Weather and Climate Extremes 21 (2018) 65–75

Contents lists available at ScienceDirect

Weather and Climate Extremes

journal homepage: www.elsevier.com/locate/wace

Impact of different microphysical parameterizations on extreme snowfall T

events in the Southern Andes
Alcimoni Nelci Comina,∗, Vanúcia Schumachera, Flávio Justinoa, Alfonso Fernándezb
a
Department of Agricultural Engineering, Federal University of Viçosa, Brazil
b
Department of Geography, Universidad de Concepción, Concepción, Chile

A R T I C LE I N FO A B S T R A C T

Keywords: This study evaluates the reliability of the Weather Research and Forecasting (WRF) model to simulate extreme
WRF snowfall events in the Southern Andes. The assessment includes comparison of seven microphysics para-
Cloud microphysical parameterizations meterizations (MPs) schemes, using two different reanalysis datasets as boundary and initial conditions, namely
Snowfall NCEP-FNL and ERA-interim. Results demonstrate the feasibility of predicting extreme snow events with rea-
Andes
sonable accuracy using WRF, but the accuracy level is dependent on the imposed initial conditions. In particular,
by computing the RMSE turned out that the WSM6 under NCEP-FNL performed better as compared with the
other schemes in the highly complex topography of the Andes. Conversely, Morrison and WDM5 ranked the
worst as both simulated excessive snowfall. For ERA-interim initial conditions, Goddard (WDM6) scheme shows
the best (weaker) performance. Despite these limitations, these modeling experiments demonstrate the feasi-
bility of using the WRF to forecast the spatial and temporal distribution of snowfall and precipitation in this
region of steep topography. Therefore, modeling experiments may reduce people losses by anticipating the
weather threat for local communities, and provide decision makers with information on which to base future
interventions for water supply hydrological hazards.

1. Introduction amount differ completely according to the prescribed SST.

McMillen and Steenburgh (2014) investigated the sensitivity of si-
Most operational weather forecasts system struggle to accurately mulations of Great Salt Lake-effect snowstorm to the choice of micro-
simulate seasonal and daily snowfall associated to deficiencies in re- physics parameterization and found that simulated precipitation varied
solving complex precipitation structures forced by topography, land in areal coverage, amount and position depending on the scheme uti-
surface heterogeneities and land-water contrasts (Dinesh et al., 2016; lized, posing a challenge for operational weather prediction. Thompson
Tao, 2013). Despite recent progress, there remain uncertainties on how et al. (2004) also demonstrated that differences in the microphysics
different parameterizations properly simulate snowfall in coupled cli- parameterization of ice affect the accuracy of surface precipitation.
mate models. It has to be noticed, however, that misrepresentation of With certain limitations, the WRF model has been shown to reasonably
complex terrain and simplified snow and cloud physics are also large simulate observed patterns of extreme precipitation in high mountains
source for model uncertainty (Jin and Wen, 2012). (Jiménez et al., 2013; Viale et al., 2013; Maussion et al., 2014; Milrad
Lee and Lee (1994) exemplified the complexity of simulating et al., 2014). Te dynamics, thermodynamics and the onset of convective
snowfall and extreme snowfall events by including heat and moisture snow bands. However, most of the studies on the impact of different
related to a low-level jet in association with the topographic effect of WRF parameterizations have focused on liquid precipitation and are
the Korean Taebaek Mountains. Miller (2012), using the Weather Re- confined to the Northern Hemisphere. Investigations for snowfall
search and Forecasting (WRF) Model also found that surface latent heat events in the Southern Hemisphere have been hampered by the scarcity
fluxes in the region upstream of the Appalachian mountain provide a of observations to validate the modeling experiments, in particular in
source of moisture that contributes for snowfall events. Furthermore, complex terrains such as the Andes.
Cha et al. (2011) in a study about the influence of the Yellow Sea's sea Jankov et al. (2009) analyzed the performance of a high-resolution
surface temperature (SST) on a snowfall event during a cold-air out- WRF model using four different microphysics schemes for significant
break demonstrated that distinct cloud streets, and resultant snowfall precipitation events during a winter season in California, and noted a

∗
Corresponding author.
E-mail address: [email protected] (A.N. Comin).

https://doi.org/10.1016/j.wace.2018.07.001
Received 17 January 2018; Received in revised form 18 June 2018; Accepted 20 July 2018
Available online 30 July 2018
2212-0947/ © 2018 Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/BY-NC-ND/4.0/).
A.N. Comin et al. Weather and Climate Extremes 21 (2018) 65–75

large sensitivity to changes in microphysics. Rajeevan et al. (2010) zone in Araucania, southern Chile. 22 thousand people were left iso-
analyzed the sensitivity of the WRF model to simulate features of a lated regions and more than 800 km of roads disappeared under the ice.
thunderstorm passing over Gadanki (India), using four different mi- With frozen pipes there was a lack of drinking water, compromising
crophysical schemes; all the simulations correctly mimicked the pas- inhabitants livelihoods (ONEMI, 2011). This kind of intense event is
sage of two convective cores, but showed large variations among the popularly nicknamed as “white earthquake” due to the widespread
schemes. damage.
Comparison of high resolution WRF simulations of seasonal snow- The other two events affected “Los Andes”, a municipality part of
fall over the Colorado Head-waters regions show very good agreement the Valparaiso administrative region in Chile, although geographically
if a grid spacing minor 6 km is used (Ikeda et al., 2010). However, in closer to capital of Santiago and Mendoza, Argentina. In both storms,
evaluating the North American Mesoscale (NAM) model forecasts for 25 between Chile and Argentina was shut down. Due to the potential da-
snow events, Evans and JurewiczSr (2009) found that correlations be- mage the Chilean department for emergencies (ONEMI) and the na-
tween simulated and observed snowfall decreased substantially as tional customs issued a number of alerts (ONEMI, 2015, 2016). It is
forecast lead-time increased over 12 h. estimated that economic loss resulted from closing the connection be-
Snow is the most important source of water and economic growth in tween Chile and Argentina amounts to about 1 million USD per day
several regions in tropical (high altitude) and subtropical/extratropical (Manterola, 2009). Vuille and Ammann (1997) showed that snowfall
latitudes. In fact, snow accumulation is crucial to maintain mountain during winter is quite common over Andes, mainly linked to northward
river discharge and to allow the permanence of glaciers, which are displacements or cut-offs of cold air-masses from the Pacific. During a
considered a vital water resource during the driest part of the year and frontal system, snowfall is most frequent in the southernmost and on
more so during droughts (Mark et al., 2015; Ohlanders et al., 2013). In the western Chilean side of the Andes (Vuille and Ammann, 1997). The
the Andes, snow ensures water resources for urban consumption, cold air interacts with more humid and warmer air on the continent and
agriculture, industrial and hydroelectric use (Cai et al., 2003; Masiokas therefore gives rise to a different snowfall distribution.
et al., 2012). It has been estimated that 2% (∼150M people) of world's Weather conditions associated with these 3 extreme snowfall events
population lives in mountain regions above 2500 m (Hornbein and in the Southern Andes are indeed interesting test cases to verify the skill
Schoene, 2001). Tebaldi et al. (2006) using a multi-model ensemble of the WRF to reproduce weather conditions in the Andes (as sum-
predict precipitation (rainfall plus snowfall) decreases in the Chilean marized in Table 2). Improving modeling experiments may reduce
region between ∼30 and 40°S. people losses by anticipating the weather threat for local communities.
In order to understand the Andean snowfall regime, the present
study aims to analyze the impact of different cloud microphysical 3. Data and methods
parameterizations of the WRF model in simulating three extreme
snowfall events in the Southern Andes. Three cases of extreme snowfall In the present study, we have used the Weather Research and
that occurred in the Chile-Argentina Andes were selected due to the Forecasting (WRF) model version 5.3.1 to compare the effect of dif-
severe damage produced, with snow depths reaching by up to 3 m in a ferent parameterizations on the snowfall events on the Southern Andes.
few hours (ONEMI, 2011), producing considerable damage to the li- The WRF was setup with two horizontal nested grids, one at 9 km
velihoods of its inhabitants. (120 × 150) and a second at 3 km (246 × 280) and vertical levels from
The Southern Andes is characterized by a complex terrain in the 1000 hPa to 50 hPa (38 levels). Both domains are centered in region of
central part of Chile and central-western Argentina (∼30-40ºS), with the snowfall event and cover part of the Central Chile shown, as shown
elevation from 3500 m and peaks of up to 6000 m altitude (Masiokas in Fig. 1a. Details on parameterizations and microphysics schemes are
et al., 2013). The regime of Andean rivers is fundamentally snow/ice provided in sub-section 3.2.
dominated, supporting numerous economic activities downstream on Our analysis of model performance involved description of the sy-
both slopes (Masiokas et al., 2013). noptic features occurring during the event using fields from the ERA-
Winter precipitation in the Southern Andes typically occurs due to interim reanalysis (http://apps.ecmwf.int/datasets/data/interim-full-
the passage of cold fronts from west to east over the Cordillera (Falvey daily/levtype=sfc/) and GOES 13 satellite based-infrared images.
and Garreaud, 2007). The low-level atmospheric jet also plays a role in Model skill was assessed from available field snow observations and
all seasons by transporting water vapor (Marengo et al., 2002). More- satellite products.
over, events of precipitation in the Southern Andes are related to in-
tense water vapor transport from the Pacific Ocean prior to a cold 3.1. Observational datasets
front/extratropical cyclonic circulation (Viale and Nuñez, 2011). The
Andes act as a topographic barrier to this atmospheric circulation, Hourly data of precipitation, snow depth and snow accumulation for
modifying the wind flow and midlatitude waves (Garreaud, 2009). The Portillo (32º50′10″S – 70º7′44″W) in Chile and daily data of snow depth
paper is organized as follows: Description snowfall events are presented for Horcones on ‘Parque Aconcagua’ (32º47′9.5″S - 69º56′42.5″W) in
in section 2. Section 3 is dedicated to a synoptic analysis of the snowfall Argentina were used to verify the WRF capability to simulate two ex-
events, comparison of the microphysics schemes and Noah LSM. Section treme snowfall events. The Mendoza's Irrigation Department
4 provides conlcuding remarks. (Departamento General de Irrigación) regularly retrieves these ob-
servations. In the available databases and recent papers (e.g. Stehr and
2. Description snowfall events Aguayo, 2017) there are new data on snow, but all we checked did not
have the sub-daily resolution needed to track the skill of the simulation.
To investigate the meteorological conditions associated with
snowfall events in the complex terrain on the Southern Andes, three Table 2
occurrences of extreme snowfall are selected in the Chilean-Argentina Snowstorm cases.
Cordillera, as summarized in Table 2. The selected events resulted in Year Date of cases Region
severe local damage with snow depth up to 3 m height reported by local
journals, as well as measured by meteorological weather stations 2011 17–20 July La Araucanía-Chile (cold front passage)
(https://goo.gl/hmk5Zn; https://goo.gl/PD6TKP; https://goo.gl/ 2015 07–09 August Los Andes-Chile and Mendoza-Argentina (passage of
intense extra-tropical cyclone)
FkkLq1; accessed June 15, 2018 ONEMI (2011).
2016 03–05 June Los Andes-Chile and Las Cuevas-Argentina (cold front
The 2011 snowfall event was considered the strongest of the last passage)
three decades so far and led the government to declare a catastrophe

66
A.N. Comin et al. Weather and Climate Extremes 21 (2018) 65–75

Table 1
Details of the microphysical schemes considered in the study.
Microphysical Mixed phase variables processes
schemes

WSM3 water vapor, cloud water/ice and rain/snow.

WSM6 cloud droplets, rain, cloud ice, snow, and graupel.
Goddard ice, snow and graupel.
Morrison mixing ratios of rain, cloud ice, snow, and graupel/hail
and mass mixing ratio of cloud droplets.
WDM5 Clouds, rain, ice, snow and graupel.
WDM6 cloud droplets, rain, cloud ice, snow, and graupel and
number mixing ratios of cloud droplets, rain, and cloud
condensation nuclei.
NSSL cloud droplets, rain, small ice, snow, graupel, and hail,
as well as the particle volume of graupel.

2005). Initial atmospheric and land surface conditions, as well as lateral

boundary conditions were taken from the National Centers for En-
vironment Prediction (NCEP) Global Final Analyses (NCEP-FNL) and
Fig. 1. a) WRF was setup with two horizontal nested grids, d1 at 9 km the ERA-interim reanalysis (Dee et al., 2011) at 1° x 1° and 0.75° x 0.75°
(120 × 150) and d2 at 3 km (246 × 280) shown for the case 2015 centered latitude-longitude grids, respectively, with output every 6 h.
over Las Cuevas; b) Spatial domain for each comparison over the mountains. The impact of cloud microphysics schemes on snowfall events is
examined using seven different microphysical parameterizations (MP):
To overcome this observational limitation, we supplemented our ana- WRF Single-Moment 3-class (WSM3), WRF Single-Moment 6-class
lysis with comparison of model output with MODIS products. (WSM6), Goddard microphysic scheme, Morrison double-moment
We first performed visual comparison using the daily snow cover scheme, WRF double-moment 5-class mixed ice microphysic (WDM5),
product at 0.05° horizontal resolution from the MOD10C1 (available at WRF Double-Moment 6-class (WDM6) and NSSL single-moment
http://nsidc.org/data/MOD10C1). Secondly, we performed a forecast scheme. Table 1 show the main species of prognostic variables in
verification analysis of the WRF output by comparing the simulated schemes.
snow accumulation with the snow cover daily data from the MOD10A1 The WSM3 is a simplified but efficient scheme with ice and snow
product (Hall and Riggs, 2016). MOD10A1 is a collection of snow-re- processes suitable for mesoscale grid sizes (Hong et al., 2004). The
lated fields derived from processing daily MODIS/Terra images. For this process related to the cloud ice and cloud water are counted as the same
analysis, we utilized the product L3 version 5, which contains a “snow category and distinguished by temperature. When the temperature is
cover map” field, a 500 m spatial resolution product that flags snow- less than or equal to the freezing point cloud ice occurs, otherwise it is
covered grid-cells. MODIS data were accessed through the Earth Data considered to be cloud water. Hong et al. (2004) also show that the
portal (https://earthdata.nasa.gov) from NASA's Earth Observing scheme has a significant improvement in high-cloud amount, surface
System Data and Information System (EOSDIS). The original files were precipitation and large-scale mean temperature through a better re-
reprojected and regridded to WRF's grid-cell size utilizing a nearest presentation of the ice cloud–radiation feedback.
neighbor algorithm. For this comparison, we reclassified the WRF's Hong and Lim (2006) in simulating heavy rainfall events in Korea,
snow output assuming that a snow-covered grid-cell was anyone ac- show that the evolution of precipitation in the WSM6 is similar to that
cumulating more than 10 cm. The comparison consisted of two forecast predicted by the WSM3 microphysics schemes, using a low resolution
verification metrics: the Threat Score (TS) and the Pierce Skill Score grid. In high resolution, distinct characteristics are noted. The rainfall
(PSS). The TS ranges from 0 to 1 in which 1 corresponds to a perfect increases and the peak intensity is stronger in WSM6 according to the
simulation of the variable of interest; the PSS allows assessment of the number of hydrometeors.
probability that the modeled event is better or worse than a random Chern et al. (2016) implemented four versions of the Goddard mi-
forecast (Hsieh, 2009). For clarity, we multiplied both metrics by 100. crophysical schemes into a multi-scale modeling framework. These si-
Acting together, these two metrics allow evaluation of the proportion of mulations were validated against CloudSat/CALIPSO cloud ice products
hits relative to the whole study area and also the skill of the simulation and other satellite products. Demonstrated that simulated mean annual
to discern areas where the event did not occur. The comparison was global precipitation is very similar in both amplitude and pattern
computed between every reprojected/regridded MOD10A1 available among the four schemes. The single moment bulk cloud microphysics
for the duration of the event and the WRF output from the hour closest scheme updates seven variables (potential temperature, water vapor,
to the local MODIS overpass time, excluding the cells flagged as cloud cloud water, cloud ice, rain water, snow and graupel/hail) based on a
covered by MOD10A1. The spatial domain for each comparison was prognostic saturation equation and microphysical interaction processes
smaller than the WRF spatial domain in order to focus on results on the (Mielikainen et al., 2014).
mountains (Fig. 1b). The Morrison scheme simulates the concentrations number and
mixing ratios of four hydrometeor species: droplets, rain, cloud ice and
3.2. WRF model microphysics schemes snow that are developed for simulating homogeneous and hetero-
geneous ice nucleation, droplet activation, and the droplet size spectra
The WRF physical configuration includes the Rapid Radiative (Morrison et al., 2004). The Morrison scheme outpaced observed pre-
Transfer Model (RRTM) long-wave radiation based on Mlawer et al. cipitation accumulation in the first 4 h, but was nearly identical to
(1997) and the Dudhia shortwave scheme (Dudhia, 1989). For the land storm total accumulation in a series of simulations in the Canadian
surface process the Noah LSM scheme is used (Niu et al., 2011). The Centre for Atmospheric Research Experiments (Molthan and Colle,
turbulent kinetic energy planetary boundary layer follows the Monin- 2012).
Obhukov-Janjic approach (Janjic, 2002). The Kain-Fritsch convective The WDM5 is a bulk microphysics scheme that simulates the mixing
parameterization (Kain, 2004) is used only in the first domain to avoid ratio of hydrometeors and number concentrations for warm rain species
precipitation sensitivity in the innermost (3 km) domain (Jankov et al., including clouds and rain, and for cold rain species including ice, snow

67
A.N. Comin et al. Weather and Climate Extremes 21 (2018) 65–75

and graupel (Lim and Hong, 2010; 2010a; Huang et al., 2015). Cloud
condensation nuclei (CCN) are also predicted. The cold rain species
follow the single-moment 5-class microphysics in which theoretical
background for WSM ice phase microphysics is based on Hong et al.
(2004).
The WDM6 scheme in comparison to WSM6, adds the prognostic
number concentration of cloud and rainwater together with the CCN
(Lim et al., 2010). The ice-phase microphysics of Hong et al. (2004) is
identical for both the WDM6 and WSM6 schemes; the former has a
wider range of cloud and rain number concentrations.
According to Mansell et al. (2010) NSSL is a the single-moment
scheme, a loss of droplet mass reduced droplet size, but not con-
centration until the minimum allowed droplet size of 5 μm was reached.
The snow and cloud ice have low mean fall speeds than the graupel
hampering the flux of ice into the melting layer and production of rain
in NSSL scheme (Morrison et al., 2014).

3.3. Noah LSM

Microphysics parameterizations represent the size spectra of dif-

ferent hydrometeor species with particle size distribution function and
predict the development of hydrometeor, their interactions, growth and
precipitation (Galligani et al., 2017). In single scheme performs uni-
formly well under all conditions, and each has predictive ability highly
dependent on climate regimes (Jankov et al., 2005; Zhu and Liang,
2007). Thus, ensemble of multiple physics configurations of a model
can be very useful to highlight superior skills as compared to those
using a single configuration (Kirtman and Min, 2009). Different regions
are identified where each model or scheme complementarily captures
certain but not all observed signals (Liang et al., 2012). Microphysics
directly impacts spatial patterns and the location with highest pre-
cipitation (ElTahan and Magooda, 2017; Maraun et al., 2010).
Livneh et al. (2010) demonstrated that the snow-albedo decay and
liquid-water refreeze results in improvements in the magnitude and
timing of peak snow water equivalent in Noah LSM, which resulted in
more realistic snow-covered albedos at large scales in the western
United States. They reported that the inclusion of a realistic and phy-
sical lower bound on decaying albedo was invoked to this effect and Fig. 2. Differences between 2011 July 17 (peak of the snowfall) and July
provided limited benefit. Noah LSM has a negative bias in snow water average climatology: (a) sea level pressure (hPa), (b) 500 hPa geopotential
equivalent in early spring (Jin and Miller, 2007). However, Noah LSM height (gpm), (c) 850 hPa air temperature (ºC), (d) 500 hPa air temperature
overestimates the precipitation due to a unstable atmosphere with in- (ºC), (e) 850 hPa specific humidity (g kg−1) and (f) 850 hPa wind vector and
creased upward air motion and elevated surface evaporation rates speed (m s−1). Dotted indicate level of significance of 95% in relation to the
climatological difference (1979–2016).
(Livneh et al., 2010). Jin et al. (2012) demonstrated that snow water
equivalent is underestimated with the Noah and Rapid Update Cycle
(RUC) land surface schemes, in which snow physics is oversimplified. 4. Analysis and discussion
They reported that the reduction in precipitation is caused by lower
surface evaporation resulting from the longer, more realistic snow 4.1. Atmospheric conditions during the snowfall cases
duration.
Noah has a bulk layer of snow and soil. Thick snowpack tends to Dynamics and synoptic features for the days with occurrence of the
underestimate the ground heat flux, leaving too much energy at the snowfall events are explored by investigating departures from clima-
snow surface and being thus too favorable to snowmelt (Niu et al., tology using ERA-interim. Figs. 2–4 show the synoptic characteristics
2011). To alleviate this problem three layers were introduced in the during the peak of the snowfall in each case. The surface features are
Noah model: percolation, retention, and refreezing of meltwater within characterized by a SLP (sea level pressure) gradient between the extra-
the snowpack. Niu et al. (2011) show that snow interception accounts tropical (∼45°–60°S) and subtropical (∼30°S) latitudes (Figs. 2a, 3a
for a greater capacity for snowfall than rainfall and improving the re- and 4a). However, this gradient is moderate in the 2015 event, which is
sults of sublimation and surface albedo. Ikeda et al. (2010) applied the dominated by a substantial reduction in the SLP over the snowfall main
SNOTEL (SNOwpack TELemetry) observation of snowfall to evaluate area (Fig. 3a). These features appear in all events as shown in Figs. 2, 3
the ability of a high resolution model to accurately simulate snowfall and 4b,c and d.
over the Colorado Headwaters region. They noted that accurate sea- There is a strong negative anomaly of geopotential height (Fig. 2b)
sonal snowfall in WRF was better simulated with horizontal resolutions and temperature at 850 and 500 hPa (Fig. 2b and c), this negative
smaller than 6 km. anomaly will further induce unstable conditions leading to precipita-
tion. Evaluation of specific humidity and the wind anomalous pattern
demonstrates that the 2011 case is very particular because it shows
reduction in the amount of water vapor during the event as compared
to climatological values. But the wind anomalies indicate an increase in

68
A.N. Comin et al. Weather and Climate Extremes 21 (2018) 65–75

Fig. 3. As in Fig. 2, but to difference between 2015 August 9 and August Fig. 4. As in Fig. 2, but to difference between 2016 June 5 and June average
average climatology. climatology.

the cold air toward the Southern Andes (Fig. 2e and f). This cold air is To provide additional information on these events Fig. 5 shows
sufficient to drive the atmosphere to saturation levels (not shown) al- GOES 13 satellite based-infrared images. For the period 17–20 July
lowing for snowfall despite reduction in the water vapor amount 2011, deep clouds are seen over the study area and indicate the po-
(Fig. 2e and f). tential for heavy precipitation/snowfall (Fig. 5a). In addition, Fig. 6
However, for the 2015 and 2016 cases (Figs. 3–4) in general the shows daily snow cover as estimated by the MOD10C1 sensors (http://
synoptic features linked to snowfall are driven by moisture advection nsidc.org/data/MOD10C1). Fig. 6 shows the large snow covered the
from the Pacific Ocean accompanied by cyclonic and increased polar region on days 17 and 18/07/2011. WRF's simulated snowfall WRF is in
advection (cold air outbreaks). The atmospheric characteristics simu- good agreement with this satellite snow cover product (see next sub-
lated by the model suggest that these snowfall events resulted primarily section for a detailed quantitative analysis). In the second case (2015),
from intensified isolated cells rather than arising solely from a typical images have more cloud cover, mainly during the day 7 and the onset of
cold front, as observed by Vuille and Ammann (1997). day 8.
The extreme events of snowfall occur due to the passage of cold Even harsher conditions characterized the 2016, since significant
fronts or extratropical cyclonic from west to east over the Andes. This convective activity over Chile and Argentina favored large amounts of
was also observed by Falvey and Garreaud (2007). In general snowfall snowfall. This convective activity is related to the passage of intense
events occur due to strong negative anomalies of geopotential height, extra-tropical cyclones (Fig. 5), and snow cover (Fig. 6d).
relativity humidity and temperature in the low and middle troposphere,
associated with ocean moisture transport. The presence of relatively
cold upper level air combined with a large humidity content determine 4.2. Analysis of the WRF model simulations
conditions to trigger snowfall.
The snowfall events share some common features but also present These cases of extreme snowfall in the Southern Andes are analyzed
distinct properties. They occurred in phase with a substantial drop in in detail based on different microphysics parameterizations of cloud
the Z500, T850 and T500. Conversely, SLP, specific humidity, and wind and precipitation available in the WRF system. Two different initial
patterns (although at a minor degree) differ among the snowfall si- conditions were used: NCEP-FNL and ERA-interim. In order to validate
tuations. the WRF simulation we have provided an in-depth evaluation of the
parameterization scheme by comparing their results with the snow

69
A.N. Comin et al. Weather and Climate Extremes 21 (2018) 65–75

Table 3
Results from the application of TSS and PSS to the comparison of MODIS and
WRF output.
Parameterization Event year Cloud cover (%) TS PSS

ERA FNL ERA FNL

Goddard 2011 74.7 78.7 77.2 94.1 93.9

2015 78.6 30 37.8 82.4 88.4
2016 63.6 86.2 87.3 91.2 92.1
Morrison 2011 74.7 78.1 76.5 93.9 93.8
2015 78.6 29.1 36.7 81.5 87.8
2016 63.6 90.2 88.8 94.1 93.2
NSSL 2011 74.7 78.1 77.4 93.9 94
2015 78.6 27 34.9 79.3 86.7
2016 63.6 84.8 84.7 90.2 90.1
WDM5 2011 74.7 77.6 76 93.8 93.6
2015 78.6 29.4 36.4 81.9 87.6
2016 63.6 88.5 89.8 92.9 93.8
WDM6 2011 63.6 88.8 76.8 93.1 93.8
2015 74.7 77.3 36.3 93.6 87.6
2016 78.6 28.7 87.5 81.3 92.2
WSM3 2011 74.7 73.6 76.5 92.1 93.7
2015 78.6 27.7 34.3 80.1 86.4
2016 63.6 89.3 85.5 93.5 90.8
WSM6 2011 74.7 77 76.8 93.6 93.8
2015 78.6 29.4 36.5 81.7 87.7
2016 63.6 89.3 87.8 93.5 92.4

Fig. 5. GOES 12 lowest cloud top temperatures (ºC) during the occurrence of of 93.7 ± 0.46 (1 standard deviation), slightly higher than the 2016
2011, 2015 and 2016 events. Panels show the onset, evolution and demise of event (92.5 ± 1.22), while the 2015 remains as the simulation with
the event. Available at (http://satelite.cptec.inpe.br/acervo/goes.formulario. least skill (84.31 ± 3.25). The general performance of TS seems in-
logic). versely related to the cloud cover; this might be product of the fact that
more cloud cover results in less effective area of comparison. The PSS,
cover daily data from the MOD10A1 satellite product. on the other hand, captures the correct negatives thereby gives a better
The metrics indicate that the WRF accurately simulates the spatial indication if the simulation is able to deliver a correct spatial dis-
distribution of snow cover during all 3 events (Table 3). The 2011, 2016 tribution of the whole event, including locations where snow should not
events present scores close to 100, with 2016 having the best perfor- have fall; this makes the analysis less sensitive to cloud cover, as de-
mance according to the TS while 2011 for the PSS. The least skill was monstrated in Table 3.
identified in simulating the 2015 event, although it is important to Figs. 7–11 show the amount of snow accumulated for the individual
highlight that for this event the TS and the PSS showed significant events using the finer grid with 3 km horizontal resolution for the MP
differences. Models with MP Morrison and Goddard generally showed schemes tested in this study. According to Leung et al. (2003) regional
highest skill scores. For the TS, the ERA-interim boundary and initial climate models resolutions (∼18 and 36 km) typically underestimate
conditions seem to show the best performance, as only the combination snowfall by 15–30%. Fine resolution models better represent non-hy-
NCEP-FNLWDM5 ranked in the highest 4 most skilled simulations drostatic features of topographically-induced motions and this con-
(> 80). For the PSS, only the combination NCEP-FNLNSSL ranks among tribute to improved snowfall simulations (Ikeda et al., 2010). Shrestha
the most skilled simulations. However, it is relevant to point out that et al. (2017) investigate the MP schemes to simulate storms over the
results from the PSS are on average higher relative to the TS. For the Nepal, Himalayas. Morrison, WDM6 and WSM6 schemes failed to re-
PSS, the best simulation corresponds to the 2011 event with an average produce ice mixing ratio that may be attributed to a rapid

Fig. 6. Snow cover daily (0.05 deg) for (a) 2011 July 17, (b) 2011 July 18, (c) 2015 August 9 and (d) 2016 June 5. Available at (http://nsidc.org/data/MOD10C1).

70
A.N. Comin et al. Weather and Climate Extremes 21 (2018) 65–75

Fig. 7. Snow accumulated (cm) for July 17, 18, 19 and 20 of 2011 using MP
schemes and initial conditions data from NCEP-FNL: (a) WSM6-WSM3; (b) Fig. 9. Differences in accumulated snowfall between the simulation driven by
WSM6-Goddard; (c) WSM6-Morrison double-moment; (d) WSM6-WDM5; (e) the NCEP-FNL and ERA-interim (mm) for August 7, 8 and 9 of 2015. (a) WSM3;
WSM6-WDM6 and (f) WSM6-NSSL and (g) WSM6. (b) WSM6; (c) Goddard; (d) Morrison double-moment; (e) WDM5; (f) WDM6
and (g) NSSL.

Fig. 8. Snow accumulated (cm) for August 7, 8 and 9 of 2015 using MP schemes
and initial conditions data from NCEP-FNL: (a) WSM6-WSM3; (b) WSM6-
Goddard; (c) WSM6-Morrison double-moment; (d) WSM6-WDM5; (e) WSM6- Fig. 10. Snow accumulated (cm) for June 3, 4 and 5 of 2016 using MP schemes
WDM6 and (f) WSM6-NSSL and (g) WSM6. and initial conditions data from NCEP-FNL: (a) WSM6-WSM3; (b) WSM6-
Goddard; (c) WSM6-Morrison double-moment; (d) WSM6-WDM5; (e) WSM6-
WDM6 and (f) WSM6-NSSL and (g) WSM6.
transformation of ice to snow and graupel. Ice mixing ratio in the
WDM6 scheme was the highest among the schemes. The Morrison
conditions and difference between the WSM6 (reference scheme,
scheme simulated highest snow mixing ratio. Morroni (2017) using
Fig. 7g) and all the other schemes (Fig. 7a–f). While the location of the
WSM6 scheme found more realistic amounts of rainfall to events of
snow accumulation maxima does not differ substantially among the
extreme precipitation in the Andean region.
tested schemes, the snowfall amount does. WSM3, Morrison double-
Fig. 7 shows accumulated snow based on the NCEP-FNL initial

71
A.N. Comin et al. Weather and Climate Extremes 21 (2018) 65–75

(Fig. 8d) schemes simulate up to 125 cm higher accumulated snowfall

than WSM6, but the snowfall regions were the same for all schemes. In
this event the amount of snow was due to an intense cyclonic system.
The WDM6 (Fig. 8e) differences as compared to WSM6 are by about
25 cm of snowfall. For the NSSL (Fig. 8f) the accumulated snowfall
differences are up to 75 cm smaller than the WSM6 scheme.
In Fig. 9 differences between the NCEP-FNL and ERA-interim initial
conditions are shown. There are no remarkable differences relative to
Fig. 9, insofar as the preferential region for snowfall occurrence is
concerned. However, snow accumulation from WRF forced with ERA-
interim is lower than those based on NCEP-FNL counterpart is most of
the region. The WRF model simulation with the WSM3, WSM6, and
WDM5 shows differences up to 40 and 30 cm (Fig. 9a,d,e). In the other
schemes the snowfall changes vary from 10 cm (NSSL) to 30 cm (God-
dard, Fig. 9c,g). The WRF using ERA-interim as initial conditions de-
livers more snowfall than that forced with NCEP-FNL for the WSM6 and
WDM5 in the region centered at 33oS (Fig. 9d and 9e).
The snowfall event in 2016 was more severe than in 2015. Up to
300 cm of snowfall covered the region of Valparaiso (Los Andes pro-
vince, specifically) in Chile and Las Cuevas in Argentina (Fig. 10).
Evaluation of the performance of WRF schemes for the 2016 simula-
tions differs in comparison to the 2015 case. The geographical dis-
tribution delivered by both set of simulations shows an extensive region
of snow from 27º to 33ºS. The differences between the WSM6-WSM3
(Fig. 10a), WSM6-Morrison double-moment (Fig. 10c) and WSM6-
WDM5 (Fig. 10d) also simulated higher amount of snowfall with
125 cm maximum magnitude compared with the WSM6 (Fig. 10g). The
Fig. 11. Same as Fig. 9 but for the 2016 event.
WSM6-Goddard (Fig. 10b) differ −25 to 50 mm snowfall. Differences
between the WSM6-NSSL (Fig. 10f) are up to 75 cm. Lower amounts of
moment and WDM5 (Fig. 7a,c,d) differences show the largest amount of snowfall are again simulated by NSSL. In Fig. 10 again the amount of
snowfall with maximum accumulation of approximately 30 cm snowfall accumulation occurred approximately in the same regions for
(∼36.5°S and 71.6°W) as compared to the WSM6 scheme. For WSM6- all schemes. All schemes seem to properly simulate that this event was
Goddard and WSM6-NSSL (Fig. 7b,f), the accumulated maximum si- caused by a frontal system.
mulated snow were 10–30 cm smaller than the WSM6 scheme. Differ- Differences between the simulations with ERA-interim and NCEP-
ences between WSM6-WDM6 (Fig. 7e) shows values of snow accumu- FNL are remarkable (Fig. 11). The later forcing induces snowfall values
lation in the range of −10 to 20 cm. WSM6 (Fig. 7g) show accumulated up to 60 cm higher than those predicted by the ERA-interim the WSM3,
maximum simulated snow 60 cm. WDM5 and WDM6 (Fig. 11a,d,e). Dong et al. (2017) compared NCEP-
The WSM3, Morrison double-moment and WDM5 schemes were FNL and ERA-interim with radiosonde data recorded at 11 sites in 6-h
very similar, since the amount of accumulated snow varies little among increments June–August (2011–2015) on regions of the Tibetan Pla-
the schemes. The snow-accumulated region does not differ substantially teau. Relative humidity at 900, 600 and 200 hPa had bias of 4, 8 and
among the schemes, showing the model capability to simulate snowfall. 3% respectively, with initial conditions NCEP-FNL, while in ERA-in-
The lack of snowfall prediction is most problematic in the NSSL. In terim 1, 2 and 5%, respectively. This shows the limitations of the NCEP-
general, WRF simulations conducted in this study have properly re- FNL in terms of the relative humidity. This may provide causes for
presented the location and maximum snowfall (Fig. 2). Similar studies differences between the simulations using the two ICs (see Fig. 13).
conducted for other regions may deliver different results in the sense Fig. 12a and b shows the comparison between the observed values
that other MP schemes can fit better observation for other regions. of snowfall and both set of simulations (ERA-interim and NCEP-FNL) in
Mcmillen and Steenburgh (2014) found that the Goddard scheme re- the Horcones region. The modeled data have been extracted at the
presents the precipitation amounts more realistically, whereas Morrison corresponding grid point of the observation. ERA-interim delivers va-
and WDM6 produced more graupel and inconclusive precipitation lues in better agreement with observations, in particular for WSM6 and
amounts in the Great Salt Lake. WDM6 relative to NCEP-FNL-based output. Simulations conducted with
Shi et al. (2010) evaluated accumulated snowfall based on the the NCEP-FNL generally underestimate snowfall and the Morrison and
Goddard scheme and argued that these estimates agree well with radar NSSL schemes show the largest deficiency when forced by NCEP-FNL
estimates in the King City radar, Canada. According to Liu e al. (2011), initial conditions.
the Morrison scheme has shown comparable skill and is apparently Different from noted in the 2015 runs, simulations conducted with
superior to schemes such as WSM6, WDM6 and Goddard in the Col- the ERA-interim initial conditions in the 2016 experiment are not al-
orado headwaters region, which substantially overestimate snowfall ways closer to observations than those delivered by NCEP-FNL forcing.
amounts. Underestimation of snow and overestimation of precipitation Comparison with on site measurements shows that simulations under
and graupel in the WSM6 have been found by Dudhia et al. (2008) in the NCEP-FNL initial conditions provide values much closer observation
the Korea. than those predicted by the ERA-interimwith the exception of Goddard
Due to the lack of field observations for the 2011 event, from now and WDM6 (Fig. 12b). Furthermore, small differences among the MP
on we discuss individual values as simulated by the NCEP-FNL, and schemes are observed with NCEP-FNL as initial conditions. In general,
differences between the simulated snowfall from the NCEP-FNL and differences in regard to observations are around to 60 (80) cm in the
ERA-interim initial conditions. In the 2015 event, all MP schemes NCEP-FNL (ERA-interim) with exception to the NSSL approach, with
(Fig. 8) under the NCEP-FNL initial conditions show higher con- anomalies as high as 100 cm (Fig. 12b). In both, the 2015 and 2016
centration of snow in the 34-37oS latitudinal belt. The WSM3 (Fig. 8a), experiments WDM6 performs best under the two initial conditions. It
Goddard (Fig. 8b), Morrison double-moment (Fig. 8c) and WDM5 should also be pointed out that the magnitude of model errors increases

72
A.N. Comin et al. Weather and Climate Extremes 21 (2018) 65–75

snowfall evolution, being the other schemes in better agreement with

observations. In the case of ERA-interim, the simulations show more
scattered results and qualitatively an ensemble mean would reproduce
more accurately those observed data.
Fig. 12e shows the snow evolution as resulted from NCEP-FNL and
ERA-interim initial conditions. The ERA-interim delivers better results
than the NCEP-FNL. Fig. 12f shows differences between ensemble and
the observations. The model presents difficulties in estimating the rate
of snow increase at 12–18 UTC on June 3, 2016, but after stabilizing (4
June) the snowfall is consistent with the observations. In general, the
simulations were able to deliver very consistent results, is difficult to
accurately estimate the snowfall in a specific weather station since
model grid-cell size corresponds to a larger area estimates. The WSM6
based on the NCEP-FNL scheme presents the smallest root mean square
error with 18 cm, while WDM5 presents the largest error reaching
42 cm of accumulated snow (see Table 4). For ERA-interim, the God-
dard Scheme delivers the smallest errors by about 13.8 cm while the
WDM6 shows weaker performance with 38 cm of snow depth differ-
Fig. 12. Observed and simulated accumulated snow depth (cm) in 72 h at
ences. Fig. 13 shows differences between air temperature and dew point
Horcones in the 2015 (a) and 2016 (b) event. (c) and (d) depict the snowfall
evolution for Portillo in 2016 for NCEP-FNL and ERA-interim, respectively. (e)
temperatures for the WSM6 simulations during the 2016 case event. It
snow evolution for Portillo as resulted from NCEP-FNL and ERA-interim initial is clear that large differences result from the simulations with the two
conditions and (f) differences between ensemble and the observations. different initial conditions. (http://weather.uwyo.edu/upperair/
sounding.html). From low levels up to 3000 m altitude, where most
of water vapor is located, ERA-interim initial conditions allows the
model to properly simulate in the lower atmosphere. Results also ob-
served by Dong et al. (2017) favorably NCEP-FNL and ERA-interim with
radiosonde data over the Tibetan Plateau.

5. Concluding remarks

This study evaluated the impact of seven microphysics para-

meterizations schemes using WRF model forced by two different initial
and boundary conditions, mimicking three extreme snowfall events in
the Southern Andes. Our simulations of snowfall show considerable
sensitivity to the choice of microphysics parameterization. For instance,
the WSM3, Morrison and WDM5 overestimate snowfall as compared to
WSM6, Goddard, WDM6 and NSSL schemes.
These WRF results shed some light on the feasibility of predicting
extreme snow events with reasonable accuracy. In particular by taking
into account the complex topography of the Andes. Evaluation of dif-
ferences among the MP schemes and observations had been very useful
to identify that the WSM6 approach is more suitable to be used in the
region as far as the snow forecast is concerned.
Other aspect that should be pointed out is that the snow depth si-
mulations forced by the ERA-interim show values closer to observations
Fig. 13. Atmospheric sounding and the WSM6 scheme in Mendoza for June 3, 4 than those delivered by NCEP-FNL for the majority of schemes. This
and 5 at 00 UTC and 12 UTC. Black line is the difference between the tem- results show that the WRF model can be a useful tool in simulating
perature (T) and the dew point temperature (Td) for the observed data (Obs); snowfall events. WRF simulations can therefore partially alleviate the
Red line T-Td (NCEP-FNL) for data simulated by the WRF using the NCEP-FNL gap related to the absence of ground observations, because the net-
conditions; and Blue T-Td using the ERA-interim conditions. Web reference: works of meteorological stations and atmospheric soundings are scarce
http://weather.uwyo.edu/upperair/sounding.html.
in the region.

in consonance with the intensity of the snow event. Acknowledgments

Fig. 12c and d shows the evolution of the 2016 snowfall event
Portillo. In 2016, 33 values are available allowing for an in-depth The authors thank CAPES-PNPD for support of this work (1671778).
analysis. The WRF model reproduces the onset of the event very sa- Chilean Water Authority (Dirección General de Aguas), Rodrigo
tisfactorily, in particular for ERA-interim initial conditions (Fig. 12d). Villarreal (Departamento General de Irrigación, Argentina), and
For the NCEP-FNL, the WDM5, Morison, WSM3 overestimate the Mariano Masiokas. Flavio Justino and Vanucia Schumacher

Table 4
Root mean square error to event 2016 in Portillo.
Schemes WSM3 WSM6 Goddard Morrison WDM5 WDM6 NSSL

RMSE (NCEP-FNL) 24.7 cm 18 cm 21.6 cm 34 cm 42 cm 22.7 cm 20.4 cm

RMSE (ERA-interim) 17.8 cm 19.6 cm 13.8 cm 22.7 cm 21.5 cm 38 cm 25.8 cm

73
A.N. Comin et al. Weather and Climate Extremes 21 (2018) 65–75

acknowledge the CAPES support through the funding 306181/2016-9. mountainous regions using MM5. J. Hydrometeorol. 8, 245–258.
Alfonso Fernández is funded by FONDECYT 11160454. Kain, J.S., 2004. The Kain–Fritsch convective parameterization: an update. Journal of
Applied Meteorology and Climatology 43, 170–181. < 0170:TKCPAU > 2.0.CO;2.
https://doi.org/10.1175/1520-0450(2004)043.
References Kirtman, B.P., Min, D., 2009. Multimodel ensemble ENSO prediction with CCSM and CFS.
Mon. Weather Rev. 137, 2908–2930.
Lee, H., Lee, T.-Y., 1994. The governing factors for heavy snowfalls in Youngdong area (in
Cai, X., Rosegrant, M.W., Ringler, C., 2003. Physical and economic efficiency of water use
Korean with English abstract). Journal of the Korean Meteorological Society 30,
in the river basin : implications for efficient water management. Water Resour. Res.
197–218.
39 (1) 1013. https://doi:10.1029/2001WR000748.
Liang, X.-Z., Xu, M., Yuan, X., Coauthors, 2012. Regional climate –weather Research and
Chern, J.-D., Tao, W.-K., Lang, S.E., Matsui, T., Li, J.-L.F., Mohr, K.I., Skofronick-
forecasting model. American meteorological society. BAMS 1362–1387.
Jacksonv, G.M., Peters-Lidard, C.D., 2016. Performance of the Goddard multiscale
Lim, K.-S.S., Hong, S.-Y., 2010. Development of an effective double-moment cloud mi-
modeling framework with Goddard ice microphysical schemes. J. Adv. Model. Earth
crophysics scheme with prognostic cloud condensation nuclei (CCN) for weather and
Syst. 8, 66–95. https://doi.10.1002/2015MS000469.
climate models. Mon. Weather Rev. 138, 1587–1612. https://doi.org/10.1175/
Cha, Y.-M., Lee, H.-W., Lee, S.-H., 2011. Impacts of the high-resolution sea surface tem-
2009MWR2968.1.
perature distribution on modeled snowfall formation over the Yellow Sea during a
Lim, K., Hong, S., 2010a. Development of an effective double-moment cloud microphysics
cold-air outbreak. Weather Forecast. 26, 487–503. https://doi.org/10.1175/WAF-D-
scheme with prognostic cloud condensation nuclei (CCN) for weather and climate
10-05019.1.
models. Mon. Weather Rev. 138, 1587–1612. https://doi.org/10.1175/
Dee, D.P., Coauthors, 2011. The ERA-Interim reanalysis: configuration and performance
2009MWR2968.1.
of the data assimilation system. Q. J. R. Meteorol. Soc. 137, 553–597. https://doi.
Liu, C., Ikeda, K., Thompson, G., Rasmussen, R., Dudhia, J., 2011. High-resolution si-
org/10.1002/qj.828.
mulations of wintertime precipitation in the Colorado Headwaters Region: sensitivity
Dinesh, K., Mohanty, U.C., Krisha, K., 2016. Sensitivity of land surface and Cumulus
to physics parameterizations. Mon. Weather Rev. 139, 3533–3553. https://doi.org/
schemes for Thunderstorm prediction. In: The International Archives of the
10.1175/MWR-D-11-00009.1.
Photogrammetry, Remote Sensing and Spatial Information Sciences XLI-B8, 2016
Livneh, B., Xia, Y., Mitchell, K.E., Ek, M.B., Lettenmaier, D.P., 2010. Noah LSM snow
XXIII.
model diagnostics and enhancements. J. Hydrometeorol. 11 (3), 721–738. https://
Dong, Y., Li, G., Yuan, M., Xie, X., 2017. Evaluation of five grid datasets against radio-
doi.org/10.1175/2009JHM1174.1.
sonde data over the eastern and downstream regions of the Tibetan plateau in
Mansell, E.R., Ziegler, C.L., Bruning, E.C., 2010. Simulated electrification of a small
summer. Atmosphere 8 (56), 1–19. https://doi.org/10.3390/atmos8030056.
thunderstorm with two–moment bulk microphysics. J. Atmos. Sci. 67, 171–194.
Dudhia, J., 1989. Numerical study of convection observed during the Winter Monsoon
Manterola, L., 2009. Controlfronterizo integrado Los Libertadores. Undergraduate thesis,
Experiment using a mesoscale two-dimensional model. J. Atmos. Sci. 46, 3077–3107.
Facultada de Arquitectura y Urbanismo. Universidad de Chile 133. Available at:
< 3077:NSOCOD > 2.0.CO;2. https://doi.org/10.1175/1520-469(1989)046.
https://goo.gl/h77GBT.
Dudhia, J., Hong, S.-Y., Lima, K.-S., 2008. A new method for representing mixed-phase
Maraun, D., Wetterhall, F., Ireson, A.M., Chandler, R.E., Kendon, E.J., Coauthors, 2010.
particle fall speeds in bulk microphysics parameterizations. J. Meteorol. Soc. Jpn.
Precipitation downscaling under climate change: recent developments to bridge the
86A, 33–44. https://doi.org/10.2151/jmsj.86A.33.
gap between dynamical models and the end user. Rev. Geophys. 48https://doi.org/
ElTahan, M., Magooda, M., 2017. Evaluation of different WRF microphysics schemes:
10.1029/2009RG000314. RG3003.
severe rainfall over Egypt case study. Atmospheric and Oceanic Physic (in press).
Marengo, J.A., Douglas, M.W., Silva Dias, P.L., 2002. The South American low-level jet
https://arxiv.org/abs/1711.04163?context=physics.
east of the Andes during the 1999 LBA-TRMM and LBA-WET AMC campaign. J.
Evans, M., JurewiczSr, M.L., 2009. Correlations between Analyses and forecasts of
Geophys. Res. 107 8079. https://doi:10.1029/2001JD001188.
banded heavy snow ingredients and observed snowfall. Weather Forecast. 24,
Mark, B.G., Baraer, M., Fernández, A., Immerzeel, W., Moore, R., Weingartner, R., 2015.
337–350. https://doi.org/10.1175/2008WAF2007105.1.
Glaciers as water resources. In: Huggel, C., Carey, M., Clague, J.J., Kääb, A. (Eds.),
Falvey, M., Garreaud, R., 2007. Wintertime precipitation episodes in Central Chile: as-
The High-mountain Cryosphere Environmental Changes and Human Risks.
sociated meteorological conditions and orographic influences. J. Hydrometeorol. 8,
Cambridge University Press, Cambridge, pp. 184–203.
171–193.
Masiokas, M., Villalba, R., Luckman, B., Montaña, E., Betman, E., Christie, D., Quesne,
Galligani, V.S., Wang, D., Imaz, M.A., Salio, P., Prigent, C., 2017. Analysis and evaluation
C.L., Mauget, S., 2013. Recent and historic andean snowpack and streamflow var-
of WRF microphysical schemes. Atmos. Meas. Tech 10, 3627–3649. https://doi.org/
iations and vulnerability to water shortages in central-western Argentina. Climate
10.5194/amt-10-3627-201.
Vulnerability 5, 213–227. https://doi.org/10.1016/B978-0-12-384703-4.00522-0.
Garreaud, R.D., 2009. The Andes climate and weather. Adv. Geosci. 22, 1–11.
Masiokas, M.H., Villalba, R., Christie, D.A., Betman, E., Luckman, B.H., Quesne, C.L.,
Hall, D.K., Riggs, G.A., 2016. MODIS/Terra Snow Cover Daily L3 Global 500m Grid,
Prieto, M.R., Mauget, S., 2012. Snowpack variations since AD 1150 in the Andes of
Version 6. NASA National Snow and Ice Data Center Distributed Active Archive
Chile and Argentina ( 30 ° - 37 ° S ) inferred from rainfall, tree-ring and documentary
Center, Boulder, Colorado USA. https://doi.org/10.5067/MODIS/MOD10A1.006.
records. J. Geophys. Res. 117, 1–11. http://dx.10.1029/2011JD016748.
Hong, S.-Y., Dudhia, J., Chen, S.-H., 2004. A revised approach to ice microphysical
Maussion, F., Scherer, D., Mölg, T., Collier, E., Curio, J., Finkelnburg, R., 2014.
processes for the bulk parameterization of clouds and precipitation. Mon. Weather
Precipitation seasonality and variability over the Tibetan plateau as resolved by the
Rev. 132, 103–120. https://doi.org/10.1175/1520-0493(2004)
high asia reanalysis. J. Clim. 27, 1910–1927. https://doi.org/10.1175/JCLI-D-13-
132<0103:ARATIM>2.0.CO;2.
00282.1.
Hong, S.-Y., Lim, O.J., 2006. The WRF single- moment 6-class microphysics scheme
Mcmillen, J.D., Steenburgh, W.J., 2015. Impact of microphysics parameterizations on
(WSM6). Journal of the Korean Meteorological Society 42, 129–151.
simulations of the 27 october 2010 great Salt Lake–Effect snowstorm. Weather
Hornbein, T., Schoene, R., 2001. High Altitude: an Exploration of Human Adaptation
Forecast. 30, 136–152. https://doi.org/10.1175/WAF-D-14-00060.1.
(Lung Biology in Health and Disease Volume 161). Marcel Dekker, New York, USA.
Mielikainen, J., Huang, B., Huang, A.H.-L., 2014. Intel xeon phi accelerated weather
pp. 42–874 ISBN 082474604X.
Research and forecasting (WRF) goddard microphysics scheme. Geosci. Model Dev.
Hsieh, W., 2009. Machine Learning Methods in the Environmental Sciences. Cambridge
Discuss. (GMDD) 7, 8941–8973. https://doi.org/10.5194/gmdd-7-8941-2014.
University Press.
Miller, D.K., 2012. Near-term effects of the lower atmosphere in simulated northwest flow
Huang, M., Huang, B., Huang, A.H.-L., 2015. Parallel implementation of WRF double
snowfall forced over the southern appalachians. Weather Forecast. 27, 1198–1216.
moment 5-class cloud microphysics scheme on multiple GPUs. Satellite data com-
https://doi.org/10.1175/WAF-D-11-00103.1.
pression. Communications, and Processing XI 9501, 95010K1–9501010. https://doi.
Milrad, S.M., Gyakum, J.R., Lombardo, K., Atallah, E.H., 2014. On the dynamics, ther-
org/10.1117/12.2180039.
modynamics, and forecast model evaluation of two snow-burst events in southern
Ikeda, K., Rasmussen, R., Liu, C., Gochis, D., Yates, D., Chen, F., Coauthors, 2010.
alberta. Weather Forecast. 29, 725–749. https://doi.org/10.1175/WAF-D-13-
Simulation of seasonal snowfall over Colorado. Atmos. Res. 97, 462–477. https://doi.
00099.1.
org/10.1016/j.atmosres.2010.04.010.
Mlawer, E.J., Taubman, S.J., Brown, P.D., Iacono, M.J., Clough, S.A., 1997. Radiative
Janjic, Z.I., 2001. Nonsingular implementation of the Mellor–Yamada level 2.5 scheme in
transfer for inhomogeneous atmosphere: RRTM, a validated correlated-k model for
the NCEP Meso model. NCEP. Tech. Rep. 437, 61.
the longwave. J. Geophys. Res. 102, 16663–16682. https://doi.org/10.1029/
Jankov, I., Bao, J.-W., Neiman, P.J., Schultz, P.J., Yuan, H.-L., White, A.B., 2009.
97JD00237.
Evaluation and comparison of microphysical algorithms in ARW-WRF model simu-
Molthan, A.L., Colle, B.A., 2012. Comparisons of single- and double-moment micro-
lations of atmospheric river events affecting the California coast. J. Hydrometeorol.
physics schemes in the simulation of a synoptic-scale snowfall event. Mon. Weather
10, 847–870. https://doi.org/10.1175/2009JHM1059.1.
Rev. 140, 2982–3002. https://doi.org/10.1175/MWR-D-11-00292.1.
Jankov, I., Gallus Jr., W.A., Segal, M., Shaw, B., Koch, S.E., 2005. The impact of different
Morrison, H., Curry, J.A., Khvorostyanov, V.I., 2005. A new double-moment microphysics
WRF Model physical parameterizations and their interactions on warm season MCS
parameterization for application in cloud and climate models. Part I: description. J.
rainfall. Weather Forecast. 20, 1048–1060. https://doi.org/10.1175/WAF888.1.
Atmos. Sci. 62, 1665–1677. https://doi.org/10.1175/JAS3446.1.
Jiménez, P.A., Dudhia, J., González-Rouco, J.F., Montávez, J.P., García-Bustamante, E.,
Morrison, H., Milbrandt, J.A., Bryan, G.H., Ikeda, K., Tessendorf, S.A., Thompson, G.,
Navarro, J., Arellano, J.V.-G., Muñoz-Roldán, A., 2013. An evaluation of WRF's
2014. Parameterization of cloud microphysics based on the prediction of bulk ice
ability to reproduce the surface wind over complex terrain based on typical circu-
particle properties. Part II: case study comparisons with observations and other
lation patterns. Journal of Geophysical Research Atmospheres 118, 7651–7669.
schemes. J. Atmos. Sci. 77, 312–339. https://doi.org/10.1175/JAS-D-14-0066.1.
https://doi.org/10.1002/jgrd.50585.
Morroni, G.Y., 2017. Using Weather Research and Forecasting (WRF) Model for Extreme
Jin, J., Wen, L., 2012. Evaluation of snowmelt simulation in the weather Research and
Precipitation Forecasting in an Andean Region with Complex Topography.
forecasting model. J. Geophys. Res. 117 D10110. https://doi.org/10.1029/
Niu, Guo-Yue, Yang, Zong-Liang, Mitchell Kenneth, E., Chen, Fei, Ek Michael, B., Barlage,
2011JD016980.
Michael, Xia, Youlong, 2011. The community Noah land surface model with multi-
Jin, J., Miller, N.L., 2007. Analysis of the impact of snow on daily weather variability in
parameterization options (Noah-MP): 1. Model description and evaluation with local-

74
A.N. Comin et al. Weather and Climate Extremes 21 (2018) 65–75

scale measurements. J. Geophys. Res.: Atmosphere 116 (D12). https://doi.org/10. 5111–5126. https://doi.org/10.5194/hess-21-5111-2017.
1029/2010JD015139. Tao, Y., 2013. High-resolution seasonal snowfall simulation over Northeast China. Chin.
Ohlanders, N., Rodriguez, M., McPhee, J., 2013. Stable water isotope variation in a Sci. Bull. 58, 1412–1419.
Central Andean watershed dominated by glacier and snowmelt. Hydrol. Earth Syst. Tebaldi, C., Hayhoe, K., Arblaster, J.M., Meehl, G.A., 2006. Going to the extremes: an
Sci. 17 (3), 1035–1050. https://doi.org/10.5194/hess-17-1035-2013. intercomparison of model-simulated historical and future changes in extreme events.
ONEMI, 2011. Nevadas Zona Sur-regiones Biobío Y Araucanía, Santiago. pp. 44. Available Climate Change 79 (3–4), 185–211. https://doi.org/10.1007/s10584-006-9051-4.
at: https://goo.gl/NcqSvC. Thompson, G., Rasmussen, R.M., Manning, K., 2004. Explicit forecasts of winter pre-
ONEMI, 2015. AlertaTempranaPreventiva para la RegiónMetropolitanaporsistema cipitation using an improved bulk microphysics scheme. Part I: description and
frontal. Available at: https://goo.gl/uzs2tr. sensitivity analysis. Mon. Weather Rev. 132, 519–542. https://doi.org/10.1175/
ONEMI, 2016. AlertaTempranaPreventiva para la RegiónMetropolitanaporsistema 1520-0493(2004)132<0519:EFOWPU>2.0.CO;2.
frontal. Available at: https://goo.gl/a481RW. Viale, M., Houze Jr., R.A., Rasmussen, K.L., 2013. Upstream orographic enhancement of a
Rajeevan, M., Kesarkar, A., Thampi, S.B., Rao, T.N., Radhakrishna, B., Rajasekhar, M., narrow cold-frontal rainband approaching the Andes. Mon. Weather Rev. 141,
2010. Sensitivity of WRF cloud microphysics to simulations of a severe thunderstorm 1708–1730. https://doi.org/10.1175/MWR-D-12-00138.1.
event over southeast India. Ann. Geophys. 28, 603–619. Viale, M., Nuñez, M.N., 2011. Climatology of winter orographic precipitation over the
Shi, J.J., Matsui, T., Tao, W.-K., Tan, Q., Peters-Lidard, C., Chin, M., Pickering, K., Guy, subtropical southern Andes and associated synoptic and regional characteristics. J.
N., Langa, S., Kemp, E.M., 2014. Implementation of an aerosol–cloud-micro- Hydrometeorol. 12, 481–507. https://doi.org/10.1175/2010JHM1284.1.
physics–radiation coupling into the NASA unified WRF: simulation results for the 6–7 Vuille, M., Ammann, C., 1997. Regional snowfall patterns in the high, arid Andes.
August 2006 AMMA special observing period. Q. J. R. Meteorol. Soc. 140, Climatic Change 36, 413–423.
2158–2175. https://doi.org/10.1002/qj.2286. Zhu, J., Liang, X.-Z., 2007. Regional climate model simulation of U.S. precipitation and
Stehr, A., Aguayo, M., 2017. Snow cover dynamics in Andean watersheds of Chile surface air temperature during 1982–2002: interannual variation. J. Clim. 20,
(32.0–39.5° S) during the years 2000–2016. Hydrol. Earth Syst. Sci. 21 (10), 218–232.

75