The Arctic surface energy budget as simulated with the IPCC AR4 AOGCMs
- First Online:
- Cite this article as:
- Sorteberg, A., Kattsov, V., Walsh, J.E. et al. Clim Dyn (2007) 29: 131. doi:10.1007/s00382-006-0222-9
- 173 Views
Ensembles of simulations of the twentieth- and twentyfirst-century climate, performed with 20 coupled models for the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment, provide the basis for an evaluation of the Arctic (70°–90°N) surface energy budget. While the various observational sources used for validation contain differences among themselves, some model biases and across-model differences emerge. For all energy budget components in the twentieth-century simulations (the 20C3M simulation), the across-model variance and the differences from observational estimates are largest in the marginal ice zone (Barents, Kara, Chukchi Seas). Both downward and upward longwave radiation at the surface are underestimated in winter by many models, and the ensenmble mean annual net surface energy loss by longwave radiation is 35 W/m2, which is less than for the NCEP and ERA40 reanalyses but in line with some of the satellite estimates. Incoming solar radiation is overestimated by the models in spring and underestimated in summer and autumn. The ensemble mean annual net surface energy gain by shortwave radiation is 39 W/m2, which is slightly less than for the observational based estimates, In the twentyfirst-century simulations driven by the SRES A2 scenario, increased concentrations of greenhouse gasses increase (average for 2080–2100 minus average for 1980–2000 averages) the annual average ensemble mean downward longwave radiation by 30.1 W/m2. This was partly counteracted by a 10.7 W/m2 reduction in downward shortwave radiation. Enhanced sea ice melt and increased surface temperatures increase the annual surface upward longwave radiation by 27.1 W/m2 and reduce the upward shortwave radiation by 13.2 W/m2, giving an annual net (shortwave plus longwave) surface radiation increase of 5.8 W/m2 , with the maximum changes in summer. The increase in net surface radiation is largely offset by an increased energy loss of 4.4 W/m2 by the turbulent fluxes.
Future climate change simulations show enhanced climate sensitivity at high latitudes, where there is also the largest spread among different models (IPCC 2001; ACIA 2005; Randall et al. 1998). The surface energy balance is an essential element of the climate and constitutes an important part of the energy available for melting/freezing the sea ice and warming/cooling the surface. For example, Fletcher (1965) argues that an advance of the onset of sea ice melt by only one week in June would result in an additional melt of 0.5–1.0 m of sea ice.
On one hand, due to the Arctic’s low moisture content, changes in CO2 and other greenhouse gasses have the potential to become more important in the Arctic than at lower latitudes. On the other hand the impact of changes in infrared absorbers depends on the vertical tropospheric temperature gradient which is small in the Arctic and therefore the impact of greenhouse gas changes could be smaller. In addition to this, the complex interactions between the atmosphere, ocean and cryosphere give rise to a variety of climate feedbacks, with the ice/snow albedo-temperature feedback (Budyko 1969) being an important factor. In a simplified climate system, the strength of the ice-albedo feedback is a function of the sea ice extent (Budyko 1969). The strength of the albedo-temperature feedback, however, is a complicated function of the initial extent of the sea ice and the responses of the horizontal energy and moisture transports, as well as clouds (Held and Suarrez 1974, Hartmann 1994; Vavrus 2003; Björk and Söderkvist 2002; Beesley 2000) to the changes in greenhouse gases. Clouds play an especially important role in arctic feedbacks because their radiative impacts are large in the solar and longwave portions of the spectrum, and these impacts depend strongly on cloud height, thickness, and hydrometeor type (liquid or ice), concentration and size. The recent changes in the permafrost (Romanovsky et al. 2002), snow cover (Frei and Robinson 1999), glaciers (ACIA 2005; Dyurgerov and Meier 1997), sea ice (Vinnikov et al. 1999), temperature (ACIA 2005; Serreze et al. 2000) and precipitation (Kattsov and Walsh 2000) show a consistent picture of an Arctic climate in rapid change. However, Arctic climate is highly variable and the causes of the changes are still debated (e.g. Polyakov et al. 2003; McBean 2005). Credible model simulations are important in attributing the changes to a cause. In addition, models are the main tool in developing physically plausible climate change scenarios, given prescribed scenarios of future greenhouse gasses and aerosol loadings.
In the present paper, we first evaluate the models ability to simulate the different energy terms for the present climate. Twenty global coupled (atmosphere–ocean–ice) climate models are compared to five observationally based estimates. The motivation for this evaluation is that a realistic simulation of the present Arctic climate may be a necessary (but not sufficient) condition for a successful simulation of future global climate. Secondly we assess to which extent projected changes in greenhouse gases and aerosols may affect the surface energy budget of the Arctic. Special emphasis is placed on the behavior of the modeled ensemble mean and the spread among the different models.
2 Models and data
2.1 The coupled models
List of models that participate in this study
Bjerknes Centre for Climate Research, University of Bergen Norway
Canadian Centre for Climate Modeling & Analysis, Canada
Meteo-France/Centre National de Recherches Météorologique, France
CSIRO Atmospheric Research, Australia
NOAA/Geophysical Fluid Dynamics Laboratory, USA
2.0° × 2.5° L24
NOAA/Geophysical Fluid Dynamics Laboratory, USA
2.0° × 2.5° L24
NASA/Goddard Institute for Space Studies, USA
3° × 4° L12
NASA/Goddard Institute for Space Studies, USA
4° × 5° L20
NASA/Goddard Institute for Space Studies, USA
4° × 5° L20
LASG/Institute of Atmospheric Physics, China
Institute for Numerical Mathematics, Russia
4° × 5° L21
Institute Pierre Simon Laplace, France
2.5° × 3.75° L19
Center for Climate System Research, National Institute for Environmental Studies, and Frontier Research Center for Global Change, Japan
Center for Climate System Research, National Institute for Environmental Studies, and Frontier Research Center for Global Change, Japan
Max Planck Institute for Meteorology, Germany
Meteorological Research Institute, Japan
National Center for Atmospheric Research, USA
National Center for Atmospheric Research, USA
Hadley Centre for Climate Prediction and Research/Met Office, UK
2.5° × 3.8° L19
Hadley Centre for Climate Prediction and Research/Met Office, UK
∼1.3° × 1.9° L38
In this study we use several groups of the archived simulations. For comparison against observed estimates we use the 20C3M simulations, which span the period starting not later than 1901 and ending not earlier than 1999. These simulations are forced with observed aerosol loadings and greenhouse gas concentrations. 20C3M simulations with some of the models include natural forcings such as solar variability and volcanic eruptions. In addition, the indirect effects of aerosols are only taken into account in a few of the models. In the sections discussing the twentyfirst-century simulations, the projected changes in greenhouse gases are taken from the Special Report on Emission Scenarios (SRES; Nakicenovic et al. 2000). Changes are calculated as differences between the 2080–2099 mean for the SRES A2 scenario and the 1980–1999 mean in the 20C3M simulation. The CO2 level in the SREAS A2 scenario increases to around 800 ppm by the late twentyfirst century (IPCC 2001).
Simulations with some of the models include several ensemble members started from different initial conditions. In this study, the entire ensembles were used only in an analysis of the twentieth century trends and variability in the 20C3M simulations. Otherwise, whenever more than one simulation was available, only the first members of the ensembles were included in the analysis.
The scenario simulations for the twentyfirst century were not available for some IPCC AR4 models, thus different subsets of the models are used in twentyfirst century estimates discussed in Sect. 4.
The reference area used in this study is 70–90°N. Area-averaged values are calculated by using the original grid of the individual models and selecting the grid squares within the chosen region and weighting the individual grid-squares by their areas. For the spatial maps of the multimodel mean and their spread, all models are interpolated into a 2.5 × 2.5° grid using Cressmann interpolation, where the weights are reduced exponentially with distance to the point on the 2.5° grid. Only grid points 600 km or less from the 2.5° grid point are used in the interpolation. The intermodel standard deviation (STD) is used as a measure of the level of agreement between the different models. Assuming that the model estimates are Gaussian distributed, 95% of the distribution is within ±2STD of the mean.
2.2 Observationally based estimates and reanalysis
With the exception of the Russian measurements made from drifting ice stations during the early 1950s through 1991, in situ observations of the different terms in the energy budget are rare and usually only available for a limited region during short-term intensive field campaigns. In this study we use observationally based estimates that depict spatial variability over the whole Arctic concurrently. We use five different observational databases to gain some insight into the uncertainty related to the different methods of observational analysis. Two of these databases are based on a state of the art data assimilation procedure used in numerical weather prediction and three are based on satellite estimates.
The ECMWF (ERA40) and the NCAR-NCEP reanalyses are both based on a three-dimensional variational assimilation of observations (Simmons and Gibson 2000; Uppala et al. 2005; Kalnay et al. 1996), but with no direct assimilation of radiative fluxes. Conventional data comes from a wide selection of sources starting with 1958 (the International Geophysical Year) and 1948, respectively. Here, we focus on the data from the last part of the century (after 1980) when TOVS satellite data and Cloud Motion Winds were used in the assimilation.
The third and fourth datasets are two versions of the surface radiation budget based on the International Satellite Cloud Climatology Project (ISCCP; Rossow and Schiffer 1991): Version 2 of the Surface Radiation Budget (SRB) and the Version 1 polar radiation fluxes (POLAR ISCCP; Key et al. 1999). The inputs for the SRB data (1983–1995) are from different satellite sources. Cloud data was taken from the DX data of the ISCCP, which provides top of atmosphere (TOA) narrowband radiances, atmospheric soundings, and cloud information. ERBE measurements provided TOA broadband clear-sky albedos. Atmospheric water vapor is taken from a 4-D data assimilation product provided by the Data Assimilation Office at NASA GSFC and were produced with the Goddard Earth Observing System model version 1 (GEOS-1). Ozone is taken from the Total Ozone Mapping Spectrometer (TOMS). The general approach was to use the ISCCP DX data supplemented by the ERBE results as input to the SRB satellite algorithms to estimate the various surface parameters. The shortwave components of the surface radiative fluxes were computed with a broadband radiative transfer model (Pinker and Laszlo 1992) and the longwave component using the Fu–Liou Model (Fu et al. 1997).
The POLAR ISCCP radiation terms (1985–1993) were calculated by training a neural net (a special implementation of Fluxnet, cf. Key and Schweiger 1998) with a small subset of the available ISCCP-D1 data. Fluxes were generated by the Streamer radiative transfer model (Key and Schweiger 1998). When available, a more accurate set of atmospheric temperature and water vapor profiles from the TOVS Pathfinder Path-P data set were used in place of the ISCCP profiles. A more detailed description is given in Key et al. (1999).
The fifth database is the Version 1 of the Extended Advanced Very High Resolution Radiometer (AVHRR) Polar Pathfinder dataset (APP-X), spans the period from 1985 to 1993. The Extended APP dataset is an extension of the standard clear sky products (Maslanik et al. 2001; Maslanik et al. 1998; Meier et al. 1997) using the Cloud and Surface Parameter Retrieval (CASPR) system (Key 2001). The calculation of cloudy sky surface skin temperature was based on an empirical relationship between the clear sky surface skin temperature, wind speed, and solar zenith angle (daytime). The cloudy sky broadband surface albedo is determined using the clear sky broadband albedo (interpolated from nearby pixels) adjusted by the APP cloud optical depth and the solar zenith angle. The all-sky radiative fluxes were computed in CASPR using FluxNet (Key and Schweiger 1998). Key (2001) and references therein provide more information on the algorithms and their validation. The APP-X data are available for the local solar times 1400 and 0400 hours. For the longwave components the two times were averaged to obtain values representative of the full day. No attempt was made to calculate the full-day shortwave components.
3 Simulations of the twentieth century
3.1 Longwave radiation
The main factor that determines the annual mean and seasonal cycle of the upwelling longwave radiation (LW) terms is the surface temperature. The primary determinants of the downwelling surface LW radiation are the boundary layer humidity and temperature; its stratification; and the amount and optical properties of clouds. LW radiation transfer in high latitudes is somewhat different from the lower latitudes. Due to the small amount of water vapour the opacity of the water vapour rotation band is smaller; also, the lower temperatures shift the maximum blackbody intensity to lower frequencies and therefore towards the low-frequency rotational band of water vapor (Stamnes et al. 1999). Zhang et al. 1997 showed that for clear sky the downwelling LW radiation reaching the surface comes from a very shallow layer of the atmosphere (90% of the accumulated contribution comes from the lowest 500–1,000 m of the atmosphere). Thus, high vertical resolution in the boundary layer may be required in order to capture both the annual mean and especially the seasonal cycle of this element, making it a challenging task for climate models. A detailed analysis of the impact of water vapor, atmospheric temperatures and stratification on the LW radiation can be found in Curry et al. (1995) and Zhang et al. (1997).
3.1.1 Downward longwave radiation
The seasonal cycle in the downward LW multimodel ensemble mean (Fig. 2b) is within the observational estimates during all months. However, some models tend to underestimate the December-April downward LW radiation as a surface energy source. A possible explanation for this is the insufficient vertical resolution of AOGCMs which may prevent a correct buildup of deep wintertime surface inversions (Byrkjedal et al. 2006). It should be noted that the strength of the seasonal cycle differs substantially among the different observational estimates, and there is a 30 W/m2 difference between the NCEP and ERA40 estimates in mid-summer. This is of the same magnitude as in a recently conducted comparison of the downward radiative fluxes in different datasets over the SHEBA site (Liu et al. 2005). Liu et al. found that the ERA40 and AVHRR-based estimates (quite similar to the APP-X dataset used here) describe the seasonal cycle of downward LW radiation quite well, and that an ISCCP-derived estimate (using the same cloud data, but another radiative transfer code than was used for the estimates given here) overestimate the wintertime downward LW flux and underestimate the summertime flux, resulting in a seasonal cycle that is too weak. The shape of the seasonal cycle reported by Lindsay (1998) for the Arctic pack ice using the NP-stations is also quite similar to the ERA40 estimates. These studies indicate that the seasonal cycle in the downward component may be more realistically represented by the ERA40 and the AVHRR based APP-X datasets. However, it should be noted that the ERA40 assimilates the SHEBA radiosondes and the good quality of the ERA40 estimates over this site may therefore lead to overconfidence in the ability of ERA40 to capture the entire arctic region.
3.1.2 Upward longwave radiation
Averaged over all five observational estimates, the annual mean upward LW flux averaged over 70–90°N is 258.4 W/m2, which is 5.0 W/m2 larger than the IPCC models’ ensemble mean (Fig. 2c), for which the across model spread (±1STD) is 13.7 W/m2. This spread is comparable to the spread in the LW downward component and is linked to the state of the sea ice and its impacts on the mean arctic surface temperature. A comparison of the individual models’ mean annual mean ice fractions and the upward LW components shows that models with a large sea ice fraction tend to have smaller upward LW radiation (correlation r = −0.68 (p = 0.02), with the IAP model removed). As with the downward component, there is a negative bias (too large an energy loss from the surface) over the Barents Sea area (20–25 W/m2 ), related to the positive biases in sea ice fraction in this region. The correlation between the individual model’s mean Barents Sea (15–65°E and 70–85°N) annual upward LW radiation and the annual mean Barents Sea ice fraction is −0.65 (p = 0.04). There is a quite large spatial discrepancy among the different observational estimates. Thus, the spatial pattern of the ensemble mean model errors is not easy to evaluate.
All observational estimates show a fairly similar seasonal cycle of upward LW radiation (Fig. 2d), although the monthly values have a spread of 10–20 W/m2. Several of the models underestimate the wintertime energy loss by 10–40 W/m2 indicating that the models have a cold surface temperature bias.
3.1.3 Net longwave radiation
As a consequence of the models’ biases in the upward and downward components, the ensemble mean net LW radiation is overestimated (the LW radiation heat sink is too small) compared to the reanalyses and in line with the satellite measurements. The across-model spread in the models is 5.5 W/m2 with a tendency for models having a large annual cloud fraction to have the smallest energy loss (a non-significant correlation of 0.36). However, the seasonal cycle of net LW radiation is the difference between two large terms and is not well known. This is an element that historically has been measured only rarely and our knowledge is therefore to a large extent based on simulations and regional field campaigns. As seen in Fig. 2f the observational estimates diverge and there is no consensus on the seasonal cycle. The two ISCCP-based estimates show the strongest LW energy loss in summer (35–45 W/m2), while the ERA40, NCEP and the AVHRR-based APP-X datasets indicate the largest loss in early spring (40–65 W/m2). Most of the models indicate a seasonal cycle in the net LW radiation similar to the ERA40, NCEP and APP-X data, and the models ensemble mean follows the APP-X dataset closely. There seems to be a tendency for many models to underestimate the summertime surface energy loss, and there is a clear relationship between summertime cloud cover and net LW radiation: models having a large cloud cover show the smallest surface energy loss (correlation r = 0.69). The summertime LW energy loss is also related to the sea ice fraction which strongly influences the surface temperatures. Models having a large sea ice fraction generally have larger LW surface energy loss (correlation r = −0.48).
It should be noted that the relationships between the downward radiation components and cloud cover should not be taken as the direct influence of the cloud cover as the correlations do not give any causal relationships. Cloud cover changes are related to changes in both heat and moisture transport which, in addition to changing the cloud cover, may change the atmospheric temperatures and water vapor content, Consequently, it is difficult to distinguish between the direct influence of cloud fraction and the influence of atmospheric water vapor and temperature, which may co-vary with the cloud cover fraction and therefore lead to too strong statistical cloud–radiation relationships.
3.2 Shortwave radiation
The incoming surface solar radiation is, relatively speaking, well documented in the Arctic. Comprehensive information on the seasonal cycle and spatial distribution can be found in a variety of studies in both the Russian (Mashunova 1961; Mashunova and Chernigovskii 1971; Atlas Arktiki 1985; Krohl 1992; see Przybylak 2003 for an excellent review of these findings) and English (Fletcher 1961; Vowinckel and Orvig 1964; 1970; McKay and Morris 1985; Serreze et al. 1998) literature. The annual mean and seasonal cycle are determined by the length of the day which gives zero direct-beam flux at the North Pole from the autumnal to spring equinoxes. The annual means of the downward fluxes have a latitudinal gradient, which is modified by the occurrence of topography, clouds and their optical properties such as liquid water content, number of droplets and their size. An overview of the topic is given by Curry and Ebert (1992), Curry et al. (1993, 1996) and Zhang et al. (1996).
The outgoing surface solar radiation is largely determined by the surface albedo and the amount of downward radiation.
The spatial pattern of the annual upward component (Fig. 3b) show central Arctic values of 45–55 W/m2 and an across-model spread that is slightly smaller than in the downward component.
3.2.1 Downward shortwave radiation
Because the biases in the model ensemble mean change when different observational estimates are used, it is hard to detect any regions of strong annual biases in the downward component. Compared to the ERA40 reanalysis, the models show an underestimation of downward SW radiation of 5–10 W/m2 over the central Arctic, while for the same area there is an overestimation of 5–10 W/m2 when compared to the ISCCP-based estimates.
The spread in the models’ summertime maximum (June) is over 100 W/m2 (Fig. 4b). This spread does not seem to be related to the models different cloud fraction. Compared to three of the four observational estimates there is a tendency of the models to overestimate the incoming SW radiation in spring (March–May) and underestimate the radiation in summer and autumn (June–September). The spring overestimation in the ensemble mean has a peak in April and May (5–30 W/m2 compared to the different observational estimates) while the summer/autumn underestimation is greatest in July (10–35 W/m2). The models’ springtime downward SW radiation is related to the model’s cloud fraction, with models having a larger cloud fraction giving less surface downward radiation (the MAM correlation is −0.41, p = 0.07). The relationship between cloud cover and summertime radiation is less clear (the correlation of −0.34 which is reduced to −0.03 when the IAP model was removed, is not statistically significant). As linkage between the model’s cloud fraction and downward SW radiation is not very strong and the seasonal cloud cover of the Arctic is not well known, it is difficult to conclude that the model’s spring and autumn biases are related to biases in the seasonal cloud cover fraction. This does not exclude any possible relationships between cloud thickness etc. and downward SW radiation which cannot be rigorously investigated with the IPCC model database.
It should also be noted that the June maximum in ERA40, and ISCCP-based estimates given here is 40–50 W/m2 smaller than the estimates for the pack ice obtained using NP-station data by Lindsay (1998) and the Artic Ocean averages of Ebert and Curry (1993). Around half of the bias can be explained by the larger area chosen here (including the cloudier Greenland and Barents Sea region). A possible explanation for the remaining bias may be the different time periods. The estimates used here are averages from the last two decades, while the NP-station estimates are based on data from the late 1950s to the beginning of the 1990s. The reported increase in spring and summer cloudiness over the last decades (Wang and Key 2003) may therefore contribute to some of the discrepancy and the time evolution of both the ERA40 and NCEP reanalyses shows large trends in the downward SW component (see Sect. 3.4).
3.2.2 Upward shortwave radiation
For the upward SW component the biases are more apparent. Averaged over the Arctic area, the model ensemble mean overestimates the upward SW radiation compared to three of the four observational estimates (Fig. 4c). The across-model spread (±1STD) is 7.3 W/m2. Much of this overestimation comes from the Barents and Greenland Sea area (15–25 W/m2), indicating a tendency for the models to overestimate the sea ice extent in this area (Arzel et al. 2006). The bias extends over the land areas of the eastern and western Arctic, implying that is associated with positive biases in the seasonal snow cover.
We expect the amount of upward SW radiation to be tightly linked to the sea ice fraction through the surface , and the annual mean sea ice fraction correlates well with the annual mean upward SW radiation (0.50, p = 0.09, with the IAP model excluded). There is no unanimity among the different observational estimates on the month of maximum SW radiation (Fig. 4d). Most models show a maximum in May, which corresponds well with the ERA40 reanalysis. There is a tendency among the models to overestimate the May maximum (compared to three of the observational estimates), consistent with the biases in the downward component that seem to be dependent on the cloud fraction for the individual models (see Sect. 3.2.1). Generally the seasonal pattern in the differences between simulated and observed upward radiation follows the pattern for the downward component, but with no clear underestimation in the summer/autumn radiation, indicating that the bias in summer/autumn downward SW radiation is counteracted by positive biases in the models’ summertime surface albedo. This albedo bias is likely linked to the extensive sea ice extent in many models (Arzel et al. 2006)
3.2.3 Net shortwave radiation
The model ensemble mean of the net SW radiation, a net energy source to the Arctic surface, is underestimated compared to all the observational estimates by 7.1 W/m2 compared to the mean of the observations (Fig. 4e). The range among the different models is comparable to that of the net LW radiation, as is the across model spread (6.1 W/m2). The net SW radiation underestimation is seen during all months (Fig. 4f) with a maximum in summer of 15–30 W/m2 compared to the different observational estimates.
3.3 Turbulent fluxes
Even more so than the radiative components, knowledge of turbulent fluxes is very limited, and only a few attempts have been made to produce spatial maps of these components (Khrol 1992 reproduced in Przybylak 2003). According to these maps the annual sensible heat fluxes are a modest surface heat source over the Arctic Ocean covered with perennial ice. Lindsay (1998) used the NP-station data to estimate the annual mean sensible heat flux over the ice pack to be a surface heat source (a downward flux) of around 3 W/m2 using the NP-station data. According to Khrol (1992) the sensible heat is a substantial surface energy sink over the eastern part of the Greenland Sea and the Barents Sea. With exception of the large fluxes over the East Greenland Current reported by Khrol, the ERA40 values are very similar. In contrast to the sensible heat flux, the ERA40 estimates show the latent heat flux to be a surface energy sink over the central Arctic Ocean. This is in accordance with the estimates of Lindsay (1998), who showed the annual average latent heat flux to be 2.3 W/m2 (positive upward) over the Arctic ice pack.
3.3.1 Sensible and latent heat flux
Latent heat fluxes are systematically underestimated in the IPCC ensemble mean during the entire year compared to ERA40 and during most of the year compared to NCEP. The underestimation is related to the tendency of the models to have excessive ice cover over the Barents and Greenland Seas.
3.4 Evolution of fluxes through the twentieth century
One of the most intriguing features of the arctic climate evolution in the twentieth century was the warming observed in the Arctic in 1930–1940s, with a magnitude comparable to the warming during last few decades (McBean 2005). Recently, the two warming periods in the Arctic, their relative magnitudes, possible causes, and implications for credibility of the state-of-the-art projections of the future arctic (and global) climate have been widely discussed in the literature (Polyakov et al. 2003; Bengtsson et al. 2004; Johannessen et al. 2004; Overland et al. 2004; McBean 2005; Wang et al. 2006). While the early twentieth century arctic warming is often attributed to the unforced natural variability of the high-latitude climate, there is no consensus on the relative importance of the increasing anthropogenic forcing for the late twentieth century arctic warming. From this point of view, multi-member ensemble simulations of the twentieth-century climate have a potential to provide a better insight into the problem.
Wang et al. (2006) analyzed the IPCC AR4 20C3M simulations of the land surface air temperature in the Arctic, and hypothesized in particular that if mid-century warm anomalies are based on intrinsic atmospheric variability, then models should not necessarily reproduce warm events in the same years as the observed warming, but they should simulate the same variability and reproduce trends associated with external forcing. Some of the IPCC AR4 models were only forced with observed atmospheric concentrations of greenhouse gases, while others included time-varying natural forcings (e.g. volcanic and solar effects). Wang et al. (2006) found the inclusion of natural forcings to be of minor importance relative to a model’s ability to reproduce the timing of the early twentieth century arctic warming, while a robust feature of model responses to the anthropogenic greenhouse gas concentrations increase were positive temperature trends both over the entire twentieth century and its last decades.
One of the foci of our study, which uses essentially the same set of climate models as Wang et al. (2006), was the evolution of the energy balance components in the Arctic region through the entire twentieth century. While the lack of observational data prevented us from directly establishing the validity of the simulated radiation and turbulent fluxes in the Arctic for the entire past century, we tried to identify common features and differences in the behavior of the models in the 20C3M experiment. Another focus was an estimation of the connections between different arctic energy budget components and the surface air temperature, whose behavior in the twentieth century is known better from the observational record.
Linear trends for the 1958–1998 period using the ERA40, NCEP reanalysis and the ensemble mean of the IPCC models
Linear trends (W/m2 per decade)
1.37 [0.25 2.69]
1.26 [0.16 2.60]
LW↓ − LW↑
0.13 [−0.03 0.24]
−0.45 [−0.88 0.03]
−0.60 [−1.34 0.06]
SW↓ − SW↑
0.15 [−0.19 0.58]
For models running an ensemble of simulations for the twentieth century, a notable feature of the evolution of the radiation budget components in each 20C3M simulation is the high similarity between LW radiation variations, especially in recent decades. A typical example is given by the two ensemble members of PCM model, for which each ensemble member shows two distinct periods of increased downward and upward LW radiation in the twentieth century, to a certain extent consistent with surface air temperature records. However, the early twentieth century maxima obtained in the two ensemble members have different shapes and are shifted in time relative to each other. It is noteworthy that PCM 20C3M runs are among the IPCC AR4 simulations of the twentieth century that include observationally based natural forcing.
The two reanalyses as well as the IPCC models show a decrease in the downward SW radiation in recent decades, with the IPCC models having weaker decrease. The decrease in the downward component is found similarly in the upward component for both the ERA40 and IPCC models, On the other hand the reduction in upward SW in the NCEP data is considerably stronger than in the downward component, indicating larger changes in the albedo than in ERA40 and the other datasets. The decrease in the downward SW radiation is consistent with the reduction reported by Wang and Key (2003) for the 1982–1999 period using AVHRR data.
An investigation of the temperatures simulated by the subset of the models that included natural forcings shows that, while some of the ensemble members show a resemblance to the twentieth century’s double maxima of observed arctic temperature, the inclusion of the natural forcings clearly does not ensure a pronounced mid-twentieth century warming, let alone its timing. On the other hand, all ensemble members generally show an increase in the upward LW radiation by the end of the twentieth century, consistent with modeled and observed temperature trends. This provides further support to the findings of Wang et al. (2006) concerning the relative importance of the unforced variability generated by the models in the mid- and late-twentieth-century climate simulations. Specifically, the mid-twentieth century warming is much more consistent with unforced variability.
4 Projections for the twentyfirst century
4.1 Longwave radiation
Arctic (70–90°N) ensemble mean changes in the LW components for the SRES A2 scenario
SRES A2 - 20C3M
LW↓ − LW↑
SW↓ − SW↑
SH + LH
The annual mean response in the net LW radiation over the Arctic (Fig. 9e) is the difference between two large terms that partly cancel. For the SRES A2 scenario, the mean Arctic (70–90°N) increase in net LW energy to the surface (a decrease in LW energy lost by the surface) is 2.8 W/m2 (Table 3). The largest increase occurs in summer (6.3 W/m2). On the other hand there is no consensus on the sign of the wintertime changes in the net LW component, and the DJF ensemble mean change is slightly negative (Table 3). This indicates that the seasonal cycle in net LW radiation is increased in the scenario simulations.
4.2 Shortwave radiation
The across-model (±1STD) spread in the annual SW changes is 30–60% smaller than the spread in the LW changes (Table 3), but the summertime spread of the SW changes is a factor of 2–3 larger than the spread of the LW changes.
The annual mean Arctic (70–90°N) downward SW radiation is reduced by 10.7 W/m2. This is counteracted by a reduction in the upward component of 13.2 W/m2 due to reduced surface albedo, giving an increase in net surface SW radiation of 3.0 W/m2 (Table 3) with a maximum is summer (7.5 W/m2). This is 20% larger than the increase in the summertime net LW radiation and is consistent with a continued reduction (melt) of sea ice in the late twentyfirst century. With the exception of one model, the IPCC models all give an increase in net SW radiation. The model giving a reduced net SW radiation is the model showing the strongest increase in summertime cloud cover (Fig. 13).
4.3 Turbulent fluxes
Annual changes in the total turbulent fluxes averaged over 70–90°N and all models are 4.2 W/m2 which is considerable less than the changes in the individual downward and upward radiative terms, but nearly as large as (but opposite in sign to) the annual changes of 5.8 W/m2 in net (SW and LW) surface radiation.
5 Summary and conclusions
Averages of Arctic (70–90°N) downward LW radiation range from 205 to 230 W/m2. The spread in monthly values is typically 20–30 W/m2 and the amplitude of the seasonal cycle is not well constrained.
Upward LW radiation estimates differ by about the same amount (annually from 248 to 267 W/m2), but there is a closer agreement on the strength of the seasonal cycle.
Longwave radiation as an Arctic energy sink ranges from 28 to 52 W/m2 for the different observational estimates and there is no consensus on the seasonal cycle of net LW radiation. Two estimates indicate a maximum in net surface energy loss in summer, and three estimates show the loss to be highest in early spring.
The NCEP reanalyses has a strong bias in downward and upward shortwave radiation relative to the other estimates
Annual downward SW radiation estimates (excluding NCEP) range from 87 to 92 W/m2, and the monthly spread is typically 20–25 W/m2 during the months April to August.
The differences in upward SW radiation estimates are somewhat smaller (ranging from 44 to 47 W/m2 annually, excluding NCEP).
Annual mean shortwave radiation as a net surface energy source ranges from 43 to 50 W/m2, with the largest spread during summertime (10–15 W/m2).
As might be expected, for all energy budget terms, the model spread is largest in the marginal ice zone of the Barents, Kara and Chukchi Seas, where sea ice varies among he models. These are also the areas where the models most strongly deviate from the observational estimates.
There is a tendency among the models to underestimate the downward LW radiation during wintertime. The DJF model ensemble mean is 10.5 W/m2 lower than the mean of the observations. The across-model spread (±1STD) is 16.6 W/m2 during wintertime (DJF) and reduced to 9.3 W/m2 in summer (JJA). The DJF bias may be related to an overestimation of the sea ice extent which may feed back to the lower atmosphere.
As with the downward component, there is a tendency for the models to underestimate the upward LW radiation in winter (DJF ensemble mean is 11.6 W/m2 lower than the mean of the observations). The across model spread is similar to that for the downward component.
The models indicate that the Arctic surface has it greatest net LW energy loss in early spring (April-May, approximately 40 W/m2) and smallest in mid summer (July–August, approximately 30 W/m2). The seasonal cycle is in line with some of the observational estimates, but with a tendency for many models to underestimate the surface energy loss in summertime. The summertime LW energy loss from the individual models is related to the summertime cloud and sea ice fraction.
Most models overestimate the downward SW radiation in spring (April and May) and underestimate the downward SW radiation in summer and early autumn (July, August) compared to four of the five observational estimates. The model ensemble mean for March-May is 14.9 W/m2 higher than the mean of the observations (excluding NCEP) and the JJA model ensemble mean is 14.3 W/m2 lower. The springtime downward SW radiation in the individual models is related to the models’ cloud fraction.
Upward SW radiation is overestimated compared to three of four observational estimates. The biases are related to the downward SW radiation and to the models’ sea ice fraction.
There is a large spread in the model’s net surface SW radiation. The net annual surface energy gain from SW radiation ranges from 25.7 to 50.2 W/m2.
Most models simulate the Arctic mean sensible heat flux to be a surface energy sink during wintertime (the DJF ensemble mean is 2.5 W/m2, upward). This is contrary to the reanalyses which show the sensible heat flux to be a energy source (−7.5 W/m2). The discrepancy implies an absence of, or unrealistically weak, surface-based inversions in the models.
Most models show the surface latent heat flux to be a smaller sink of energy than the reanalyses suggests. (The annual model ensemble mean is 4 W/m2 lower than the mean of the reanalyses). This discrepancy is consistent with a positive biases in the models wintertime sea ice extent (Arzel et al. 2006).
It should be stressed that without controlled experiments with individual models it is difficult, if at all possible, to specify what particular features of model physics or/and numerics and resolution are responsible for the across-model scatter in simulation of the arctic surface energy budget. Hopefully, such experiments with model resolution; atmospheric boundary layer, cloud and sea-ice parameterizations; and other process formulations will help to sort out the possible causes in the coming years. The more complete documentation and assessment of recent model improvements in the IPCC’s upcoming Fourth Assessment Report will likely serve as a stimulus for such experiments.
The second objective of the paper was to assess the changes in the arctic surface energy budget due to increased greenhouse gasses. For the entire twentieth century, the full set of the 20C3M simulations (including all ensemble available for each individual model) shows a robust positive twentieth century trend in the net radiation budget (with few exceptions). The increase of the total cloudiness, while accompanied by a decrease in the downward SW radiation, results in an increase of the surface air temperature due to increase in the downward LW radiation.
Increased concentrations of greenhouse gasses increase the annual mean LW downward radiation by 30.1 W/m2, with a maximum of 40.7 W/m2 in autumn (SON) and a minimum during summertime (JJA) of 16.2 W/m2. The across-model spread (±1STD) of the annual mean changes is 7.7 W/m2; the spread of the monthly changes follows the strength of the mean response. There is a positive correlation between the models’ changes in downward LW radiation and the models’ changes in cloud fraction.
The increase of downward LW radiation is counteracted by the surface thermal radiation feedback, which increases the annual surface upward LW emission by 27.1 W/m2. The maximum change is in autumn (38.9 W/m2) and minimum is in summer (9.9 W/m2). The across-model spread of the annual values is close to the spread in the downward changes (6.2 W/m2). There is a negative correlation between the models’ changes in upward LW radiation and the models’ changes in sea ice fraction.
The ensemble mean changes in the net LW radiation are slightly negative in winter and positive in all other months. Averaged over the year the net increase in LW radiation at the surface is approximately 3 W/m2.
Increases of water vapor and cloud cover reduce the annual downward SW radiation by 10.7 W/m2. The changes in the summertime downward SW radiation in the individual models are related to the models’ changes in cloud fraction.
Together with changes in the surface albedo, the reductions in downward SW radiation lead to a decrease of 13.2 W/m2 in the annual upward SW radiation. The summertime changes are strongly correlated to the models’ changes in the summertime sea ice fraction.
Twelve of the 14 models show a twentyfirst-century increase in the net SW radiation. Annually, the absorbed SW radiation increases by 2.5 W/m2, with a maximum in summer (7.5 W/m2) and the changes in the annual net SW radiation are comparable to the changes in the annual net LW radiation.
The net surface radiation loss (RN = LW↓ + SW↓ − LW↑ − SW↑) is slightly increased (−1.5 W/m2) in winter due to the warmer temperatures while the surface energy gain in summer is substantially increased (13.8 W/m2), giving an annual increase in net radiation of 5.5 W/m2. The changes in the summertime radiation gain exhibits an large spread in the different models and is a strong function of the models’ changes cloud fraction (Fig. 18).
The annual increased net radiation is largely counteracted by a increased surface energy loss by turbulent fluxes of 4.4 W/m2. This is primarily due to increased latent heat loss during autumn and winter.
The choice of the 20-year reference period followed the recommendation for the IPCC AR4 and 20-year periods were also used in earlier evaluations by the Arctic Climate Impact Assessment (ACIA, Kattsov and Källén 2005). Thus the use of 20-year periods allows direct comparison of the results from this paper with results from other papers supporting the IPCC 4th Assessment (e.g. Arzel et al. 2006; Zhang et al. 2006; Kattsov et al. 2006; Chapman and Walsh 2006) and the ACIA. Additionally, this period is linked to that of sufficiently high-quality (satellite) observations, which are available only since the late 1970s.
It should also be noted that the relationship between the radiation components and cloud fraction should not be taken as the direct influence of the cloud cover. Regression and correlations do not provide any causal relationships. Cloud cover changes are related to changes in both heat and moisture transport which, in addition to changing the cloud cover, may change the atmospheric temperatures and water vapor content. These concurrent changes make it difficult to distinguish the direct influence of cloud changes from the changes in water vapor and temperatures, which may co-vary with the cloud cover changes and therefore inflate statistical relationships between cloudiness and radiative fluxes.
This work was supported by the Norwegian Research Council’s NORKLIMA program through the RegClim project and the BCCR SSF grant, by the U.S. National Science Foundation’s Office of Polar Programs through Grant OPP-0327664, and by the Russian Foundation for Basic Research through Grant 05-05-65093. We acknowledge the international modeling groups for providing their data for analysis, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) for collecting and archiving the IPCC model data, the JSC/CLIVAR Working Group on Coupled Modelling (WGCM) and their Coupled Model Intercomparison Project (CMIP) and Climate Simulation Panel for organizing the model data analysis activity, and the IPCC WG1 TSU for technical support. The IPCC Data Archive at Lawrence Livermore National Laboratory is supported by the Office of Science, US Department of Energy. This is publication No. A150 from the Bjerknes Centre for Climate Research.