Very high resolution regional climate model simulations over Greenland: Identifying added value

Authors


Abstract

[1] This study presents two simulations of the climate over Greenland with the regional climate model (RCM) HIRHAM5 at 0.05° and 0.25° resolution driven at the lateral boundaries by the ERA-Interim reanalysis for the period 1989–2009. These simulations are validated against observations from meteorological stations (Danish Meteorological Institute) at the coast and automatic weather stations on the ice sheet (Greenland Climate Network). Generally, the temperature and precipitation biases are small, indicating a realistic simulation of the climate over Greenland that is suitable to drive ice sheet models. However, the bias between the simulations and the few available observations does not reduce with higher resolution. This is partly explained by the lack of observations in regions where the higher resolution is expected to improve the simulated climate. The RCM simulations show that the temperature has increased the most in the northern part of Greenland and at lower elevations over the period 1989–2009. Higher resolution increases the relief variability in the model topography and causes the simulated precipitation to be larger on the coast and smaller over the main ice sheet compared to the lower-resolution simulation. The higher-resolution simulation likely represents the Greenlandic climate better, but the lack of observations makes it difficult to validate fully. The detailed temperature and precipitation fields that are generated with the higher resolution are recommended for producing adequate forcing fields for ice sheet models, particularly for their improved simulation of the processes occurring at the steep margins of the ice sheet.

1. Introduction

[2] Remote sensing observations show that the Greenland ice sheet is thinning and losing mass at an accelerating rate in the recent years [Luthcke et al., 2006; Pritchard et al., 2009; Velicogna, 2009]. In the literature, considerable discrepancies appear between the mass balance estimates of the Greenland ice sheet and their inherited uncertainties [Dahl-Jensen et al., 2009, Table 2.3]. Even different estimates applying similar methodology and data sets show discrepancies [Sørensen et al., 2011; Zwally et al., 2011], which may be partly explained by the limited knowledge of the Greenland climate. Without robust and validated climate forcing, it is not possible to compute realistic estimates of the surface mass balance and surface evolution of the Greenland ice sheet.

[3] Coastal weather stations have been operated in Greenland by the Danish Meteorological Institute (DMI) since the late 19th century [Cappelen, 2010]. However, the Greenland ice sheet itself suffers from both poor spatial and temporal coverage of observations. Although, this situation has recently improved considerably with an increasing number of weather stations on the ice sheet, from the GC-Net stations [Steffen and Box, 2001], the K transect [van de Wal et al., 2005] and the monitoring project PROMICE [Ahlstrøm et al., 2008]. (GC-Net data are available online athttp://cires.colorado.edu/science/groups/steffen/gcnet/.) There are still large regions without any weather data, especially in the ablation zone where the Greenland ice sheet loses most mass.

[4] Climate models are thus the only physically sound tools that are capable of filling the spatial gap between the weather stations and to enable computation of the past and future evolution of the climate over Greenland. Global climate models (GCMs), with horizontal resolution of about 100–200 km [Randall et al., 2007], are too coarse to represent the detailed topography that controls the surface forcing, such as the steep slopes of the ice sheet and the description of the fjords necessary to simulate the mesoscale climate in the coastal regions. For the last 20 years (see the review in the work of Giorgi [2006]), regional climate models (RCMs) have been used to efficiently increase the resolution of climate simulations to the scale of a few tens of kilometers. At this higher resolution, the RCMs describe better the steep topography in the ablation zone and the effects of orographically enhanced precipitation generated on the upstream slopes of the mountains, with less precipitation on the lee side [e.g., Dahl-Jensen et al., 2009]. The improved description of these effects is important to accurately compute the accumulation and the surface mass balance, especially when the climate model is coupled with, or otherwise used to drive, an ice sheet model. The RCM employs a limited area domain that is driven at the boundaries by large-scale atmospheric fields from a GCM or by reanalysis data. RCMs can therefore be considered as smart physically consistent interpolators because they are dynamically downscaling the large-scale atmospheric fields provided at their boundaries.

[5] A number of RCMs have been used to simulate the recent past climate over Greenland. In their validation of the climate simulated with the HIRHAM4 RCM, Box and Rinke [2003] recognized the importance of using an accurate description of the topography of the Greenland ice sheet in order to reduce the biases in the climate simulation. Fettweis [2007] computed the surface mass balance of the Greenland ice sheet by using the regional climate model MAR for the period 1979–2006, but without validating the RCM climate output against observations. Later, Ettema et al. [2009] computed higher surface mass balance than MAR over the ice sheet with RACMO2 during the period 1958–2007. According to Ettema et al. [2009], the higher surface mass balance with RACMO2 is likely to be resulting from the higher precipitation and melt computed with a higher spatial resolution of 11 km compared to the 25 km resolution that MAR used. The higher resolution facilitates capturing snow accumulation peaks that coarser RCMs miss because of poorer representation of the topography [Ettema et al., 2009].

[6] In another study, Ettema et al. [2010]demonstrated that RACMO2 can simulate the present-day near-surface characteristics of Greenland by doing a systematic comparison of the model output against observations from coastal and ice sheet weather stations.Burgess et al. [2010] used the RCM Polar MM5 to derive an accumulation field over Greenland for the period 1958–2007 using a large number of ice cores and the DMI coastal weather station measurements as input. They highlighted the paucity of the in situ data in the southeastern part of Greenland in particular, which leads to an uncertain correction of the simulated accumulation in this region. An analysis of the precipitation and temperature output of HIRHAM4 showed that the RCM simulated precipitation is smaller than the one observed over the main ice sheet and larger on the coast [Stendel et al., 2007; Aðalgeirsdóttir et al., 2009]. Forcing ice sheet models with this RCM output results in a thinner than observed ice sheet with a reduced extent toward the north and west coasts, while the simulated ice sheet is in good agreement with observations in the south.

[7] The European Union Framework-7 project ice2sea (seehttp://www.ice2sea.eu) that started in 2009 has the main goal to estimate the future contribution of continental ice to sea level rise. It is the first coordinated effort to project long-term ice sheet surface mass balance changes with coupled ice sheet and regional climate models designed for this purpose. Moreover, the Greenland climate research center, established in 2009, has a research project focusing on climate system simulations over Greenland (seehttp://www.natur.gl/en/climate-research-centre/research-projects/climate-simulations). In this project, the feedback processes within the RCM will be improved with a more complete description of fjords, lakes and open seas with a target spatial resolution of 1–2 km.

[8] The work presented here is a first step toward the achievement of the goals of these two projects leading toward a Greenland model system suitable for ice sheet and permafrost studies. It consists of a robust validation of a new multidecadal (1989–2009) regional climate model simulation. This simulation has the novelty of being computed at an unprecedented horizontal spatial resolution of 0.05° (∼5.55 km) with the most up-to-date Danish regional climate model (HIRHAM5) driven at its boundaries by the latest ECMWF reanalysis (ERA-Interim). Following model validation, the climate of Greenland is described for different regions and elevations. Then, the added value of the higher resolution is assessed by comparing the results with a lower-resolution (0.25°) simulation. The analysis focuses on 2 m air temperature and precipitation, which are the most important variables for the ice sheet models to compute the surface mass balance and the dynamics controlling the extent of the Greenland ice sheet.

[9] Section 2 describes the experimental setup where the HIRHAM5 RCM, the climate simulations and the observations are introduced. In section 3, the RCM simulated 2 m temperature and precipitation are compared to observations. Section 4describes the simulated climate for different regions and elevations. The added value of the high-resolution RCM simulation is assessed insection 5. Finally, discussions and conclusions are presented in section 6.

2. Experimental Setup

2.1. Regional Climate Model HIRHAM5

[10] The climate model used in this study is the Danish regional climate model (RCM) HIRHAM5 [Christensen et al., 2006], which is a hydrostatic RCM developed at the Danish Meteorological Institute. It is based on the HIRLAM7 dynamics [Eerola, 2006] and the ECHAM5 physics [Roeckner et al., 2003] using the Tiedtke [1989] mass flux convection scheme, with modification after Nordeng [1994], and the Sundqvist [1978] microphysics. The land surface scheme is unmodified from that used in the ECHAM5 model [Roeckner et al., 2003], which employs the rainfall-runoff scheme described in the work ofDümenil and Todini [1992]. At the lateral boundaries of the model domain, a relaxation scheme according to Davies [1976] is applied with a buffer zone of ten grid cells.

[11] As in ECHAM5, the land surface scheme used in this study does not include snow processes, including sublimation and snowmelt over land based ice (glaciers and ice sheets), although these are included where glaciers are not present. Instead, a snow layer of 10 m water equivalent is prescribed on all glacier surfaces. The energy and moisture flux interactions at and below the surface are determined by this thick snowpack. The albedo of the ice sheet is a linear function of surface temperature with a minimum albedo of 0.6 at the melting point and a maximum albedo of 0.8 at surface temperatures of −5°C and lower [Roesch et al., 2001; Roeckner et al., 2003]. Because most of the Greenland ice sheet is snow covered year round, we assess the effect of this approximation on air temperature and radiative and turbulent fluxes to be small. The main bias is limited in time and space to the ablation zone at low elevations around the margins during the melt season.

[12] Precipitation and evaporation are simulated by the model and it is possible to compute the surface mass balance (SMB) offline by combining these with a separate melt model. We use a linear relationship identified by Ohmura et al. [1996] and also applied in a study by Kiilsholm et al. [2003]. Computing the surface mass balance is not the main purpose of this study and simplifications in the surface scheme are likely to reduce the accuracy of such calculations. However, a comparison of the estimated SMB at the two different resolutions can be used to further assess the added value of high-resolution runs, since the input fields to calculate SMB are both likely to be affected by resolution. Also, it is important to understand how these effects are summed together.

[13] Snow processes are important for calculating the surface mass balance of glaciers. Most ice sheet modeling studies use dedicated schemes to calculate the surface mass balance and these schemes usually include sophisticated snow processes, often driven by output from atmospheric regional climate models. Recent model development of HIRHAM5 includes implementing an interactive surface scheme in which the surface mass budget over glaciers and ice sheets is explicitly computed. The new scheme will take into account snowmelt, sublimation, retention and refreezing in the snowpack. This will allow an online coupling between the RCM and an ice sheet model (R. Mottram et al., Surface mass balance of the Greenland ice sheet 1989–2009 using the Regional Climate Model HIRHAM5, manuscript in preparation, 2012).

2.2. Domain and the RCM Simulations

[14] The HIRHAM5 model is used to simulate the climate over Greenland at two horizontal resolutions, 0.05° (∼5.55 km) and 0.25° (∼27.75 km). The two domains are of similar size with 31 vertical levels and use a rotated map projection of 402 × 602 and 92 × 122 grid cells to reduce grid cell distortion at higher latitudes. The topography of the Greenland ice sheet is determined from the database of Bamber et al. [2001], interpolated on the HIRHAM5 grid. The two domains and the land-sea-glacier mask are shown inFigure 1. The HIRHAM5 domain sizes and limits were chosen such that the whole of Greenland and Iceland are included when the relaxation zone has been removed. This explains why the 0.25° resolution domain is larger than the one at 0.05°. The corresponding fields from the ERA-Interim reanalysis, interpolated on a 0.75° rotated grid, are also shown inFigure 1a. With increasing resolution, there is a better description of the topography and the land-sea contrast around Greenland. InFigure 1d, a small part of the southwestern coast of Greenland at 0.05° resolution is shown in greater detail. This shows that the fjord systems, in this case near the capital Nuuk, are now almost resolved in the land-sea mask. This is not the case for the lower-resolution representations. This is the main reason for our attempt to use the very high resolution.

Figure 1.

Topography (in meters) and land-sea-glacier mask for (a) ERA-Interim interpolated on a 0.75° grid and the HIRHAM5 simulations at a resolution of (b) 0.25° and (c) 0.05°. The elevation is indicated with the color scale and with red contours from 1000 to 3000 m. The black horizontal lines indicate the location of the cross sections shown inFigure 2. (d) More detailed view of southwestern Greenland at 0.05° resolution as indicated by the black box in Figure 1c.

[15] Figure 2shows a cross section of Greenland for the two HIRHAM5 resolutions and the ERA-Interim reanalysis data set interpolated on a 0.75° grid. The incremental rise in the topography between two grid cells within the ablation zone is 300 to 500 m at 0.25° and even larger (∼800 m) at 0.75°. Such a steep increase is potentially in conflict with the formulation of the vertical structure of the climate model and introduces systematic surface temperature errors [e.g.,Dahl-Jensen et al., 2009]. The incremental rise is considerably smaller in the 0.05° resolution simulation and does not introduce inconsistencies. The high resolution of the RCM may therefore contribute to accurate simulation of the temperature and precipitation gradients that are required for a realistic description of the surface mass balance in the ablation zone. Moreover, Figure 2 shows that the models at lower resolutions than 0.05° do not depict the fjords on the east coast of Greenland.

Figure 2.

Cross section of Greenland and its ice sheet showing the topography of the HIRHAM5 simulations at 0.05° resolution (red line) and 0.25° resolution (blue line). The black line shows the topography used by the ERA-Interim reanalysis interpolated on a 0.75° resolution grid. The location of the cross section is indicated onFigure 1 by the black lines.

[16] For each resolution (0.05° and 0.25°), a continuous simulation was realized for the period 1989–2009 using the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA-Interim reanalysis data at T255 (∼0.7°) horizontal resolution [Dee et al., 2011] as lateral boundary conditions. The sea surface temperature and the sea ice distribution from the ERA-Interim reanalysis are prescribed daily in the model. The horizontal wind components, the atmospheric temperature, the specific humidity and the surface pressure are transmitted to the RCM every 6 h for each of the 31 atmospheric levels. A climate simulation at ∼6 km is challenging the hydrostatic assumption. The numerical weather prediction forecast system HIRLAM, using the same dynamical core as HIRHAM5, is used for operational weather forecast at a few kilometers (∼3–6 km) over Greenland and Denmark with satisfactory results (seehttp://www.dmi.dk/eng/index/research_and_development/dmi-hirlam-2009.htm). A comparison with a new nonhydrostatic model planned to be operational within a few years indicates that the hydrostatic assumption is not a critical limitation for these simulations. We have therefore confidence in using the model at this high resolution.

2.3. Weather Station Data Sets

[17] The two HIRHAM5 simulations are validated against two observational data sets. The first comes from weather stations located around the coast of Greenland, maintained by the Danish Meteorological Institute (DMI) [Cappelen et al., 2001; Cappelen, 2010] and measuring the 2 m temperature and precipitation among other variables. The measured precipitation can be affected by many factors such as the wind speed and the type of precipitation [Allerup et al., 1997, 2000; Yang et al., 1999]. Therefore, the observed precipitation used in this study has been corrected with respect to evaporation, wetting losses and wind forcing such as drifting snow [Wulff, 2010]. The second data set comes from a network of 19 automatic weather stations (AWS) distributed on the Greenland ice sheet [Steffen and Box, 2001]. The AWS also measure other weather parameters, but in this study, only the temperature will be considered. Each AWS has four temperature sensors at different heights (1 and 2 m). In order to have the most consistent data set and to avoid data gaps, a mean of the four temperature sensors for each weather station was computed for the available period. A careful inspection of the time series temperature of the four sensors for all the AWS was done in order to remove the temperature of sensors far from the four sensor ensemble mean. The source, position, elevation and period considered for each station is described in Table 1.

Table 1. List of Stations Used for the Validation of the Simulations
Station NameSourceLatitude (ºN)Longitude (ºW)Altitude (m)Time Period
Crawford Point 1GC-Net69.8746.9820221996–2009
Crawford Point 2GC-Net69.9046.8519901997–2000
DYE-2GC-Net66.4746.2720991996–2009
GITSGC-Net77.1361.0318691996–2007
HumboldtGC-Net78.5256.8219951996–2008
JAR 1GC-Net69.4849.709321996–2009
JAR 2GC-Net69.4050.085071999–2008
JAR 3GC-Net69.3850.302832001–2004
NASA-EGC-Net75.0029.9826141997–2008
NASA-SEGC-Net66.4742.4823731998–2007
NASA-UGC-Net73.8349.5023341996–2008
NGRIPGC-Net75.0842.3229411997–2005
PetermannGC-Net80.6860.28372002–2006
SaddleGC-Net69.9844.5029011997–2009
Swiss CampGC-Net69.5549.3211761996–2006
South DomeGC-Net63.1344.8229011997–2008
SummitGC-Net72.5738.5031991996–2009
Tunu-NGC-Net78.0033.9820521996–2008
Aasiaat (4220)DMI68.7052.75431989–1999
Danmarkshavn (4320)DMI76.7718.66111989–2009
Illoqqortoormiut (4339)DMI70.4821.95651989–2009
Ilulissat (4221)DMI69.2251.05291989–2009
Kangerlussuaq (4231)DMI67.0250.70501989–1999
Narsarsuaq (4270)DMI61.1745.42341989–2009
Nuuk (4250)DMI64.1751.75801989–2009
Pituffik (4202)DMI76.5368.75771989–1999
P.C. Sund (4390)DMI60.0543.17881989–1999
Qaqortoq (4272)DMI60.4346.05321989–1999
Station Nord (4310)DMI81.6016.65361989–1999
Tasiilaq (4360)DMI65.6037.63501989–2009
Upernavik (4211)DMI72.7856.171261989–2009

3. Validation With Observations

3.1. Spatial Distribution of Temperature and Precipitation

[18] Figures 3 and 4give a general impression of the climate simulated by the RCM HIRHAM5 over Greenland, showing the average 2 m temperature and precipitation, for summer (JJA) and winter (DJF), at 0.05° and 0.25° resolution, respectively, during the period 1989–2009. The same variables from the ERA-Interim reanalysis, interpolated to a 0.75° grid, are also shown. The temperature from ERA-Interim is derived directly from the assimilated measurements, while the precipitation is simulated with the integrated forecast system (IFS) model.Figure 3shows that for the same season, the large-scale spatial distribution of the 2 m temperature is similar from one resolution to another. HIRHAM5 is therefore simulating the climate realistically by being forced at the lateral boundaries and by the sea surface temperature from the ERA-Interim reanalysis. As the resolution increases, more details in the spatial pattern can be observed, especially near the coast where the topography and the land-sea mask are complex and sensitive to the model resolution. The precipitation at the southeast coast of Greenland during winter is more intense in the fjords and has a higher spatial variability at 0.05° resolution (Figure 4f) than at 0.25° (Figure 4e). A more detailed analysis of the added value of the higher-resolution simulation is presented insection 5.

Figure 3.

Average 2 m (a, b, c) summer (JJA) and (d, e, f) winter (DJF) temperature (in degrees Celsius) for the period 1989–2009. ERA-Interim reanalysis interpolated on a 0.75° grid (Figures 3a and 3d). HIRHAM5 simulation at 0.25° resolution (Figures 3b and 3e). HIRHAM5 simulation at 0.05° resolution (Figures 3c and 3f).

Figure 4.

Average (a, b, c) summer (JJA) and (d, e, f) winter (DJF) precipitation (in millimeters per day) for the period 1989–2009. ERA-Interim reanalysis interpolated on a 0.75° grid (Figures 4a and 4d). HIRHAM5 simulation at 0.25° resolution (Figures 4b and 4e). HIRHAM5 simulation at 0.05° resolution (Figures 4c and 4f).

3.2. Comparison of Simulated 2 m Temperature With Data From Stations on the Coast and Ice Sheet

[19] Figure 5presents an overview of the comparison of the 2 m winter (DJF) and summer (JJA) temperatures after matching the time periods between the RCM HIRHAM5 simulations and the available data from the coastal DMI stations and the GC-Net ice sheet stations. The temperature of the closest land grid cell is used to compare to the weather stations and a lapse rate correction of 6°C km−1 is applied to take into account any elevation difference between the HIRHAM5 cells and the weather stations.

Figure 5.

Temperature bias (in degrees Celsius) between the HIRHAM5 simulations at 0.05° (top color cells) and 0.25° (bottom color cells) resolution and observations from DMI (blue) and GC-Net (red) stations for the period 1989–2009. The left column of color cells is for winter (DJF), and the right column of color cells is for summer (JJA). The stations JAR1, JAR2, JAR3, Swiss Camp, Crawford Point 1 (CP1), and Crawford Point 2 (CP2) are not shown in their exact locations but are presented next to each other for clarity.

[20] In general, the biases between the HIRHAM5 simulations and the observations are between −2 and +2°C during the summer season (JJA). In winter (DJF), there is a warm bias compared to the GC-Net stations on the ice sheet and a cold bias on the southwest coast compared to the DMI stations.Figure 6 shows this opposite winter temperature bias more clearly with a scatter diagram of the 2 m temperatures simulated at 0.05° and 0.25° resolution against the observations. The correlation between the observed and simulated values is good in summer with a small warm bias. In winter, the correlation is also good but the simulated values are up to 5°C too warm for the coldest stations on the ice sheet and up to 5°C too cold for the warmest stations on the coast. There is only a small difference in the biases at 0.05° and 0.25°. For spring (MAM) and fall (SON) (not shown), the biases are similar to winter with an overestimation of temperature on the ice sheet and an underestimation on the coast.

Figure 6.

Simulated 2 m average temperature on (left) 0.05° and (right) 0.25° resolution against the average observed temperature for the period 1989–2009. Winter (DJF) temperature is shown in blue, and summer (JJA) temperature is shown in red. The observations from the DMI stations are shown with squares and from the GC-Net stations with crosses. The slopes and the correlations of the best linear fits are indicated on the graph.

[21] Several model specifications can contribute to the winter warm temperature bias. The vertical resolution of the model is too low (at standard atmosphere the lowest three model levels are at 33 m, 106 m, 189 m) to resolve the boundary layer processes that cause strong temperature inversion as well as the katabatic winds, which prevail in winter over the ice sheet. The temperature bias can also be related to biases in the incoming longwave radiation (likely due to errors in the cloud cover) during the dark and cloudless winter conditions and errors in the turbulent exchange near the surface, that is poorly resolved in the model.

[22] The cold bias of the HIRHAM5 simulations on the southwest coast, compared to the DMI coastal stations, is probably partly related to the sea ice distribution that is prescribed by the ERA-Interim reanalysis data set in the Labrador Sea and the Davis Strait. The sea ice distribution plays a critical role for the atmospheric conditions, as the surface fluxes are dependent on whether the sea surface is ice covered or not.Kauker et al. [2010]reported an overestimation of the sea ice extent in the ERA-Interim reanalysis data set, which could explain the cold bias in the HIRHAM5 simulation. The model does not include fractional land or sea points but does allow fractional sea ice cover. The model even at 0.05 degree resolution tends to be dominated by land points close to the coast, while many coastal observational sites are mainly influenced by nearby oceanic conditions and therefore tend to be much warmer than just a few tens of kilometers further inland.

[23] Some care should be taken when comparing a climate variable from a grid cell, representing the mean conditions over an area of many km2 (∼6 × 6 or ∼28 × 28 km2in our case), with a point measurement from a weather station. Climate models represent the full climate system, on the basis of physical principles valid on large scales, and employ parameterization to solve sub grid cell processes. Therefore, to validate a climate simulation, it is common to consider many grid cells over a region like a watershed or a climatic zone. In our case, high-quality gridded observations are not available and therefore measurements from weather stations, that represent weather conditions at a very local scale (of a few meters), have to be considered. The local-scale measurements are sensitive to the immediate surrounding factors and are thus not directly comparable to the weather conditions simulated by a climate model on a grid cell of many km2. This may partly explain why the simulations do not have smaller biases at 0.05° than at 0.25° in Figure 5 (see section 5).

3.3. Comparison of Simulated Precipitation With Data From Stations on the Coast and on the Ice Sheet

[24] As with temperature, the validation of simulated precipitation is limited owing to the fact that there are only a few stations that measure precipitation. These stations are located on the coast [Cappelen, 2010] and the measured precipitation is affected by a number of factors. In most cases, the measured precipitation does not reflect the “actual” precipitation. To obtain an estimate of the “actual” precipitation, the measurements have to be corrected with respect to evaporation, wetting losses, wind speed and the type of precipitation (snow, rain). Further details on the method to correct the precipitation are described in the work of Wulff [2010].

[25] Accumulation values measured from ice cores [Andersen et al., 2006; Banta and McConnel, 2007; Bales et al., 2009] can be used to extend the validation of the simulated precipitation to the whole ice sheet. Ice core derived accumulation includes deposition, sublimation, melt and snow transport by the wind. All the ice cores are taken high on the ice sheet where the mean temperatures are below freezing and melting is therefore negligible. Transport of snow by wind is difficult to estimate, but as the ice cores are located in areas with no obstacles, it is assumed that this transport is evenly distributed. Sublimation and evaporation is estimated from the model. To make the precipitation measurements on the coast and the simulated precipitation comparable to the accumulation measurements on the ice sheet, the simulated evaporation is subtracted from both the observations and the simulated precipitation.

[26] Figure 7 shows the relative difference (r) (see equation (1)) between the simulated accumulation (Asim), for each of the closest land grid cell to the stations, and the observed accumulation (Aobs) from both the ice cores [Andersen et al., 2006; Banta and McConnell, 2007; Bales et al., 2009] and the accumulation computed for the DMI weather stations.

display math

The relative difference is generally closer to 0 for the 0.05° simulation than for the 0.25° simulation, especially for the high elevations in the center of Greenland, suggesting the higher-resolution simulation is capturing the precipitation and evaporation better than the lower. On the southwest coast, both simulations underestimate the accumulation while the other regions show both over and underestimates.Figure 8 shows a scatter diagram of the simulated against the observed accumulations. There is a similar correlation for the comparison of the accumulation from the ice cores data and the one estimated with the 0.05° and 0.25° resolution simulations. However, the correlation of the accumulation at the DMI weather stations and the one estimated by the simulations is lower at 0.05° than at 0.25°. This could be related to the high variability in precipitation between 0.05° grid cells.

Figure 7.

Relative difference (r; see equation (1)) between the HIRHAM5 simulated accumulation at 0.05° (small circle) and 0.25° (large circle) resolution and the observed accumulation from ice core measurements and the DMI weather stations.

Figure 8.

Simulated average accumulation at (a) 0.05° and (b) 0.25° resolution against the average observed accumulation for the period 1989–2009. The observations from ice cores are shown in red, and observations from the DMI stations are shown in blue. The slopes and the correlations for the best linear fits are indicated on the graph.

[27] Figure 9shows a comparison of land-sea-glacier mask, topography and precipitation at the two resolutions for the region surrounding Narsarsuaq in southwest Greenland. The increase in resolution improves the description of the fjords and the topography at the coast. The more detailed description of the topography depicted in the 0.05° run has a strong influence on the precipitation pattern (Figure 9f). The 0.05° grid cells contained within one 0.25° grid cell closest to Narsarsuaq show a large variability of precipitation. Figure 10 shows the average 1989–2009 monthly mean precipitation for the 0.05° and 0.25° grid cell closest to the Narsarsuaq station and the seven 0.05° neighboring land grid cells. The simulated precipitation can double or even triple from one grid cell to another. Figure 10also shows the uncorrected and corrected precipitation measurements at the DMI station Narsarsuaq (4270). The correction factor for the precipitation is close to two in the winter months. The simulated precipitation at the closest 0.05° grid cell is considerably higher than the corrected measured values. However, the comparison is better at a few of the neighboring grid cells. The height of the closest and neighboring 0.05° grid cells varies from 124 to 622 m. This is far from the elevation of the station at 34 m. The neighboring 0.05° cell with the lowest elevation is not the one, which shows the closest time series to the observations. The large cell-to-cell variability amply demonstrates why validation is difficult for the precipitation on the coast of Greenland. The precipitation field in the interior of Greenland has a smaller cell-to-cell variability than on the coast. Therefore, more confidence should be put on the precipitation validation over the ice sheet.

Figure 9.

Comparison of (a and d) land-sea-glacier mask (sea in blue, land in green, glacier in white); (b and e) topography (in meters); and (c and f) the 1989–2009 winter (DJF) precipitation (millimeters per day) simulated with HIRHAM5 at a resolution of 0.25° (Figures 9a–9c) and 0.05° (Figures 9d–9f). The closest 0.05° and 0.25° grid cells to the DMI station Narsarsuaq are shown with a red and black squares, respectively.

Figure 10.

Simulated mean monthly precipitation from HIRHAM5 at 0.05° resolution for the period 1989–2009 at the closest grid cell to the DMI station Narsarsuaq (station 4270; thick black line) and its seven neighboring land grid cells (gray lines). The corresponding values for the closest grid cell of HIRHAM5 at 0.25° are indicated in green. The uncorrected and corrected observed precipitation measurements at Narsarsuaq are shown in red and blue, respectively.

4. Description of the Simulated Climate Over Greenland

[28] The model validation presented above confirms that the RCM HIRHAM5 is doing a fair job simulating the 2 m temperature and precipitation over Greenland. These simulations can therefore be used to describe the climate over Greenland, which is to a large extent unknown due the small amount of in situ observations. To examine the regional variability, four regions are defined on the basis of the surface topography of the Greenland ice sheet [Bamber et al., 2001]. Ice sheet drainage basins were determined using the method described by Schwanghart and Kuhn [2010] and the location of the major outlet glaciers identified [Hardy and Bamber, 2000]. The resulting nine drainage basins were later joined together according to the similarities in their climate to form the final four drainage basis shown in Figure 11a.

Figure 11.

(a) Division of the Greenland ice sheet into four drainage regions. (b) Mean monthly 2 m temperature (in degrees Celsius) and (c) the mean monthly precipitation (in millimeters per day) for the four regions (all Greenland and all ocean cells are shown in Figure 11a) averaged over the period 1989–2009 for the 0.05° HIRHAM5 simulation. The numbers in the graphs show the annual mean temperature and precipitation for each region.

[29] Region 1 covers the north of Greenland with cold (<−30°C) and dry (<0.5 mm d−1) winters as shown in Figures 11b and 11c. Region 2 on the southeast coast has a milder climate with wet conditions, especially in the winter. Region 3 located on the southern tip of Greenland presents the warmest conditions with an annual mean 2 m temperature around −10°C and wet conditions (>4 mm d−1 annually), with higher precipitation in the winter. Region 4, on the west coast, is cold and dry with wetter conditions in the summer and autumn. Similar analysis for the whole of Greenland including land points not covered with ice, as well as the surrounding sea is shown in Figure 11.

[30] The mean annual temperature and precipitation simulated with HIRHAM5 at 0.05° are shown in Figures 12a and 12b, respectively. Region 1 in the north shows the strongest warming trend with a mean annual increase of 0.11°C yr−1(significant above the 99% significance level according to a two-tailedttest). The other regions show lower positive trends for the temperature, also significant above 99%. The precipitation trends are weak and not significant (above 99%) according to a two-tailedttest. The trends of the 0.05° simulation are not statistically different from those of the 0.25° simulation and do not reflect any added value. The mean temperature and the warming trend determined by HIRHAM5 for the ERA-Interim period over Greenland are consistent with those reported by other model studies includingFettweis [2007], Box et al. [2009], and Ettema [2010].

Figure 12.

Mean annual (a) 2 m temperature (in degrees Celsius) and (b) precipitation (in millimeters per day) for the period 1989–2009 for the four drainage regions (all Greenland and all ocean cells are shown in Figure 11a) for the 0.05° HIRHAM5 simulation. The numbers in the graphs show the 21 year average and the trend for this period.

[31] In Figure 13, the analysis of the mean annual temperature and precipitation is extended to examine whether different elevation bands of Greenland have evolved differently during the period 1989–2009. The results indicate that the lower elevations (0–1000 m) have warmed the most during the period 1989–2009, with a linear trend of 0.13°C yr−1. The temperature trends decrease toward the higher elevations. All trends for the temperature are significant above 99% according to a two-tailedt test. For the precipitation, the trends are weak and not significant. Moreover, it is interesting to notice the strong temporal correlation between the 2 m temperature and the precipitation in Figures 13a and 13b. A warm year has generally stronger precipitation than a cold year.

Figure 13.

Mean annual (a) 2 m temperature (in degrees Celsius) and (b) precipitation (in millimeters per day) for the period 1989–2009 for different height bins of 500 m for the 0.05° HIRHAM5 simulation over Greenland. The numbers in the graphs show the 21 year average and the trend for this period.

5. Impact of Higher Resolution: Assessment of Added Value

[32] The comparison of the model output at the two resolutions with observations in section 3 shows that the precipitation and temperature biases do not decrease significantly with increased resolution. However, the validation is not comprehensive owing to the few available observations, their short duration and the inherent measurement errors, especially the precipitation measurements, which have to be corrected for various factors. For a more robust comparison between the two simulations and to assess the added value of the higher resolution, the two HIRHAM5 simulations are compared directly to one another.

[33] Figure 14shows the difference of the elevation, the 1989–2009 mean annual 2 m temperature, precipitation and cloud cover between the 0.05° and 0.25° simulations. Here, no correction was done on the temperature to take into account the different heights. With higher resolution, the elevation of the small mountains on the coast of Greenland is generally higher than at lower resolution. Consequently, the 0.05° simulation is mainly colder and wetter on the coast of Greenland compared to the 0.25° simulation. The increase in resolution allows a better description of the topography, which increases the orographically enhanced precipitation on the coast. The increase in precipitation on the coast of Greenland at 0.05° compared to 0.25° resolution simulation dries out the atmosphere and may explain the lower cloud cover at 0.05° generating less precipitation. The colder conditions over the main ice sheet at 0.05° are mainly linked to the reduced downward longwave radiation at the surface associated to the lower cloud cover for the higher-resolution simulation.

Figure 14.

Difference over Greenland between the HIRHAM5 simulations at 0.05° and 0.25° for (a) elevation (meters), (b) 1989–2009 mean 2 m temperature (in degrees Celsius), (c) 1989–2009 mean precipitation (in millimeters per day) and (d) 1989–2009 mean cloud fraction [0,1]. The inner black contour indicates the limit of the ice sheet at 0.25°.

[34] This effect is shown more clearly in Figure 15, which shows a cross section of the 1989–2009 2 m temperature and precipitation for the summer months June, July and August (JJA). Figure 15b shows that the maximum precipitation on the west coast of Greenland is located closer to the ice sheet edge at 0.05° than at 0.25°. This leads to drier and colder conditions at higher elevations for the simulation at higher resolution. Figure 15 also shows the difference between the two simulations on the east side of Greenland where the topography is sensitive to the resolution (see also Figure 2).

Figure 15.

Cross sections of the 1989–2009 summer (JJA) (a) 2 m temperature (in degrees Celsius), (b) precipitation (in millimeters per day), and (c) offline estimate of the surface mass balance (SMB) (in millimeters per year). The simulated values at 0.05° and 0.25° are shown in red and blue, respectively. The topography of Greenland at 0.05° is shown with a black line. The location of the cross section is indicated on Figure 1.

[35] Further to this, we show in Figure 15c the surface mass balance calculated offline using annual snow fall and evaporation, plus summer temperature output from the two runs. To estimate ablation, we use an empirical relationship [Ohmura et al., 1996; Kiilsholm et al., 2003] between the annual loss of mass and the seasonal mean air temperature above −2°C for June, July and August. A summer mean temperature of −2°C is the minimum temperature where mass loss due to ablation can be expected and a linear relation for the trend has been calculated on basis of observations by Ohmura et al. [1996] and Wild and Ohmura [2000]. Refreezing at the surface is taken into account by this method, although the complete physical understanding of the relationship behind it is not fully accounted for. With a resolved topographical gradient near the ice edge, the ablation zone is well constrained. Note for example that within a few tens of kilometers from the ice sheet margin, the altitude jump between neighboring grid points is on the order of a hundred meters even at 0.05° resolution. This influences the SMB quite substantially as both the local ablation and precipitation is strongly altitude dependent. As a result, the local SMB on the western part of the ice sheet in this cross section is reduced in the higher-resolution experiment compared to the coarser one with a similar pattern on the east coast, likely reflecting the precipitation pattern shown inFigure 15b. Despite comparable performance between the two experiments (0.05 and 0.25° resolution) when validated against station data, we therefore assess that the higher resolution gives the best estimate of the true SMB.

[36] The impact of the resolution over the whole ice sheet is shown in Figure 16where the elevation dependency of the 2 m temperature and the precipitation for the two HIRHAM5 simulations and ERA-Interim reanalysis is presented.Figure 16b shows that below 1200 m, the simulation at 0.05° is generating more precipitation than the simulation at 0.25°. This is probably due to the steeper slopes at 0.05° resolution that increase the precipitation amount due to the orographic enhancement. Above 1200 m, the opposite is observed. The 0.25° simulation has more precipitation than the 0.05° simulation. This is mainly caused by wetter and warmer atmospheric conditions in the 0.25° resolution simulation compared to the simulation at 0.05°, which loses most of the atmospheric moisture at the ice sheet edges as discussed. As indicated in Figure 16b, it is interesting to note that the mean precipitation over Greenland is larger at 0.25° (1.42 mm d−1) than at 0.05° (1.37 mm d−1). This may be associated with the fact that more precipitation is falling over the ocean rather than on the land in the 0.05° resolution simulation owing to the larger topographic gradients. On the ice sheet, the drier conditions and the higher elevation of the 0.05° simulation also lead to lower temperatures (Figure 16a).

Figure 16.

Elevation dependency of the HIRHAM5 simulated (a) 2 m temperature (in degrees Celsius) and (b) precipitation (in millimeters per day) at 0.05° (red) and 0.25° (blue) using 200 m bins over Greenland. The same fields for the ERA-Interim reanalysis are presented in green. The numbers in the graphs correspond to the average over all the levels.

[37] The total amount of precipitation simulated with HIRHAM5 at 0.05° and 0.25° resolutions over the whole of Greenland, including land points not covered with ice, is almost the same for both (3% difference; see Figure 17). However, the total amount of precipitation simulated over the ice sheet only is 11% larger at 0.25° than at 0.05°. More precipitation is simulated with the 0.05° resolution model at the coast of Greenland owing to the steeper slopes. The total amount of precipitation over the Greenland ice sheet in the year 1992 computed with HIRHAM5 at 0.05° (753 × 1012 kg) and 0.25° (856 × 1012 kg) is in close agreement with that computed by Ettema et al. [2009] with the RCM RACMO2 at 0.1° (∼757 × 1012 kg) and 0.15° (∼729 × 1012 kg). The conclusion of Ettema et al. [2009], that more mass accumulates on the Greenland ice sheet with higher resolution, is not supported by the comparison of the 0.25° and 0.05° resolution simulations with HIRHAM5. In the HIRHAM5 model, it appears that the orographically enhanced precipitation reduces the amount of moisture over the main ice sheet and thereby reduces the total amount of precipitation for the higher-resolution simulation.

Figure 17.

Total precipitation summed (GT per year) over all Greenland (thin lines) and over the Greenland ice sheet (thick lines) for the two HIRHAM5 simulations for the period 1989–2009 (0.05° resolution in red; 0.25° in blue); ERA-Interim is shown in green. The numbers in the graph indicate the 21 year mean total precipitation for HIRHAM5 at 0.05° and 0.25° and ERA-Interim over the ice sheet and over all Greenland.

6. Discussions and Conclusions

[38] A robust validation of two 1989–2009 HIRHAM5 simulations (0.05° and 0.25° resolution) over Greenland using the ERA-Interim reanalysis as lateral boundary conditions is presented. The model output is compared with observed 2 m temperature and precipitation from the DMI meteorological stations on the coast and the GC-Net automatic weather stations on the ice sheet. The simulated 2 m temperature is in good agreement with observations over Greenland in summer, while in winter the temperature is lower than observed in the southwest and higher at higher elevations of the ice sheet. The increase in resolution from 0.25° to 0.05° does not reduce the temperature bias for the weather stations on the coast and on the ice sheet. This is in agreement with the study ofMass et al. [2002], who found no significant improvement of the objectively scored accuracy of the forecasts as the grid spacing decrease to less than 10–15 km. A high-quality 2 m gridded temperature field based on observations over Greenland would be necessary to assess whether a higher-resolution model improves the temperature simulation. The accumulation bias is reduced with higher resolution at the higher elevation on the ice sheet where the precipitation field is homogeneous. On the coast of Greenland, the increase in resolution increases the spatial variability of the topography, which has a strong impact on the simulated precipitation. Therefore, it is not feasible to make a fair comparison to the point measurements when the spatial variability in the model output is so large.

[39] An analysis of the simulated climate over Greenland for the period 1989–2009 shows that the largest warming trend is in the north of Greenland where the climate is dry and cold. Moreover, the warming trend is larger at lower elevations than at higher elevations. No significant trends for the precipitation were found for the different regions or height intervals during the period 1989–2009. Direct comparison of the HIRHAM5 simulations at 0.05° and 0.25° resolution shows that the 0.05° simulation is significantly wetter close to the coast and drier at high elevations than the 0.25° simulation. Also, the 0.05° simulation has less precipitation than the 0.25° simulation in the ablation zone where the highest relief is found. This could have important consequences when the climate outputs are given to an ice sheet model.

[40] This study shows the usefulness of the high-resolution regional climate model simulation for ice sheet studies as the high spatial variability is not captured in the low-resolution driving model. With the higher-resolution simulation, the description of the climate fields over Greenland appears to be more physically plausible owing to the additional details in the precipitation and temperature patterns simulated at the margins of the ice sheet where most of the ablation occurs. A good description of the climate of the ablation zone is critical when calculating surface mass balance for the full ice sheet and when coupling climate models to ice sheet models in order to simulate realistically the current and future responses of the Greenland ice sheet to climate change.

[41] A number of improvements are currently planned, including adding the computation of the surface mass balance within the RCM for which more realistic snow processes are required. Along with such improvements, a better treatment of the surface processes is planned, which will improve the radiative balance on the ice sheet surface, by including a correct snowmelt scheme, an improved albedo scheme and allowing the melting of glacier ice (R. Mottram et al., manuscript in preparation, 2012).

[42] This study provides baseline high-resolution simulations of the recent climate of Greenland. The small biases in the simulations show that HIRHAM5 generates a realistic climate over Greenland, which is suitable to drive ice sheet models. These results illustrate that the dynamical downscaling of reanalysis is necessary to correctly characterize the climate of Greenland at the regional scale. Furthermore, these results underline the sensitivity to the model resolution. Additionally, this study highlights both the difficulties that the lack of observations pose for validation of RCMs and the advantages of using an RCM in areas where observations are sparse in order to create a regional climatology. RCMs allow us to make projections of future climate on a regional scale, but also to study the immediate causes and effects of those climate changes. The output of those simulations will be used to drive ice sheet models participating in the project ice2sea. The response of the ice sheet models to the climate forcing from the RCM will give additional information on the accuracy of the climate simulated by the RCMs.

Acknowledgments

[43] We acknowledge the ice2sea project funded by the European Commission's 7th Framework Programme through grant 226375. This is ice2sea manuscript 034. In addition, this study was partially supported by the Greenland Climate Research Center (project 6504).

Ancillary