Using MODIS land surface temperatures and the Crocus snow model to understand the warm bias of ERA-Interim reanalyses at the surface in Antarctica

Introduction Conclusions References Tables Figures


Introduction
Ice-sheet melt is the largest potential source of uncertainties for future sea level rise, which has led to a growing interest in the observation and modeling of the interactions between the ice sheets and their environment.While it is clear 2 Data and methods

MODIS land surface temperature
Clear-sky LST values are derived from observations of the MODIS instruments on board Terra and Aqua spacecrafts.The two MODIS instruments view the entire surface of the Earth every 1-2 days at least thanks to a large swath (i.e., the cross track size of the image) of 2330 km.We used Terra MOD11 and Aqua MYD11 products in version-5.These products were evaluated by Wan (2014) with the radiancebased method: using atmospheric temperature and water profiles and surface emissivity, the radiance-based method calculates MODIS LST values from brightness temperatures in band 31 through radiative transfer simulations.By applying this method on 42 sites, Wan (2014) found MODIS T s errors within ±2 • C for all the sites but six bare soil sites (not including South Pole).For the South Pole site, MODIS LST error is only −0.5 • C. The accuracy of MODIS LST values depends primarily on the quality of the detection of clouds (Hall et al., 2008).When clear-sky conditions are detected, the generalized split-window land surface temperature algorithm of Wan and Dozier (1996) is used to retrieve LST values for each MODIS pixel along with emissivities in bands 31 (10.78 to 11.28 mum) and 32 (11.77 to 12.27 µm) (http:// modis.gsfc.nasa.gov/about/specifications.php).For the comparison with reanalysis and model outputs, the 1 km resolution product was projected onto a 25 km grid in stereographic polar projection.Once the interpolation on the stereopolar grid and the time binning are made, the MODIS LST product is referred to as MODIS T s , as MODIS snow surface temperature.To create an hourly data record of clear-sky T s values, all data of a sufficient quality acquired within a given hour in a grid cell were averaged.The MODIS LST algorithm provides two indicators of quality: quality assurance (QA) and quality control (QC).To minimize cloud detection errors, we selected only the pixels produced with "good quality" or "fairly calibrated" according to the MODIS quality nomenclature.The data set extends from March 2000 to December 2011, as Aqua data are available only from July 2002 onward.Figure 1 shows the mean rate of available data in Antarctica over the period, for the annual mean (Fig. 1a), for winter (JJA) (Fig. 1b) and for summer (DJF) (Fig. 1c).The data availability depends on -the revisit time of MODIS, which presents two areas of maximum controlled by the swath width (2230 km) and the orbit inclination.The maximum is in the area south of 87 • S centered around the South Pole; a second local maximum is in the area extending from 71 to 87 • S; -cloudiness, which is less marked over the Antarctic Plateau than over West Antarctica or coastal regions.
Both variables explain that the availability of hourly clearsky MODIS T s is at its maximum around the South Pole, with approximately 14 hourly T s values available per day on average and more than 9 over a large area of the Antarctic Plateau.In the coastal areas and West Antarctica, hourly data are available for fewer than 5 h per day on average.MODIS LST (land surface temperature) products are not produced in ice shelf areas because of MODIS land definition.On the plateau, data availability is higher in summer than in winter.This can be explained by more frequent clouds in winter (Bromwich et al., 2012) and by more frequent failures in cloud detection during the polar night (Comiso, 2000).

In situ observations
Several sites in Antarctica provide near-continuous, longterm data sets of variables relevant to monitoring boundarylayer conditions.To assess the accuracy of MODIS T s values, we processed upwelling and downwelling long-wave radiation observations (LW up and LW down ) provided at an hourly time step by three BSRN (Baseline Surface Radiation Network) stations (Ohmura et al., 1998) BSRN data were collected at stations with permanent staff, thereby ensuring regular cleaning of the pyrgeometers and limiting the perturbations due to riming, which is very frequent on the Antarctic Plateau.In contrast to BSRN, AWS pyrgeometers are visited once a year at most.van den Broeke et al. ( 2004) performed an evaluation of the quality of longwave radiation measured at these AWSs, which revealed frequent errors due to riming of the pyrgeometers, especially in winter.To detect erroneous measurements, the authors proposed to reject any data with LW down larger than LW up .We further analyzed these observations and decided to use a more conservative filter: AWS data are rejected when LW down is larger than LW up -5 W m −2 .This selection reduces the amount of data available for the winter.Nevertheless, we believe that some of the filtered data are still affected by riming.In order to assess the possible impact of such processing, we also analyzed observations from two additional pyrgeometers at Dome C which do not benefit from the standard BSRN cleaning procedure.They were available only for the year 2012.
Snow surface temperature T s at the stations was derived according to the following: where is snow surface emissivity, σ is the Stefan-Boltzmann constant, and T s is snow surface temperature.
In the long-wave domain (5-40 µm), snow behaves almost as a blackbody.Snow emissivity has been found to range from 0.98 and 0.99 for grain size larger than 75 µm and close to 0.985 for fine-grain snow with grain size equal to 50 µm (Dozier and Warren, 1982).In order to derive T s , has been set to the constant value 0.99 as in Brun et al. (2011).Sensitivity tests with values set to 0.98 and 1.0 were made, leading to differences in surface temperature smaller than 0.1 • C on average (0.080 and 0.079 • C respectively).

ERA-Interim surface temperature
The ERA-Interim reanalysis includes a comprehensive set of variables describing the surface and the ABL.We focused on the ERA-Interim skin temperature, hereafter ERA-i T s , which forms the interface between the soil and the atmosphere in the Integrated Forecast System (IFS) (European Centre for Medium-Range Weather Forecasts (ECMWF), http://www.ecmwf.int).IFS is the meteorological model and assimilation scheme used in the ERA-Interim analysis to assimilate observations.ERA-i T s is the temperature used in the derivation of the heat budget between the atmosphere and the surface.ERA-i T s was extracted at 0.5 • resolution in latitude and longitude and then projected at 25 km resolution on the same grid as the MODIS T s using bilinear interpolation.This step of projection aims to facilitate the comparison between ERA-Interim data set and MODIS data set.ERA-i T s is not produced by the ERA-Interim analysis scheme and hence does not benefit from the assimilation of any surface temperature observations.It is the result of the resolution of the energy balance equation during the forecast step of IFS.We extracted ERA-i T s at a 3 h time step, at analysis time and at the 3 to 9 h forecast from the 0 and 12 h analyses.We also extracted the skin temperature from a new ECMWF product named ERA-Interim/Land (Balsamo et al., 2012).This temperature, hereinafter referred to as ERA-i/land T s , was derived from a stand-alone land-surface model simulation using the land surface model HTESSEL (Hydrology Tiled ECMWF Scheme for Surface Exchanges over Land) (Balsamo et al., 2009), with meteorological forcing from ERA-Interim.ERA-i/land T s was regridded exactly as ERA-i T s .It is available every 6 h.

Surface temperature simulation using Crocus snowpack model
Within the modeling platform SURFEX (EXTernalized land and ocean SURFace platform) (Masson et al., 2013), the Crocus snowpack model (Brun et al., 1992;Vionnet et al., 2012) was run in a stand-alone mode, using meteorological forcing data from ERA-Interim.The ERA-Interim total precipitation was turned into snowfall at temperatures below 1 • C and into rainfall above.Crocus simulates the time evolution of snow properties in homogeneous layers.The number of layers evolves adaptively and is limited to 50 in this study.Heat exchanges, solar energy absorption, snow metamorphism and compaction, phase changes, and water percolation are simulated within the snowpack along with the energy and mass exchanges in the atmosphere.The albedo is a prognostic variable, depending on the type, size and age of the crystals of the snow surface layers.To run the model, near-surface meteorological data including T 2 m , 2 m air humidity, 10 m wind velocity, precipitation rate, LW down , SW down and air pressure were extracted at 0.5 • resolution from ERA-Interim and then projected onto the 25 km grid (21 499 points on the ice sheet) as for ERA-i T s .Crocus results are not independent of ERAi T s since the latter strongly influences ERA-i T 2 m which is used as a forcing data.All the grid points were initialized with the same snow profile (20 m snow water equivalent), deduced from observations made at Dome C in 2009, as in Brun et al. (2011).Since this is a rough approximation, a spin-up of the model was obtained by running Crocus three times over 1 decade (forcing conditions were taken from July 1999 to July 2009) in a loop mode in order to produce an initial profile in equilibrium over each grid point with the local climate conditions.A last run was then performed from July 1999 to December 2012, in order to produce the hourly time series of snow surface temperatures used in the following, hereinafter named Crocus T s , covering the whole Antarctic continent at 25 km resolution.
3 Evaluation results

LST MODIS evaluation
Hourly cloud-free MODIS T s at 25 km resolution, described in Sect.2.1, was evaluated with respect to in situ observations (Sect.2.2).The year 2009 was chosen to obtain the maximum number of available stations.Figure 2 shows the comparison between in situ T s and MODIS T s over the whole data set.

Evaluation with respect to stations located over the plateau
At all stations located on the plateau, MODIS T s values exhibit biases ranging from −1.8 to 0.1  is a good performance considering the fact that the comparison is performed at an hourly time step.A significant part of the RMSE comes from a few largely underestimated MODIS T s .The examination of LW down provided by the BSRN stations (South Pole and Dome C in 2009) reveals that the largest errors are due to erroneous detection of clearsky conditions.This is consistent with the likely underestimation of cloudiness on the plateau reported by Bromwich et al. (2012).The regression slope is very close to 1, showing almost no seasonal variability in the MODIS T s bias.The conservative filter described in Sect.2.2 in AWS stations suppresses cases with a false detection of clear-sky conditions by MODIS as well.This is illustrated with Dome C 2012 better scores (Fig. 2h) compared to Dome C 2009 (Fig. 2g).
In spite of the possible accumulation of several error sources -MODIS cloud detection, time shift of the MODIS data to the closest full hour time step, difference in the representative scale of the different data sets (25 km for the projected MODIS T s against a few meters for the in situ observations), occasional errors in in situ observations due to riming -MODIS T s exhibits quite good performances on the Antarctic Plateau.This confirms and extends, in space and time, the results reported in Brun et al. (2011).In this previous study, MODIS T s over Dome C was successfully compared over a 11-day period with several independent hourly time series of snow surface temperature.The MODIS T s data set clearly has great potential for the evaluation of surface temperature produced by model or analysis outputs in Antarctica.

Evaluation with respect to coastal stations
MODIS T s exhibits larger errors at coastal stations.Syowa has only 37 data points, due to the quasi-permanent detection of clouds by MODIS, which limits the significance of the results.The bias is low and the 7.5 • C RMSE is mainly due to two erroneous measurements.Princess Elisabeth Station provides a much larger data set.The cold bias (−2.7 • C) and the RMSE (4.8 • C) mainly stem from erroneous cloud detection leading in this specific case to a severe underestimation of the surface temperature.The physiographic heterogeneity around the station may also be a contributing factor.
In the rest of the paper, we focus on the Antarctic Plateau, where hourly MODIS T s values are more frequent and of better quality than in coastal regions.

ERA-Interim and Crocus surface temperature analysis
ERA-i T s (3-hourly data described in Sect.2.3) and Crocus T s (hourly data described in Sect.2.4) were evaluated with respect to the MODIS T s data set over the period 2000-2011.
It must be kept in mind that the latter includes only observations under meteorological conditions analyzed as cloud free by the MODIS cloud detection algorithm, hereinafter referred to simply as clear-sky conditions.Figure 3  In Fig. 2d, comparing MODIS T s to in situ T s at Plateau Station B, we observe twofold behavior at higher T s which are not shown in Fig. 5e, f, e and f.It probably reveals cases where false cloud detections induce underestimated MODIS T s .Crocus T s RMSE is remarkably low (from 3.1 to 4.4 • C) considering the errors in the hourly in situ observations and the uncertainties in the ERA-Interim forcing.Consequently, the evaluation against in situ observations is very consistent with the one made against MODIS T s on the whole plateau, increasing the confidence in the ERA-i T s warm bias.

Possible impact of the ERA-i T s warm bias on ERA-i T 2 m over the Antarctic Plateau
The warm bias in ERA-i T s undoubtedly has an impact on the ERA-Interim T 2 m .Indeed, the latter is not produced by the analysis scheme but by a diagnosis from the surface temperature and the air temperature at the lowest atmospheric vertical level of IFS.Air temperature in the lowest atmospheric levels is constrained very weakly by the observations The Cryosphere, 8, 1361-1373, 2014 in the Antarctic Plateau because of both the scarcity in radiosoundings and the absence of low-level observations by satellite sounders (Rabier et al., 2010).Hence ERA-Interim T 2 m evolves almost freely under the combined influence of both surface temperature, derived from the surface energy budget, and temperature of higher levels in the atmosphere.In Fig. 7 we compare ERA-Interim T 2 m with in situ T air observed at Kohnen (AWS9), Plateau Station B (AWS12), Pole of Inaccessibility (AWS13) and Princess Elisabeth (AWS16).Measuring air temperature over the Antarctic Plateau at stations where no permanent staff can perform maintenance is challenging.There are specific issues, among which the riming of the shelter and the changing elevation of the sensor between visits, due to snowfall accumulation, riming/sublimation and occasional snow drift deposits or erosion.In the left column (Fig. 7a, c, e and f) the air temperature (T air ) is the temperature as measured on the AWS at a level above the surface between 2 and 4 m, changing with the accumulation.In the right column (Figs.7b, d and g) the 2 m air temperature (T 2 m ) is determined from the AWS observations based on an energy balance model.Stability values determined in this model are used to correct the air temperature values to a fixed height of 2 m above the surface.Due to problems with the wind speed sensor, we were not able to correct the air temperature at the Pole of Inaccessibility (AWS13).The height above the surface of that sensor is about 4 m, and not changing much.
Note that the energy balance calculations developed at Institute for Marine and Atmospheric Research Utrecht, University of Utrecht, are not published.Figure 7 unambiguously shows that ERA-Interim air temperature at 2 m exhibits a positive bias from 1.93 to 3.68 • C at all stations located on the plateau.This bias is consistent with a recent study from Jones and Harpham (2013) showing that the main discrepancy between ERA-Interim and HadCRUT4 T 2 m over all continental surfaces is a warm bias in East Antarctica.It is also consistent with the positive bias of T s revealed both from MODIS T s and in situ T s .For Kohnen, Plateau Station B and Pole of Inaccessibility, the cloud of points exhibits a particular feature: for the coldest observed temperatures, ERA-i T 2 m stays systematically warmer than in situ observations.The same feature appears also for T s (Fig. 5).Indeed, these cases correspond to meteorological situations with the more stable boundary layer, showing once more the mentioned weakness in the ERA-Interim parameterization of turbulent fluxes.In contrast to the stations located on the plateau, Princess Elisabeth Station is located on a coastal and mountainous region which strongly limits the validity of comparing ERA-Interim interpolated T 2 m with a local station.

Causes of the ERA-Interim warm bias
In order to identify the origin of this warm bias, Fig. 8 shows the time series of the difference between observed and   During periods of increase, which generally correspond to increasing cloudiness, ERA-i T s increases at a realistic rate.
In the latter case, ERA-i T s starts to decrease at a realistic rate but this rate slows down rapidly, leading to an overestimation of ERA-i T s , often higher than +5 • C. Crocus T s does not show the same behavior, except when the difference between ERA-i T s and the actual surface temperature is too high, as illustrated around 18 August.This is due to the warm bias in ERA-Interim T 2 m which is used by Crocus and hence impacts Crocus T s as well.Most meteorological models parameterize the effects of stability in the calculation of the surface exchange coefficients which are used to derive the turbulent fluxes between the surface and the lowest atmospheric level.For the combined reasons described below, we think that the detected overestimation of ERA-i T s stems from this parameterization in IFS: -We evaluate the product ERA-i/land T s which is produced by a stand-alone simulation of HTESSEL (Balsamo et al., 2009), the new surface scheme developed at ECMWF.Similarly to Crocus T s simulations, ERAi/land T s values were simulated from ERA-Interim forcing.They exhibit a warm bias similar to ERA-i T s bias.HTESSEL uses the same parameterization of the surface exchanges as IFS (stability function from Högström (1988)), in contrast to the formulation used in SURFEX and consequently in Crocus (Louis (1979) modified by Mascart et al. (1995), including a limitation of the maximum Richardson number).
-Though HTESSEL has a better description of snow processes than the land scheme TESSEL (Tiled ECMWF Scheme for Surface Exchanges over Land) used in IFS, ERA-i/land T s does not significantly differ from ERAi T s .There are several differences between Crocus and HTESSEL, especially in terms of albedo and snow density, but they cannot explain the differences between their respective simulations shown in Fig. 8: there is almost no solar radiation at this time of the year, and differences in the snow heat capacity and conductivity cannot lead to a long-lasting constant difference in surface temperature, as observed from 13 to 19 August.Sub-surface flux in ERA-Interim has been derived from the temperature difference between the surface and the thermally active snow layer.Figure 8c shows that it is always very low (absolute value less than 5 W m −2 ).This is due to the large depth of this layer (1 m at the South Pole), which leads to an overestimation of the thermal resistance and an underestimation of the conductive heat fluxes between the sub-surface and the surface.While this TESSEL feature cannot explain the overestimation of ERA-i T s , it must be noticed that it should introduce significant error on the sub-daily timescales.
- explain the overestimation of ERA-i T s .In general, the surface sensible heat fluxes from the atmosphere towards the surface are very high in ERA-Interim, as shown in Fig. 9 for August 2009.On most of the plateau, the mean fluxes are higher than 20 W m −2 and even higher than 30 W m −2 over large areas.Such permanent high fluxes at high-elevation and low-wind sites are incompatible with those reported in the literature.Reijmer and Oerlemans (2002)  -ERA-i T s overestimation cannot be due to an overestimation of LW down because it would impact Crocus T s similarly, which is not the case.Furthermore, Fig. 8 shows that during August 2009, the largest bias occurred at the South Pole under clear-sky conditions at periods when LW down was perfectly represented in ERA-Interim.
-Figure 5 clearly shows a systematic overestimation of ERA-i T s during the coldest periods at each individual station.It seems that ERA-i T s cannot drop low enough during these situations, as already discussed from the time series in Fig. 8.These periods correspond to an extremely low LW down which induces a strong radiative cooling at the surface and thus leads to very stable conditions.The shape of the cloud of points is very similar to the shape of the cloud of points in the left column of figure 3 in Jones and Harpham (2013) which compares ERA-Interim and HadCRUT4 T 2 m in Antarctica.This is also the case for the lowest T 2 m during DJF north of 60 • N (Jones and Harpham, 2013, left column, raw 4), revealing that the ERA-Interim warm bias is not specific to Antarctica.It also affects northern Eurasia in winter, an additional element calling into question the representation of the surface turbulent exchanges under very stable conditions, especially when the ground is covered by snow.
The difficulty of properly estimating the surface turbulent fluxes under very stable conditions has been extensively documented (for example Brun et al. (1997), Martin and Lejeune (1998), Essery and Etchevers (2004), Anderson and Neff (2008), Sukoriansky et al. (2006), Town andWalden (2009), Genthon et al. (2010), Holtslag et al. (2013)).Ad hoc treatments are often introduced in meteorological and snow models to solve the problem, as was done in SURFEX/Crocus with the introduction of the limitation in the Richardson number.This is treated differently in IFS, which could explain the warm bias.Holtslag et al. (2013) show that T 2 m forecasted in winter over snow-covered areas with IFS is much more sensitive to slight changes in the stability functions with the current version of IFS than they were in previous versions.In order to demonstrate the sensitivity of T The last experiment clearly shows how a small change in the parameterization of the effects of stability on the surface exchange coefficients drastically changes the snow surface temperature.

Conclusions
Thanks to its orbital characteristics and to its large swath width, MODIS shows great potential in the observation of the surface temperature of the Antarctic Plateau under clearsky conditions.Thus, more than 9 hourly observations per day are retrieved on average on the plateau, and they compare very well with in situ surface temperature observations, in terms of both bias and RMSE.To our knowledge, no previous study has performed an evaluation of MODIS LST with as much detail and as many in situ observations.Further, by comparing in situ surface temperature instead of in situ nearsurface air temperature, we avoid the uncertainties in the observation of T 2 m over the Antarctic Plateau as documented in Genthon et al. (2010).
Hourly MODIS T s from 2000 to 2011 was used to evaluate the accuracy of snow surface temperature in the ERA-Interim reanalysis and the one produced by a stand-alone simulation with the Crocus snowpack model using ERA-Interim forcing.It reveals that ERA-Interim has a widespread warm bias on the Antarctic Plateau, ranging from +3 to +6 • C depending on the location.This is consistent with a recent comparison of ERA-Interim T 2 m with the HadCRUT4 data set (Jones and Harpham, 2013).Considering the very low constraint by the observations of the analyzed ABL temperature in ERA-Interim, the warm bias 2 m above the surface is due mainly to the bias at the snow surface.Comparison with in situ surface temperature shows that this bias is not limited to clear-sky conditions.At this stage, it is difficult to estimate the impact of this bias on other ERA-Interim variables, such as temperature and humidity in the ABL and snow accumulation.A detailed comparison with Crocus outputs and with the ERA-Interim/land stand-alone outputs by the new ECMWF land scheme (Balsamo et al., 2012) shows that the warm bias may be due primarily to the overestimation of the surface turbulent sensible heat fluxes in very stable conditions.Numerical experiments with Crocus show that small changes in the turbulent flux parameterization strongly impact surface temperature, highlighting the sensitivity of simulated surface temperatures to poorly known parameters.
According to the method developed in this study, hourly MODIS T s in Antarctica is particularly well suited for evaluating the surface temperature simulated by various types of models: meteorological models, global or regional climate models and stand-alone snow models.This should help in the identification of current model weaknesses and lead to improved future reanalyses, which are necessary for a better detection and understanding of climate change in Antarctica.
Figure 1d, e and f show the 2000-2011 annual, winter and summer mean value of the hourly clear-sky MODIS T s values, respectively.

Figure 2 .
Figure 2. Comparisons of MODIS T s and in situ T s at (a) South Pole, (b) Syowa, (c) Kohnen base (AWS9), (d) Plateau Station B (AWS12), (e) Pole of Inaccessibility (AWS13), (f) Princess Elisabeth station (AWS16), (g) Dome C in 2009 and (h) Dome C in 2012."Pole of I." means Pole of Inaccessibility and "Pr Elisabeth" means Princess Elisabeth.The number N of simultaneous MODIS T s and in situ T s used in the evaluation primarily depends upon satellite overpasses and cloudiness.In AWS stations, N also depends upon the filter used to select data not affected by riming.The green line represents the 1 : 1 line.
Figure 3. (a) 2000-2011 averaged ERA-i T s bias, (b) ERA-i T s bias in winter (JJA) and (c) ERA-i T s bias in summer (DJF), with respect to MODIS T s .(d) 2000-2011 averaged Crocus T s bias, (e) Crocus T s bias in winter (JJA) and (f) Crocus T s bias in summer (DJF), with respect to MODIS T s .

Figure 4 .
Figure 4. (a) 2000-2011 averaged ERA-i T s RMSE, (b) ERA-i T s RMSE in winter (JJA) and (c) ERA-i T s RMSE in summer (DJF), with respect to MODIS T s .(d) 2000-2011 averaged Crocus T s RMSE, (e) Crocus T s RMSE in winter (JJA) and (f) Crocus T s RMSE in summer (DJF), with respect to MODIS T s .

Figure 5 .
Figure 5. Comparisons of ERA-i T s and in situ T s at (a) Dome C, (c) South Pole, (e) Plateau Station B and (g) Pole of Inaccessibility.Same comparisons but only when MODIS T s values are available at the same time: (b) Dome C, (d) South Pole, (f) Plateau Station B and (h) Pole of Inaccessibility.The green line represents the 1 : 1 line.

Figure 6 .
Figure 6.Comparisons of Crocus T s and in situ T s at (a) Dome C, (c) South Pole, (e) Plateau Station B and (g) Pole of Inaccessibility.Same comparisons but only when MODIS T s values are available at the same time: (b) Dome C, (d) South Pole, (f) Plateau Station B and (h) Pole of Inaccessibility.The green line represents the 1 : 1 line.

Figure 7 .
Figure 7. Left column: comparisons of ERA-i T 2 m and in situ T air at (a) Kohnen (2892 m a.s.l., z ERA−i = 2867 m), (c) Plateau Station B (3619 m a.s.l, z ERA−i = 3617 m), (d) Pole of Inaccessibility (3718 m a.s.l., z ERA−i = 3746 m) and (f) Princess Elisabeth (1372 m a.s.l, z ERA−i = 1316 m) during 2009.Right column: comparisons of ERA-i T 2 m and in situ T 2 m at (b) Kohnen, (d) Plateau Station B and (g) Princess Elisabeth during 2009.The green line represents the 1 : 1 line.

Figure 8 .
Figure 8.(a) Comparison between different observations of surface temperature at the South Pole: BSRN T s (solid black curve), Crocus T s (solid pink curve), ERA-i T s (red point), LST MODIS (solid blue curve) and ERA-i/land T s (green point).(b) Comparison between thermal radiations: BSRN LW down (solid black curve), ERA-i LW down (red point) and ERA-i LW down_cloud (blue point).ERA-i LW down_cloud was obtained by the difference between ERAi LW down and ERA-i clear-sky LW down .(c) Comparison between turbulent fluxes of sensible heat: Crocus H (violet point), ERA-i H (red point) and ERA-i sub-surface flux (blue point).
Figure 8c shows that the sensible heat fluxes are much larger in ERA-Interim than in Crocus simulations when ERA-i T s and Crocus T s are close.Latent heat fluxes are not shown in Fig. 8c because they are almost negligible during the period (absolute value less than 2 W m −2 in both ERA-Interim and Crocus simulation) and cannot derived a mean sensible heat flux at Kohnen (AWS 9) around 12 W m −2 in August, which is in agreement with the sensible heat fluxes calculated in winter at Kohnen by Van den Broeke et al. (2005a, b), while mean ERA-Interim fluxes reach 25 W m −2 at the corresponding points during August 2009.
s to air temperature and to the representation of surface turbulent fluxes, four numerical experiments were made with Crocus under the following configurations: three experiments with a constant change in ERA-Interim T 2 m of +2, −2 and −4 • C, respectively, and an additional experiment with a change in ERA-Interim T 2 m of −4 • C and a change in the maximum Richardson number from its original 0.2 value to 0.1, which enhances the turbulent fluxes towards the surface in very stable conditions.The comparison with the control experiment in July 2009 leads to the following conclusions: -The impact of the sole changes in the forcing air temperature leads to a change in Crocus T s equal to only about half of the air temperature change.It shows how very stable conditions attenuate the impact of T 2 m on the surface temperature.-A lower maximum Richardson number used in Crocus almost balances the air temperature decrease of −4 • C over a large part of the Antarctic Plateau.

Figure 9 .
Figure 9. Averaged ERA-i sensible heat fluxes from the atmosphere towards the surface in August 2009.