeated ∼50-km-long tracks, perpendicular to the
mean wind and upwind of the observational domain. Exact track
locations and altitudes dependended on wind direction, speed,
and cloud conditions. Tracks, with approximately the same
amount of AgI, were repeated six times on 19 January, eight
times on 20 January, and two times on 31 January (11).
Results and Discussion
Snow Accumulation at Gauge Sites. On 19 January, airborne cloud
seeding began at 1619 UTC. Two seeding lines passed over the
precipitation gauges at Five Corners and Silver Creek, located
within the area covered by the radars, termed the radar observational domain (ROD; Fig. 1) and eventually over the Banner gauge
site, located 67 km east of the Packer John radar site outside of the
ROD. Although, a network of precipitation gauges was deployed,
gauge data were included in this study only if they met the quality
standards described in the SI Appendix and seeding lines passed
over the gauge. Between 1640 and 1710 UTC light snowfall with
a mean rate of about 1.3 mm h−1 (0.6 mm h−1) was observed at
Five Corners (Silver Creek and Banner; Fig. 2A). Patches of
natural, light precipitation surrounded the seeding lines as they
passed through the ROD. These patches were readily detected
by the radars with equivalent radar reflectivity factor (Ze) values

ranging between 1 and 15 dBZe (Fig. 1A and SI Appendix, Movie
S1). Based on radar analyses, the two seeding lines passed over
the Five Corners site between 1710 and 1722 UTC (Fig. 2B and
SI Appendix, Movie S1). Between 1719 and 1724 UTC, a steep
increase in accumulation rate was observed by the Five Corners
gauge (Fig. 2A, red line). Since the increase in precipitation accumulation from the gauge coincides with the time that the seeding
lines passed over the gauges, we conclude that the precipitation
accumulation observed by the gauge was related to the precipitation generated through cloud seeding. A total of 0.2 mm accumulated during the 5 min when the seeding lines passed over the
instrument. Removing the mean accumulation rate of 1 mm h−1
measured 5 min prior and after the passage of the seeding lines,
we attribute about 0.1 mm (1.2 mm h−1) to the effect of cloud
seeding. SD in snowfall rate before and during seeding is about 0.3
and 0.4 mm h−1, respectively, indicating that the increase due to
seeding is three to four times greater than the apparent natural
variability.
As the lines propagated downwind, the first seeding line passed
over the Silver Creek gauge between 1734 and 1740 UTC based
on the radar analysis (Fig. 2B and SI Appendix, Movie S1). For
20 min prior, until 1736 UTC, no precipitation was observed by
the Silver Creek gauge. The second seeding line passed over the
instrument between 1741 and 1746 UTC as seen in Fig. 2B.
Between 1736 and 1746 UTC, the gauge showed a slight increase
in precipitation accumulation (Fig. 2A) amounting to a total of

Fig. 1. Maximum Ze between the surface and 1 km AGL within the ROD for selected times on (A–C) 19 January based on the merged Ze from both radars; (D–F)
20 January from the Packer John radar; and (G–I) 31 January 2017 from the Packer John (PJ) radar. Gauge data were included in this study only if they met the
quality standards and seeding lines passed over the gauge (SI Appendix). Location of the precipitation gauges at Silver Creek on 19, 20, and 31 January and
Five Corners on 19 January and radars at Packer John and Snowbank (SB) are indicated as square and star symbols, respectively. Gray circles centered on the
radars indicate the maximum range. Seeding aircraft flight track, time of seeding, and wind at flight level appear on A, D, and G (half barb: 2.5 m s−1 and full
barb: 5 m s−1).

2 of 6   www.pnas.org/cgi/doi/10.1073/pnas.1917204117

Friedrich et al.

 EARTH, ATMOSPHERIC,
AND PLANETARY SCIENCES

Fig. 2. Radar and precipitation observations at the gauge sites for (A and B) 19 January showing 1-min accumulated precipitation at Silver Creek, Five
Corners, Banner and 6-min Ze observed by PJ radar at 1° at the Five Corners site (Left) and Ze at 1° elevation angle in the vicinity of Five Corners and Silver
Creek during seeding line passage (Right); (C and D) 20 January showing 1-min accumulated S and 1-min Ze observed by PJ radar at 1° at Silver Creek (Left) and
Ze at 1° around the Silver Creek site during seeding line passage (Right); and (E and F) 31 January showing 1-min S from the Geonor and ETI gauge and 6-min
Ze at 1° at Silver Creek (Left) and Ze at 1° around Silver Creek during the seeding line passage (Right). (Left) Periods when seeding lines passed over the gauge
sites based on radar analysis are color coded. Times of Ze observations (extrapolations) are indicated as open (closed) circles in Left. Times of radar observations at Five Corners in B and Silver Creek in D and F shown in the Right are indicated as yellow circles in the Left. Panels B, D, and F only show Ze associated
with the seeding lines; method to isolate the seeding lines is described in the SI Appendix and shown in SI Appendix, Movies S1–S3.

about 0.05 mm. It is not possible to attribute this increase to the
passage of either the first or second seeding line (or both) as
these increases are small and lie within the uncertainty bounds of
the precipitation observations (±0.1 mm). Airborne radar observations (10) indicate that the first seeding line may not have
reached the surface at Silver Creek, while the second seeding
line, producing light surface snowfall with Ze < 5 dBZe, had already reached the surface at 1730 UTC, upwind of Silver Creek.
By 1750 UTC, the seeding lines moved out of the ROD and
approached the Banner gauge site. Here, we use the propagation
speed of the seeding lines within the ROD to estimate the time
the lines passed over Banner. Assuming a radar-estimated propagation speed of 43 km h−1, the lines were expected to pass over
the Banner gauge between 1750 and 1810 UTC (Fig. 2A, green
shaded area). The gauge shows a distinct increase in precipitation
accumulation of 0.2 mm (0.86 mm h−1 with an SD of 0.4 mm h−1)
between 1746 and 1800 UTC compared to the mean precipitation
accumulation rate of 0.43 mm h−1 (SD of 0.2 mm h−1) prior to and
after the seeding lines’ passage (Fig. 2A, green line). The passage
of the seeding lines took 14 min, during which time 0.1 mm of natural
precipitation occurred. Therefore, subtracting that from the 0.2-mm
accumulation during seeding line passage yields 0.1 mm of precipitation attributed to cloud seeding. Although the precipitation accumulations at Banner and Five Corners are quite similar (0.1 mm),
the duration of the increase in precipitation lasted longer (∼14 min)
at Banner compared to ∼5 min at Five Corners, a consequence of
the horizontal dispersion of the precipitation particles with distance
downwind from the seeding location. As a result, the precipitation
Friedrich et al.

rate was less at Banner compared to Five Corners. While light
snowfall would have been observed without cloud seeding on 19
January, the gauge analysis indicates that snowfall from the
seeded clouds almost doubled the snowfall rate at Banner and
Five Corners.
On 20 January, radar observations indicate almost no snowfall
(Ze < 0 dBZe) from natural clouds. Patches of enhanced reflectivity (Ze < 10 dBZe) surrounding the northern end of the lines
were observed between 0000 and 0200 UTC (SI Appendix, Movie
S2). Two of the eight seeding lines observed by the radar passed
over the Silver Creek site, the only gauge site with a seeding line
passage and that met the quality standards on 20 January (SI
Appendix). The lines merged on the northwesternmost end of the
lines before passing over the gauge site. The remaining six seeding
lines precipitated out before reaching Silver Creek (Fig. 1 D–F and SI
Appendix, Movie S2). The set of merged lines passed over the gauge
site between 0154 and 0232 UTC (Fig. 2 C and D). The Silver Creek
gauge measured an increase in precipitation between 0156 and 0230
UTC, before and after which no precipitation was detected (Fig. 2C,
blue line). A total of 0.28 mm accumulated during the 38 min the
seeding lines passed over the instrument, resulting in a mean
precipitation rate of 0.44 mm h−1. Gauge and radar analyses
indicate that snowfall between 0000 and 0300 UTC in the ROD
can be attributed to cloud seeding. Natural clouds would not
have produced any snowfall in the ROD during this time.
On 31 January, enhanced Ze from two seeding lines (>20 dBZe)
were detected. The lines are surrounded by natural clouds with Ze
mainly ranging between 1 and 10 dBZe. Isolated areas of enhanced
PNAS Latest Articles   3 of 6

 Ze up to 20 dBZe associated with natural clouds were observed in
the northeastern part of the ROD. The northern end of the two
seeding lines passed over the Silver Creek site between 2120 and
2142 UTC (Figs. 1 G–I and 2 E and F and SI Appendix, Movie S3).
Only observations at Silver Creek met the quality standards described in SI Appendix. The lines were propagating quickly and at
that time had already merged. As such, the radar analysis was unable to distinguish the passage of individual lines. During this event
each seeding leg utilized both burn-in-place and ejectable flares
(11). Strong vertical wind shear greater than 10−2 s−1 initially led to
a horizontal separation of about 10 km (Fig. 1G) between areas of
enhanced Ze resulting from burn-in-place flares (wide lines of
continuous enhanced Ze; Fig. 1G and SI Appendix, Movie S3) and
those resulting from ejectable flares (comma shapes trailing the
main line of enhanced Ze; Fig. 1G and Movie S3). On this day, two
types of gauges were deployed at Silver Creek (SI Appendix) both
measuring light snowfall prior to the passage of the seeding lines
with slight differences in their measured precipitation rate. The first
gauge indicated snowfall of <1 mm h−1 between 2020 and 2115
UTC (Fig. 2E), before the snowfall stopped at 2115 UTC. Corresponding to the time of seeding line passage, the gauge measured
increased snowfall between 2125 and 2150 UTC before snowfall
stopped again at 2150 UTC. The gauge did not appear to resolve
precipitation-related burn-in-place and ejectable flares separately.
Inspection of the radar data suggests that radar echoes resulting
from the ejectable flares, visible as comma-shaped maxima in Ze,
were not reaching the ground as they passed over the gauge site.
Since no snowfall occurred before 2125 UTC, nor after 2150 UTC,
the increase of 0.25 mm (0.6 mm h−1) in snow accumulation is
attributed solely to the passage of the seeding lines. The SD of
snowfall rate during seeding is 0.4 mm h−1. The collocated second
gauge measured precipitation accumulation that compared very
well with the first (Fig. 2E). However, the second gauge observed
almost continuous snowfall (0.6 mm h−1) between 2100 and 2150
UTC with a slight decrease in accumulation rate after 2150 UTC,
wich was the end of the seeding line passage. There was no noticeable increase in snow accumulation rate during the seeding line
passage between 2125 and 2150 UTC, as observed by the first gauge.
This suggests that the snow accumulation signal from these seeding
lines is quite small and often within the uncertainty of the gauge
measurement. Nonetheless, given the clear passage of these lines in
the radar data, we are confident with the times of their passage and
that precipitation in particular from the first gauge from this time
was due to seeding. Gauge and radar analyses suggest that natural
clouds produced light snowfall of <1 mm h−1 on 31 January.
There was increase in snowfall rate in the first gauge as the
seeding lines passed through the ROD.
Spatial Distribution and Total Amount of Snow Accumulation. To
quantify the amount of precipitation generated through cloud
seeding and determine its distribution across the ROD, measurements from two ground-based radars (Fig. 1) are combined
to estimate snowfall rates based on relationships between Ze and
liquid equivalent snowfall rate, S. Only Ze associated with seeding
lines between the surface and 1 km above ground level are used.
Signals not attributed to seeding (i.e., natural precipitation) were
removed manually. This technique minimizes the impact of snowfall
from natural clouds outside the seeding lines. We acknowledge
that the seeded clouds could have produced snowfall had they
not be seeded. However, gauge and radar analyses showed that the
signal from the natural clouds nearby is less than that from the
seeded clouds. Animations of each of the three cases of the observed
and processed Ze are provided in SI Appendix, Movies S1–S3; details
on radar operations and data processing appear in the SI Appendix.
Distribution maps of accumulated snowfall were generated by
applying a Ze–S relationship (Ze = aSb) to each radar scan
timestep and integrating S with each timestep as the seeding
lines pass through the ROD. A thorough analysis was conducted
4 of 6   www.pnas.org/cgi/doi/10.1073/pnas.1917204117

to establish a relationship for each day that best represents
snowfall from the seeding lines based on precipitation gauge
observations, hereafter referred to as the best-match relationship
(SI Appendix, Fig. S1). Since Ze–S relationships depend on snow
characteristics, a single relationship is unlikely to accurately represent the snowfall rate (12). To address the Ze–S related uncertainty,
an ensemble of Ze–S relationships was also developed by varying
the coefficient a between 100 and 500 (in steps of 1) and the
exponent b between 2 and 2.2 (in steps of 0.1). The range of these
coefficients is based on Ze–S relationships from the literature (13–
17). From the ensemble it is possible to calculate percentile values,
providing an uncertainty estimate for the radar-derived precipitation. We calculated ensemble percentiles based on Ze measured at
the gauge sites to complement the best-match Ze–S relationship
(Fig. 3 and Table 1) and Ze measured over the entire ROD (Fig. 4).
Domainwide accumulated snowfall resulting only from the
seeding lines are shown in Fig. 3. On 19 January, seeding-induced
snowfall was almost equally distributed from each line as the
seeding lines passed through the ROD with slightly greater accumulations over the higher terrain. Note that the seeding lines
were detected prior to 1705 UTC by the airborne radar, but were
not visible by the ground-based radar due to radar beam blockage south of Packer John. Gaps between lines are related to the
6-to-7-min sampling interval of the radars on this day. Holes in
accumulated snowfall, such as those apparent just west of Five
Corners (Fig. 3A), are due to the removal of data contaminated
by ground clutter. Most of the seeding-induced snowfall on 20
January appeared confined to the southeastern portion of the
ROD (Fig. 3B). On this day, lighter winds and faster fallout of the
precipitation provided a much better resolved surface precipitation field compared to 19 January. For the day with the strongest
winds (31 January; Fig. 3C), much of the snowfall occurred in the
eastern most portion of the ROD and one might conjecture that
additional snow fell to the east, outside the ROD.
On 19 January, snow resulting from the seeding lines began to
reach the surface at 1705 UTC and continued through 1812 UTC
(Fig. 3A). Using the best-match Ze–S relationship for that day,
less than 0.15 mm accumulated at each radar pixel (100 × 100 m)
through the entire period. The highest accumulations occurred
over higher terrain suggesting that snowfall was enhanced
through orographic lift. As the seeding lines interacted with the
higher terrain, the area of enhanced Ze also widened (SI Appendix,
Movie S1), covering an area of about 40 to 90 km2 between 1705
and 1716 UTC (Fig. 1 A–C and SI Appendix, Movie S1) to
130 km2 at 1804 UTC. Using the best-match Ze–S relationship,
a total of 123,220 m3 (∼100 ac ft) of liquid equivalent snowfall

Fig. 3. Distribution of accumulated liquid equivalent snowfall (S) attributed
to cloud seeding over the observational period between (A) 1705 and 1806
UTC on 19 January; (B) 0042 and 0315 UTC on 20 January; and (C) 2117 and
2151 UTC on 31 January 2017 using the best-match Ze–S relationship for that
day. Data are shown on a 100 × 100 m grid. Total accumulations over the
entire domain and observational period are highlighted. Corresponding Ze
are shown in SI Appendix, Movies S1–S3.

Friedrich et al.

 19 January
Best match
5th percentile
25th percentile
75th percentile
95th percentile

2.1

228 S
495 S2
387 S2.1
162 S2.1
81 S2

20 January
234
491
352
171
71

2

S
S2.1
S2.0
S2.2
S2.0

31 January
165
445
420
169
72

S2.2
S2
S2.2
S2.2
S2.1

Best match is the minimum between radar-based S and gauge observation with <0.05-mm differences

accumulated over the 67 min over an area of 2,327 km2
resulting from two seeding lines or 20 min of seeding (Fig. 3A).
Uncertainties related to the choice of Ze–S relationship, estimated
from the 25th and 75th (5th and 95th) percentile (Table 1), range
from 95,776 to 144,994 m3 (78,582 to 194,261 m3). Comparing the
Ze–S relationship ensemble to the best match leads to differences
in snow accumulation of about ±20% (±47%) using the 25th/75th
(5th/95th) percentiles Ze–S relationship compared to the bestmatch Ze–S relationship (Table 1).
Next, we investigate how snowfall rate varies by examining the
percentile of snowfall estimates based on all of the Ze–S relationships within the ensemble for each radar scan time. As the
seeding lines passed through the ROD on 19 January, the total
areal snowfall accumulation within the ROD ranged between
about 4,000 and 19,000 m3 (3.2 and 15.4 ac ft) following the
seven Ze–S relationships (Fig. 4A, colored lines). The 5th and
95th percentile for every radar scan from the Ze–S ensemble
ranged between 4,000 and 24,000 m3 (3.2 and 19.4 ac ft). Total
accumulations increased between 1705 and 1746 UTC with
larger accumulations between 1722 and 1746 UTC. For instance,
the total accumulation based on the best-match relationship
doubled from 7,000 m3 (5.7 ac ft) at 1722 UTC to 14,000 m3

(11.4 ac ft) at 1746 UTC. After 1753 UTC, accumulations slowly
decreased as the lines left the ROD. This suggests that the
seeding-induced snowfall increases with time and as the seeding
lines passed over the higher terrain. As the snowfall rate within
the lines increased, the areal coverage also increased (Fig. 4, star
symbols). For this day, the area covered by the seeding lines
increased from 200 to 410 km2 between 1705 and 1758 UTC
(Fig. 4A, star symbols) and decreased to 224 km2 as the seeding
lines moved out of the ROD at 1806 UTC.
On 20 January, snow accumulated mostly within the ROD
about 5 to 30 km east of Packer John within a ∼20-km swath perpendicular to the southwesterly flow (Fig. 3B). Total accumulation
was mostly less than 1.5 mm at each individual radar pixel. Initially,
seeding lines were narrow. They started to broaden quickly and
merge with other seeding lines due to the low wind speed. Most
of the seeding lines precipitated out before reaching the highest
terrain. A total of 241,260 m3 (∼196 ac ft) of liquid equivalent
snowfall accumulated during the 160 min over an area of 1,838 km2
resulting from eight seeding lines or 86 min of cloud seeding. The
total snowfall based on the ensemble varied by 20% to 80% compared to the best-match relationship. This relates to total S
ranging between 196,319 and 312.521 m3 (179,986 and 437,125 m3)
considering the 25th and 75th (5th/95th) percentile of the ensemble.
Compared to 19 January, almost double the amount of snow was
distributed over a much smaller area and throughout a much longer
time period on 20 January.
Contrary to 19 January, most of the snow fell within the ROD
with up to 280,00 m3 of accumulations (22.7 ac ft; Fig. 4B) following the published Ze–S relationships (Fig. 4B, colored lines).
Considering the Ze–S ensemble, the 5th/95th percentile estimates
reached up to 34,000 m3 (27.6 ac ft). Accumulations increased
between 0042 and 0123 UTC, remained almost steady until 0212
UTC and decreased until all snow reached the ground at 0315
UTC. The areal coverage of precipitation at a given time from the
seeding lines steadily increased from 140 to 550 km2 between 0034
and 0123 UTC, remained above this value until 0158 UTC, and

Fig. 4. Total accumulated S over the ROD for each radar volume on (A) 19 January using the SB and PJ radars; (B) 20 January using the PJ radar; and (C) 31 January
using the PJ radar. Color-coded lines indicate known Ze–S relationships for dry and wet snowflakes from the literature (13–17). Gray boxes represent Ze–S ensemble
with the 25th/75th percentile, whiskers extend to the 5th/95th percentile, and the median. Star symbols show the total area of the seeding lines reaching the ground.

Friedrich et al.

PNAS Latest Articles   5 of 6

EARTH, ATMOSPHERIC,
AND PLANETARY SCIENCES

Table 1. List of Ze–S ensemble members (Ze = a Sb) based on Ze
observed after the seeding lines passed Five Corners on 19
January and Silver Creek on 20 and 31 January

 then decreased to 60 km2 at 0301 UTC (Fig. 4B, star symbols).
Spatial distribution of precipitation and the accumulation amounts
are all attributed to cloud seeding as radars and gauges observed no
natural precipitation (Fig. 2C and SI Appendix, Movie S2).
On 31 January, most of the snow accumulated in the ROD
∼20 to 50 km east of Packer John (Fig. 3C) with a total accumulation
of 339,540 m3 (275 ac ft) assuming the best-match Ze–S relationship
over an area of 2,410 km2 through a ∼25-min period resulting from
two seeding legs or 24 min of cloud seeding. The total accumulations range from 222,051 to 335,863 m3 (187,004 to 480,420 m3)
when considering the 25th and 75th (5th/95th) percentiles from the
Ze–S relationship ensemble resulting into an average difference of
20% (43%) compared to the best-match estimate. Due to the high
winds of about 30 m s−1 at flight level, the lines remained in the
ROD for only about 25 min with some of the precipitation from the
lines falling outside of the ROD. The largest amount of snow over
the shortest time and over the largest area was produced on this day
compared to 19 and 20 January.
The total accumulation for each radar scan time ranged between 10,000 and 80,000 m3 (8.1 and 64.8 ac ft) following the
published Ze–S relationships (13–17) (colored lines in Fig. 4C).
For the Ze–S ensemble, the 5th and 95th percentiles of accumulations ranged between 10,000 and 108,000 m3 (8.1 and 87.5
ac ft) for each radar scan. Using the best match Ze–S relationship
shows total accumulations for each scan between 15,000 and
75,000 m3 (12.2 to 60.8 ac ft) throughout the precipitating time
period. The seeding lines coverage steadily increased from 830 km2
to over 1,000 km2 between 2110 and 2117 UTC, then remained
above this value until 2135 UTC, and decreased to 400 km2 at
2146 UTC (Fig. 4C, star symbols) as the lines propagated out of
the ROD.
Conclusion
Snow accumulations from cloud seeding were quantified using
precipitation gauge and radar analyses as seeding lines passed
through the ROD in three cases. As the seeding lines passed over
the precipitation gauges, snow accumulations attributed to cloud
seeding ranged from 0.05 to 0.28 mm with precipitation rates of
0.4 to 1.2 mm h−1 (Fig. 2). Ground-based radar data were used to
analyze the spatial distribution and estimate the total amount of
snowfall from these seeding lines. During weaker wind conditions,
such as on 19 and 20 January, precipitation associated with cloud
seeding reached the ground within 50 km range from the release of
the seeding material (Fig. 3). As seeding lines moved across the
terrain, seeding lines widened and snowfall intensity increased as
shown on 19 January (Fig. 3A). However, on 20 January most of the
precipitation reached the ground before approaching the highest
terrain. During strong wind conditions on 31 January, precipitation
accumulated farther downwind over the higher terrain.
1. B. Vonnegut, The nucleation of ice formation by silver iodide. J. Appl. Phys. 18, 593–
595 (1947).
2. R. M. Rauber et al., Wintertime orographic cloud seeding. Appl. Meteor. Climatol 58,
2117–2140 (2019).
3. A. W. Huggins, Another wintertime cloud seeding case study with strong evidence of
seeding effects. J. Weather Modif. 39, 9–36 (2007).
4. A. B. Super, B. A. Boe, Microphysical effects of wintertime cloud seeding with silver
iodide over the Rocky Mountains. Part III: Observations over the Grand Mesa, Colorado. J. Appl. Meteorol. 27, 1166–1182 (1988).
5. A. B. Super, E. W. Holroyd, III, Some physical evidence of AgI and liquid propane
seeding effects on Utah’s Wasatch Plateau. J. Weather Modif. 29, 8–32 (1997).
6. A. B. Super, J. A. Heimbach, Microphysical effects of wintertime cloud seeding with
silver iodide over the Rocky Mountains. Part II: Observations over the Bridger Range,
Montana. J. Appl. Meteorol. 27, 1152–1165 (1988).
7. P. V. Hobbs et al., Radar detection of cloud-seeding effects. Science 213, 1250–1252 (1981).
8. M. J. Manton et al., Further analysis of a snowfall enhancement project in the Snowy
Mountains of Australia. Atmos. Res. 193, 192–203 (2017).
9. R. M. Rasmussen, Evaluation of the Wyoming Weather Modification Pilot Project
(WWMPP) using two approaches: Traditional statistics and ensemble modeling. J.
Appl. Meteorol. Climatol. 57, 2639–2660 (2018).

6 of 6   www.pnas.org/cgi/doi/10.1073/pnas.1917204117

Cloud seeding generated the most amount of snow on 31
January when the strongest wind was observed with total accumulations of 339,540 m3 (considering the best-match relationship)
over only 25 min resulting from two seeding lines or 24 min of
seeding. During the days with low wind speed, snow accumulated
mainly within the eastern part of the ROD. On 19 January, a total
of 123,220 m3 of liquid equivalent snowfall generated by two
seeding lines or 20 min of cloud seeding accumulated over
2,327 km2. As the seeding lines propagated over higher terrain,
snowfall was orographically enhanced. On 20 January, snow fell
out more rapidly compared to 19 January accumulating mostly
over a smaller area. Total accumulations generated by eight
seeding lines or 86 min of cloud seeding over 160 min reached
241,260 m3, double the amount observed on 19 January. The
uncertainty in the total accumulation is related to the choice of
Ze–S relationship. Based on a Ze–S relationship ensemble the
total accumulation for all three seeding events can vary between
20% and 47% using 25th/75th and 5th/95th percentiles from the
Ze–S relationship ensemble.
Questions remain how cloud seeding efficacy depend on the
type of cloud seeding operations, the amount of AgI released,
and the temporal and spatial evolution of a variety of atmospheric
variables that control microphysical and dynamical processes in
clouds and precipitation affect. As a next step, we will quantify ice
and snow production in these seeded clouds and study the environmental conditions and cloud dynamic and microphysical processes for the three cases presented here. These results are a
fundamental step forward toward being able to answer the overall
question about glaciogenic cloud seeding efficacy. These findings set
the stage for validating numerical models that simulate the microphysical impacts of cloud seeding (2, 18) and improving interpretation
of precipitation observations during cloud seeding operations. After
that, the AgI seeding effect over a target can be quantified with more
confidence at different time scales using the ensemble approach (9).
Materials and Methods
All data presented here are publicly available through the SNOWIE data
archive website maintained by the Earth Observing Laboratory at the National
Center for Atmospheric Research and the SNOWIE radar data archive maintained by the Center for Severe Weather Research (CSWR). Data are archived in
NetCDF or ASCII format, and each \ contains an accompanying text file providing necessary metadata. Description of calibration and uncertainties for
DOW and precipitation gauges are also provided in SI Appendix.
ACKNOWLEDGMENTS. We thank the crews from the Doppler on Wheels
(DOW) radars from the CSWR and students from the University of Colorado,
University of Wyoming, and University of Illinois for their help operating and
deploying instruments during the campaign. Funding for CSWR-DOWs was
provided through National Science Foundation (NSF) AGS-1441831. Funding
for seeding aircraft was provided by Idaho Power Company. The research
was supported under NSF Grants AGS-1547101, AGS-1546963, and AGS-1546939.

10. J. R. French et al., Precipitation formation from orographic cloud seeding. Proc. Natl.
Acad. Sci. U.S.A. 115, 1168–1173 (2018).
11. S. A. Tessendorf et al., A transformational approach to winter orographic weather
modification research: The SNOWIE Project. Bull. Am. Meteorol. Soc. 100, 71–92 (2019).
12. R. M. Rasmussen et al., How well are we measuring snow: The NOAA/FAA/NCAR
winter precipitation test bed. Bull. Am. Meteorol. Soc. 93, 811–829 (2012).
13. K. L. Gunn, J. S. Marshall, The distribution with size of aggregate snowflakes.
J. Meteorol. 15, 452–461 (1958).
14. J. Imai, “Raindrop size distributions and the Z-R relationship” in Proceedings of the
Eighth Weather Radar Conference (American Meteorological Society, Boston, MA,
1960), pp. 321–326.
15. R. S. Sekhon, R. C. Srivastava, Snow size spectra and radar reflectivity. J. Atmos. Sci. 27,
299–307 (1970).
16. T. Puhakka, “On the dependence of the Z-R relation on the temperature in snowfall”
in Preprints, 16th Conference on Radar Meteorology (American Meteorological Society,
Boston, MA, 1975), pp. 504–507.
17. J. P. Wolfe, J. R. Snider, A relationship between reflectivity and snow rate for a highaltitude S-band radar. J. Appl. Meteorol. Climatol. 51, 1111–1128 (2012).
18. L. Xue, Implementation of a silver iodide cloud seeding parameterization in WRF: Part
1: Model description and idealized 2D sensitivity tests. J. Appl. Meteorol. Climatol. 52,
1433–1457 (2013).

Friedrich et al.