Interactive comment on “ Updated cloud physics improve the modelled near surface climate of Antarctica of a regional atmospheric climate model ”

The paper describes the effect of the implementation of a new physics package on the performance of the RACMO regional model in Antarctica, with a focus on the surface climate. This physics update results in increased moisture and clouds over the Antarctic continent, which has a positive impact on the surface radiation budget and surface temperature but little effect on the surface wind field. While not particularly original in its design, the study certainly fits well within the scope of The Cryosphere. It is in my view worth publishing because 1) it sheds further light on the skill of a regional model used in many prominent Antarctic studies; and 2) global models and even regional models still often struggle to properly reproduce some fundamental aspects of Antarctic climate. C1243


Introduction
Regional atmospheric climate models (RCMs) are important tools to improve our understanding of atmospheric processes and their relation to climate change.They provide a physically coherent representation of the climate in areas with a low spatial and temporal coverage of observations.RCMs are also capable of resolving detailed features that are not captured by global circulation models (GCMs).Specifically, RCMs have been successfully applied to remote areas such as Antarctica (e.g.Van Lipzig et al., 2002) and Greenland (Fettweis, 2007;Ettema et al., 2010a) to assess the climate and surface mass balance of the ice sheet (e.g.Van de Berg et al., 2005;Lenaerts et al., 2012).
Moreover, the RCM output can be used to enhance the interpretation of remote sensing data such as GRACE (Chen et al., 2006) and InSAR (Rignot et al., 2008).In combination with a firn densification model, RCM output provides a correction for firn densification in support of radar/laser altimetry (Ligtenberg et al., 2012).All techniques combined have recently provided a synthesis of mass balance estimates for the Greenland and Antarctic ice sheets (Shepherd et al., 2012).
The Regional Atmospheric Climate Model RACMO2, which has been adapted for specific use over the polar regions, has recently undergone a major update of its physics package.In the present study we show whether this update from version RACMO2.1 to RACMO2.3 has improved the representation of the Antarctic climate, with its extreme temperatures and winds.Even though RACMO2.1 has proved to realistically simulate the Antarctic near-surface climate ( Van de Berg et al., 2005;Lenaerts et al., 2012), previous model evaluations showed that downward longwave radiation is generally underestimated by the model (Van de Berg et al., 2007) resulting in a significant cold surface bias ( Van den Broeke, 2008).To see whether this has improved in RACMO2.3we will assess the changes for Antarctica in the modelled surface energy balance (SEB), near-surface wind speeds and surface temperatures and compare these to available observations.Section 2 discusses the model, the changes in model formulation and the observational data used for evaluation.In Sect. 3 the effects of the model changes on clouds, the SEB and the near-surface temperature and wind are presented and a comparison is made to observational data (Sect.3.3 -3.5), followed by conclusions in Sect. 4.

RACMO2 physics update
RACMO2 combines the dynamical processes of the High Resolution Limited Area Model (HIRLAM) (Undén et al., 2002) with the physics package of the European Centre for Medium-range Weather Forecasts (ECMWF) Integrated Forecast System (IFS).RACMO2 has been specifically adapted for use over the large ice sheets of Greenland and Antarctica (e.g.Reijmer et al., 2005).It is interactively coupled to a multilayer snow model that calculates melt, percolation, refreezing and runoff of meltwater (Ettema et al., 2010b;Greuell and Konzelmann, 1994).Surface albedo is based on a prognostic scheme for snow grain size (Kuipers Munneke et al., 2011) and a drifting snow routine simulates the interactions of drifting snow with the surface and the lower atmosphere (Lenaerts et al., 2012).A horizontal resolution of ∼27 km and a vertical resolution of 40 levels is used.The model is forced by ERA-Interim re-analysis data (January 1979-December 2011, Dee et al., 2011) at the ocean and lateral boundaries, while the domain interior is allowed to evolve freely.
Here we analyse changes in the modelled Antarctic near-surface climate after the ECMWF IFS physics package cycle CY23r4 in RACMO2.1 (White, 2001) has been updated to cycle CY33r1 in RACMO2.3 (ECWMF-IFS, 2008).The updates that have the most impact on Antarctic applications are the changes in the cloud scheme, the cloud microphysics and the radiation and turbulence schemes.All changes will be described below and are discussed in more detail in relation with the results in Sect.3.5.
An important change in the cloud scheme is the inclusion of a parameterization for ice supersaturation as described by Tompkins and Gierens (2007).As a result, the specific humidity of cold air parcels (at temperatures where the difference between liquid water and ice saturation pressure is large) has to reach a higher value in order for condensation to occur.This leads to an improved representation of clouds and moisture concentrations in the (upper) troposphere (Tompkins and Gierens, 2007).Aircraft observations with the Microwave Limb Sounder (MLS) have shown that supersaturation frequently occurs over the steep coastal regions of Antarctica (Spichtinger et al., 2003).
Simulations with the ECMWF IFS have already shown that the new parameterization leads to a better global distribution of super-saturated atmospheres, albeit a slight underestimation for Antarctica remains (Tompkins and Gierens, 2007).
Another change in the physics is the introduction of the McRad radiation scheme (Morcrette et al., 2008).It describes short-and longwave radiation transfer through clouds, based on the Monte Carlo Independent Column Approximation (McICA, Barker et al., 2008), and a revision of cloud optical properties making the parameterizations that use these properties more accurate.This improves the interaction of multi-layer cloud cover with short-and longwave radiation, but is believed to be of minor importance for Antarctica, considering the low occurrence frequency of these cloud types in this region.In the shortwave radiation scheme (SRTM, Mlawer and Clough, 1997) the Fouqart-Bonnel scheme is replaced by a scheme that is based on the correlated k-method (Lacis, 1991).
The latter is shown to lead to an overall improved accuracy in calculated fluxes and heating rates (ECWMF-IFS, 2008).
The last relevant physics change is the newly implemented Eddy-Diffusivity Mass Flux (EDMF, Siebesma et al., 2007) scheme for boundary-layer turbulence/shallow convection.This scheme distinguishes between large-scale (updraughts) and small-scale (turbulence) mixing processes in the surface and boundary layer by describing them with either mass fluxes or diffusion.The surface flux relies on Monin-Obukhov similarity theory but takes into account form drag (Beljaars et al., 2004) that is dependent on subscale orography.For topographically rough areas like the Antarctic Peninsula these changes are expected to be especially important.
There are other minor changes in the model but these are of little significance in the context of this study.For instance the RACMO2 model update also incorporates changes in the HIRLAM dynamical core.These are mostly of numerical nature and are not addressed here.For a more detailed and complete description of the entire RACMO2 update the reader is referred to Van Meijgaard et al. (2012) and ECWMF-IFS (2008) and references therein.

Observational data
The near-surface wind, temperature and SEB are evaluated using observational data from nine automatic weather stations (AWS).These AWSs were selected because they measure all four radiation components as well as humidity, and therefore enable a reliable closure of the SEB. Figure 1 shows the locations of the AWSs.They are located in different climate regimes: from relatively mild and wet coastal sites (AWS 4 and 11) to the steep escarpment region of Dronning Maud Land (DML) (AWS 5,6 and 16), the South Dome of Berkner Island (AWS 10) and the high and cold East Antarctic plateau (AWS 8, 9 and 12).Due to instrumental problems and icing of the sensors some months of the data are of lower quality.Observation lengths range from 4 to 15 yr of data.A summary of the location and data records of the AWSs is provided in Table 1.All AWSs are of similar design: single level measurements of wind speed/direction, temperature and relative humidity are performed at a height of approximately 3 m.The individual radiation components (SW↓, SW↑, LW↓, LW↑) are measured with a single sensor.For more details see Van den Broeke et al. (2005a,b) and Reijmer and Oerlemans (2002).
The SEB can be written as: where fluxes directed towards the surface are defined positive with units W m −2 , M is melt energy (M = 0 if the surface temperature T s < 273.15 K), SW net and LW net are the net shortwave and longwave radiative fluxes, SHF and LHF are the sensible and latent heat fluxes and G is the subsurface conductive heat flux.The sensible-and latent heat fluxes are calculated using Monin-Obukhov similarity theory using the bulk method (Van den Broeke et al., 2005b).Treatment of the radiation fluxes is as in Van den Broeke (2004).All data from the nine AWSs are monthly averaged (resulting in 770 months) and compared with data from the same months of the two RACMO cycles.An assessment of the quality of the observational data can be found in Reijmer and Oerlemans (2002); Van den Broeke et al. (2004).Note that the above implies that the turbulent fluxes and surface temperatures are calculated values and not direct measurements.
As the AWSs only cover a limited part of East Antarctica, 64 snow temperature observations (Fig. 1) are additionally used to evaluate the spatial performance of RACMO2 for T s .The modelled surface temperature, averaged over 1979 to 2011, is compared to snow temperature measurements at 10 m depth similar to Van de Berg et al. (2007).The 10 m snow temperature is assumed to be representative for the annual mean surface temperature.Note that at 27 km, the observational data are compared with data from the nearest model grid point.For the 10 m snow temperatures, this causes four locations to fall outside of the ice mask.For these points the nearest grid point that does fall within the ice mask is used.

General climate characteristics
In Antarctica a negative to zero net radiation budget prevails year round.In summer the radiation budget regularly becomes positive due to absorption of shortwave radiation at the surface.In winter, the radiation budget is balanced mainly by a positive (downward directed) SHF, as LHF is generally small due to the low humidity and thus small near-surface moisture gradients exist over the Antarctic Ice Sheet (AIS).The negative radiation budget prevails and cools the surface, resulting in a quasipermanent surface-based temperature inversion.In combination with a sloping surface, this leads to the characteristic persistent katabatic winds over the AIS.As the cooling is stronger in winter, the katabatic winds increase in strength in winter.Stronger katabatic winds enhance downward sensible heat transport, which counteracts the strength of the surface temperature inversion by increasing the surface temperature.This results in a weaker seasonality of (near) surface temperature in high wind speed areas.
To illustrate these interactive processes, Fig. 2 shows the monthly mean values of 10 m wind speed (V 10m ), surface temperature (T s ), net longwave radiation (LW net ) and sensible heat flux (SHF) for four AWSs in different climate zones of the AIS (4, 5, 6 and 9). Figure 2a shows that monthly mean wind speeds for AWS 5 and 6, in the steep escarpment region, exhibit a strong seasonal cycle due to stronger katabatic forcing in winter, with monthly wind speeds up to 9 m s −1 .These katabatic winds mix warm air downward to the surface (large SHF), increasing surface temperature (Fig. 2b), and hence upward longwave radiation (Fig. 2c).For AWS 4 and AWS 9, located on the relatively flat coastal ice shelf and interior ice sheet, respectively, wind speeds are lower, weak (AWS 4) and nonkatabatic (AWS 9) and show no seasonal cycle.At these sites the seasonal amplitude in temperature is larger than at the sites dominated by katabatic winds (AWS 5 and 6) mainly because the wintertime surface temperature inversion is stronger.

Changes in cloud properties and impact on simulated near-surface variables
The new parameterization for cloud ice super-saturation, included in the new RACMO2 physics cycle, changes the total amount of modelled clouds over Antarctica, most notably over the East Antarctic plateau.To illustrate this effect, Fig. 3 shows a latitudinal cross-section of the vertical distribution of total cloud water/ice content, averaged over the period 2007-2010 (representative for the entire simulation).A significant increase of modelled cloud content is found over the East Antarctic plateau, while cloud content has decreased along the coastline and over the ocean.With the new parameterization, moist air that reaches the continent has to exceed 100% relative humidity up to 150% in order to form clouds.As a result, clouds form further inland and higher up in the troposphere, resulting in more clouds simulated by RACMO2.3 in the interior.
The increase in clouds has caused more downward longwave radiation to be emitted as seen in Fig. 4a, where the difference fields (RACMO2.3− RACMO2.1)for LW↓, T s , V 10m and SHF are shown.The increase in LW↓ is found on most of the AIS, but is strongest on the East Antarctic plateau, and has led to higher surface temperatures (Fig. 4b), reducing the temperature gradients in the surface layer and resulting in lower SHF values (Fig. 4d).A related pattern in near-surface wind speed is not seen (Fig. 4c), with changes being smaller than 5%.
3.3 Impact on simulation of the Surface Energy Balance and SW net are small.For SW net the bias increased from −1.3 Wm −2 to −2.0 Wm −2 but the correlation remains high (r 2 0.93, with significance level p < 0.0001).For LHF the slight improvement from 0.6 Wm −2 to 0.4 Wm −2 is of little significance due to its small magnitude and the uncertainty of the observed fluxes.The standard deviation σ bias has the same order of magnitude as the bias, suggesting that the improvements are not statistically significant because of the significant noise.
To investigate the seasonal effects, Figure 6 shows the monthly mean difference of SEB fluxes, LW↓ and SW↓ for AWS 4, 5, 6 and 9. LW net is underestimated at the four AWS locations, and most significantly at AWS 4 and 9.The overestimated slope at these two sites is responsible for activating a katabatic feedback in winter, resulting in overestimated SHF.In RACMO2.3, the improved LW↓ reduces this problem for the stations located more inland, as can be seen in Figure 6, most significantly for AWS 9.For AWS 4 the improvement is not related to the increased LW↓, but to a decreased LW↑ (not shown).For this coastal site, V 10m that was overestimated has slightly decreased (the performance of the near-surface wind speed will be discussed in Sec.3.4), resulting in a lower T s and less longwave cooling.The difference in SW↓ has increased, most notably for the coastal AWSs where changes in cloud content are small.Here, changes in the radiation scheme have led to increased atmospheric transmissivity.For the more inland AWSs the effect is increasingly balanced by increased snowfall, raising the surface albedo and SW↑ (not shown).
The underestimated surface slope results in too low wind speeds at AWS 5 and 6, resulting in an overestimation of the surface temperature inversion (Fig. 8d).As a result, SHF is reasonably well represented, because stability effects remain small even at the underestimated wind speeds (Van den Broeke et al., 2005b).The problem is further reduced in RACMO2.3, in which the SEB at AWS 5 and 6 is well represented.
In summer the SHF bias at AWS 4 is smaller because SW net and LHF help to balance the excess longwave cooling.For AWS 9 however, there is an increased negative bias in SW net in RACMO2.3due to an overestimated albedo (Van de Berg et al., 2007).To conclude, for most of the climate zones of the AIS the improved representation of SEB components is induced by a better representation of LW↓, but also changes in the turbulence scheme (section 3.5) and small local changes contribute to the improvement.
Figure 5a confirms that high wintertime SHF values in the escarpment zone are well represented, and that RACMO2 generally overestimates SHF in flatter areas.This leads to overestimated T s and hence too negative LW net (Fig. 5d).Both biases are significantly reduced in RACMO2.3, by 33 % (SHF) and 39 % (LW net ) respectively.In summer, RACMO2 underestimates SW net in the high interior (Figs.5c and 6d).As a result of the underestimation of SW net , T s is underestimated (Figs. 7c, d and 8c), the surface temperature inversion overestimated as well as SHF (Figs. 5a, 6d).As a result of too low T s , sublimation (negative LHF) is underestimated and summertime convection (upward SHF) is not modelled on the ice sheet.

Impact on simulation of temperature and near-surface wind speed
Figure 7 shows modelled values (Fig. 7a, c) and difference (model-observation) (Fig. 7b, d) of the monthly averaged wind speed and surface temperature of all AWSs as a function of the observed value.The figure also shows correlation coefficient r 2 and average bias b, also denoted in Table 2. Figure 7a, b shows that the wind speed representation has not improved in RACMO2.3 when compared to RACMO2.1 (for both datasets: bias 0.5, r 2 0.27, p < 0.0001).Both model cycles generally underestimate high wind speeds and overestimate low wind speeds.Since near-surface winds over the AIS are dominated by katabatic forcing, this is caused by an overestimation of surface slopes in relatively flat areas (AWS 4,8,9,10,12) and an underestimation of surface slopes in steep areas (AWS 5,6,11) owing to the steep terrain of the escarpment region in DML, in combination with the limited horizontal resolution of the model (Reijmer et al., 2005).The small differences in V 10m are due to the combined effect of model changes but the errors in wind speed with respect to the observations are dominated by the model topography (note that the model topography is the same in both model cycles).
Figure 8a shows the monthly mean difference (model − observation) of V 10m for AWS 4, 5, 6 and 9 respectively.For AWS 5 and AWS 6 (slope > 10 m km −1 ) wind speed is underestimated year round.For AWS 4 and AWS 9 the wind speed is overestimated in winter, when katabatic forcing is overestimated, and therefore shows a seasonality that is too pronounced.This has a strong effect on the temperature, as shown in Fig. 8b,c, where temperature is underestimated at AWS 5 and 6.The surface temperature inversion, defined here as T inv = T 2m − T s , is underestimated when wind speed is overestimated (AWS 4 and 9) (Fig. 8d), which is intuitively expected.
In contrast to wind speed, a clear improvement in surface temperature T s (Fig. 7c, d) in RACMO2.3 over RACMO2.1 is seen, due to the increased LW↓, where the cold bias has been reduced from 3.2 K to 1.9 K while the correlation has not changed (r 2 = 0.91, p < 0.0001).This improvement occurs year-round for all the AWSs except for the coastal AWSs (see AWS 4), where the representation was already good due in part to the overestimated wind speed.A comparison of monthly averaged V 10m and T 2m from the Reference Antarctic Data for Environmental Research (READER, Turner et al., 2004) AWS and surface station data showed similar results and will not be discussed in this study.
Since the AWS are located mainly in DML, which has a topography that is not typical for the entire ice sheet, we use 10 m snow temperature data to obtain a better spatial coverage for the T s evaluation (Fig. 1). Figure 9 shows the difference between modelled and observed T s as a function of the latter, averaged over the model timespan.Average values, bias, σ bias , RM SD and correlation are given in Table 2. Surface temperatures in RACMO2.1 are underestimated at almost all locations and on average are too low by 2.3 K. RACMO2.3 reduces this bias to −1.3 K. Overall the spatial variability of surface temperatures is well represented by both model versions (r 2 = 0.96 for RACMO2.1 and r 2 = 0.98 for RACMO2.3,p < 0.0001) although there seems to be a tendency towards larger underestimations at higher temperature locations.Figure 9 shows that the best agreement is found on the cold East Antarctic plateau, where the overestimated wind speeds compensate the bias in temperature caused by the underestimated LW↓.Because of the improved LW↓, T s on the plateau has changed from being slightly underestimated in RACMO2.1 to being slightly overestimated in RACMO2.3.In coastal and West Antarctica, T s is underestimated the most, due to both V 10m and LW net being underestimated, and it is here that RACMO2.3produces the largest improvement.

Impact on SHF regimes
To assess the impact of changes in the atmospheric surface layer scheme Figure 10 shows monthly averaged SHF for all AWSs, RACMO2.3 and RACMO2.1 as a function of the surface temperature inversion and wind speed (color scheme).The filled circles represent winter conditions (April-September), open circles conditions for October-March. Figure 10a clearly shows four regimes.
Regime I represents the katabatic wind zone where SHF increases quadratically with the katabatic wind forcing (inversion strength).Regime II represents the exceptional conditions at AWS 16 (Thiery et al., 2012;Gorodetskaya et al., 2013), where despite stable conditions and low wind speeds, SHF values are high probably due to large scale circulation effects.The model (Fig. 10b,c) does not simulate this regime accurately due to the limited spatial resolution as the station is positioned in a topographically complex region.Regime III represents the stable conditions of the AWSs in flat areas where static stability effects become important at high values of the temperature inversion, suppressing SHF.
For lower stabilities and towards summer conditions the branches join and SHF shows a linear dependence on T inv , indicating the convective summertime conditions at plateau stations AWS 9 and 12 (Regime IV).This regime is exclusively found on the plateau, where the low summertime temperatures prevent sublimation (LHF) to act as a surface energy sink (King et al., 2006).Because RACMO2 overestimates albedo and underestimates atmospheric transmissivity, a positive radiation balance is not simulated (at least not in the monthly mean sense) and convection does not occur.
Figure 10b, c shows the inability of RACMO2 to simulate regimes II and IV.The behaviour in the katabatic wind zone is represented well, although the branch is less pronounced due to the underestimation of the slope and hence wind speeds.RACMO2.3simulates an improved separation of the most important regimes II and III compared to RACMO2.1, because of the improved surface layer turbulence scheme and general changes in the simulated results.

Conclusions
The physics package of the regional atmospheric climate model RACMO2 adopted from the ECMWF-IFS has been upgraded from cycle CY23r4 (RACMO2.1) to CY33r1 (RACMO2.

Figure 5
Figure5shows the difference of the monthly averaged modelled and observed SEB fluxes for all nine AWSs (>750 months) for RACMO2.3 and RACMO2.1.The average bias and correlation coefficients are summarized in Table2, as well as the bias standard deviation σ bias and root-meansquare deviation RM SD.Most biases (and RM SD) are reduced in RACMO2.3:for SHF from 10.5 Wm −2 to 7.1 Wm −2 , for LW net from −10.4 Wm −2 to −6.3 Wm −2 .The changes in LHF 3).This study evaluates the effects of this change on the surface energy balance (SEB), 10 m wind speed V 10m and surface temperature T s by comparing both cycles with observational SEB data gathered from 9 automatic weather stations in East Antarctica and 64 deep snow temperature sites.The model has improved in several aspects.Due to the inclusion of a parameterization for cloud ice supersaturation, more clouds and increased moisture content are simulated in the upper troposphere.As a result, more clouds and an increased cloud optical thickness in the interior have resulted in more downward longwave radiation.Consequently, in RACMO2.3 the biases in the sensible heat flux (SHF) and the net longwave radiation (LW net ) have decreased from 10.4 to 6.3 Wm −2 and −10.5 to −7.1 Wm −2 , respectively.The change in longwave radiation has improved the bias in the SHF through its tight coupling with T s and wind speed: the bias in T s , based on the deep snow temperature observations, has decreased from −2.3 K to −1.3 K. Near-surface air temperatures have also increased but less so than T s , decreasing the surface-based temperature inversion.The bias in V 10m , which is mainly due to the flattened ice sheet topography, remains unchanged.

Fig. 1 .
Fig. 1.Map of Antarctica with locations of the AWS (red diamonds) the 64 coring sites (black dots) and the position of the latitudinal cross-section used in Fig. 3. Also shown are the ice-shelf edge and grounding line (solid lines) and height intervals every 500 m (dashed lines) based on a digital elevation model from Liu et al. (2001).

Fig. 7 .
Fig. 7. Modelled and difference (modelled − observed) as a function of the AWS observations of (a, b) monthly averaged 10 m wind speed (V10m) and (c, d) surface temperature (Ts).Shown are RACMO2.3(red) with correlation r 2 new and bias bnew and RACMO2.1 (blue) with correlation r 2 old and bias b old .Biases are averages over all data with units [m s −1 ] for V10m and [K] for Ts.

Fig. 10 .
Fig. 10.Monthly averaged SHF as a function of the surface temperature inversion (Tinv = T2m − Ts) for (a) all AWSs, (b) RACMO2.3 and (c) RACMO2.1.The color scheme represents wind speed V10m and model data is from the same months and locations as the observational data.Filled circles are winter values (months 4-9) and open circles represent months (10-3).Four climate regimes are denoted, and explained in Sect.3.5.