The Atmospheric Radiation Measurement (ARM) West Antarctic Radiation Experiment (AWARE) provided a highly detailed set of remote-sensing and surface observations to study Antarctic clouds and surface energy balance, which have received much less attention than for the Arctic due to greater logistical challenges. Limited prior Antarctic cloud observations have slowed the progress of numerical weather prediction in this region. The AWARE observations from the West Antarctic Ice Sheet (WAIS) Divide during December 2015 and January 2016 are used to evaluate the operational forecasts of the Antarctic Mesoscale Prediction System (AMPS) and new simulations with the Polar Weather Research and Forecasting Model (WRF) 3.9.1. The Polar WRF 3.9.1 simulations are conducted with the WRF single-moment 5-class microphysics (WSM5C) used by the AMPS and with newer generation microphysics schemes. The AMPS simulates few liquid clouds during summer at the WAIS Divide, which is inconsistent with observations of frequent low-level liquid clouds. Polar WRF 3.9.1 simulations show that this result is a consequence of WSM5C. More advanced microphysics schemes simulate more cloud liquid water and produce stronger cloud radiative forcing, resulting in downward longwave and shortwave radiation at the surface more in agreement with observations. Similarly, increased cloud fraction is simulated with the more advanced microphysics schemes. All of the simulations, however, produce smaller net cloud fractions than observed. Ice water paths vary less between the simulations than liquid water paths. The colder and drier atmosphere driven by the Global Forecast System (GFS) initial and boundary conditions for AMPS forecasts produces lesser cloud amounts than the Polar WRF 3.9.1 simulations driven by ERA-Interim.
West Antarctica is among the most rapidly warming locations on Earth, and its warming is closely linked with global sea level rise (Rignot, 2008: Turner et al., 2006; Steig et al., 2009; Bromwich et al., 2013a, 2014). Recent paleoclimate work links temperature increases of a few degrees with past sea level increases of several meters due to disintegration of parts of the Antarctic Ice Sheet (DeConto and Pollard, 2016). Additional rise in Antarctic summer temperatures could lead to more frequent and extensive surface melting of the West Antarctic Ice Sheet (WAIS) (e.g., Nicolas and Bromwich, 2014). Conversely, increased temperatures can result in greater evaporation over the oceans and increased snowfall over Antarctica (Nicolas and Bromwich, 2014). The observational evidence shows West Antarctic warming since the 1950s (Bromwich et al., 2013a). Unlike the elevated ice mass of East Antarctica, West Antarctica is highly prone to intrusions of moist air from the Southern Ocean (Nicolas and Bromwich, 2011; Scott et al., 2017). Thus, the West Antarctic climate is much more ocean dominated than that of the colder and drier East Antarctica.
Moisture flux over West Antarctica leads to cloud formation. Clouds alter the net surface radiative flux and can thus impact the onset, extent, intensity, and duration of surface melting, refreezing, and ultimately meltwater control on cryospheric dynamics or runoff into the ocean (van Trincht et al., 2016). Modeling studies have shown that changes in cloud properties over Antarctica may impact regions of the globe well beyond high southern latitudes (Lubin et al., 1998). Moreover, Antarctic clouds have different characteristics than Arctic clouds (Hogan, 1986; Bromwich et al., 2012; Grosvenor et al., 2012; O'Shea et al., 2017). Silber et al. (2018a) show that cloud thickness at McMurdo Station peaks in austral winter, possibly due to cyclone activity, while Arctic cloud thickness peaks in boreal summer (Shupe, 2011). O'Shea et al. (2017) note significantly different types and concentrations of cloud condensation nuclei (CCN), and ice nuclei (IN) are expected between the Arctic and Antarctic due to the minimal anthropogenic sources at high southern latitudes. Aerosols tend to peak in winter and spring in the Arctic with a minimum during summer, while Antarctic aerosols tend to peak in austral summer and fall and are reduced during winter (e.g., Wagenbach et al., 1988; Schmeisser et al., 2018). Consequently, it is uncertain how well the findings of the various Arctic field programs and modeling experiments translate to Antarctica.
Clouds, including liquid water clouds, have a strong modulation on the local climate (Nicolas and Bromwich, 2011; Bromwich et al., 2012; Scott et al., 2017; Silber et al., 2018a). A supercooled liquid cloud is likely to be more optically thick than a fully glaciated ice cloud (Shupe and Intrieri, 2004; Grosvenor et al., 2012; McCoy et al., 2015). Arctic cloud modeling studies find that cloud liquid water is frequently underrepresented in simulations with bulk microphysics schemes, and this can result in too little longwave radiation and too much shortwave radiation reaching the surface (e.g., Morrison and Pinto, 2006).
Unfortunately, there have been few Antarctic field programs to detail cloud
microphysical properties (e.g., Bromwich et al., 2012; Lachlan-Cope et al., 2016; Scott and Lubin, 2016). One study in the past decade by the British
Antarctic Survey examined clouds over the Antarctic Peninsula (e.g.,
Grosvenor et al., 2012; Lachlan-Cope et al., 2016). Lachlan-Cope et al. (2016) found large differences in ice crystal concentrations between the
clouds on the eastern and western sides of the peninsula, while Grosvenor et
al. (2012) found elevated ice crystal concentrations with relatively warm
temperature between
The prime motivation for this work, as noted by Witze (2016), is that there has been little in-place atmospheric science or climatological field work over interior West Antarctica since 1967, when a weather balloon program ended. A few automatic weather stations there have provided direct meteorological information since 1980 (Lazzara et al., 2012). There is a need to quantify the impact of continental and oceanic air masses on the local hydrology and surface energy balance. Furthermore, there is a need for observations that can enable improved numerical simulations, both regional and global, through better representation of Antarctic clouds. The scarcity of cloud observations and well-tested simulations has so far inhibited significant progress. The work presented here may contribute to improvements to the AMPS simulations of clouds being sought by the National Center for Atmospheric Research (NCAR) if computational efficiency can be achieved (Jordan Powers, personal communication, 2018). Furthermore, we seek to evaluate and improve the numerical weather prediction for Antarctica, where the sparse observational network, the physics of the polar atmosphere, and the steep terrain challenge model capabilities (Bromwich et al., 2012). The need for accurate weather forecasting to support logistical and scientific activities has been important since the earliest Antarctic explorations
The Atmospheric Radiation Measurement (ARM) West Antarctic Radiation Experiment (AWARE, Witze, 2016) is a recent field program to study clouds and their impacts on atmospheric radiative transfer over the Antarctic continent. AWARE used the joint capabilities of the United States Antarctic Program, managed by the National Science Foundation, and the Department of Energy's second ARM Mobile Facility (AMF2) to provide quantitative data about energy components, changing air masses, and cloud microphysical data to improve model simulations of the ice sheet as influenced by Earth system processes. The AMF2 consists of a collection of lidars, radars, and radiometers taking remote-sensing observations of the Antarctic clouds combined with in situ instruments documenting the atmospheric state, but more comprehensive observations are needed.
Beginning late November 2015, AMF2 was deployed to Antarctica to make the
most extensive suite of measurements in more than 40 years (Witze, 2016). The primary AWARE site was McMurdo Station (77.85
The WAIS Divide component of the AWARE field campaign ran from 4 December
2015 through to 18 January 2016. A suite of ARM Mobile Facility instruments
(Mather and Voyles, 2013) optimized for surface energy budget observations
was moved from McMurdo to the WAIS Divide site during this period. Estimates
of upper-air temperature and moisture were obtained from 6-hourly
rawinsonde launches and continuous retrievals from a profiling microwave
radiometer (MWR, Morris 2006). Liquid water path (LWP) was extracted from a
co-location of the MWR with a G-Band Vapor Radiometer Profiler (Cadeddu,
2010). The uncertainty of observed LWP is 10 g m
Upwelling shortwave and longwave radiative flux components were measured by a Surface Energy Balance System (SEBS, Cook, 2018). Downwelling flux components were measured by a Sky Radiation System, which consists of a normal incidence pyrheliometer, shaded pyranometers, and pyrgeometers (Dooraghi et al., 1996). The global downwelling shortwave flux was computed as in Nicolas et al. (2017). Surface fluxes for sensible and latent heat are derived according to the algorithm of Andreas et al. (2010). Near-surface measurements of temperature, moisture, and wind speed were measured by the ARM surface meteorological instrumentation (Holdridge and Kyrouac, 1993). Instruments at the WAIS Divide were unable to obtain reliable measurements of the heat flux within the ice pack. As an alternative, estimates of the conductive heat flux from the ice surface and the underlying ice were taken from Nicolas et al. (2017), who calculated the residual of other terms in the surface energy balance.
A cloud mask (derived from detected hydrometeor-bearing air volumes) is used to determine the cloud and liquid occurrence fractions at the WAIS Divide associated with the method of Silber et al. (2018a). In brief, depolarization micropulse lidar (MPL; Flynn et al., 2007) observations are used to generate a linear depolarization ratio (LDR) versus log-scaled particulate backscatter cross-section two-dimensional histogram that can identify the hydrometer categories (Silber et al., 2018b, c).
Hourly time series of total hydrometeor and liquid-cloud fractions were calculated from the processed cloud and liquid masks (with column integration). The occurrence fractions were normalized relative to the hourly MPL data availability, under the assumption that the measured period provided an acceptable representation of the whole hour. It should be noted that the MPL pulse can occasionally be completely attenuated by optically thick cloud layers (for example, as part of a frontal system). Therefore, the real cloud top, geometrical cloud thickness, and potentially the liquid occurrence are underestimated by the MPL in these situations.
The advanced research Weather Research and Forecasting Model (WRF) is an extensively used community numerical weather prediction model for numerous applications world-wide (e.g., Skamarock et al., 2008). Most of the polar optimizations for Polar WRF are added in the Noah land surface model (Barlage et al., 2010) and improve the representation of heat transfer through snow and ice (Hines and Bromwich, 2008; Hines et al., 2015). Fractional sea ice was implemented in Polar WRF by Bromwich et al. (2009), followed by the addition of specified variable sea ice thickness, snow depth on sea ice, and sea ice albedo. These updated options were developed by the Polar Meteorology Group (PMG) at Ohio State University's Byrd Polar and Climate Research Center and were included in the standard release of WRF (NCAR Mesoscale and Microscale Meteorology, 2019) with the help of the Mesoscale and Microscale Meteorology Division at NCAR (Hines et al., 2015). Hines et al. (2011) made comparisons for cloud and radiation quantities between Polar WRF 3.0.1.1 simulations and observations at the north slope of the Alaska ARM site.
Recently, Deb et al. (2016) evaluated Polar WRF 3.5.1 versus near-surface observations from West Antarctica. They found that pressure is simulated with high skill, and wind speed is generally well represented. The timing and amplitude of strong wind events were well captured. There were weaknesses in the diurnal cycle of temperature, especially denoted by a cold summertime minimum temperature bias. This was attributed to a negative bias in downwelling longwave radiation, consistent with clouds over Antarctica being poorly represented by models (e.g., Bromwich et al., 2012, 2013b; King et al., 2015; Listowski and Lachlan-Cope, 2017). Arctic modeling studies, however, suggest reason for optimism as Hines and Bromwich (2017) improved the representation of low-level liquid clouds by Polar WRF 3.7.1 with adjustments to the microphysics for simulations of the Arctic Summer Cloud Ocean Study (ASCOS, Tjernström et al., 2014) near the North Pole during the period August–September 2008.
To improve forecasting support for the United States Antarctic Program, the National Science Foundation's Office of Polar Program initiated the Antarctic Mesoscale Prediction System (AMPS, Powers et al., 2012) in the year 2000. The AMPS is a real-time numerical weather prediction with Polar WRF through a collaboration between the National Center for Atmospheric Research (NCAR) and the PMG. The AMPS supports a variety of scientific and logistical needs for its international user base and has reduced costly flight turn-arounds between Christchurch, New Zealand, and the McMurdo Station (Powers et al., 2012).
For the time of the AWARE WAIS case study, the AMPS grid system consists of a series of nested domains with 60 vertical levels between the surface and the model top at 10 hPa. There are 12 layers in the lowest 1 km, with the levels nearest the surface at 10, 37, 73, and 119 m . The outermost domain had 30 km horizontal resolution and covered Antarctica and much of the Southern Ocean (Fig. 1a). Grid 2 had 10 km resolution and covered the Antarctic continent. Four additional higher-resolution nested domains (3.3 or 1.1 km) covered the Antarctic Peninsula, the South Pole and the region near the McMurdo Station. For the present study, grid 2 fields from the AMPS forecasts are used, and results are bilinearly interpolated to the WAIS Divide from the four nearest grid points. Sea ice fraction is provided by the National Snow and Ice Data Center analyses. The mesoscale representation in the initial fields is enhanced by the assimilation with 3-D variational data assimilation (Barker et al., 2004). Ingested fields include surface data, upper-air soundings, aircraft observations, geostationary and polar-orbiting satellite atmospheric motion vectors (AMVs), Constellation Observing System for Meteorology, Ionosphere, and Climate (COSMIC) GPS radio occultations, and Advanced Microwave Sounding Unit (AMSU) radiances. Twice daily AMPS forecasts are begun from analyses at 00:00 and 12:00 UTC of the Global Forecast System (GFS, NOAA Environmental Modeling Center, 2003), a global forecast system run by the U.S. National Centers for Environmental Prediction. For the current study we use the AMPS output for forecast hours 12–21 at 3 h intervals. Thus, our AMPS fields have a spin-up of a minimum of 12 h, with the possibility of jumps every 12 h due to the change toward a more recent initialization time. AMPS forecast fields in original WRF format are available from NCAR Mesoscale and Microscale Meteorology (2019). Selected AMPS output fields for March 2006–December 2016 for grids 2–6 can be downloaded from the Polar Meteorology Group (2017).
Antarctic Mesoscale Prediction System (AMPS) grids
The scarcity of Antarctic meteorological observing stations and satellite blackout periods that can coincide with peak aircraft flight times increase the need for AMPS accuracy. Wille et al. (2017) note that unpredicted fog, low ceilings, and high winds lead to costly flight mission failures over Antarctica, thus accurately predicting acceptable flight windows is essential to prevent delays for science missions and cargo transportation. Unfortunately, the AMPS has been shown to underestimate low clouds over the Antarctica (Wille et al., 2017). According to Pon (2015) the cloud fraction product in the AMPS is so unreliable that most forecasters rely more on the AMPS relative humidity as a proxy for cloud predictions. Therefore, addressing the cloud prediction in the AMPS is a primary concern of this work.
Simulations for 3 December 2015 to 21 January 2016.
The AMPS simulations used for the period 00:00 UTC 1 December 2015–12:00 UTC 19 January
2016 employ Polar WRF 3.3.1 as described by Wille et al. (2017). Afterward,
the AMPS forecast system was upgraded to Polar WRF 3.7.1 (Table 1). The
update has no impact on our analyses for the WAIS Divide, where all of the
observations concluded prior to the change. Grid 2 at 10 km resolution has
Additional numerical simulations during the time of the AWARE field program
are conducted with Polar WRF version 3.9.1 (Table 1). These are
single-domain simulations with the same grid and topography as the AMPS grid 2
(Fig. 1b). The 60 vertical layers are identical to the AMPS simulations. In
addition to the AMPS, prior simulations of Polar WRF guide the selection of
physical parameterizations (e.g., Wilson et al., 2011, 2012; Bromwich et
al., 2013b; Cassano et al., 2017; Hines and Bromwich, 2017). The
Mellor–Yamada–Nakanishi–Niino (MYNN; Nakanishi and Niino, 2006) level-2.5
scheme is used for the atmospheric boundary layer and the corresponding
atmospheric surface layer. We use RRTMG for longwave and shortwave
radiation. Cloud liquid water, cloud ice, and snow impact the shortwave and
longwave radiation, but rain water is not used in the radiation
calculations. Cumulus is parameterized with the Kain–Fritsch scheme (Kain,
2004). The polar-optimized Noah land surface model is also used. The Polar WRF
3.9.1 simulations presented here input fractional sea ice concentrations
from gridded fields at 12.5 km resolution processed by
l'Institut Français de Recherche Pour l'Exploitation de La Mer
(
Plots of
One simulation, referred to as WRF GFS, is conducted with initial and boundary conditions taken from the GFS final analysis. The AMPS forecasts use the GFS forecasts by the same model which are available at the time. Thus, the AMPS and WRF GFS are conducted with the same forecast system, although the products used will not be identical. Additional observations are assimilated into the final analysis. Initial and boundary conditions of meteorological fields for the other Polar WRF 3.9.1 simulations are interpolated from ERA-Interim reanalysis (ERA-I; Dee et al., 2011) fields available every 6 h on 61 sigma levels and the surface at T255 resolution. We have made this change to obtain the best available agreement with observed clouds and radiation. Bracegirdle and Marshall (2012) found that, among the reanalyses they evaluated, ERA-I best represented the atmospheric circulation near Antarctica. Bromwich et al. (2013b) found that the boundary layer temperature fields were better represented in WRF simulations driven by ERA-I. Nudging toward analysis fields or observations is not performed on grid 2 during the forecast segment of the AMPS forecasts, and no nudging is included for the Polar WRF 3.9.1 simulations. Besides the microphysics schemes that are of interest to us, some differences between the AMPS and Polar WRF 3.9.1 simulations will occur due to the different base versions of WRF, the source for driving initial and boundary conditions, and the data assimilation used for AMPS initialization. Strict equality between the AMPS and Polar WRF simulations is not required for the goals of this paper, as we are interested in testing the sensitivity to the microphysics parameterization.
As shown in Table 1, four different schemes are employed for the cloud microphysics to see how the schemes impact the atmospheric hydrology and cloud radiative effect. Listowski and Lachlan-Cope (2017) previously tested five schemes with Polar WRF 3.5.1 for simulations over the central Antarctic Peninsula; however, we are interested in two newer schemes that have become available in more recent versions of WRF. Furthermore, the WAIS Divide is more southerly and colder, and the local atmosphere is likely to be more pristine than over the Antarctic Peninsula, where the oceanic influence is strong.
First, we consider WSM5C as it is the microphysics scheme used for the AMPS.
This widely used scheme is computationally efficient and considers cloud
water, cloud ice, rain, and snow as hydrometer classes. Cloud water and
cloud ice are suspended, while rain and snow gradually precipitate out with
a fall speed. Supercooled water is allowed to exist, and falling snow
gradually melts at temperatures above 0
Three more recent schemes are also tested. Following Hines and Bromwich
(2017), the two-moment Morrison scheme (e.g., Morrison et al., 2005, 2009)
is used as it has been extensively tested in the Arctic and is known for its
ability to simulate supercooled liquid water (e.g., Morrison et al., 2008;
Klein et al., 2009; Solomon et al., 2011, 2014, 2015). It was amongst the
best performing schemes in the simulations from Listowski and Lachlan-Cope (2017) .
This two-moment bulk microphysics scheme predicts mixing ratios for cloud
water, cloud ice, rain, snow, and graupel, and predicts number concentrations for cloud
ice, snow, rain, and graupel. Particle size distributions are specified with
gamma functions. The IN are parameterized according to the Cooper curve, with
greater ice crystal concentrations at lower temperatures (Cooper, 1986). The
prediction of two-moments (number concentration and condensate mixing ratio)
allows a more robust treatment of the particle size distributions that are
important for the microphysical process rates and cloud and precipitation
evolution. The liquid water droplet concentration for clouds, however, is
specified in the WRF implementation. The standard setting with WRF is 250 cm
Simulations are also performed with the aerosol-aware Thompson microphysics scheme (Thompson and Eidhammer, 2014), which is an advance over the earlier Thompson
et al. (2008) bulk microphysics scheme that was one-moment for cloud water
and two-moment for cloud ice. This microphysics scheme accounts for cloud
nucleating aerosol particles and five water species: cloud water, cloud ice,
rain, snow, and graupel. The scheme includes 1st-order aerosol treatment
with interactive IN and CCN concentrations. Nucleation or complete
evaporation of hydrometeors deplete or add to condensation nuclei. Thus, the
CCN process is now more interactive on a local scale. Cloud water,
cloud ice, and rain are treated with two-moment predictions but snow with
only single moment (mixing ratio) predictions. We refer to this scheme as the Thompson scheme. All cloud ice with diameters exceeding 200
The final microphysics scheme is the Morrison–Milbrandt P3 scheme (Morrison
and Milbrandt, 2015), hereafter called the P3 scheme. The use of the WRF 3.9.1 in our simulations is motivated by the addition of the very recent P3
scheme to the microphysics options. The new scheme avoids the previous
arbitrary categorization of frozen hydrometers into cloud and precipitation,
and thus allows for a continuum of particle properties. Fall speed is now
applied across the continuum, rather than being limited to precipitation.
There are four ice mixing ratio variables: total mass, rime mass, rime
volume, and number, allowing for 4 degrees of freedom. Liquid hydrometers
use a standard two-moment approach with cloud and rain categories. The
constant liquid droplet number, 400 cm
Both the P3 scheme and the Thompson scheme were unavailable in Polar WRF 3.5.1 when Listowski and Lachlan-Cope (2017) ran simulations for the Antarctic Peninsula. They tested the WSM5C, the WRF double-moment scheme, the Morrison scheme, the older Thompson scheme (Thompson et al., 2008), and the Milbrandt scheme (Milbrandt and Yau, 2005). The older Thompson scheme lacks the aerosol predictive ability of the newer Thompson scheme, and is single moment in cloud water. The last three schemes simulated clouds in best agreement with observations (Listowski and Lachlan-Cope, 2017). All schemes were unsuccessful in representing the supercooled water for some temperature ranges, but the results show that some schemes with more complicated microphysical parameterizations show improvements in representing Antarctic clouds.
The six simulations for this study are shown in Table 1. The AMPS 3 h output was retrieved for 1 December 2015 to 31 January 2016. Five Polar WRF 3.9.1 simulations were then performed. The AMPS has the same microphysics as the WSM5C and WRF GFS simulations. Unlike the AMPS forecasts, we used 24 hourly points of the Polar WRF 3.9.1 run segments. A 12 h spin-up is taken for each segment initialized at 00:00 UTC each day from 3 December 2015 to 19 January 2016. Output each hour for hours 12–35 is combined into fields spanning 12:00 UTC 3 December 2015 to 11:00 UTC January 2016. Polar WRF output is bilinearly interpolated from the four nearest grid points to the location of the WAIS Divide.
The time period of the December 2015–January 2016 field program at the WAIS
Divide includes a major melting event over the Ross Ice Shelf and the
adjacent Siple Coast of West Antarctica (Nicolas et al., 2017). Temperature over the Ross Ice Shelf and West Antarctica increased after 10 January, and
many observing sites there experienced maximum temperatures above freezing
for several days during the melting event. Figure 2a shows meteorological
fields near the onset of the melting event, including the sea level pressure
field, 2 m temperature, and 10 m wind speed from the WSM5C simulation at 12:00 UTC 10 January. Nicolas et al. (2017) discuss the contribution of a blocking
high between 90 and 120
Time series of
Time series of the 2 m temperature at the WAIS Divide for 7–15 January reveal
large warming after 12:00 UTC 10 January (Fig. 3a). The observed temperature
increases by 13.6
Model performance at WAIS for the period 4 December 2015–19 January 2016.
Table 2 shows statistics of simulations compared to observations. A total of 1099
hourly observations are available for most meteorological variables from
06:00 UTC on 4 December to 00:00 UTC on 19 January. Only values every 3 h are
used for AMPS statistics, since output was available at these intervals, and
means, biases and other statistics are impacted by the reduced number of
values (367). For each variable, Table 2 shows observed averages, and the
following rows show the AMPS, WSM5C, Morrison, Thompson, and P3 statistics. The
largest magnitude temperature bias is for the AMPS, which has a negative bias of
1.6
The reduced negative bias for WSM5C can be understood following the
sensitivity tests by Bromwich et al. (2013b), with driving by the GFS final
analysis (FNL) and ERA-I. They found the sensitivity to the source for
initial and boundary conditions varied depending upon season and the choice
of physical parameterizations. Their comparison using Polar WRF 3.2.1 with
the MYNN planetary boundary layer and the RRTMG radiation scheme has the closest model
configuration to that used for the AMPS and the Polar WRF 3.9.1 simulations.
They found that the 2 m temperature bias changed from
Scatter plots of observed values (horizontal axis) and simulated results (vertical axis) of 2 m temperature (
Figure 4 shows scatter plots of the 2 m temperature and downwelling longwave
radiation. The AMPS, WSM5C, and WRF GFS are shown in Fig. 4a. Morrison, Thompson,
and P3 have similar scatter fields for the 2 m temperature; therefore only Morrison
is shown (Fig. 4b). The cold bias for the AMPS is increased for temperature
warmer than
The warmer and more moist atmosphere in the Polar WRF 3.9.1 simulations is
demonstrated by vertical profiles of temperature and specific humidity
biases compared to radiosonde observations (Fig. 5). There is a general
negative bias except near 1900 m a.s.l., where the positive biases
reach up to 0.8 to 0.9
Vertical profiles of average
An especially striking difference between the AMPS simulation forced with
GFS and the simulations driven with ERA-I is shown in Fig. 5b. The AMPS is dryer
than the radiosonde observations at the WAIS Divide at all levels shown,
especially in the lowest 3000 m a.s.l. The WSM5C simulation is slightly drier
than the other Polar WRF simulations. The simulations with the newer
microphysics schemes are more moist than the observations just above the
surface with biases as large as 0.13 g kg
Figures 3b and 6b help to explain the near-surface temperature results.
Downwelling longwave radiation shows a clear negative bias for both the AMPS and
WSM5C, but the magnitude is much larger for the former. Table 3, with
contribution from SEBS observations for 7 December to 16 January, shows that
the downwelling longwave bias is quite large,
Average diurnal cycles over 4 December 2015–19 January
2016 for
Surface energy balance at WAIS for the period 7 December 2015–16 January 2016.
The deficit in longwave radiation is contributing to the negative
temperature bias. Even though the downwelling shortwave biases are positive
for the AMPS and WSM5C (Table 3), most of the solar flux is reflected by the ice
surface. Thus, the net radiation flux bias is negative,
Figure 6 shows the diurnal cycles of average fields for 2 m temperature,
downwelling longwave radiation, downwelling shortwave radiation, and
upwelling shortwave radiation. The time periods for averaging are 4 December
2015–19 January 2016 for the temperature and 7 December 2015–16 January 2016
for the radiation terms. Simulated biases in these fields vary
with time of day, with local noon near 19:30 UTC. To provide an idea of the
statistical significance of differences in Fig. 6a, we use the Student's
The AMPS mean temperature in the daily cycle is less than the observed value
at all AMPS output times. Only 03:00 UTC is not statistically significant.
The observations have an earlier minimum of
For downwelling shortwave radiation (Fig. 6c), the AMPS has statistically
significant positive biases at all hours, with the bias peaking at 106 W m
The shortwave results are encouraging and suggest that changing the microphysics scheme can greatly alleviate and perhaps even reverse Antarctic radiation biases in numerical simulations. It may appear odd, however, that the upwelling shortwave radiation shows negative biases for all the Polar WRF 3.9.1 simulations that do not coincide with downwelling biases. The difference can be explained by the specified snow albedo in the WRF Noah routine. The specified maximum snow albedo is 0.8 for Noah, and average simulation albedos are slightly below this value. The average observed albedo, however, is 0.843. Therefore, a higher fraction of solar insolation is reflected at the WAIS Divide than in these simulations. This results in a deficit of upwelling shortwave radiation (Table 3, Fig. 6d). The deficit increases the net radiation and contributes to the positive temperature bias for the Morrison, Thompson, and P3. The impact of the albedo can be seen in the slope of the temperature curves after 12:00 UTC in Fig. 6a.
We ran a sensitivity test with segments initialized at 00:00 UTC each day
between 6 and 16 January 2016. The active period for analysis is
12:00 UTC on 6 January until 11:00 UTC on 17 January. The settings were equal
to the WSM5C; however, the albedo over glacial ice was increased to 0.84,
closer to the observed albedo at the WAIS Divide. For the used part of the
segments (hours 12–35), the 2 m temperature average was
The observed downwelling longwave radiation (see Fig. 6b) has a mean value
of 210.6 W m
Figure 7 shows the average diurnal cycle over 7 December–17 January of
longwave and shortwave cloud forcing at the surface for the simulations.
Cloud forcing (CF) is defined following Eq. (1):
Average diurnal cycles over 7 December 2015–16 January
2016 for
Polar WRF 3.9.1 produced clear-sky flux values for longwave and shortwave
radiation, so cloud forcing could be readily calculated. Clear-sky shortwave
fluxes were not available from the AMPS. Figure 7a clearly shows that the
longwave cloud forcing for the AMPS is weak, while the longwave cloud forcing
for WSM5C is less than that of the more recent schemes. The results for the AMPS
and WSM5C are consistent with the negative temperature biases during these
simulations. P3 produces the greatest overall longwave cloud forcing, but
the impact varies somewhat with time of day. Thompson produces nearly as
much longwave cloud forcing as P3. The overall averages are 12.2, 31.9,
37.1, 44.8, and 46.1 W m
Figure 7b shows shortwave cloud forcing which has a cooling effect on the
surface. There are considerable differences between the more recent
microphysics schemes. The overall averages are
The average diurnal cycles of sensible heat flux and the conductive heat
flux into the ice at the WAIS Divide are shown in Fig. 8. The conductive flux
was not directly measured by Nicolas et al. (2017); however, the flux was
estimated from the residual of other terms in the surface energy balance.
The diurnal cycle of sensible heat flux was greatly amplified in the
simulations compared to the observations (Fig. 8a). The positive sensible
heat fluxes into the atmosphere are especially large near the time of maximum temperature, with a maximum of 32.3 W m
Average diurnal cycles over 7 December 2015–16 January
2016 for
The conductive flux into the ice is a critical term for mass balance of West
Antarctica. Therefore, it is important for modeling studies to be able to
well represent this quantity. Positive values are expected during December
and January when insolation is large. The overall average for the residual
estimate of Nicolas et al. (2017) is 7.5 W m
Time series of
While the previous analysis has concentrated on radiation fields and the surface energy balance, we now more directly examine the observed and simulated clouds. Figure 9 shows the LWP for the period 2–18 January 2016. Modeled LWP includes both suspended liquid cloud droplets and falling rain. LWP values above zero are observed at most times, but the AMPS and WSM5C simulate non-zero values only during the period 11–12 January (Fig. 9a). The results demonstrate the known difficulty of the WSM5C microphysics in simulating liquid water for polar clouds (e.g., Listowski and Lachlan-Cope, 2017). The more advanced microphysics schemes simulate liquid water much more frequently than WSM5C but do not well represent the instantaneous observed liquid water (Fig. 9b). Therefore, we suggest that the simulation of liquid water in polar clouds remains problematic (e.g., King et al., 2015; Hines and Bromwich, 2017; Listowski and Lachlan-Cope, 2017).
Table 4 shows the average condensate from 00:00 UTC on 2 January to 00:00 UTC on 18 January. The average observed LWP, 23 g m
Mean hydrometers (g m
Caution should be applied in comparing the distributions of suspended and
precipitation hydrometers between schemes since the definitions of such
categories are arbitrary and poorly defined physically (Morrison and
Milbrandt, 2015). The distribution of hydrometers can be helpful, however,
in understanding the inner workings of a microphysics scheme and comparing
the simulated amounts of liquid and ice. Simulated cloud water tends to be 1 or 2 orders of magnitude larger than rain water. Little ice is simulated
as graupel or rime. Morrison simulates 1 order of magnitude more snow than
cloud ice, while the difference is 2 orders of magnitude for Thompson. In
contrast, the simulations with the WSM5C microphysics produced high amounts
of cloud ice but little amounts of snow. The total ice condensate in the
WSM5C simulation, 21 g m
Time series of cloud fraction for
Figure 10 shows times series of cloud occurrence fraction at the WAIS Divide
during the MWR availability period. Figure 10a shows values determined from
the MPL observations. For the model, however, cloud fraction requires a
definition. One earlier method was widely used and defined clouds
diagnostically. The cloud fraction was determined based upon factors such as
relative humidity, statistic stability, and vertical velocity (Slingo, 1987).
With prognostic cloud schemes, cloud fraction is not necessarily a simple
function of the condensate, and thus we must consider what value is used for the comparison with observations. One formula that has been used for comparison
between model and observations is the cloud fraction formulation of Fogt and
Bromwich (2008) calibrated to manual McMurdo Station cloud fraction observations:
The observed cloud occurrence fraction is frequently 1, and the average is 0.77 during this time (Fig. 10a). Cloud-free times are more common for the AMPS, and thus the average is 0.32 (Fig. 10b). The Polar WRF 3.9.1 simulations show some similarity in their time series of cloud fraction, with the average varying from 0.59 for WSM6C to 0.71 for P3. Microphysics schemes with stronger cloud radiative forcing have larger average total cloud fraction (Figs. 7 and 10).
Liquid cloud occurrence fraction is shown in Fig. 11. Only the first term on the right-hand side of Eq. (2) is used to define modeled liquid cloud fraction. Liquid clouds are frequently observed but are rarely simulated by the AMPS (Fig. 11a). The Morrison scheme simulates liquid clouds much more frequently than WSM5C but not as frequently as the observations. The Thompson and P3 schemes simulate liquid clouds more frequently than the Morrison scheme. Average liquid cloud fractions are 0.65, 0.01, 0.05, 0.20, 0.26, and 0.34 for the observations, AMPS. WSM5C, Morrison, Thompson, and P3, respectively.
Time series of liquid cloud fraction for
Figure 12 shows the vertical distribution of cloud fraction. While the
observed cloud fraction is again determined by surface-based MPL
observations, Eq. (2) is inappropriate for point values of cloud fraction in
a column. We select the mixing ratio 0.001 g kg
Time–height plots of total cloud fraction (color scale)
for
Remote sensing at the WAIS Divide detects clouds that are frequently present
below 650 hPa (Fig. 12a). Detectable clouds can decrease with height due to
attenuation of the lidar pulse at lower altitudes. Thus, it is not
surprising that simulated clouds are appear deeper (Fig. 12b–f).
Furthermore, the minimum threshold of 0.001 g kg
Figure 13 shows liquid cloud occurrence fraction to be more confined to the lower troposphere than total cloud occurrence fraction (Fig. 12). The simulated liquid clouds, when present, are near the surface for the simulations with the WSM5C microphysics (Fig. 13b, c). The more recent microphysics schemes simulate deeper liquid clouds.
Time–height plots of liquid cloud fraction (color scale)
for
Figure 14 shows the mean cloud fraction profiles above sea level (a.s.l.) for the period 2–16 January. As noted earlier, the MPL pulse attenuation likely results in some underestimation of both the total cloud and liquid occurrence fractions at higher elevations. Returning to Fig. 12 that shows shallow clouds with variable vertical structure observed by the MPL, while the simulations have deep, vertically aligned clouds, the means shown in Fig. 14 display this difference in vertical structure. The averaging of frequent deep cloud structures results in high mean values for the simulations, compared to the means of the more variable observations. Therefore, a vertically aligned cloud overlap better represents the simulated clouds than a random overlap. These stacked clouds reduce the modeled cloud fraction shown in Fig. 10, as the middle cloud layer is on of top of the low cloud layer, rather than additive to the cloud fraction. The observed average total cloud fraction peaks at 0.51 at 1985 m a.s.l. (Fig. 14a). The fraction decreases to 0.30 near 2300 m a.s.l. then decreases to 0.10 above 3300 m. The profiles suggest that there could be slightly elevated (liquid-bearing) cloud occurrence at 3935 m a.s.l. The observed liquid cloud fraction is more surface based with a peak of 0.28 at both 1915 and 1985 m a.s.l., and values decreasing to 0.06 at 2210 m (Fig. 14a).
Vertical profiles of average cloud fraction over the period 2–16 January 2016 for
The simulated cloud fraction profile peaks near the surface for the AMPS and WSM5C (Fig. 14b). For the AMPS (WSM5C), the maximum is 0.50 (0.64) at 8 m (84 m) above the surface. The cloud fraction is higher for the recent microphysics schemes, with all having maxima above 0.64 at heights below 2400 m a.s.l. The largest cloud fraction is 0.69 at 2165 m for Thompson.
The mean simulated liquid and ice cloud fractions are shown in Fig. 15. The values are from 2 to 16 January 2016, the same period used for the profiles in Fig. 5. Similar to the profile displayed in Fig. 14, the observations show a more shallow peak in the lower troposphere than in the simulations. (Fig. 15a). The fractions are based upon the total liquid or ice content. P3 has a unique liquid profile that peaks at 0.26 at 2376 m a.s.l.. The Morrison and Thompson simulations have similar liquid cloud fraction profiles with double peaks between 2160 and 2500 m. Figure 15b shows that ice is frequently present in the lowest 1000 m above the surface for the simulations. All the simulations show maxima for ice in the lowest 500 m above the surface, varying from 0.50 at 8 m for the AMPS and 0.49 for P3 at 365 m to 0.64 for 85 m with WSM5C. The Polar WRF 3.9.1 simulations produce more ice cloud fractions than the AMPS.
Vertical profiles of average cloud fraction over 2–16 January 2016 for
A sensitivity test referred to as P3-50, was based upon P3 to see if the
setting of 400 cm
Times series of
The recent 2015–2017 AWARE field program provides a highly detailed set of remote-sensing and surface observations that can be used to study the simulation Antarctic clouds and the surface energy budget. We focus on the December 2015–January 2016 test period when observations were taken at the WAIS Divide. These observations are used for comparison with the AMPS forecasting system and new simulations with Polar WRF 3.9.1. The AMPS uses the WRF single-moment 5-class microphysics, while the new Polar WRF 3.9.1 simulations are run with WSM5C and three more recent microphysics schemes. These are the Morrison 2-moment microphysics, the Thompson–Eidhammer aerosol-aware microphysics, and the new Morrison–Milbrandt P3 microphysics.
The AMPS simulates few liquid hydrometers during austral summer at the WAIS Divide, even though liquid clouds are frequently observed by the MPL, primarily a consequence of the WSM5C microphysics in the AMPS. Consequently, downwelling shortwave radiation is excessive at the surface, while downwelling longwave radiation is too small. The WSM5C simulation with Polar WRF 3.9.1 has reduced biases of the same sign. The decreased magnitude in WSM5C appears due to GFS forcing of initial and boundary conditions for the AMPS while ERA-I is used for WSM5C. Simulated hydrometers are overwhelmingly composed of ice with WSM5C.
The more advanced microphysics schemes show considerable improvement in the simulation of overall cloud fraction, liquid hydrometers, and cloud radiative effects. The instantaneous simulation of liquid remains somewhat problematic even given the improvements. The Morrison scheme simulates less LWP and weaker cloud radiative forcing than the Thompson and P3 schemes. P3 simulates the greatest LWP and cloud radiative effect. All schemes appear to underestimate total cloud fraction and liquid cloud fraction at the WAIS Divide. The vertical distribution of simulated cloud properties differs from observed profiles, with deeper clouds simulated than observed, although the MPL may not detect the upper regions of clouds due to attenuation.
In the near future, the more extensive AWARE cloud observations at McMurdo Station over the full seasonal cycle will provide a basis for sensitivity tests designed to seek Antarctic optimizations to the advanced microphysics schemes used for the WAIS Divide. In particular, we plan to work with two more advanced implementations of the P3 microphysics (Milbrandt and Morrison, 2016). Sensitivity tests will also vary the background IN concentrations in simulations with the Thompson microphysics, as the limited observational evidence suggests that the contributing aerosol concentrations may vary or are unknown over a range of orders of magnitude.
The standard release of WRF can be downloaded from NCAR Mesoscale and Microscale Meteorology (2019). The polar optimizations can be requested from Polar Meteorology Group (2019).
All the observations from the AWARE field campaign (including the
reprocessed MPL dataset) can be downloaded from the ARM Data Discovery
website (
KMH was the primary author and coordinated with the other authors. KMH conducted the Polar WRF simulations, downloaded and processed AWARE data, and processed the model data.
DHB coordinated the Ohio State component of AWARE and was the primary co-author. DHB read and provided input on all drafts of the manuscript and helped plan the simulations.
SHW processed the AMPS data for the manuscript.
IS processed the MPL and cloud mask data. IS provided advice on the use of these data.
JV coordinated the Penn State component of AWARE. JV provided advice on Antarctic clouds and the AWARE data.
DL was the overall coordinator of AWARE. DL provided advice on Antarctic clouds and the AWARE project and helped to coordinate the use of CERES data.
The authors declare that they have no conflict of interest.
Any opinions presented here are those of the paper authors alone and are not necessarily those of
Numerical simulations were performed on the Intel Xeon cluster at
the Ohio Supercomputer Center, which is supported by the state of Ohio. We
thank Julien Nicolas for providing rawinsonde, surface energy balance and
LWP observations for the WAIS Divide and thank Ryan Scott for providing cloud forcing
derived from CERES. All the observations from the AWARE field campaign
(including the reprocessed MPL dataset) can be downloaded from the ARM Data
Discovery website (
This research has been supported by the U.S. Department of Energy, Office of Science (DOE (grant no. DE-SC0017981)) and National Science Foundation (NSF (grant no. PLR 1443443)).
This paper was edited by Timothy J. Dunkerton and reviewed by two anonymous referees.