1. Introduction
The Antarctic continent, which is almost fully covered by snow and ice, is not only the coldest, windiest, highest, and driest, but also the most remote landmass on Earth, entirely surrounded by the Southern Ocean. Owing to the remote location and harsh climate conditions, especially outside austral summer, we know relatively little about the Antarctic surface climate.
Currently, only about 100 automatic weather stations (AWS) are installed on the Antarctic Ice Sheet (AIS), the majority of which (≈90%) are in East Antarctica (Lazzara et al. 2012). Those AWSs are unstaffed, and most of them are irregularly serviced, which increases the likelihood of data loss through instrument failure and/or power shortage.
There are around 100 staffed stations, but fewer than one-half provide year-round observations, and most are situated along the Antarctic coast. The lack of observations complicates the detection of climate variability and change on Antarctica. In addition, near-surface climate affects surface mass balance (SMB) processes (and vice versa), such as precipitation, surface melt, and snow redistribution. The scarcity of in situ observations currently prevents us from accurately estimating the AIS SMB, its spatial and temporal variability, and its impact on AIS mass balance and consequent contribution to sea level rise, from observations alone. In recent years, satellite remote sensing datasets are emerging, but the polar night or limitation to clear-sky conditions impedes measurements and large uncertainties limit the retrieval algorithms, and, although they provide multiyear datasets, time series are still limited for long-term climate research.
Alternatively, we can use atmospheric reanalysis products, which now give us four decades (they typically start in 1979) of continuous, gridded atmospheric data at high temporal resolution. Reanalyses are essentially global climate models that are assimilated with atmospheric observations to obtain a “reanalysis” of the global atmosphere, including that at the high latitudes.
Because of the scarcity of observational data over Antarctica, little atmospheric information is assimilated in the global climate model that underlies the reanalysis. This leads to a stronger sensitivity of the simulation to the actual global model physics, and biases therein. Global climate models are optimized for midlatitude atmospheric conditions, which enhances model biases in high-latitude regions, including Antarctica. Recent studies have highlighted deficient performance of atmospheric reanalyses in clouds and radiation (Lenaerts et al. 2017; Miller et al. 2018), precipitation (Bromwich 1988; Bromwich et al. 2012; Chen et al. 2011; Medley et al. 2013, 2014; Nicolas and Bromwich 2011), (near-) surface temperature (Fréville et al. 2014; Jones and Lister 2015; Nicolas and Bromwich 2014; Bracegirdle and Marshall 2012), and turbulent heat exchange (Miller et al. 2018) over ice sheets. Newly emerging reanalysis products will likely improve on many of these issues. In particular, the new European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis product ERA-5 combines unparalleled horizontal resolution (0.25°) with refined atmospheric physics and a revised snow scheme (Dutra et al. 2010).
Existing studies comparing reanalysis products over Antarctica have generally concluded that ERA-Interim shows the smallest biases (Bracegirdle and Marshall 2012). However, these comparisons are mostly focused on one atmospheric parameter, such as SMB or temperature (Bromwich et al. 2011; Wang et al. 2016). Here, we present a comprehensive assessment of the capacity of several reanalysis products, including the new ERA-5, to reproduce the (near-) surface climate conditions over the AIS. To do so, we analyze several surface atmosphere parameters and compare several reanalysis products with available in situ data from weather stations.
Section 2 presents the dataset compiled to evaluate the reanalyses, and the comparison method. Results are presented in section 3, and section 4 concludes with the strengths and weaknesses of each of the reanalyses.
2. Material and methods
a. Overview of the reanalyses
In an atmospheric reanalysis, observations of Earth’s atmosphere, land, and ocean surface are assimilated in global climate model simulations in order to produce a spatially complete, multivariate, three-dimensional dataset of Earth’s atmosphere that is as close to reality as possible. To minimize the bias of the modeled atmosphere compared to observations, a wide variety of observational products are used, including satellite data, ground stations, sea buoys, radiosondes, and airplane observations. Reanalysis products have been developed and improved since the start of the satellite era, circa 1979, and have proven invaluable tools for understanding our climate system. In this paper, we evaluate four reanalysis products: ECMWF’s ERA-5 and ERA-Interim, the Climate Forecast System Reanalysis (CFSR) from the National Centers for Environmental Prediction (NCEP), and the National Aeronautics and Space Administration (NASA) Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2). We use the 1979–2017 time period for the entire analysis. We investigate the near-surface temperature, wind speed, and relative humidity for the four reanalyses. We use the reanalysis output at time steps 0000 and 1200 UTC. For SMB evaluation, 6-hourly precipitation and evaporation output are used.
1) ERA-5
ERA-5 is the most recent reanalysis product produced at ECMWF. In the near future, it will fully replace its predecessor ERA-Interim as ECMWF’s most state-of-the-art reanalysis product, covering the time period of 1979 up to the present day. One of the main improvements over ERA-Interim is a higher horizontal and vertical resolution, as ERA-5 uses a horizontal resolution of 31 km (0.28°) and 139 vertical levels, covering the atmosphere from the surface up to 0.01 hPa (≈80 km). In addition, ERA-5 uses a more recent and improved version of the Integrated Forecast System (IFS) Earth system model and associated observational assimilation system (Cycle 41r2; Hersbach and Dee 2016). Improvements have been made not only to the assimilation of satellite observations, but also to the assimilation of ozone, and aircraft and surface pressure data (Albergel et al. 2018). Relative to ERA-Interim, additional observational products are assimilated that were not available earlier (Hersbach and Dee 2016). Moreover, the updated snow scheme enables a reduction of the snow albedo bias and an improved hydrological cycle (Dutra et al. 2010). Model forcings were also improved, and now include CMIP5 greenhouse gases and volcanic eruptions as well as improved sea surface temperature and sea ice cover. ERA-5 data products are available on an hourly time scale, compared to 6-hourly for ERA-Interim. In addition, uncertainty estimates are available, calculated by means of a 10-member ensemble at reduced resolution (horizontal resolution of 62 km; 0.5625°).
2) ERA-Interim
ERA-Interim is the predecessor of ERA-5. This reanalysis product covers the time period of 1979 up to the present day. The production of ERA-Interim started in 2006, as the successor to ECMWF’s previous reanalysis product ERA-40. It uses the spectral model IFS Cycle 31r2 assimilation system, a 4D-var data assimilation and 12-h analysis window. The prognostic equations that determine atmospheric flow are solved using the semi-Lagrangian method with a finite element method in the vertical (Hodges et al. 2011). ERA-Interim uses a horizontal spatial resolution of 80 km and a vertical resolution of 60 levels, covering the surface up to 0.1 hPa. Reanalysis output is available on a 6-hourly time scale (0000, 6000, 1200, and 1800 UTC). For further details on ERA-Interim, refer to Dee et al. (2011).
3) NCEP CFSR
The Climate Forecast System Reanalysis is produced by NCEP and the National Center for Atmospheric Research. CFSR is a third-generation reanalysis product and uses the NCEP Coupled Forecast System model, which is a spectral atmospheric model (Saha et al. 2010). It has a horizontal resolution of 38 km and a vertical resolution of 64 levels. Atmospheric, oceanic, and land surface output products are available at an hourly time resolution. As compared with previous versions of CFSR, climate is now forced with observed estimates of evolving greenhouse gas concentrations, aerosols, and solar variations (Wang et al. 2011).
4) NASA MERRA-2
The Modern-Era Retrospective Analysis for Research and Applications, version 2, is a reanalysis product provided by NASA. It uses the GEOS-5 model physics and a grid point statistical interpolation data assimilation system, and has a 6-hourly observational assimilation cycle (Hodges et al. 2011). The horizontal resolution employed is 0.5° in the latitudinal direction and 0.66° in the longitudinal direction. In the vertical direction, 72 layers are used. Output variables are available on different time resolutions depending on the variable: hourly for the 2D diagnostic fields, 3 hourly for the 3D diagnostic fields, and 6 hourly for the prognostic fields. Among others, MERRA-2 contains several improvements to hydrology, as this aspect was found lacking in previous reanalysis produced by NASA (Rienecker et al. 2011).
b. Observations and method
To evaluate the performance of the different reanalysis products, an observational database was compiled. Basic ground-based meteorological observations over the AIS were gathered from several sources: the Scientific Committee on Antarctic Research (SCAR) database (Turner et al. 2004), the Antarctic Meteorological Research Center (AMRC) program (http://amrc.ssec.wisc.edu/), the Australian Antarctic AWS dataset (http://aws.acecrc.org.au/), and the Italian Antarctic Research Program (http://www.climantartide.it). In total, monthly near-surface temperature and wind speed observations are available for more than 10 years for 79 individual sites over the AIS since 2000 (Fig. 1). A detailed overview of the stations, including their height and temporal coverage, is given in the appendix (see Table A1). In general, temperature observations are available at 2 m above ground level but wind speed observations are taken at various heights. Therefore, all wind speed observations are extrapolated to 10 m above ground level by assuming a Monin–Obukov logarithmic vertical profile at neutral conditions (Sanz Rodrigo 2011).
The ground-based meteorological observation network. Only stations with data availability of more than 10 years are included. Yellow dots denote stations that have relative humidity measurements available.
Citation: Journal of Climate 32, 20; 10.1175/JCLI-D-19-0030.1
As discrepancies between model and reanalysis elevation at the AWS locations can reach up to several hundreds of meters, we apply a lapse-rate correction to the surface temperature before making comparisons, following the method described in van Lipzig et al. (1999) taking into account the lapse rate based on the eight neighboring model grid cells for each individual station. Lapse rates vary from 6.3°C km−1 for CFSR up to 7.2°C km−1 for ERA-5. The average height difference between the station locations and the corresponding model pixel is equal to 122 m for ERA-5, 152 m for CFSR, 171 m for MERRA-2, and 199 m for ERA-Interim.
Near-surface relative humidity observations are obtained by AWS measurements of the Institute for Marine and Atmospheric Research Utrecht (IMAU) Antarctic AWS Project (https://www.projects.science.uu.nl/iceclimate/aws/antarctica.php) and from the AMRC AWS network (accessible at https://amrc.ssec.wisc.edu/data/archiveaws.html). Long-term observations are available for 31 sites, which are mainly located in the Dronning Maud Land region and the Antarctic Peninsula (IMAU) and the Ross Ice Shelf and Victoria Land (AMRC) (yellow dots in Fig. 1). A similar correction as for temperature is performed in order to take into account elevation differences between the station and the models. Detailed information about these AWS is given in the appendix (see Table A2). These wind speed, near-surface relative humidity, and temperature observations are not assimilated in the reanalysis and therefore constitute independent measurements to perform the evaluation.
For each of the stations, observations of temperature, wind speed, and relative humidity are compared with the value at the corresponding location in the different reanalyses. The reanalysis pixels that are in the vicinity of 50 km from the station are equally considered. We compare the average annual values, and the mean absolute error (MAE) and the Pearson correlation coefficient are calculated by comparing the individual monthly registered values of each observation to the corresponding month in the reanalyses. Furthermore, the annual cycle of the variables is also evaluated. As such, in addition to getting an overview of the average performance of the reanalysis, information about the temporal variability is obtained. For this, we compared the average monthly values for each variable of interest (temperature, wind speed, and relative humidity) between all reanalyses and the observations. Next to that, also the biases between the reanalyses and observations were calculated. To test whether the mean biases of the reanalyses differ significantly (at the 0.05 significance level), a paired sample t test was performed (Nwogu et al. 2016).
For all near-surface atmospheric variables, we distinguish between three zones: coastal Antarctica [<1500 m above mean sea level (MSL)], the interior (>1500 MSL) of the continent, and the Antarctic Peninsula [the classification can be retrieved from Table A1 in the appendix (the “Elev class” column)]. AWSs are very useful instruments, but it must be noted that some of their sensors (e.g., the aerovanes) might suffer from the harsh winter conditions, such as freezing temperatures and very strong wind speeds. In addition, the AWSs need to be raised every so often because the sensor gets buried under precipitation and/or blowing snow (Costanza et al. 2016). The loss of power can lead to limited data availability, biasing the climatological means. These limitations must be kept in mind when evaluating the performance of the reanalyses against the AWS results.
SMB over Antarctica is defined as precipitation minus meltwater runoff, surface sublimation, and redistribution and sublimation of blowing snow particles (van den Broeke et al. 2004). Precipitation is the dominant source of AIS SMB. Meltwater runoff is only important for areas below 1000 m MSL, and a small component of the AIS-integrated SMB. Blowing snow processes are not represented by the reanalyses. We therefore use accumulation (i.e., precipitation minus evaporation/sublimation) in the reanalysis output as proxy for SMB. To evaluate the performance of the reanalyses, we use the SMB dataset compiled by Favier et al. (2013). This quality-controlled observational dataset contains single-year measurements from various sources, such as ice cores, ground penetrating radar, and stake measurements. In addition to the quality check performed by Favier et al. (2013) (using only the A-rated observations), we discard data with a mismatch of more than 150 m between the elevation given in the dataset and the elevation of the reanalysis pixel. We compare each of the observation to the accumulation value of the corresponding model grid cell of the same year. If more than one observation is located within the same pixel or if an observation covers several years, each of the single year observations is treated individually and compared to the model value. We then bin the resulting observations in four classes, according to elevation and roughly corresponding to different zones of the AIS. Elevations below 500 m MSL represent the coastal zones, including the ice shelves. Elevations from 500 to 2000 m MSL represent the escarpment zone. Elevations from 2000 to 3000 m MSL correspond roughly to the higher-elevation peripheral zones, and elevations over 3000 m MSL represent the interior of Antarctica.
To calculate the reanalysis mass change [(dm/dt)|RA] for each month, we attribute the sources of mass to snowfall minus sublimation for each reanalysis, and compute the anomaly by subtracting the monthly mean amount of SMB (snowfall minus sublimation, calculated on a base period of 22 years prior to the launch of GRACE) to the snowfall amount. In that way, since the anomalies are mostly explained by the large snowfall events and the sink term can be neglected.
3. Results and discussion
a. Near-surface temperature
A comparison between the annual average near-surface temperature in the four reanalysis and the observations is presented in Figs. 2 and 3. Overall, all reanalyses are able to cover the observed temperature range on the AIS to some extent. However, lowest temperatures (<−40°C) are overestimated by ERA-Interim and CFSR (up to 5°–10°), while ERA-5 and MERRA-2 better represent these low temperatures. ERA-5 performs best in simulating the near-surface temperature with an MAE of 2.0, followed by ERA-Interim, with an MAE of 2.6 (Fig. 2). MERRA-2 and CFSR perform less good, showing MAEs of 3.1 and 3.4, respectively. These reanalyses also show the largest spread of annual average temperature values around the identity line.
Annually averaged 2-m temperature observations compared with the different reanalyses. Mean absolute error MAE and Pearson correlation coefficient r are calculated from individual monthly values for each station to take temporal variability into account.
Citation: Journal of Climate 32, 20; 10.1175/JCLI-D-19-0030.1
Averaged annual cycles of 2-m temperature for all reanalyses compared with the observations.
Citation: Journal of Climate 32, 20; 10.1175/JCLI-D-19-0030.1
Next, we compare the seasonal variability of the four reanalyses temperature with the observations, over the three Antarctic climate zones (Fig. 3). Note that because of the different seasonal temperature range for the different Antarctic regions the ranges of the y axes differ among the three panels of Fig. 3. There are large differences in performance for the reanalyses, both for the winter [June–August (JJA)] and summer [December–February (DJF)] seasons and between the different Antarctic regions. In the interior, MERRA-2 shows a very good correspondence to the observations during the Antarctic fall, winter, and spring. The other reanalyses strongly overestimate surface temperature during these seasons, and their biases are found to be significantly different from MERRA-2, especially for the fall and winter months (Tables 1–3). During the winter season, the differences in temperature between these reanalyses and the observations reach values up to 5°C or more. The highest bias is found especially for CFSR, which significantly overestimates the interior winter temperatures by almost 10°C. Furthermore, the annual cycle of the variables is also evaluated. As such, in addition to getting an overview of the average performance of the reanalysis, information about the temporal variability is obtained. For this, we compared the average monthly values for each variable of interest (temperature, wind speed, and relative humidity) between all reanalyses and the observations. Next to that, also the biases between the reanalyses and observations were calculated. To test whether the mean biases of the reanalyses differ significantly (at the 0.05 significance level), a paired sample t test was performed (Nwogu et al. 2016). In the summer season, the spread in performance is much smaller for the Antarctic interior (differences of maximally 2°C) with ERA-5 performing equally well as CFSR and MERRA-2. Only ERA-Interim shows a significant overestimation (of about 5°C) during summer relative to the other reanalyses and observations (Tables 2, 4, and 5).
Overview of the statistical significance of the differences between MERRA2 and CFSR for temperature, wind speed, and relative humidity. An X for a given variable, region, and month indicates statistical significance at the 0.05 level or better.
Over coastal Antarctica, ERA-5 performs best over all seasons. CFSR and ERA-Interim also perform very well, especially in the winter season, and MERRA-2 underestimates the temperature during winter (up to 5°C). The difference between MERRA-2 and the other reanalyses during winter was found to be statistically significant, except for ERA-Interim (Tables 1–3).
Similar to over coastal Antarctica, ERA-5 captures the seasonal cycle of surface temperature rather well over the Antarctic Peninsula. The other reanalyses show a positive bias of a few degrees over this region during winter, although the difference in biases with ERA-5 is found to be not statistically significant (Tables 1–6). For the summer season, all reanalyses are in line and show a good correspondence with the observations.
The reanalysis performance in representing the seasonal cycle of near-surface temperature depends on the season as well as on the region, as was already mentioned by Bracegirdle and Marshall (2012). ERA-5 performs very well over coastal Antarctica and the Antarctic Peninsula, while MERRA-2 comes out as the best-performing reanalysis product over the interior. All other reanalyses show a substantial warm bias over the interior, especially during the winter season. The interior warm bias was already found for ERA-Interim (e.g., Fréville et al. 2014; Jones and Lister 2015) and seems to be still present in ERA-5, based on our results. Note that although some differences appear between ERA-Interim and ERA-5, most differences are not statistically different. Only for the summer season over the interior, ERA-Interim and ERA5 differ significantly for temperature (Table 5).
b. Near-surface wind speed
The near-surface wind speed performance is quite similar for the different reanalyses, with a good representation of the spatial variability: The observed annual average wind speed ranges from 2 to ~15 m s−1, and a similar range is found in the reanalyses. However, there is a general underestimation of the strongest winds, which might be at least partly related to the comparison between point observations with gridcell averages (Fig. 4). MAE values range from 2.4 to 2.9 m s−1, reflecting the scatter in Fig. 4. Part of this is likely explained by the effect of local conditions (especially in regions of complex orography) on the measured wind speed and to errors that might occur in the observational datasets, especially during the winter season when sensors may freeze.
As in Fig. 2, but for 10-m wind speed.
Citation: Journal of Climate 32, 20; 10.1175/JCLI-D-19-0030.1
Highest wind speeds are observed during the winter season, when the katabatic forcing is strongest (Fig. 5). During the summer season, the large-scale pressure gradient force (PGF) becomes the dominating driver of the wind (van den Broeke and van Lipzig 2003). The seasonal cycle on the Antarctic Peninsula is controlled by the large-scale PGF, more specifically by the climatological low pressure system (the Amundsen Sea low) that moves from the Ross Sea in winter (generating strong winds over the peninsula) toward the Bellingshausen Sea (generating weak winds over the peninsula) (van den Broeke and van Lipzig 2003).
As in Fig. 3, but for 10-m wind speed.
Citation: Journal of Climate 32, 20; 10.1175/JCLI-D-19-0030.1
All reanalyses are successful in capturing the typical annual cycle described above. The strong wintertime katabatic winds in the interior and the coast are well represented by the mean MERRA-2 wind speeds, whereas the other three products depict a mean underestimation, though not statistically different from MERRA-2. During the summer season, all reanalyses have an excellent correspondence to the observations in the interior, but underestimate the winds in coastal Antarctica. This is consistent with Sanz Rodrigo et al. (2013), who found a strong underestimation of the coastal winds in ERA-Interim, which was partly reduced in the regional climate model RACMO. The best preforming models over the Antarctic Peninsula are ERA-5 and CFSR, although not significantly different (at the 0.05 level) from ERA-Interim. The negative biases in MERRA-2 are significantly different from the other three reanalyses. Overall, the performance of the reanalyses for wind strongly depends on the region and season considered and there is no clear-cut best-performing reanalysis for wind.
c. Near-surface relative humidity
For relative humidity, a difference in performance exists between CFSR and the other models. While CFSR lies closest to the observations for highest relative humidities, its performance for smaller values is worse than that of other reanalyses. It overestimates relative humidity, especially in the interior and at the coast (see Fig. 7), leading to the highest MAE of all models [18.9, as compared with 13.5, 11.9, and 13.6 for ERA-5, ERA-Interim, and MERRA-2 (Fig. 6), respectively]. It therefore comes out as the worst-performing model for relative humidity.
As in Fig. 2, but for 2-m relative humidity.
Citation: Journal of Climate 32, 20; 10.1175/JCLI-D-19-0030.1
Of the three reanalyses, MERRA-2 appears to be the model that most closely captures the observed annual cycle of relative humidity, and does so for all three geographical regions (interior, coast, and peninsula; see Fig. 7). ERA-Interim and ERA-5 strongly overestimate the amplitude of the annual cycle, with significantly lower relative humidity values during winter than the other reanalyses, which is not supported by the observed data. CFSR significantly overestimates the relative humidity values relative to the other reanalyses (Tables 1–6).
As in Fig. 3, but for 2-m relative humidity.
Citation: Journal of Climate 32, 20; 10.1175/JCLI-D-19-0030.1
Note that for relative humidity, measurement accuracy is relatively high: the measurement accuracy for the Vaisala HMP35AC instrument used is 2% for humidities of <90% and 3% for humidities of >90% (van den Broeke et al. 2004; Lazzara et al. 2012). Therefore, it is likely the biases in relative humidity seen in these models represent genuine model biases.
Underestimations of relative humidity, as seen here for three of the four reanalyses (ERA-Interim, ERA-5, and, to a lesser degree, MERRA2) have been reported for other model experiments, for example, in Polar WRF simulations evaluated over the Ross Ice Shelf (Wille et al. 2016, 2017). Here, the authors conclude that the likely source of the underestimation of relative humidity is the fact that katabatic flow from the Transantarctic mountains is too dry, as well as the fact that sublimation from blowing snow is not parameterized.
d. Surface mass balance
Figure 8 shows the performance of the SMB (approximated by accumulation) of the reanalyses in comparison with the observations, according to four classes: the coastal zones and ice shelves (<500 m MSL), the escarpment zone (from 500 to 2000 m MSL), the higher-elevation peripheral zones (from 2000 to 3000 m MSL), and the interior of Antarctica (>3000 m MSL).
Accumulation of each reanalysis against the stratified mean observed accumulation from the dataset of Favier et al. (2013) (circles). The error bars indicate the standard deviation of the measurements (horizontal) and reanalysis (vertical) accumulation values.
Citation: Journal of Climate 32, 20; 10.1175/JCLI-D-19-0030.1
In general, all reanalyses display the same pattern regarding the interior of the continent, with varying biases regarding the coastal areas and the escarpment zone. ERA-5 shows the closest agreement to the observations [MAE of 48.1 mm yr−1, in water equivalent (w.e.)] but underestimates the coastal SMB. All reanalyses underestimate the accumulation except for MERRA-2. ERA-Interim (MAE of 54.2) tends to overestimate areas between 500 and 2000 m MSL, which is also the case for MERRA-2 (MAE of 58.5). CFSR underestimates both elevation classes between 500 and 2000 m MSL (MAE of 59.5). The coastal areas and ice shelves are underestimated by all reanalyses except MERRA-2 and CFSR lies closest to the observations for that elevation class. The escarpment zone is overestimated, especially by ERA-Interim and MERRA-2 while CFSR underestimates the accumulation over areas between 500 and 2000 m MSL. ERA-5 and CFSR underestimate the accumulation in the 2000–3000-m MSL elevation class, while ERA-Interim and MERRA-2 overestimate it. ECMWF’s reanalyses lie closest to the observations for this elevation class. Finally, all reanalyses simulate the SMB in the interior with the least bias, except for ERA-5, which underestimates the accumulation on the plateau. The relative biases are higher for the elevation classes above 2000 m MSL, as they receive a limited amount of precipitation compared to the other areas, where the relative bias is limited below 50%. Overall, the gradation of reanalyses mean accumulation rates over the ice sheet compare well to Medley et al. (2013): the largest rate over the 1979–2017 time period is displayed by CFSR, followed by ERA-Interim and then MERRA-2 with decreasing accumulation rates. ERA-5 shows the lowest accumulation rate.
In general, the spread of the modeled SMB is narrower for all reanalyses, in comparison to the SMB dataset. This can be explained by the smoothening of SMB variability in the reanalyses due to their relatively coarse horizontal resolution. As a result, SMB variations smaller than the reanalysis grid size are not resolved. In addition, since the modeled SMB only represents precipitation minus sublimation, melt and blowing snow processes are not taken into account. This leads to under- and overestimations of local SMB, mainly over the ice shelves and at the coastal locations, where these processes are most effective. The interior of the continent, less prone to melt and blowing snow, receives a very limited amount of precipitation.
The comparison of the mass change derived from the reanalyses with the mass anomaly retrieved from GRACE for the Dronning Maud Land area (Fig. 9) show that sharp increases due to the surface mass anomalies of 2009 and 2011 are present in all reanalyses. ERA-5 and ERA-Interim show the smallest MAE (24 and 30 Gt yr−1, respectively) and closely follow the cumulative mass change derived from GRACE gravimetry. The signatures of the atmospheric rivers, although present in CFSR and MERRA-2, are less pronounced and lead to a larger bias to the observed mass anomalies (MAE of 91 and 68 Gt yr−1, respectively). Since the comparison is only valid for the years 2009 and 2011, no conclusion can be drawn from the rest of the time series.
Comparison of mass anomaly derived from GRACE, and the mass anomaly computed using the accumulation (precipitation minus sublimation) from each reanalysis minus the climatological mean accumulation from a 22-yr time period previous to the launch of GRACE.
Citation: Journal of Climate 32, 20; 10.1175/JCLI-D-19-0030.1
The differences in performance between SMB and the dataset of Favier et al. (2013) and the mass anomaly derived from GRACE gravimetry are due to different time resolutions and different localizations.
4. Conclusions
We present here an evaluation of near-surface atmospheric variables and SMB over the Antarctic Ice Sheet in four reanalysis products, namely ECMWF’s ERA-5, its predecessor ERA-Interim, NCEP CFSR, and NASA MERRA-2, against a set of in situ observations over the 2000–17 time period. The mean near-surface temperature, wind speed, and relative humidity are investigated, as well as their annual cycle.
The results for near-surface temperature show that all reanalyses are able to some extent to cover the temperature range of the AIS, and they highlight that the latest state-of-the-art reanalysis product ERA-5 overall captures the mean temperature climatology best when compared with the other reanalyses.
The typical annual cycle of wind speed, with a strong influence of the katabatics in winter and the pressure gradient force in summer, is captured by all reanalyses. However, performance of the reanalyses decreases during winter when the katabatic forcing is prevailing: the strongest wind speeds are strongly underestimated by ERA-5 and CFSR in the coastal and interior regions during winter while MERRA-2 overestimates the wind in the interior during winter.
For relative humidity, MERRA-2 apparently represents the seasonal cycle of relative humidity best, whereas it is overestimated by ECMWF’s models, which have significantly lower values during winter. It seems that, except for CFSR, all reanalyses are unable to simulate the relative humidity accurately at low temperatures.
For SMB, the performance of the individual models vary according to the zones investigated. Despite a general similar pattern for all reanalyses, ERA-5 shows the smallest bias to the observations but tends to underestimate SMB, whereas all other reanalyses overall overestimate accumulation over the AIS. In addition, all are able to represent the atmospheric rivers, but ECMWF’s reanalyses show the smallest bias in accumulation in Dronning Maud Land. SMB is very challenging to model accurately, and includes processes not accounted for in the reanalyses. Therefore, inclusion of blowing snow and melt processes, as well as a better representation of snowfall over the AIS, is crucial to represent the observed SMB in an accurate way.
Even though the reanalyses perform very differently and there is no best model for all variables, the analysis performed in this paper enables the users to choose the best-performing reanalysis, depending on the area of investigation, the season, and the variable to represent (see Fig. 10).
Mean absolute error for each reanalysis for each of the studied variables: temperature (T; °), wind speed (WS; m s−1), relative humidity (RH; %), and SMB (mm w.e. yr−1) from both methods [validation against Favier et al. (2013) and GRACE mass anomaly derived from altimetry]. The color denotes the relative performance in comparison with the other reanalysis. The mean value of all MAE is computed for each variable, and green indicates better performance than 1 standard deviation to the mean, yellow is within 1 standard deviation to the computed mean, and orange is higher than 1 standard deviation.
Citation: Journal of Climate 32, 20; 10.1175/JCLI-D-19-0030.1
Acknowledgments
This work was supported by the Belgian Science Policy Office (BELSPO; Grant R/143/A2/AEROCLOUD) and the Research Foundation Flanders (FWO; Grant G0C2215N). It was also supported by the EOS programme of the National Fund for Scientific Research (NFSR) through the project FNRS EOS 30454083 “Decadal predictability and variability of polar climate: The role of atmosphere–ocean–cryosphere multiscale interactions (PARAMOUR)”. We thank Wim Boot, Carleen Reijmer, and Michiel van den Broeke (Institute for Marine and Atmospheric Research Utrecht) for the development of the automatic weather stations, technical support, and raw data processing. Part of the observational data was obtained from “MeteoClimatological Observatory at Mario Zucchelli Station and Victoria Land” of PNRA (http://www.climantartide.it) and the Australian Antarctic Division Glaciology Program. The authors appreciate the support of the University of Wisconsin–Madison Automatic Weather Station Program for the dataset, data display, and information (NSF Grant ANT-1543305).
APPENDIX
Information about the Observations
A detailed overview of the ground-based observational stations, including their height and temporal coverage, is given in Table A1, binned by elevation class. Detailed information about data from the ground-based AWS observations is given in Table A2.
Overview of the ground-based observational data availability per elevation class (C = coast, I = interior, and P = peninsula) of near-surface temperature (label T) and 10-m wind speed (label WS) for the period 1984–2016. AUS denotes the Australian Antarctic AWS dataset (http://aws.acecrc.org.au/), ITA is the Italian Antarctic Research Program (http://www.climantartide.it), and USAP/SCAR is the meteorological data obtained from the University of Wisconsin–Madison and the British Antarctic Survey (Turner et al. 2004). Here, MAM indicates March–May and SON is September–November.
Overview of the ground-based radiative flux and relative humidity observational data availability (label “Rad”). All data are retrieved from the IMAU Antarctic AWS Project (https://www.projects.science.uu.nl/iceclimate/aws/antarctica.php) and the AMRC AWS network (https://amrc.ssec.wisc.edu/data/archiveaws.html).
REFERENCES
Albergel, C., E. Dutra, S. Munier, J. C. Calvet, J. Munoz-Sabater, P. de Rosnay, and G. Balsamo, 2018: ERA-5 and ERA-Interim driven ISBA land surface model simulations: Which one performs better? Hydrol. Earth Syst. Sci., 22, 3515–3532, https://doi.org/10.5194/hess-22-3515-2018.
Boening, C., M. Lebsock, F. Landerer, and G. Stephens, 2012: Snowfall-driven mass change on the East Antarctic Ice Sheet. Geophys. Res. Lett., 39, L21501, https://doi.org/10.1029/2012GL053316.
Bracegirdle, T. J., and G. J. Marshall, 2012: The reliability of Antarctic tropospheric pressure and temperature in the latest global reanalyses. J. Climate, 25, 7138–7146, https://doi.org/10.1175/JCLI-D-11-00685.1.
Bromwich, D. H., 1988: Snowfall in high southern latitudes. Rev. Geophys., 26, 149–168, https://doi.org/10.1029/RG026i001p00149.
Bromwich, D. H., J. P. Nicolas, and A. J. Monaghan, 2011: An assessment of precipitation changes over Antarctica and the Southern Ocean since 1989 in contemporary global reanalyses. J. Climate, 24, 4189–4209, https://doi.org/10.1175/2011JCLI4074.1.
Bromwich, D. H., and Coauthors, 2012: Tropospheric clouds in Antarctica. Rev. Geophys., 50, RG1004, https://doi.org/10.1029/2011RG000363.
Chen, J. L., C. R. Wilson, and B. D. Tapley, 2011: Interannual variability of Greenland ice losses from satellite gravimetry. J. Geophys. Res., 116, B07406, https://doi.org/10.1029/2010JB007789.
Costanza, C. A., M. A. Lazzara, L. M. Keller, and J. J. Cassano, 2016: The surface climatology of the Ross Ice Shelf, Antarctica. Int. J. Climatol., 36, 4929–4941, https://doi.org/10.1002/joc.4681.
Dee, D. P., and Coauthors, 2011: The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553–597, https://doi.org/10.1002/qj.828.
Dutra, E., G. Balsamo, P. Viterbo, P. M. A. Miranda, A. Beljaars, C. Schär, and K. Elder, 2010: An improved snow scheme for the ECMWF land surface model: Description and offline validation. J. Hydrometeor., 11, 899–916, https://doi.org/10.1175/2010JHM1249.1.
Favier, V., and Coauthors, 2013: An updated and quality controlled surface mass balance dataset for Antarctica. Cryosphere, 7, 583–597, https://doi.org/10.5194/tc-7-583-2013.
Fréville, H., E. Brun, G. Picard, N. Tatarinova, L. Arnaud, C. Lanconelli, C. Reijmer, and M. van den Broeke, 2014: Using MODIS land surface temperatures and the Crocus snow model to understand the warm bias of ERA-Interim reanalyses at the surface in Antarctica. Cryosphere, 8, 1361–1373, https://doi.org/10.5194/tc-8-1361-2014.
Gorodetskaya, I. V., N. P. M. van Lipzig, M. R. van den Broeke, A. Mangold, W. Boot, and C. H. Reijmer, 2013: Meteorological regimes and accumulation patterns at Utsteinen, Dronning Maud Land, East Antarctica: Analysis of two contrasting years. J. Geophys. Res. Atmos., 118, 1700–1715, https://doi.org/10.1002/jgrd.50177.
Gorodetskaya, I. V., M. Tsukernik, K. Claes, M. Ralph, W. Neff, and N. P. M. van Lipzig, 2014: The role of atmospheric rivers in anomalous snow accumulation in East Antarctica. Geophys. Res. Lett., 41, 6199–6206, https://doi.org/10.1002/2014GL060881.
Groh, A., and M. Horwath, 2016: The method of tailored sensitivity kernels for GRACE mass change estimates. Geophysical Research Abstracts, Vol. 18, Abstract 12065.
Hersbach, H., and D. Dee, 2016: ERA5 reanalysis is in production. ECMWF Newsletter, No. 147, ECMWF, Reading, United Kingdom, 7, https://www.ecmwf.int/en/newsletter/147/news/era5-reanalysis-production.
Hodges, K. I., R. W. Lee, and L. Bengtsson, 2011: A comparison of extratropical cyclones in recent reanalyses ERA-Interim, NASA MERRA, NCEP CFSR, and JRA-25. J. Climate, 24, 4888–4906, https://doi.org/10.1175/2011JCLI4097.1.
Jones, P. D., and D. H. Lister, 2015: Antarctic near-surface air temperatures compared with ERA-Interim values since 1979. Int. J. Climatol., 35, 1354–1366, https://doi.org/10.1002/joc.4061.
Lazzara, M. A., G. A. Weidner, L. M. Keller, J. E. Thom, and J. J. Cassano, 2012: Antarctic Automatic Weather Station Program: 30 years of polar observation. Bull. Amer. Meteor. Soc., 93, 1519–1537, https://doi.org/10.1175/BAMS-D-11-00015.1.
Lenaerts, J. T. M., M. R. van den Broeke, W. J. van de Berg, E. van Meijgaard, and P. Kuipers Munneke, 2012: A new, high-resolution surface mass balance map of Antarctica (1979–2010) based on regional atmospheric climate modeling. Geophys. Res. Lett., 39, L04501, https://doi.org/10.1029/2011GL050713.
Lenaerts, J. T. M., and Coauthors, 2017: Meltwater produced by wind–albedo interaction stored in an East Antarctic ice shelf. Nat. Climate Change, 7, 58–63, https://doi.org/10.1038/nclimate3180.
Medley, B., and Coauthors, 2013: Airborne-radar and ice-core observations of annual snow accumulation over Thwaites Glacier, West Antarctica confirm the spatiotemporal variability of global and regional atmospheric models. Geophys. Res. Lett., 40, 3649–3654, https://doi.org/10.1002/grl.50706.
Medley, B., and Coauthors, 2014: Constraining the recent mass balance of Pine Island and Thwaites glaciers, West Antarctica, with airborne observations of snow accumulation. Cryosphere, 8, 1375–1392, https://doi.org/10.5194/tc-8-1375-2014.
Miller, D. J., Z. Zhang, S. E. Platnick, A. S. Ackerman, F. Werner, C. Cornet, and K. Knobelspiesse, 2018: Comparisons of bispectral and polarimetric cloud microphysical retrievals using a LES–satellite retrieval simulator. Atmos. Meas. Tech., 11, 3689–3715, https://doi.org/10.5194/amt-11-3689-2018.
Nicolas, J. P., and D. H. Bromwich, 2011: Climate of West Antarctica and influence of marine air intrusions. J. Climate, 24, 49–67, https://doi.org/10.1175/2010JCLI3522.1.
Nicolas, J. P., and D. H. Bromwich, 2014: New reconstruction of Antarctic near-surface temperatures: Multidecadal trends and reliability of global reanalyses. J. Climate, 27, 8070–8093, https://doi.org/10.1175/JCLI-D-13-00733.1.
Nwogu, E., I. Iwueze, and V. Nlebedim, 2016: Some tests for seasonality in time series data. J. Mod. Appl. Stat. Methods, 15, 382–399, https://doi.org/10.22237/jmasm/1478002920.
Rienecker, M. M., and Coauthors, 2011: MERRA: NASA’s Modern-Era Retrospective Analysis for Research and Applications. J. Climate, 24, 3624–3648, https://doi.org/10.1175/JCLI-D-11-00015.1.
Rignot, E., I. Velicogna, M. R. van den Broeke, A. Monaghan, and J. T. M. Lenaerts, 2011: Acceleration of the contribution of the Greenland and Antarctic Ice Sheets to sea level rise. Geophys. Res. Lett., 38, L05503, https://doi.org/10.1029/2011GL046583.
Saha, S., and Coauthors, 2010: The NCEP Climate Forecast System Reanalysis. Bull. Amer. Meteor. Soc., 91, 1015–1058, https://doi.org/10.1175/2010BAMS3001.1.
Sanz Rodrigo, J., 2011: On Antarctic wind engineering. Ph.D. thesis, Université Libre de Bruxelles, 218 pp.
Sanz Rodrigo, J., J. Buchlin, J. van Beeck, J. Lenaerts, and M. van den Broeke, 2013: Evaluation of the Antarctic surface wind climate from ERA reanalyses and RACMO2/ANT simulations based on automatic weather stations. Climate Dyn., 40, 353–376, https://doi.org/10.1007/s00382-012-1396-y.
Shepherd, A., and Coauthors, 2012: A reconciled estimate of ice-sheet mass balance. Science, 338, 1183–1189, https://doi.org/10.1126/science.1228102.
Tapley, B. D., S. Bettadpur, J. C. Ries, P. F. Thompson, and M. M. Watkins, 2004: GRACE measurements of mass variability in the Earth system. Science, 305, 503–505, https://doi.org/10.1126/science.1099192.
Turner, J., and Coauthors, 2004: The SCAR READER project: Toward a high-quality database of mean Antarctic meteorological observations. J. Climate, 17, 2890–2898, https://doi.org/10.1175/1520-0442(2004)017<2890:TSRPTA>2.0.CO;2.
van den Broeke, M. R., and N. P. M. van Lipzig, 2003: Factors controlling the near-surface wind field in Antarctica. Mon. Wea. Rev., 131, 733–743, https://doi.org/10.1175/1520-0493(2003)131<0733:FCTNSW>2.0.CO;2.
van den Broeke, M. R., C. H. Reijmer, and R. S. van de Wal, 2004: A study of the surface mass balance in Dronning Maud Land, Antarctica, using automatic weather stations. J. Glaciol., 50, 565–582, https://doi.org/10.3189/172756504781829756.
van Lipzig, N., E. van Meijgaard, and J. Oerlemans, 1999: Evaluation of a regional atmospheric model using measurements of surface heat exchange processes from a site in Antarctica. Mon. Wea. Rev., 127, 1994–2011, https://doi.org/10.1175/1520-0493(1999)127<1994:EOARAM>2.0.CO;2.
Wang, W., P. Xie, S.-H. Yoo, Y. Xue, A. Kumar, and X. Wu, 2011: An assessment of the surface climate in the NCEP Climate Forecast System Reanalysis. Climate Dyn., 37, 1601–1620, https://doi.org/10.1007/s00382-010-0935-7.
Wang, Y., D. Zhou, A. Bunde, and S. Havlin, 2016: Testing reanalysis data sets in Antarctica: Trends, persistence properties, and trend significance. J. Geophys. Res. Atmos., 121, 12 839–12 855, https://doi.org/10.1002/2016JD024864.
Wille, J. D., and Coauthors, 2016: Evaluation of the AMPS boundary layer simulations on the Ross Ice Shelf with tower observations. J. Appl. Meteor. Climatol., 55, 2349–2367, https://doi.org/10.1175/JAMC-D-16-0032.1.
Wille, J. D., and Coauthors, 2017: Evaluation of the AMPS boundary layer simulations on the Ross Ice Shelf, Antarctica, with unmanned aircraft observations. J. Appl. Meteor. Climatol., 56, 2239–2258, https://doi.org/10.1175/JAMC-D-16-0339.1.
Zwally, H. J., M. B. Giovinetto, M. A. Beckley, and J. L. Saba, 2012: Antarctic and Greenland Drainage Systems. NASA, accessed 10 May 2019, https://icesat4.gsfc.nasa.gov/cryo_data/ant_grn_drainage_systems.php.
