Climate Dynamics

, 34:61

Validation of a limited area model over Dome C, Antarctic Plateau, during winter

Authors

    • Laboratoire de Glaciologie et de Géophysique de l’Environnement, CNRS
  • Irina V. Gorodetskaya
    • Laboratoire de Glaciologie et de Géophysique de l’Environnement, CNRS
Article

DOI: 10.1007/s00382-008-0499-y

Cite this article as:
Gallée, H. & Gorodetskaya, I.V. Clim Dyn (2010) 34: 61. doi:10.1007/s00382-008-0499-y

Abstract

The limited area model MAR (Modèle Atmosphérique Régional) is validated over the Antarctic Plateau for the period 2004–2006, focussing on Dome C during the cold season. MAR simulations are made by initializing the model once and by forcing it through its lateral and top boundaries by the ECMWF operational analyses. Model outputs compare favourably with observations from automatic weather station (AWS), radiometers and atmospheric soundings. MAR is able to simulate the succession of cold and warm events which occur at Dome C during winter. Larger longwave downwelling fluxes (LWD) are responsible for higher surface air temperatures and weaker surface inversions during winter. Warm events are better simulated when the small Antarctic precipitating snow particles are taken into account in radiative transfer computations. MAR stratosphere cools during the cold season, with the coldest temperatures occurring in conjunction with warm events at the surface. The decrease of saturation specific humidity associated with these coldest temperatures is responsible for the formation of polar stratospheric clouds (PSCs) especially in August-September. PSCs then contribute to the surface warming by increasing the surface downwelling longwave flux.

Keywords

Antarctic PlateauDome CRegional climate modelCloud radiative properties

1 Introduction

The aim of this paper is to validate the Modèle Atmosphérique Régional (MAR) over the East Antarctic Plateau during winter. A comparison of MAR outputs will be made with meteorological observations from Dome C, a new permanent station over a dome of East Antarctica. Detailed meteorological observations made at Dome C have already been reported (Argentini et al. 2005). Several reasons may explain the need of such observations. Among them one may cite (1) the need to better know the very stable boundary layer and (2) the need to infer optical properties of the atmosphere at Dome C, since that site is a good candidate for the development of astronomical observations (Aristidi et al. 2005).

Winter is the season when meteorological conditions are the most extreme over the East Antarctic Plateau. The simulation of such situations may contain errors affecting model climatology and the subsequent interpretation of the link between climate over the East Antarctic Plateau and global climate. At least the parameterizations of two processes are still debated: turbulence under very stable conditions and cloud formation (Parish and Bromwich 2002). First, the low troposphere over the East Antarctic Plateau is extremely stable during winter, with a time mean surface temperature inversion as large as 25°C (Connolley 1996). A good estimation of the surface inversion is important when retrieving climate information from ice cores (Masson-Delmotte et al. 2008; Fujita and Abe 2006; Helsen et al. 2005). Second, clouds may be very thin over Antarctica. Antarctic meteorological models are sensitive to the parameterizations of cloud microphysical processes and in particular to their impact on the radiative transfer through the atmosphere (Lubin et al. 1998; Guo et al. 2003; Hines et al. 2004). Since the cloud cover may affect the surface energy balance by modulating the longwave downward radiation (LWD), it is important to know the model sensitivity to the representation of cloud radiative properties. Finally properties of precipitating snow are important when retrieving the climatic signal from the ice cores.

Since Dome C is a permanent station situated on the East Antarctic Plateau one of the goals of this paper is to identify a possible need of new observations and improvements of model parameterizations. Systematic errors in the model outputs will be analyzed and possible shortcomings in the parameterizations will be identified. Since the parameterizations used in this study are not definitive the results discussed here must be considered as preliminary.

The rest of the paper is divided in 5 parts: a short description of already available observations is given in the second part. Then the model is described with a particular attention to the parameterizations of clouds. The period 2004–2006 is simulated in part 4. Indeed more observations were made at Dome C since 2004: AWS observations (temperature, wind speed and direction), surface radiative fluxes, atmospheric soundings. Model sensitivity to the parameterization of cloud radiative properties is determined in part 5. The last part is reserved for a discussion and some conclusions.

2 Observations

Dome C is a site on the East Antarctic Plateau where a permanent station allows performing detailed meteorological observations during the Antarctic winter. The first wintering occurred in 2005. Observations from AWS (Automatic Weather Station) Dome C (Stearns and Weidner 1990) are used here with observations from surface radiometers (shortwave and longwave, G. Lanconelli 2007, personal communication) and aerological soundings made by the IPEV-PNRA (Institut Paul Emile Victor—Programma Nazionale Ricerche in Antartide) ’Concordia’ Cooperative Programme (Tomasi et al. 2006). Temperature and wind data are available for the entire period (2004–2006) while radiation measurements—during 2006 and soundings—during 2005. The AWS is located at 75.12°S, 123.37°E (3250 m above sea level) and measures air temperature and wind speed at 3 m above ground level (a.g.l.) (Aristidi et al. 2005). Instantaneous values at 10 min frequency are provided by the Antarctic Meteorological Research Center in Wisconsin (http://uwamrc.ssec.wisc.edu). For comparison with the model the data were subsampled at 6-h frequency, which has filtered out most of the extreme value outliers. Soundings were not assimilated in the European Centre for medium-range weather forecast (ECMWF) operational analyses in 2005 (A. Pellegrini personal communication).

3 MAR

Modèle atmosphérique régional (MAR) has been developed for Polar Regions. It is a primitive equations hydrostatic model. Atmospheric dynamics is fully described in Gallée and Schayes (1994). The turbulence scheme is based on the E-ε model of Bintanja (2000) and on the Monin-Obukhov similarity theory (MOS) respectively outside and inside the lowest model layer, assumed to be the surface boundary layer (SBL). A dependence of the Prandtl number on the Richardson number has been included in order to take into account the less efficient turbulent transport of heat under very stable conditions (Sukoriansky et al. 2005). The first model level is assumed to be the SBL and is situated roughly 8 m a.g.l. at Dome C. The radiation transfer scheme is that of Morcrette et al. (2001) and includes the RRTM longwave radiation transfer model of Mlawer et al. (1997). It is also used by ECMWF in European Re-Analyses ERA-40 (Uppala et al. 2005). The cloud microphysical scheme is described in Gallée (1995) and is improved in Gallée et al. (2001, 2005). It includes 6 prognostic equations for specific humidity, cloud droplet concentration, cloud ice crystals (concentration and number), concentration of precipitating snow particles and rain drops. Ice microphysical processes are included based on the work of Lin et al. (1983). The Fletcher (1962) equation for ice nuclei concentration is replaced with the more realistic parameterization of Meyers et al. (1992). The conversion from cloud ice crystals to precipitating snow and the prognostic equation for the ice crystal number are based on Levkov et al. (1992). Cloud radiative properties are computed from the concentration of cloud droplets and cloud ice crystals qi (Ebert and Curry 1992). Here the concentration of snow particles q* is also included in the computation of cloud radiative properties in order to take into account their small effective radius re* in the Antarctic interior (Walden et al. 2003; Ellison et al. 2006). Since re* is roughly 3 times larger than the effective radius rei of Antarctic ice crystals, it is assumed that the ice concentration used to force the radiative transfer scheme is \(q_{i}+{\frac{q_{*}}{3}}\) rather than qi. The inclusion of snow particles into the cloud ice crystal concentration scaled using the particle radius ratio contributes similarly to the cloud emissivity of thin clouds, which are typical over Antarctic plateau (ice water path not exceeding 1 g m−2). This modification is discussed in Sect. 5. MAR is coupled with a snow model and an interactive blowing snow model (Gallée et al. 2001). Snow particles may be eroded from the snow pack by the wind. The erosion threshold of snow particles depends on their dendricity, sphericity and size. The strong negative feedback of snow erosion by the wind on atmospheric turbulence is taken into account (Gallée et al. 2001). Blowing snow particles are included in the precipitating snow particle concentration.

4 Control experiment

4.1 Model Set-up

MAR is set-up over the whole Antarctic ice sheet for the period 2004–2006. The horizontal resolution is 80 km and is probably sufficient for simulating Dome C climate. Indeed the area around Dome C is flat and the terrain is homogeneous. Model sentivity to a change of the horizontal resolution from 80 to 20 km is not significant at Dome C. MAR simulation is made by initializing the model once on the 1st of January 2004 and by forcing it through its lateral boundaries by the ECMWF operational analyses (Marbaix et al. 2003). Note that Dome C is situated roughly 2,000 km away from the closest lateral boundary (Fig. 1). A Newtonian relaxation is also included for wind and temperature in the upper sponge of the model (sigma levels σ = 0.0157, 0.0283, 0.0437, 0.0618, 0.0825, 0.1055 corresponding to z = 29,558, 25,756, 23,000, 20,826, 19,034 and 17,512 m at Dome C). The Newtonian relaxation factor decreases downwards from the model top.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0499-y/MediaObjects/382_2008_499_Fig1_HTML.gif
Fig. 1

Antarctic topography in MAR for a grid spacing of 80 km. C refers to Dome C (75°S, 123°E)

MAR is first validated by comparing model outputs with available observations. The behaviour of some relevant variables (temperature, wind, LWD, specific humidity, concentration of cloud particles–ice crystals, droplets–and snow) is then discussed. In the last two sections we discuss validation of MAR meteorological variables, its capability to simulate the alternation of warm and cold regimes, and sensitivity to representation of cloud radiative properties.

4.2 Validation

MAR temperature at Dome C is compared with AWS observations on Fig. 2 and Table 1. Model temperature is interpolated from lowest model level (8 m) to AWS level (3 m) by using MOS theory. Some agreement is found between observation and simulation both for the timing and amplitude of temperature variations. Synoptic scale events are slightly better simulated than shorter time scale events. The correlation between simulated and observed temperatures is improved up to 0.66, 0.34 and 0.49 for 2004, 2005 and 2006 respectively when applying a 5-day running mean to the data. MAR simulates the warm events over an atmospheric layer which is much thicker than the SBL. Figure 3 contains a Hovmöller diagram of the temperature vertical profile in the lowest 500 m (50 dam) during winter 2006. The boundary layer is generally thinner than 30 m (not shown) while the atmospheric layer affected by the warming is much deeper. A look at levels higher than 500 m indicates that the warmings occur over a much deeper part of the atmosphere (not shown). This suggests a synoptic scale influence. Note also the high variability in time of the temperature inversion strength at Dome C (blue line), with a weaker inversion strength and a higher wind speed (dark line) when the low troposphere warms.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0499-y/MediaObjects/382_2008_499_Fig2_HTML.gif
Fig. 2

Comparison between simulated (solid line) and observed (dashed line) temperature 3 m above the surface at Dome C during winter for 2004 (upper), 2005 (middle) and 2006 (lower panel), sampled at 6-h time step

Table 1

Time average of observations (OBS), bias, correlation and RMSE of MAR temperature and wind speed at Dome C

Temperature (K)

OBS mean

MAR bias

MAR RMSE

Corr.

Wind (m/s)

OBS mean

MAR bias

MAR RMSE

Corr.

2004

212.1

+0.1

8.6

0.62

 

3.17

0.64

2.8

0.37

2005

214.0

−0.6

10.5

0.33

 

3.60

0.65

2.8

0.35

2006

209.6

−0.7

8.5

0.43

 

3.05

0.87

3.6

0.38

All correlations are significant at 95% level

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0499-y/MediaObjects/382_2008_499_Fig3_HTML.gif
Fig. 3

Temperature profile in the lowest 50 dam (1 dam = 10 m) at Dome C during winter 2006. Wind speed averaged over that layer (dark line, in m/s) and surface temperature inversion strength (blue line, in K) are also plotted. Surface inversion strength is the temperature difference between the surface and the warmest level below 13 km

The good representation by MAR of the warming 3-m above the surface in conjunction with the simulation of a warming over a deep atmospheric column suggests that the model is able to capture correctly the synoptic scale meteorological forcing through its boundaries. In particular the Newtonian relaxation towards the ECMWF analyses in the upper sponge of the model could be able to better capture the timing of the synoptic forcing. A sensitivity simulation was also made using 60 levels, with an upper relaxation zone including the layers located at 55,200, 42,770, 37,770, 34,550, 32,100 and 30,085 m at Dome C. No significant sensitivity was found to that modification, indicating coherence between the numerical solution of MAR and the forcing data.

Looking at the cold events it is found that the amplitude of the temperature minima is well simulated. Since cold events occur for clear sky situations it may be deduced that the surface energy budget is well simulated in this case. The surface energy budget during polar night is forced by LWD fluxes, sensible heat fluxes and heat accumulation in the snow pack. It will be shown later that the LWD fluxes are no more underestimated in MAR, in contrast with that was obtained when using in MAR the Morcrette (1984) parameterization of the radiation transfer (not shown).

MAR wind speed at Dome C is compared with AWS observations on Fig. 4 and Table 1. The simulated wind speed is interpolated to AWS level (3 m) by using MOS theory. As for temperature some agreement is found between observation and simulation both for the timing and amplitude of wind speed variations. Again synoptic scale events are slightly better simulated than shorter time scale events. The correlation is now improved up to 0.54, 0.42 and 0.49 for 2004, 2005 and 2006 respectively when applying a 5-day running mean to the data. However the time average of the wind speed is overestimated (Table 1). It is possible that this overestimation is due to a slight destabilisation of the boundary layer through the long-term overestimation of the LWD. Note that higher wind speeds are generally observed/simulated in conjunction with higher observed/simulated temperatures.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0499-y/MediaObjects/382_2008_499_Fig4_HTML.gif
Fig. 4

Comparison between simulated (solid line) and observed (dashed line) wind speed 3 m above the surface at Dome C during winter for 2004 (upper), 2005 (middle) and 2006 (lower panel), sampled at 6-h time step

Simulation of vertical profiles in temperature and humidity is crucial for longwave radiative fluxes and thus for correct representation of the warm and cold regimes at Dome C. Vertical profiles of the simulated bias of temperature, specific humidity, wind speed and direction are shown for MAR and ECMWF operational analyses in Fig. 5, for winter 2005. The biases were computed relative to the vertical profiles obtained from the radiosonde measurements at Dome C station (Aristidi et al. 2005; Tomasi et al. 2006). The winter profiles are averaged during June–August. This period has been chosen since the soundings made at Dome C were not assimilitated in ECMWF model. The agreement is generally good beyond 100 hPa above the surface. The wind speed bias is smaller in MAR than in ECMWF operational analyses. MAR temperature bias is significant in the first 100 hPa above the surface. It may be partly explained by a temperature difference between the sounder and the air outside when the sounder is launched (Mahesh et al. 1997). Also note a possible dry bias in the Vaisala RS90 radiosonde humidity profiles over Antarctica (Rowe et al. 2008). Nevertheless MAR is too warm and too moist in particular in the layer below 600 hPa. An analysis of individual events shows that MAR sometimes simulates a warm contribution to the bias while sometimes it simulates a cold contribution. Warm contributions to the bias are the largest above 20 m a.g.l. Vertical extent of the cold contributions to the bias is smaller. This is why a negative bias is simulated near the surface while a positive bias is simulated between 642 and 600 hPa. Each individual contribution to the bias is generally simulated over a layer extending well above 600 hPa. It is possible that because the timing of cold and warm events is not fully simulated by the model a small delay during a large temperature variation may generate a strong impact. Finally it may be argued that the impact of each individual contribution is maximum at the surface through LWD heating. The strong response of the surface rapidly weakens upwards due to the weakening of turbulent processes under stable conditions. Integrating the LWD over all individual events it is found that the contribution of the warm/moist bias situated below 600 hPa is masked by the contribution of the cold bias occurring in the two lowest atmospheric layer of the model.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0499-y/MediaObjects/382_2008_499_Fig5_HTML.gif
Fig. 5

Vertical profiles of simulated bias at Dome C of (a) temperature (K), (b) specific humidity (g/kg), (c) wind speed (m/s) and (d) wind direction (°) in MAR (red) and ECMWF operational analyses (green) compared to radiosonde measurements

The aim is now to analyze the mechanisms responsible for warming events at Dome C. Observed and simulated LWD are compared on Fig. 6. May–September 2006 averages amount respectively to 84 and 61 W/m2 in MAR and in the observations. MAR overestimates LWD but simulates rather fairly well its variability in time and in amplitude. The overestimation of LWD could be partly explained by the overestimation of the simulated precipitable water vapour (PWV) and temperature in the layer between the surface and 300 hPa (see Fig. 5, panels (a) and (b)). Due to the inclusion of snow particles into the radiative scheme, some overestimation of LWD also occurs during some relatively large snowfall events simulated by MAR when observations show small LWD values, as for example in mid June.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0499-y/MediaObjects/382_2008_499_Fig6_HTML.gif
Fig. 6

Comparison between simulated (solid line) and observed (dashed line) surface downward longwave radiation at Dome C, for winter 2006. Bias and RMSE amount respectively to 13.9 and 27.4 W/m2. Correlation is 0.40 (significant at 95% level)

The simulation of LWD and near surface temperature is compared on Fig. 7. The strong correlation (0.88) is explained by the determining influence of LWD on the surface energy budget at Dome C during winter. The sensible heat flux (SH) is much weaker than LWD (May–September 2006 average amounting to 16 and 84 W/m2 respectively). SH is better correlated with the wind speed (correlation amounting to 0.86) than with the near surface temperature (correlation amounting to 0.34). Note that SH could be overestimated at Dome C by the model since the wind speed is overestimated.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0499-y/MediaObjects/382_2008_499_Fig7_HTML.gif
Fig. 7

Comparison between simulated air temperature (solid line) for the lowest model level (roughly 8 m a.g.l.) and surface longwave downward radiation (dashed line) at Dome C for winter 2006

As the variability in time of LWD is mainly due to that of water vapour and cloudiness, a similar comparison is made between LWD and (1) PWV (Fig. 8) and (2) cloud optical thickness (Fig. 9). Again the correlation with LWD is significant, amounting to 0.81 for PWV (Fig. 8) and to 0.77 for cloud optical thickness (Fig. 9). Note that cloudiness depends on PWV, explaining partly the coherence between both correlations. Cloudiness may also depend on temperature in the stratosphere (see the following paragraph).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0499-y/MediaObjects/382_2008_499_Fig8_HTML.gif
Fig. 8

Comparison between simulated precipitable water vapour (PWV, solid line) and surface downward longwave radiation (dashed line) at Dome C for winter 2006

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0499-y/MediaObjects/382_2008_499_Fig9_HTML.gif
Fig. 9

Comparison between simulated cloud optical thickness (solid line) and surface longwave downward radiation (dashed line) at Dome C for winter 2006

The vertical distribution of cloud particles (ice crystals and droplets) and snow particles over Dome C for 2006 is shown on Fig. 10. In MAR, during the entire year cloud droplet concentration dominates over the cloud ice crystals in the troposphere over Dome C. Liquid layers in low tropospheric clouds have been occasionally observed at Dome C (M. Del Guasta 2008, personal communication). Note that such clouds have been also observed over South Pole during summer (Walden et al. 2005). Although MAR probably overestimates the frequency of liquid occurrence in tropospheric clouds, routine measurements of cloud properties are needed to make any conclusions. From June until October very high (stratospheric) clouds sometimes form with ice crystals. This is partly due to the important radiational cooling of the stratosphere during polar night which decreases saturation specific humidity. Short term variations of stratospheric clouds concentration are correlated with the short term variations of the temperature rather than by those of specific humidity (not shown). These short term stratospheric cooling occur in conjunction with warm events in the lower part of the atmosphere. Since MAR does not include chemical processes, stratospheric clouds in MAR only form with water vapour and could be classified as Polar stratospheric clouds of type II (PSC II). Snow flakes concentration may increase downwards, in agreement with the observed fact that PSCs may occur in conjunction with clouds in the troposphere (Spinhirne et al. 2005; Wang et al. 2008). Note that snow particles generated in MAR PSCs may fall down to the surface, but their contribution to the surface mass balance (SMB) is small (see panel (b) of Fig. 10). The most important part of the SMB at Dome C is rather due to a few snow fall events occurring the 2 or 3 first kilometres above ground level. Note that 2006 SMB simulated by MAR is 22 mm and could be underestimated. Indeed long term SMB at Dome C is roughly 30 mm w.e. (Urbini et al. 2008).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0499-y/MediaObjects/382_2008_499_Fig10_HTML.gif
Fig. 10

Vertical distribution of simulated hydrometeors at Dome C for 2006: (a) cloud ice crystals and droplets, (b) snow particles. Snow accumulation is also plotted (cyan line)

The comparison between panels (a) and (b) of Fig. 10 reveals that snow particles occupy a deeper atmospheric column than cloud ice crystals. The small size of snow particles during the Antarctic winter observed by Walden et al. (2003) suggests that their contribution to the radiative properties of the atmosphere is not negligible. In order to infer the respective influence of cloud particles and snow flakes on cloud radiative properties sensitivity experiments are discussed in the following section.

5 Model sensitivity to cloud radiative properties

Sensitivity of LWD to the presence of hydrometeors in the atmosphere is assessed first by performing off-line simulations of the longwave radiative transfer, using data from the control experiment. For example LWD standard deviations from time average amount respectively to 29.7 or 10.1 W/m2 for May–September 2005 (22.1 or 10.3 W/m2 for May–September 2006) when hydrometeors are included in or removed from the radiative computations.

The contribution of ice crystals to LWD is not significant except when a large amount of PSCs form (e.g., in July–September 2006, see Fig. 10). In that case the contribution to LWD of ice crystals is roughly similar to that of snow particles. The impact of PSCs on LWD has been assessed by removing from radiative computations hydrometeors situated above 10 km altitude. The LWD becomes smaller during winter in this case, with a difference reaching 6.3 W/m2 for the August-September 2006 average (14.9 W/m2 for August–September 2005). The difference increases during the winter indicating an increasing influence of PSCs on SBL temperature variability when stratosphere cools.

In order to take into account the relatively tiny size of Antarctic snow particles their concentration is taken into account in the computation of cloud radiative properties (see Sect. 3). Here that assumption is tested by performing a simulation in which that contribution has been switched off. Simulated temperatures for both the control and the sensitivity experiments are compared to those of the standard in Fig. 11.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0499-y/MediaObjects/382_2008_499_Fig11_HTML.gif
Fig. 11

Comparison between observed (thin dash) and simulated temperature 3 m above the surface at Dome C. Solid line: control experiment; thick dashed line: sensitivity experiment (no contribution of the snow particles to the cloud radiative properties)

Surface temperatures are underestimated in the sensitivity experiment. May–September average is 205.2 K while it is 209.5 K in the observations and 208.9 K in the control (Table 2). Note in particular that the amplitude of the simulated temperature is underestimated during warm events in the sensitivity test. The same behaviour is found in the simulated LWD, see Fig. 12). Similar underestimations are found for winter 2004 and 2005 (Table 2). In contrast a smaller sensitivity is found in the time average of the simulated wind speed, although the slightly larger simulated wind speed in the control simulation was expected. Indeed the larger influence of clouds on the radiative transfer in the control is responsible for a larger LWD, a larger heating of the surface, a less strong vertical stability of the low atmosphere, more turbulence and a subsequent stronger transfer of momentum downwards in the SBL. Thus the sensitivity test shows that inclusion of snow particles in the computation of cloud radiative properties improves the simulation of temperature during the warm events. On the other hand the simulation of the wind speed is degraded due to the long-term overestimation of the LWD.
Table 2

May - september averages of temperatures at Dome C at AWS - level. OBS refers to Dome C II AWS observations and CRP to “Cloud Radiative Properties”

Temperature

2004 (K)

2005 (K)

2006 (K)

Wind speed

2004 (m/s)

2005 (m/s)

2006 (m/s)

OBS

212.1

214.0

209.6

 

3.17

3.60

3.05

\(CRP(q_{i}+{\frac{q_{*}}{3}})\)

212.2

213.4

208.9

 

3.81

4.25

3.92

CRP(qi)

206.4

207.5

205.2

 

3.48

3.84

3.43

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0499-y/MediaObjects/382_2008_499_Fig12_HTML.gif
Fig. 12

Comparison between simulated (solid line: control experiment; thick dashed line: sensitivity experiment in which snow particles do not contribute to cloud radiative properties) and observed (thin dashed line) surface downward longwave radiation at Dome C

6 Discussion and conclusions

MAR has been validated over Dome C, the new permanent station situated over the East Antarctic Plateau. The model behaves fairly well at the AWS level. In particular it simulates accurately the amplitude and to some extent the timing of cold and warm events in the troposphere during winter. Wind speed is larger during warm tropospheric events, as in the observations. Temperature and wind speed variations associated with these events are also simulated over a tropospheric column much thicker than the SBL. The main forcing of the surface energy budget at Dome C during winter is the LWD. Precipitable water vapour, cloudiness and precipitating snow particles increase during warm events. Polar Stratospheric Clouds (PSCs) are also simulated during warm events and may influence LWD significantly.

The short term variations of PSCs are due to additional stratospheric cooling associated with the troposphere warming. MAR PSCs are responsible for the precipitation of snow particles over a large thickness of the atmosphere. These snow particles may reach the surface but their contribution to the surface mass balance at Dome C is small. Nevertheless an overestimation of the vertical extension of PSCs is possible and could be due to an overestimation of the snow sedimentation velocity. Indeed snow sedimentation velocity is parameterized in the model from the characteristics of heavier mid-latitude snow particles. MAR vertical extension of PSCs should be validated in the future. Observed Antarctic snow particles are small during winter and their observed effective radius is larger than that of ice crystals by roughly a factor 3 only. This is why their contribution to cloud radiative properties has been taken into account in MAR. Model sensitivity to that contribution is important. In particular the simulation of warm events is significantly degraded when switching it off. As chemical processes are not represented in MAR, only water PSCs are allowed to form, so that the PSCs concentration could be underestimated. Thus the possibility exists that the inclusion of snow particles in the computation of cloud radiative properties could partly compensate an underestimation of PSCs concentration. On the other hand it could be argued that overestimation of the snow sedimentation rate depletes hydrometeors originating from the low troposphere but does not deplete those originating from the stratosphere. A subsequent underestimation of optical thickness and emissivity should result for low clouds, but not for PSCs. Finally the overestimation of PWV below 300 hPa during warm events could suggest that the formation rate of tropospheric clouds is underestimated by the model. A detailed comparison between observation and simulation of (low) tropospheric clouds will help in clarifying that point.

MAR is too warm between 640 and 600 hPa and too cold below 640 hPa. This temperature bias is the sum of warm and cold contributions, probably because the model does not fully reproduce the timing of cold and warm events and the associated strong variations of temperature and humidity. Positive contributions to the bias are the strongest in the bulk of the boundary layer and dominate there for the period which was considered (June–August 2005). In contrast the negative contribution to the bias is the strongest just near the surface.

A consequence of the strong variability in time of atmospheric conditions at Dome C during winter is the strong variability of the simulated temperature inversion, which ranges from 0°C up to 40°C. More precisely variations of the inversion strength TaTs, with Ta the inversion temperature and Ts the surface temperature, are negatively correlated with those of the atmospheric temperatures. The time average of TaTs(21.7°C), is not representative of any typical situation, which makes difficult the interpretation of the link between the ice core record and the surface temperature, at least at short climatic time scales. Such behaviour was also found for drilling site Kohnen in Queen Maud Land (Helsen et al. 2005). It is noted in particular that precipitation events occur mainly when the temperature inversion is less strong than its time average. This may be explained by an increase of surface warming by LWD due to the occurrence of clouds.

This paper has shown the high sensitivity of the Antarctic climate to cloud radiative properties. This high sensitivity could be partly explained by the fact that Antarctic clouds are rather thin, so that taking into account an additional small amount of hydrometeors has a large effect on cloud optical thickness and emissivity. Of course observations of hydrometeors above the Antarctic plateau (Walden et al. 2003, 2005; Ellison et al. 2006) are scarce and new observations are needed to base more firmly our assumption. Accurate representation of cloud microphysics is essential for simulating surface radiative budget and temperature variability on synoptic time scale over the Antarctic Plateau.

Acknowledgments

We thank Steve Colwell for the quality check of the AWS data. I. Gorodetskaya was supported by Agence Nationale de la Recherche (France) grant OTP 232 333. Computation were realised with IDRIS computing resources. The IPEV-PNRA ‘Concordia’ Cooperative Programme—Routine Meteorological Observations is acknowledged for providing atmospheric soundings data. French LEFE/IDAO project Charmant is acknowledged for providing support for publishing the present paper.

Copyright information

© Springer-Verlag 2008