New estimates of the large-scale Arctic atmospheric energy budget


  • David F. Porter,

    1. Cooperative Institute for Research in Environmental Sciences, University of Colorado at Boulder, Boulder, Colorado, USA
    2. Also at Department of Atmospheric and Oceanic Sciences, University of Colorado at Boulder, Boulder, Colorado, USA.
    Search for more papers by this author
  • John J. Cassano,

    1. Cooperative Institute for Research in Environmental Sciences, University of Colorado at Boulder, Boulder, Colorado, USA
    2. Also at Department of Atmospheric and Oceanic Sciences, University of Colorado at Boulder, Boulder, Colorado, USA.
    Search for more papers by this author
  • Mark C. Serreze,

    1. Cooperative Institute for Research in Environmental Sciences, University of Colorado at Boulder, Boulder, Colorado, USA
    2. Also at National Snow and Ice Data Center, University of Colorado at Boulder, Boulder, Colorado, USA.
    Search for more papers by this author
  • David N. Kindig

    1. Cooperative Institute for Research in Environmental Sciences, University of Colorado at Boulder, Boulder, Colorado, USA
    2. Also at National Snow and Ice Data Center, University of Colorado at Boulder, Boulder, Colorado, USA.
    Search for more papers by this author


[1] New estimates of the current energy budget of the north polar cap (the region north of 70°N) are synthesized by combining data from new atmospheric reanalyses and satellite retrievals. For the period 2000–2005, monthly means from the Clouds and the Earth's Radiant Energy System (CERES) satellite data set are considered to provide the most reliable top-of-atmosphere (TOA) radiation budget. The remaining components of the energy budget, comprising of the energy storage, horizontal convergence of energy, and the net surface flux between the atmospheric and subsurface columns, are compiled using data from the Japanese 25 Year Reanalysis Project (JRA) and the National Centers for Environmental Prediction (NCEP) /National Center for Atmospheric Research (NCAR) Reanalysis (NRA). The annual cycles of energy budget components for the polar cap are fairly consistent between the JRA and NRA, but with some systematic differences. JRA depicts an annual mean surface flux of 14 W m−2 (upward), compared to only 5 W m−2 in NRA. Most of this disparity appears to be due to differences in sea ice and albedo. Horizontal atmospheric energy flux divergence calculated using mass-corrected flux values contains artifacts leading to unphysical results. We argue that backing out the energy flux convergence as a residual from the net surface heat flux and time change in energy storage from each reanalysis, and the TOA radiation budget from CERES, provides for more physically realistic results in the Arctic. Monthly mean anomalies of budget terms, used to examine conditions leading to the extreme seasonal sea ice extent minimum of September 2005, point to the importance of albedo feedback.

1. Introduction

[2] The Arctic is characterized by a deficit in solar radiation received at the top of the atmosphere (TOA) when compared to lower latitudes. This latitudinal gradient in incoming energy drives complex interactions between the atmosphere, land, ocean, and forcing from lower latitudes. The atmosphere and ocean respond to the gradient in radiative heating by transporting energy poleward via both transient and stationary eddies. The differential radiative heating is strongest in winter, during the Arctic polar night. The maximum high-latitude poleward transport of atmospheric energy in winter is centered along the longitudes of the Greenland and Norwegian seas [Tsukernik et al., 2004]. This area represents the northern end of the primary North Atlantic storm track, where a combination of strong transient eddy activity and open water overlain by a fairly cold atmosphere encourage large sensible and latent heat fluxes from the ocean to the atmosphere, increasing the convergence of energy transport in the Arctic region. While the Arctic climate system has a rich complexity, much can be learned about its behavior by considering the basic components of the energy budget, specifically, the time change of storage of atmospheric energy, the radiation budget at the top of the atmosphere, the horizontal convergence of atmospheric energy transports, and the net surface heat flux.

[3] In this paper, two atmospheric reanalyses are used, in part, to provide a more robust assessment of uncertainty in atmospheric transports and the net surface flux. New satellite remote sensing products provide quality TOA radiative fluxes that are used as a constraint, aiding in comparisons between representations of budget components in the reanalyses. Through these comparisons and efforts to assemble the most meaningful representation of the budget, our paper compliments past efforts to improve understanding of the Arctic and global energy budget [Nakamura and Oort, 1988; Serreze et al., 2007; Trenberth et al., 2001; Trenberth and Stepaniak, 2004] and the Arctic freshwater budget [Serreze et al., 2006]. This study provides new Arctic energy budget estimates using the current best observations of TOA radiative fluxes and next generation reanalyses. Also, monthly mean anomalies of budget terms are applied to examine conditions leading to the extreme seasonal sea ice extent minimum of September 2005, part of a new approach of using the energy budget framework to understand the changes in energy flow during an anomalously low sea ice summer.

2. Methods: Energy Budget Framework

[4] Consider the energy budget of a column extending from the surface to the top of atmosphere (TOA), represented by each grid cell over the analysis domain. In the framework of Nakamura and Oort [1988] and Trenberth [1997], the budget of this atmospheric column may be denoted by:

equation image

where the time change of atmospheric energy storage E equals the sum of the net radiation budget at the top of the atmosphere (FRAD), the vertically integrated horizontal energy flux convergence (FWALL), and the net surface heat flux (FSFC). Following Serreze et al. [2007], all terms are defined as positive when they contribute to atmospheric energy gain; hence positive downward for the TOA net radiation, positive upward for the net surface flux, and positive for horizontal convergence of atmospheric energy transport. The atmospheric column gains energy (positive time tendency) if the sum of the three right-hand side terms is positive, and the column loses energy if the sum is negative. For steady state long-term annual mean conditions, the tendency of atmospheric energy storage is zero.

[5] The time change in atmospheric energy storage can be expanded as:

equation image

where p is pressure, cp is the specific heat of the atmosphere at constant pressure (1005.7 J K−1 kg−1) [American Meteorological Society, 2000], T is temperature in Kelvin, k is the kinetic energy, L is the latent heat of evaporation (2.501 × 106 J kg−1), q is the specific humidity, g is acceleration due to gravity (9.81 m s−2) and Φs is the surface geopotential (not a function of pressure [Trenberth et al., 2001]). The energy in a column of the atmosphere is hence comprised of four forms: sensible, kinetic, latent, and geopotential energy.

[6] The net radiation at the TOA, FRAD, is expressed as:

equation image

where FSW is the net shortwave (solar) and FLW is the net longwave (thermal) radiation.

[7] FWALL, the convergence of atmospheric energy transport is written as:

equation image

where v is the horizontal wind vector, cPT + Φ is the dry static energy, and cPT + Φ + Lq is moist static energy. The contribution of kinetic energy, k, is sometimes ignored because of its comparatively small magnitude [Nakamura and Oort, 1988] but we include it here for completeness.

[8] The net surface heat flux is defined as the net transfer of heat between the subsurface column and the atmosphere:

equation image

where SWSFC and LWSFC are the net surface shortwave and longwave radiation fluxes and QH and QE are the turbulent sensible and latent heat fluxes, respectively. If the sum of the four terms is positive (upward), the atmospheric column gains energy through a heat flux from the subsurface column to the atmosphere. The opposite is true if the sum is negative.

3. Data

[9] Reanalyses from the National Center for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) and Japan Meteorological Agency (JMA) are used to assess the atmospheric budget. We use vertical integrals of the energy storage and transport convergence terms compiled by K. Trenberth and colleagues at NCAR (data are located at Atmospheric reanalyses provide gridded values of the atmospheric state variables and fluxes. Unlike operational numerical weather prediction systems that undergo constant refinement in an effort to improve forecast skill, reanalyses use fixed versions of an atmospheric model and data assimilation system. This mitigates introducing nonclimatic jumps and trends in archived fields. Because reanalyses are retrospective, as opposed to operational forecast systems, assimilated observations can undergo more quality control, providing better constraints on the atmospheric state. While reanalysis fields are widely used in the atmospheric community, they must be used with caution, particularly when assessing trends, which can be influenced by changes in the observational network through time [Bromwich et al., 2007].

[10] The NCEP/NCAR reanalysis (NRA) [Kalnay et al., 1996] is based on a numerical weather prediction model with 28 sigma (terrain-following) levels in the vertical, of which five are in the boundary layer, and T62 spectral resolution in the horizontal, equivalent to about 209 km resolution. Data are available at 6 h intervals for the period 1948 onward. Updated fields are provided through the Climate Data Assimilation System (CDAS).

[11] The JMA, along with the Central Research Institute of Electric Power Industry in Japan, developed a more modern global atmospheric reanalysis, the JRA [Onogi et al., 2007]. Several new products and techniques were developed specifically for inclusion in the JRA assimilation system, including the TIROS Operational Vertical Sounder (TOVS) radiance data assimilation method, sea surface temperatures (SST) and sea ice data from the Centennial in situ Observation Based Estimates (COBE), and three-dimensional daily ozone profiles. The philosophy of the JRA is to use as many data types as possible not included in previous reanalyses. These include wind profile retrievals around tropical cyclones, atmospheric motion vector data from geostationary satellites, and historical snow data from printed records [Onogi et al., 2007]. The model has T106 spectral resolution, which is equivalent to 125 km horizontal grid spacing, and 40 vertical levels. After 2005, the JMA CDAS (JCDAS) has been used to update the JRA.

[12] As outlined earlier, our study builds upon the recent Arctic energy budget study of Serreze et al. [2007], which made use of data from the NRA and the European Centre for Medium-Range Weather Forecasts ERA-40 reanalysis. Here, we contrast results from NRA and JRA, and make use of TOA and surface radiation data from CERES (Clouds and the Earth's Radiant Energy System). Of the four primary terms of the energy budget (1), radiative fluxes used in FSFC and FRAD show the least agreement between atmospheric reanalyses [Serreze et al., 2007]. TOA fluxes from the CERES instrument on the Terra satellites (CER_SRBAVG_Terra-FM1,2-MODIS_Edition2D, available from the NASA Langley Research Center Atmospheric Science Data Center at are used for comparison with those depicted by NRA and JRA.

[13] The CERES instruments, which can be viewed as improved versions of those used in the Earth Radiation Budget Experiment (ERBE) [Wielicki et al., 1996], measure broadband shortwave, total, and window radiances. To determine the global TOA irradiances, measurements of reflected and emitted radiation are needed from all angles. The instruments were designed to independently rotate the azimuth viewing angle in order to measure radiances from several viewing geometries. These multiple viewing angles are inputs to an empirical angular distribution model to estimate irradiances. Scene identification is crucial to correct calculations of irradiances. Cloud radiances from collocated Moderate Resolution Imaging Spectroradiometer data, along with sea ice concentration fields provided by the National Snow and Ice Data Center [Hollinger et al., 1990], are used to distinguish clouds from underlying high-albedo sea ice and snow cover. This improved discrimination is a major advancement over ERBE [Kato and Loeb, 2005]. Global TOA and surface radiation fluxes for CERES are provided on a 1° × 1° grid.

[14] Our study focuses on the “polar cap,” defined by the area poleward of 70°N (Figure 1). This focus retains continuity with previous Arctic energy and moisture budget studies [Oort and Peixoto, 1983; Nakamura and Oort, 1988; Serreze et al., 2007; Overland and Turet, 1994; Semmler et al., 2005; Trenberth and Stepaniak, 2004]. The polar cap domain has a total surface area of 0.15 × 1014 m2 or 6% of the hemisphere, of which 72% is ocean covered and 28% is land covered. A simple polar cap domain aids in the interpretation of longitudinal variations of meridional atmospheric energy transports contained in FWALL.

Figure 1.

The “polar cap domain” shown in a map of the Arctic with 70°N latitude circle highlighted.

[15] We make use of the overlapping period of all data sets, which is constrained by CERES, which is available for March 2000 through October 2005. To have an integer number of years, all climatologies discussed below are for the 60 month period from November 2000 through October 2005. This contains the very low sea ice extent observed in summer 2005 which will be examined as a case study.

4. Components of the Energy Budget

4.1. Radiation Budget at the TOA

[16] The largest energy fluxes, both in terms of magnitude and spatiotemporal variability, are included in the FRAD term (1). Figure 2 shows the polar cap monthly mean FRAD, FSW, and FLW climatologies from the NRA, JRA, and CERES data. Monthly and annual means of FRAD are also listed in Table 1. All products depict the same basic seasonal cycles. Net shortwave radiation declines through autumn to nearly zero in winter. The longwave flux becomes less negative (less longwave emission to space) as atmospheric and surface temperatures fall. The net total radiation FRAD is most negative in January, but starts to become less negative as downwelling shortwave rises through the spring to its summer maximum. FRAD is slightly positive (downward) in June and July. There are nevertheless large differences between depictions from CERES and the two reanalyses from month to month.

Figure 2.

Annual cycle of area mean FRAD and its components in the Japanese 25 Year Reanalysis Project (JRA), NCEP/NCAR Reanalysis (NRA), and the Clouds and the Earth's Radiant Energy System (CERES).

Table 1. Area-Averaged Monthly Mean Energy Budget Terms From the 5 Year Climatology Computed From November 2000 Through October 2005a
MonthE/∂tFRADFSFCFWALL (residual)
  • a

    Given for storage changes and fluxes (W m−2). JRA, Japanese 25 Year Reanalysis Project; NRA, NCEP/NCAR Reanalysis; Clouds and the Earth's Radiant Energy System, CERES.


[17] The reanalyses assimilate satellite data into the atmospheric models with better quality control than is possible for real-time forecasting. Satellite-derived retrievals (profiles) are assimilated in the first-generation NRA system, while raw satellite radiances are assimilated into the next generation JRA. The temporal span of existing reanalyses exceeds the lifetime of any single satellite mission, requiring careful cross calibration of different data streams to avoid artificial trends [Trenberth et al., 2005]. While CERES is making a direct measurement of the TOA radiances desired, TOA retrievals and radiances assimilated into NRA and JRA, respectively, become boundary conditions for the model, which then use a radiative transfer model to output the fluxes as diagnostic variables. The inherent approximations and other shortcomings in these models, especially when clouds are present, reduce our confidence in the reanalyses radiation fluxes compared to those from CERES. Kato et al. [2006] find several shortcomings in TOA fluxes from ERBE when compared to CERES, including a difference of about 5 W m−2 in the annual net allwave TOA flux for the region 60°N - 90°N. While CERES is taken as the best approximation of TOA radiation fluxes over the analysis time period, CERES is not without its own shortcomings [Trenberth et al., 2009]. The CERES global mean net TOA radiation imbalance is estimated by Loeb et al. [2009] to be 6.5 W m−2, much larger than the estimate by Hansen et al. [2005] of 0.85 W m−2 using ocean heat content data and model simulations, with the major sources for uncertainty in CERES arising from instrument calibration (4.2 W m−2) and a total solar irradiance 1 W m−2 too high. To further put this uncertainty into perspective, the anthropogenic radiative forcing in 2005 relative to preindustrial conditions is estimated at 1.66 W m−2 [Solomon et al., 2007].

[18] While all three products capture a similar area-averaged annual cycle of the TOA net longwave radiation (FLW), there is a fairly consistent difference in magnitude of about 5% between JRA and the other two products (NRA and CERES). Throughout the annual cycle, the JRA has the most outgoing longwave radiation, pointing to either a warmer surface or atmosphere or more emission from cloud cover. Because the FLW annual cycle is similar for each product, with only an offset in magnitudes, any annual cycle differences in FRAD arise from differences in the net shortwave radiation, FSW. As would be expected, the downwelling component of the TOA solar radiation flux is quite similar for all three products (not shown), meaning that disparities in the annual cycle of the net shortwave flux FSW are mostly due to differences in the upwelling shortwave component associated with differences in column albedo. In the spring, the JRA FSW increases faster than in either NRA or CERES. This difference in springtime FSW, at least between JRA and NRA, is consistent with a lower surface albedo in the JRA (not shown).

[19] Differences in FSW result in substantial differences in the TOA net total radiation between the three products. From April through June, FRAD in the JRA is 20–40 W m−2 larger than depicted in the other two products. NRA and CERES by contrast agree quite well at this time of year. Figure 3 illustrates the spatial patterns of FRAD, down to 60°N, for the months with the largest discrepancies in the polar cap mean (May, June, and July). Since CERES provides a direct measurement of FRAD we view it as providing the best estimate. In the NRA, there is a sharp, unphysical meridional gradient over Greenland from May through July that we are unable to explain. In May, JRA depicts higher values (less negative or more positive) than CERES for most of the polar cap domain, particularly in the Greenland and Norwegian Seas. By June, the budget has turned positive over much of the plotted map area in all three products. This reflects the peak in incoming solar radiation at the TOA along with a decrease in albedo linked to the melt of snow and sea ice. However, the June FRAD pattern for JRA has some unrealistic features over the Arctic Ocean. In the absence of significant forcing by changes in cloud cover, FRAD along the Arctic Ocean margins should turn positive before the interior parts as the sea ice edge retreats to expose open water with its low albedo. This expected pattern is evident in the June fields from both CERES and NRA. That JRA instead shows negative fluxes over the marginal seas and positive fluxes over the central Arctic Ocean points to problems in surface albedo or cloud cover. In July, FRAD in JRA and NRA has turned negative over most of the central Arctic Ocean, while in CERES the flux remains positive over much of the area. These differences are reflected in the polar cap mean annual cycle of FRAD shown in Figure 2, with the peak in FRAD occurring in June for both of the reanalyses, but occurring a month later in CERES, suggesting an earlier and prolonged seasonal reduction in surface albedo.

Figure 3.

Net total top of atmosphere (TOA) radiation, FRAD, for CERES (left), JRA (center), and NRA (right) for (a) May, (b) June, and (c) July. The 70°N latitude circle is highlighted.

4.2. Time Change of Atmospheric Energy Storage

[20] Differences between NRA and JRA in the monthly time change of atmospheric energy storage ∂E/∂t are small (Table 1). This holds with respect to both the polar cap averages and spatial patterns. The reanalyses show the expected pattern of the atmospheric column gaining energy in the spring, peaking in April (a tendency of 31 W m−2 in both products) and losing energy through autumn, most strongly in September (−26 W m−2 in JRA and −25 W m−2 in NRA). In both reanalyses, the annual mean of energy tendency is approximately zero.

4.3. Net Surface Heat Flux

[21] Direct measurements of the surface energy budget components in the Arctic are sparse. Herein lies the attraction of fluxes from atmospheric reanalyses [Bromwich et al., 2007]. The annual mean net surface flux for the polar cap is positive, i.e., there is a net heat transfer from the subsurface to the atmospheric column. Over land areas, the annual net surface heat flux must be zero for a steady state climate. The nonzero surface heat flux over the Arctic cap must then come from the oceanic portion of the study area through a convergence of oceanic sensible heat and sea ice transport [Nakamura and Oort, 1988].

[22] Polar cap–averaged annual mean values of the net surface heat flux from JRA and NRA are +14 W m−2 and +5 W m−2, respectively. The polar cap estimate by Nakamura and Oort [1988], calculated as a residual from rawinsonde-based atmospheric transports and satellite retrievals of FRAD from 1966 to 1977, is much smaller at 2.4 W m−2. Serreze et al. [2007] calculated an annual mean value of 11 W m−2 using ERA-40 data for the period of 1979–2001, while Semmler et al. [2005] cite a flux of 6 W m−2 based on a simulation with the regional model, REMO 5.1. Differences of this magnitude in the annual means are important, as a sustained net surface heat flux of 1 W m−2 over a year equals about 0.1 m of sea ice melt (at its melting point) [Serreze et al., 2007]. The spread of annual mean net surface heat flux from these different estimates is hence equivalent to about 1 m of ice melt over a year.

[23] The annual cycle of the net surface heat flux closely follows that of the net surface shortwave flux, as the area-averaged turbulent fluxes are small and the net longwave flux is fairly constant throughout the year (Figure 4). The area-averaged net surface heat flux in NRA is smaller than in JRA from July through March by about 10–15 W m−2, mainly because of NRA's more negative (downward in our convention) turbulent sensible heat flux in summer and autumn, and less net surface longwave radiation in the winter. In spring, FSFC is more similar between the reanalyses, but net surface shortwave radiation is more strongly downward in JRA, consistent with a larger TOA FSW in JRA (Figure 2).

Figure 4.

Annual cycle of FSFC and its components for the JRA and NRA. SWSFC and LWSFC are the net shortwave and longwave radiation fluxes, respectively. QH and QE are the turbulent sensible and latent heat fluxes, respectively.

[24] For months July through October, much of the difference in area-averaged FSFC is due to a systematic difference in the turbulent sensible heat flux, QH. Discrepancies in QH arise from differences in the representation of the vertical wind shear and stability in the planetary boundary layer [Tjernström et al., 2005]. In the spring, both reanalyses have decreasingly negative (downward) QH, consistent with reduced stability during the transition from winter to summer conditions and the increasing downwelling shortwave radiation at the surface. In June, the sensible heat flux in NRA diverges sharply, remaining strongly negative, while in JRA it increases to zero and remains small until polar night returns in November. When compared with estimates by Serreze et al. [2007], the JRA is in better agreement with ERA-40 than NRA. In July, both JRA and NRA have positive (upward) sensible heat fluxes over the open waters of the North Atlantic (Figure 5), where warm sea surface temperatures result in unstable conditions. By contrast, sensible heat fluxes are downward over the Arctic Ocean, where ice melt keeps the surface skin temperature at the freezing point, helping to maintain a surface-based temperature inversion. The difference field (JRA-NRA) is dominated by large positive values as high as 40 W m−2 over both ocean and land areas, meaning considerably more positive (or less negative) surface heat fluxes in the JRA.

Figure 5.

July turbulent sensible heat flux, QH in W m−2, for JRA (left), NRA (center), and the difference, JRA minus NRA (right). The 70°N latitude circle is highlighted.

4.4. Convergence of Atmospheric Energy Transport

[25] The final term of the budget is the convergence of atmospheric energy, FWALL, the major mechanism by which the atmosphere compensates for the imbalance in TOA solar radiation between the middle and high latitudes. Area-averaged FWALL, always positive (convergence) in monthly means, is largest in the winter when the meridional atmospheric temperature gradient is largest [Serreze et al., 2007]. While latent heat flux convergence plays an important role in the Arctic energy budget, including how it influences cloud and snow cover which alter the radiation budget, the largest components of FWALL are the convergence of sensible heat and geopotential associated with transient and stationary eddies [Overland and Turet, 1994]. Previous studies using atmospheric reanalyses have relied on atmospheric transports that have been corrected to conserve mass [e.g., Serreze et al., 2007; Trenberth, 1997; Trenberth et al., 2001]. The reanalysis fields do not conserve mass because of a multitude of reasons, including the blending of a short-term forecast with observations and the interpolation from model coordinates to pressure surfaces [Trenberth, 1991]. As discussed in the Data section, Trenberth and colleagues have provided mass-corrected monthly mean components of FWALL for both the JRA and NRA.

[26] Figure 6a shows spatial patterns of mass-corrected FWALL from JRA for the months of January and July down to 60°N, truncated from the native T106 resolution to T42 using a tapered weighting function to remove excessive power from higher wave numbers [Trenberth and Solomon, 1993]. Despite filtering of the data, much spatial variability remains and there is a “blotchy” pattern, with an odd dipole at the pole. This feature is present in the majority of monthly mean fields. As a test of the degree of closure in the Arctic energy budget in each reanalysis, the residual is calculated using polar cap averages for all four terms of (1). A zero value implies perfect closure. The energy budget residuals compare well to those of Serreze et al. [2007], however, the same patterns, including the odd dipoles seen in the mass-corrected FWALL, are present.

Figure 6.

(a) Mass-corrected FWALL and (b) Residual FSFC for JRA for January (top) and July (bottom). The 70°N latitude circle is highlighted.

[27] These features point to shortcomings in the mass correction approach as applied to high latitudes. This is further evident when the mass-corrected JRA and NRA FWALL data are used to calculate FSFC as a residual from the three other reanalysis-based energy budget terms. For the JRA in January and July (Figure 6b), the dipole over 90°N is more exaggerated in the residual FSFC (residSFC) field. More glaring are positive (upward) residual net surface fluxes over Greenland in July, unrealistically implying a source of heat to the atmosphere. The net surface heat flux should be downward at this time of year over Greenland [Serreze et al., 2007; Trenberth and Stepaniak, 2004].

[28] In light of these problems, we calculate the convergence of atmospheric energy transport as a residual of the three remaining large-scale energy budget terms: ∂E/∂t, FRAD, and FSFC. Residual FWALL values are obtained using ∂E/∂t, FRAD, and FSFC from each reanalysis as well as using ∂E/∂t and FSFC from each reanalysis but with FRAD values from CERES. Annual cycles of area-averaged FWALL from the four residual estimates are shown in Figure 7. FWALL from all four estimates has a maximum in winter and a warm season minimum, as found in previous studies [Nakamura and Oort, 1988; Serreze et al., 2007]. However, there is considerable spread between the residual estimates for individual months. Differences between the two reanalyses based on a given residual calculation method are large (up to 22 W m−2), as are those between residual calculation methods for a given reanalysis (up to 32 W m−2 for NRA and 18 W m−2 for JRA) (Table 1, Figure 7). These differences can be broadly viewed as placing bounds on the uncertainty in FWALL. Peak energy transport convergence is larger in winter in both NRA-based estimates of FWALL. The seasonal minimum in both NRA estimates of FWALL as well as the JRA reanalysis-only estimate occurs in May, two months earlier than depicted from the JRA residual estimate using CERES FRAD. That the difference in FWALL between residual calculation methods for a given reanalysis is close to zero on the annual mean follows from the close similarity in annual FRAD from CERES and the two reanalyses.

Figure 7.

Annual cycle of area-averaged residual FWALL for JRA and NRA. The FWALL residuals are calculated from each reanalysis individually, and also using FRAD from CERES and the FSFC and ∂E/∂t from the JRA and NRA, respectively.

[29] With due consideration of problems identified by Loeb et al. [2009], we take the approach that constructing the best estimate of each individual energy budget is best approached by calculating FWALL as a residual using FRAD from CERES, and using ∂E/∂t and FSFC from each individual reanalysis. With the assumption that the CERES global imbalance in FRAD of 6.5 W m−2 [Loeb et al., 2009] applies to the polar cap region, the effect of this imbalance in the residual calculation would be to reduce FWALL in both reanalyses.

[30] By backing out the FWALL term using the CERES FRAD data set, and recognizing that ∂E/∂t is similar in both reanalyses, the differences in the residual FWALL values from the two reanalyses must be primarily due to differences in the FSFC terms. In July, JRA area-averaged QH is considerably higher than the NRA value while the other components of FSFC are more similar (Figure 4). This implies that, using our method of calculating FWALL as a residual, larger July turbulent sensible heating in JRA is linked with the later minimum in convergence of atmospheric energy transport. Perhaps the increased July QH in JRA is decreasing the meridional energy gradient and thereby reducing the requirements for energy convergence. Conversely, a larger meridional temperature gradient in July because of smaller energy convergence in the JRA would lead to a cooler lower atmosphere, which would then affect QH through changes in planetary boundary layer stability.

[31] Figure 8 shows spatial patterns of residual FWALL for January and July. For both reanalyses, residual FWALL is positive over most of the polar cap in January. However, there are strong negative values (energy transport divergences) in the Greenland and Norwegian Seas, on the equatorward side of the sea ice margin. This points to the influence of migrating baroclinic disturbances in these regions that often grow along the ice margin where there are large meridional temperature and moisture gradients, transferring atmospheric energy poleward. In July, energy convergence over the central Arctic Ocean is smaller, consistent with the reduced meridional temperature gradient at this time of year. While the same large-scale features in residual FWALL are present in each reanalysis, there are differences in the location and magnitude of smaller-scale features.

Figure 8.

Maps of FWALL calculated for JRA (left) and NRA (right) for January (top) and July (bottom). FWALL is calculated as a residual from FRAD from CERES, and FSFC and ∂E/∂t from both JRA and NRA. The 70°N latitude circle is highlighted.

4.5. New Estimates of the Arctic Energy Budget

[32] We consider CERES to provide the best quality TOA FRAD fields. The two reanalyses give very similar depictions of the time change in atmospheric energy storage. As noted above, spurious features are present in the mass-corrected FWALL fields, especially evident when calculating the net surface heat flux, FSFC, as a residual from the other terms in the reanalyses. Consequently, and acknowledging uncertainties in FSFC, arguably the most viable approach for obtaining the best estimate of the Arctic energy budget is to calculate FWALL as a residual, using CERES data for the TOA radiation fluxes. The seasonal cycles of budget components for the polar cap based on this approach are given in Figure 9. Recall that differences in the FWALL annual cycle in the two reanalyses closely mirror those of FSFC because only one FRAD product is used and monthly values of ∂E/∂t in the JRA and NRA are within 1 W m−2 of each other. With these issues in mind, we explore the annual cycle of the energy budget depicted in Figure 9.

Figure 9.

Annual cycle of all energy budget terms of JRA, NRA, and CERES, where FRAD is the TOA radiation budget, FWALL being atmospheric energy convergence calculated as a residual, FSFC being the net surface heat flux, and ∂E/∂t being the time change in energy storage.

[33] As downwelling shortwave radiation at the TOA decreases through autumn to near zero by October, and with the longwave loss to space (FLW) decreasing fairly slowly through this period (Figure 2), FRAD becomes strongly negative (more upward). FRAD then remains rather steady until March when downwelling shortwave radiation returns to the polar cap. FRAD follows the seasonal increase of FSW and reaches its annual maximum in July. That the FRAD maximum in July lags the June maximum in TOA downward solar radiation by a month reflects the lower July surface albedo.

[34] Starting in September, the atmospheric column is heated from the bottom via a positive net surface heat flux, linked largely to loss of sensible heat in the ocean mixed layer and sea ice growth. This upward flux grows through October, stays fairly steady through winter, declines in spring, then turns negative (downward) in summer, linked to ice melt and energy uptake in the ocean mixed layer. That FSFC is most strongly downward in July is consistent with the seasonal maximum in FRAD.

[35] In August, the time change in atmospheric energy storage turns negative, i.e., the atmosphere begins to lose energy. The tendency of energy storage remains negative until January, when atmospheric temperatures attain their seasonal minimum. The atmosphere then accumulates energy, until July, with the strongest energy gain in April when the solar flux is rapidly increasing.

[36] As the Arctic atmosphere begins to cool in autumn, the meridional temperature gradient increases, driving an increase in the poleward atmospheric energy transport that offsets the cooling. The fairly shallow annual cycle of FWALL peaks in winter when this gradient is largest, characterized by a winter maximum in both stationary and transient eddies in the North Atlantic sector (around 0°E) [Serreze et al., 2007]. Polar cap FWALL then decreases to its spring or early summer minimum (depending on the reanalysis data set used).

[37] Differences in the annual cycle of the Arctic energy budget as depicted in the JRA and NRA highlight the interplay between the budget terms. From July through August, JRA depicts a smaller downward net surface flux than NRA (less atmospheric cooling from the bottom), while from September through March the upward JRA net surface flux is comparatively larger (more atmospheric heating from the bottom). To compensate, this means that JRA has smaller FWALL values for these months. For most months, differences in FSFC and the residual FWALL between the reanalyses are at least 10 W m−2. The large spread in estimates of FSFC and FWALL is particularly disturbing given how it impacts our ability to quantitatively diagnose links between the energy budget and change in related elements of the Arctic climate system, such as the strong downward trend in September sea ice extent [Stroeve et al., 2007].

4.6. Case Study for 2005

[38] As an application of the energy budget framework presented here, and recognizing limitations just discussed, we examine conditions leading to and attending the extreme seasonal sea ice minimum of September 2005, which was characterized by strong negative ice concentration anomalies along the Eurasian coast (Figure 10). We focus on monthly anomalies of surface albedo, FRAD, FSFC, and residual FWALL for June, July, August, and September (Figures 11121314) calculated with respect to the 5 year study period monthly means. FSFC and residual FWALL are taken from JRA.

Figure 10.

September 2005 sea ice concentration anomalies (National Snow and Ice Data Center).

Figure 11.

June 2005 monthly mean anomalies, calculated with respect to the 5 year study period monthly means, for (a) JRA surface albedo, (b) CERES TOA radiation budget, (c) JRA net surface heat flux, and (d) total atmospheric energy convergence, calculated as a residual from CERES FRAD, JRA ∂E/∂t, and FSFC.

Figure 12.

July 2005 monthly mean anomalies, calculated with respect to the 5 year study period monthly means, for (a) JRA surface albedo, (b) CERES TOA radiation budget, (c) JRA net surface heat flux, and (d) total atmospheric energy convergence, calculated as a residual from CERES FRAD, JRA ∂E/∂t, and FSFC.

Figure 13.

August 2005 monthly mean anomalies, calculated with respect to the 5 year study period monthly means, for (a) JRA surface albedo, (b) CERES TOA radiation budget, (c) JRA net surface heat flux and (d) total atmospheric energy convergence, calculated as a residual from CERES FRAD, and JRA ∂E/∂t and FSFC.

Figure 14.

September 2005 monthly mean anomalies, calculated with respect to the 5 year study period monthly means, for (a) JRA surface albedo, (b) CERES TOA radiation budget, (c) JRA net surface heat flux and (d) total atmospheric energy convergence, calculated as a residual from CERES FRAD, JRA ∂E/∂t, and FSFC.

[39] An important feature of the sea ice system is the albedo feedback, whereby melting of ice reduces the surface albedo, leading to heat gain in the ocean mixed layer which then fosters further melt. Negative monthly mean surface albedo anomalies, linked to negative sea ice concentration anomalies along the coastal seas, are already substantial by June of 2005 (Figure 11a). Negative albedo anomalies of 0.4 are found in the Laptev, Kara and eastern Barents seas. The albedo anomalies are collocated with positive (downward) FRAD anomalies (Figure 11b). Negative (downward) FSFC anomalies (Figure 11c) are also found in the same regions as a result of enhanced absorption of shortwave radiation at the surface that heats the ocean mixed layer. Note in turn how the positive albedo anomalies in northern Baffin Bay and along the east central coast of Greenland are expressed as positive (upward) anomalies in FSFC. The basic pattern of anomalies in surface albedo, FRAD, and FSFC seen for June persists into July (Figures 12a, 12b, and 12c).

[40] In June, when the sea ice anomalies are starting to form, there is a dominance of positive (convergence) FWALL anomalies over the polar cap region (Figure 11d). While there is some correspondence between positive anomaly maxima in the convergence and the strong negative albedo anomalies in the Kara Sea and eastern Beaufort Sea, the relationship between the two fields is in general weak. Relationships between FWALL and albedo anomalies are similarly weak for July (Figure 12d).

[41] The areas of negative albedo anomalies continue to grow and are very pronounced along the Eurasian coast in August, partly balanced by positive anomalies north of Alaska and east of Greenland (Figure 13a). Note the general correspondence with the September anomaly pattern in ice concentration shown in Figure 10. However, as downwelling TOA solar radiation is fairly small by August, the FRAD anomalies (Figure 13b) are considerably less pronounced compared to June and July. The smaller TOA downward solar flux also helps to explain the smaller anomalies in FSFC (Figure 13c). The FWALL anomalies are negative over the eastern half of the Arctic Ocean, where the albedo is also anomalously low (Figure 13d). This suggests that, instead of anomalous energy flux convergence (FWALL), the main mechanism by which negative sea ice concentration anomalies were reinforced and expanded in late summer is through melt fed by anomalous heat in the upper ocean, heat that was gained primarily in June and July in growing open water areas when the solar flux was fairly strong. This can be viewed as an expression of albedo feedback.

[42] Negative surface albedo anomalies along the Eurasian coast are strongest in September (Figure 14a), but in accord with the small solar flux, FRAD anomalies are small (Figure 14b). In a typical September, SWSFC is small enough for FSFC to turn positive because of net upward longwave radiation and QE. Locally, especially at the sea ice edge where there are large surface temperature gradients when airflows onto or off the ice pack, QH can be quite significant. September of 2005 saw a northward shift in these areas of large QH. Note the tongue of positive FSFC anomalies extending eastward from Svalbard, with negative anomalies to the south (Figure 14c), the latter indicating reduced upward heat transfer in regions that are normally adjacent to the sea ice edge, but in 2005 were more distant from the ice pack. The region between Svalbard and Severnaya Zemlya, usually covered by ice but open in 2005 is also a region with positive anomalies in FWALL (Figure 14d). This analysis of Arctic energy budget anomalies for the summer of 2005 suggests that the reinforcement of sea ice melt through albedo feedback played an important role in maintaining anomalous sea ice extent until the September minimum. In addition, there is some evidence that anomalous horizontal energy flux convergence early in the melt season helped to increase the downward net surface flux.

5. Summary and Conclusions

[43] To obtain meaningful results, atmospheric energy transports from reanalyses must be corrected for mass imbalance. However, unphysical features are apparent in the fields used in the present study, suggesting that the current correction technique has shortcomings in high latitudes. We therefore calculated the convergence of the total transport as a residual using the CERES TOA radiation (FRAD), and the reanalyses estimates of the net surface flux and time change in atmospheric energy storage. By using the CERES FRAD to back out FWALL, some of the discrepancies between the JRA and NRA are reduced. This approach of course has its own shortcomings, for by using the CERES FRAD, differences in FRAD from the two reanalyses propagate to the backed out FWALL term.

[44] As an application of our best estimates of the energy budget, we examined the evolution of the extreme September 2005 sea ice minimum. Results suggests a strong role of albedo feedback, in which anomalous ocean heat gain through June and July in open water areas led to further ice melt, with this ocean heat then fostering strong ice melt through August despite the fairly low solar flux. In addition, several recent studies analyzing trends and variability in the Arctic sea ice suggest that the role of increased downwelling longwave flux has become more important in recent years [Francis et al., 2005; Stroeve et al., 2007; Cavalieri and Parkinson, 2008].

[45] The primary message from our study is that considerable uncertainty still exists regarding the atmospheric energy budget of the Arctic. While JRA is a more advanced system than the NRA, there is not enough evidence from our study to conclude that it actually provides a better representation of the budget. Differences in individual monthly mean budget terms for polar cap averages from the two reanalyses are as large as 23 W m−2. The disparity in the annual net surface heat flux for the polar cap of 9 W m−2 equates to about 0.9 m of sea ice loss per year (at its melting point) [Serreze et al., 2007]. Calculating FWALL as a residual from CERES FRAD and the reanalyses estimates of FSFC and ∂E/∂t circumvents artifacts in mass corrected transports at high latitudes; however, as discussed, the CERES data also have shortcomings, in large part linked to instrument calibration and an assumed solar irradiance that is too high.


[46] This study was supported by National Science Foundation grants ARC-0805821, ARC-0531040, and ARC-0732986. Kevin Trenberth and colleagues kindly supplied the JRA and NRA energy budget data. We acknowledge the helpful comments of three anonymous reviewers and the editor who helped improve this manuscript.