model simulations of the arctic atmospheric boundary layer from the sheba year

© Royal Swedish Academy of Sciences 2004http://www.ambio.kva.se

Ambio Vol. 33 No. 4–5, June 2004 221

Model Simulations of the Arctic Atmospheric Boundary Layer from the SHEBA Year

Michael Tjernström, Mark Zagar and Gunilla Svensson

We present Arctic atmospheric boundary-layer modeling with a regional model COAMPSTM, for the Surface Heat Budget of the Arctic Ocean (SHEBA) experiment. Model results are compared to soundings, near-surface measure-ments and forecasts from the ECMWF model. The near-surface temperature is often too high in winter, except in shorter periods when the boundary layer was cloud-capped and well-mixed due to cloud-top cooling. Temperatures are slightly too high also during the summer melt season. Ef-fects are too high boundary-layer moisture and formation of too dense stratocumulus, generating a too deep well-mixed boundary layer with a cold bias at the simulated boundary-layer top. Errors in temperature and therefore moisture are responsible for large errors in heat flux, in particular in solar radiation, by forming these clouds. We conclude that the main problems lie in the surface energy balance and the treatment of the heat conduction through the ice and snow and in how low-level clouds are treated.

INTRODUCTIONModel projections of anthropogenic climate change indicate large climate sensitivity in the Arctic (e.g. 1, 2). The Arctic warming, averaged among several models, is about 2.5 times the global value. At the same time, the scatter between the in-dividual models is much larger in the Arctic than elsewhere on earth (3). The global models used for these projections also have significant problems in reproducing the current Arctic climate. They are generally too warm, have systematic biases in surface pressure fields and surface radiative fluxes vary widely between models (4). Moreover, the observed recent warming appears to have occurred mostly over the northern Asian continent, rather than over the central Arctic as has been suggested by many cli-mate models. Climate change simulations also indicate the larg-est warming in autumn, followed by winter, while observations indicate that the largest recent warming has occurred during spring (5). An important contributor to the large climate sensitivity in the Arctic has been related to physical processes peculiar to the Arctic, e.g. melting of sea ice and snow and the consequent al-bedo feedback. Many physical processes in climate models are parameterized; development of such schemes always involves an empirical component. Detailed observations are sparse in the Arctic and consequently, the ensemble of observations forming the empirical basis for development of reliable parameteriza-tions may therefore be inadequate in the Arctic. The Arctic en-vironment, with its semipermanent sea ice, sets up atmospheric boundary-layer conditions found only here. The annual cycle is very large, while the diurnal cycle, that helps in shaping the boundary layer at most other places, is most often absent. The Arctic winter makes long periods of very high static stability possible. The very stable boundary layer is poorly understood in general (6). During long-lived episodes, the interplay between gravity waves and turbulence becomes relatively more impor-tant (7). These conditions are poorly treated, or not treated at all, in most numerical models.

In summer, when the sun is constantly above the horizon, there is a weak diurnal variability in the incoming solar radia-tion. However, the corresponding variability in boundary-layer stability is cancelled by the lower boundary conditions being characterized by the melting and freezing of water at the sur-face-ice or snow. Boundary-layer clouds also play an important role to the surface energy balance, usually warming the surface (8). During summer, there is a semipersistent cover of fog and low stratus clouds but a relatively low aerosol concentration compared to mid-latitude counterparts, alters their radiation properties. In winter, SHEBA results indicate that Arctic clouds retain liquid-water droplets at much lower temperatures than are assumed in most models (8, 9). These and other important physi-cal processes involving Arctic clouds and processes in the shal-low Arctic boundary layer are poorly treated by global models. Many previous modeling studies of the Arctic climate system have been focussed on special areas, like the marginal ice zone (10), on shorter periods (11) or using dynamically incomplete models (12). Model development needs to compare model re-sults to in situ measurements; this has been a core task in the Swedish Regional Climate Modeling Programme (SWECLIM). Moreover, an underlying strategy has been to use regional mod-els to not only provide scenarios to impacts assessment, but also as a tool to improve global modeling. In a regional model, the larger-scale climate can be prescribed at the lateral boundary us-ing results from global analyses. Remaining systematic errors are likely related to deficiencies in descriptions of physical pro-cesses in the regional model. These are more easily isolated in a controlled regional-model experiment than in global climate model control experiments and can more easily be dealt with in a regional model. SWECLIM participated in ARCMIP (Arctic Regional Climate Model Intercomparison Project (13) http://curry.eas.gatech.edu/ARCMIP/index.html) (for example Jones et al. (14) and this paper). ARCMIP aims to isolate model de-ficiencies and to improve the description of Arctic climate in numerical models by carrying out controlled regional-model ex-periments for the Arctic. Several models are intercompared in ARCMIP and are also compared to observational data. All con-tributing models are run for the same model domain, using the same analyzed lateral-boundary forcing and observed surface forcing. The first ARCMIP experiment is a 13-month long simu-lation, from September 1997 through September 1998, for the western Arctic. The domain is centered on the SHEBA (Surface Heat and Energy Balance of the Arctic (15)) ice-drift track. The purpose of this paper is to present some first results, il-lustrating some typical problems with the boundary-layer repre-sentation. We have selected to study boundary layer parameters and surface energy fluxes for this first study, as they are critical to the ice/albedo feedback. We compare two simulations with a regional model against the results from a global forecast mod-el and against the observations collected at the SHEBA camp. These are first results out of the basic model setup. From the er-rors in the model results, we can learn how to improve the mod-els, and possibly also how to launch experiments in the Arctic with the aim to better understand important processes.


222 Ambio Vol. 33 No. 4–5, June 2004

THE MODEL EXPERIMENTSThe US Navy Coupled Ocean/At-mosphere Mesoscale Prediction System (COAMPSTM (16)) is one model in SWECLIM used in ARC-MIP (also see (14)). It was set up on a 3500 x 2750 km2 domain over the western Arctic. The domain cov-ered most of Alaska, the Beering Strait and eastern Asia, reaching to about 85°N (14). The horizontal resolution was 50 km, with 70 x 55 grid points. In the vertical, 30 levels were used, nine of which were be-low 1 km with the first level at 15 m. The time step was 90 seconds. Lat-eral boundary forcing was derived from six-hour ECMWF operational analyses. Sea-surface temperature (SST) and ice fraction were taken from satellite observations (using AVHRR and SSMI satellite data, see the ARCMIP home page), while the surface temperature over land is derived from the model’s own sur-face energy balance calculations. Two experiments were performed differing only in how sea ice was treated at the model surface. In the first experiment (hereafter C0), the model calculated ice-surface temperatures according to the lo-cal surface energy balance, while in the second simulation (hereafter C1) the ice-surface temperatures were prescribed from AVHRR data at a six-hourly resolution. The model results are com-pared to the measurements from the SHEBA Atmospheric Surface Flux Group (ASFG) instrumented tower (17) and to data from radio soundings performed through the whole year at the SHEBA site. For all the comparisons, we have used the model grid point closest to the SHEBA track at any given time. COAMPSTM results are also compared to the corresponding re-sults from the ECMWF operational forecasts valid at the SHE-BA site (hereafter EC). The ECMWF results used here are from a sequence of 36-hr forecasts starting at 12 UTC every day, us-ing only the last 24 hrs. Note here that the EC results are from an operational forecast model running in a data assimilation mode, where standard observations from SHEBA were ingested. Sta-tistics for November 1997 to January 1998 suggest that roughly 85% of the soundings reached ECMWF and entered the analysis (Bretherton, pers. comm.). In contrast, the COAMPSTM simula-tions are run continuously through the whole year and are forced only by the specification of the boundary conditions. This means that possible systematic errors are allowed to develop freely.

RESULTSErrors are defined as the model results minus the observations. Statistical results based on hourly data are summarized in

Table 1, while most of the results in the Figures are based on diurnal (for scatter plots) or weekly (for timeseries) averages for clarity. In addition to the standard error measures, such as the mean bias, root-mean-square error, and the correlation co-efficient, we also use the Index of Agreement, IoA2 (18). This can be viewed as an alternative correlation coefficient that takes into account phase differences between the compared signals. As an example, we note that while the correlation coefficient is zero between 2 sinus functions with the same amplitude that are a quarter of a wavelength out of phase, the IoA between these signals is still ~ 0.4. This indicates that existing similarity can be detected even in poorly correlated signals. Error evaluation is not entirely straightforward; here we use the principles outlined by Hanna (19). A good result is signified by a small bias, similar standard deviations of the model result and the corresponding observation, a root-mean-square error that is smaller than both these standard deviations, and finally a high correlation coef-ficient and/or IoA.

Figure 1. In (a–c) time series comparing simulated (red C0, blue C1 and green EC) and observed (black) weekly averaged temperature (a), specific humidity (b) and wind speed (c). Panels (d–f) show the corresponding scatter-plots, comparing diurnally averaged simulation and observation results.


Ambio Vol. 33 No. 4–5, June 2004 223

Mean Variables

Figures 1–3 show examples of comparisons of some mean me-teorological parameters between different models and observa-tions, see also Table 1. Figure 1 shows comparisons of the sim-ulated near-surface air temperature, humidity and wind speed with the corresponding observations from the ASFG tower at the SHEBA site. It is obvious that the C0 temperatures are too high, with an annual bias of ~ 3.5ºC. The bias is largest during winter but is also significant during summer, while spring and autumn results are better. The C1 temperatures are obviously closest to the observations, with an annual bias close to zero. This is not surprising as the ice-surface temperature is after all prescribed, although from a different set of measurements. The EC temperatures are also too high, on average by ~ 1.7ºC, more constantly during the whole year. There is a period in late spring/early summer (around day 150) when C0 and EC temperatures are close and both are closer to the observations than the C1 temperatures. Before this (around day 120) there is also a brief episode when the C0 and EC temperatures are similar but both deviate strongly from the observations. The EC results are from a short forecast initialized with the same analysis that is used to force both C0 and C1 simulation. This error is thus likely due to a failure in the ECMWF forecast system to capture the larger-scale flow. The C1 results are close to the observed values because C1 is forced closer to the observations at the surface. There are brief week long episodes in winter when the observed temperatures rise abruptly; during these, all the simulations are closer to the measurements. These are episodes with a low cloud cover, when the net radiation is close to zero at the surface and the boundary layer is relatively well mixed (17). This would in-dicate that all the models have a problem generating the very stable conditions that occur during the Arctic night when

clouds are not present. This could be due to problems with the surface energy balance or with high boundary-layer stability in general. It could also be that the models produce too dense wintertime clouds, or produce clouds also when clouds are not present in reality. It is intriguing that although the C1 tempera-ture error is close to zero as an annual average, its diurnally averaged error has the same annual structure as that from C0. It is too high for the mild (close to zero) and the quite cold conditions, and close to the observed or even slightly too low for the intermediate temperatures during spring and autumn. This could indicate a systematic error in COAMPSTM. The EC temperatures are systematically somewhat too high through-out most of the year, being only slightly too low in summer when the low-level temperature is governed by melting snow and ice. The errors in low-level moisture are small, but increase when the temperature increases during summer. This simply reflects its relationship to the saturation value, as a function of temperature, close to the surface with respect to ice in winter and liquid water in summer (20). The increasing error in hu-midity with increasing temperatures is illustrated by the C1 humidity, which is too low during a large part of the spring when the C1 temperature is marginally low and vice versa during the summer. We expect that most of this problem will disappear when the temperature biases are corrected. There is a risk of a positive feedback, with too much moisture causing too much boundary-layer clouds in particular during summer. This could alter the whole vertical structure of the boundary layer; see below. The winds are systematically too high in the C0 and C1 simulations, by ~ 2 m s-1, while the EC simulation is relatively accurate.Evaluating vertical structure in model simulations is much more difficult, due to the lack of high quality measurements; here

Table 1. Error statistics for several parameters, compared to observations. Bias is the mean error, is the standard de-viation, RMSE is the root-mean-square error, r is the correlation coefficient and IoA is the so-called Index of Agreement.

Tempera-ture (ºC)

Humidity (g kg-1)

Wind speed (m s-1)

Total heat flux (W m-2)

Sensible heat flux (W m-2)

Latent heat flux (W m-2)

Net shortwave(W m-2)

Net longwave (W m-2)

BiasC0 3.5 0.3 2.1 -23.6 1.8 2.9 -21.6 -17.5

C1 0.1 -0.1 2.3 -22.7 -2.5 0.6 -21.2 -10.6

EC 1.7 0.4 0.2 -4.4 -5.2 2.1 29.5 -19.3

C0 10.7 1.7 3.6 43.5 10.7 2.9 22.1 23.8

C1 11.6 1.4 3.9 38.8 11.5 0.6 22.5 24.6

EC 11.1 1.2 2.7 75.0 12.3 2.1 65.6 30.4

Obs 12.9 1.4 2.6 52.5 8.1 3.6 54.1 21.4

RMSEC0 5.0 0.7 2.9 39.2 11.8 5.6 41.1 29.0

C1 3.5 0.5 3.7 40.4 12.5 5.0 41.4 31.3

EC 4.7 0.4 1.7 46.3 13.1 6.1 46.8 29.0

rC0 0.93 0.91 0.58 0.68 0.23 0.35 0.73 0.18

C1 0.98 0.97 0.41 0.64 0.22 0.34 0.71 0.07

EC 0.96 0.98 0.80 0.79 0.11 0.31 0.79 0.43

IoAC0 0.93 0.93 0.67 0.76 0.51 0.52 0.64 0.47

C1 0.98 0.97 0.56 0.73 0.47 0.56 0.64 0.43

EC 0.96 0.98 0.89 0.86 0.43 0.49 0.78 0.59


224 Ambio Vol. 33 No. 4–5, June 2004

this evaluation is based on the regular radio soundings. Figure 2 shows, as an example, time-height cross-sections for January 1998, comparing temperature and specific moisture from C1 with soundings. Frequent appearances of intrusions of warm and moist air from more southerly latitudes are clearly seen here and are also present during most other months (not shown); this seems to be an inherent feature in lower troposphere in the Arc-tic. During the month of January, shown here, the C1 results ap-pear quite satisfactory, and the model simulates the magnitude, vertical structure and timing of these events, although details may still show slight differences. The warm bias close to the surface is evident during 12–25 January and there is also a cold bias around 4 km during the same period. The model appears to be somewhat too humid and the variability of the humidity in

the soundings is much larger than in the model. The nature and structure of some of that variabil-ity seems, however, to indicate problems with the measurements rather than with the model. During a few other months, mostly in summer, the model results are significantly poorer. The reason for this is unclear, unless it is related to the forcing from the lateral boundaries. As long as this forcing is strong, one can argue that deficiencies in the mod-el physics play a smaller role; mid-latitude influ-ence in the Arctic is typically smaller in summer. Vertical temperature-bias profiles are shown in Figure 3. Both the C0 and the C1 simulations have a small positive bias in the lower free tropo-sphere on average, but in both cases, the winter season is different with a slight cold bias. In the lowest troposphere, there is a double bias struc-ture in both C0 and C1, with a negative bias, on average around ~ 800 m, and the previously dis-cussed warm bias closer to the surface. For the C0 simulation the latter is ~ 3.5ºC, consistent with the evaluation against the ASFG mast data. Ex-amining the different seasons, all are too warm in the C0 boundary layer. The winter season (DJF) C0 warm bias is by far the largest, being twice or more than that during the other seasons. Autumn temperatures are closest to the measurements. This may be because the winds are in general stronger in autumn than in the winter and the conditions at the surface are thus more strongly affected by the

large-scale conditions. These are easily captured by the model as they are prescribed at the lateral boundaries. The negative bias around ~ 800 m is exclusively a summer phenomenon. The C1 vertical temperature structure is similar, only now the bias at the surface is forced closer to zero by the prescribed ice-surface temperature. This brings the spring temperatures closest to the measurements, autumn becomes slightly cold and winter is still the warmest at the surface, but too cold in the free tro-posphere. The cold bias peak around ~ 800 m is even larger in C1 than in C0. One mechanism that could cause this kind of error structure is if the summertime boundary layer becomes too deep and more well mixed than observed. The maximum cold bias would then appear at the top of the well-mixed boundary

Figure 3. Vertical profiles of the average temperature bias profile for the (a–b) C0 and (c–d) C1 simulations, with the annual average in (a, c) and divided into sea-sonal averages in (b, d).

Figure 2. Time-height cross-sections of (a) temperature and (b) specific humidity from soundings at the SHEBA site and the C1 model simulations from the month of January.


Ambio Vol. 33 No. 4–5, June 2004 225

layer in the simulation. A possible cause for such a structure is if the model, in the presence of the too high low-level moisture, develops too much or too dense (in an optical sense) stratocumulus clouds. This would generate a deeper boundary layer where the mixing is driven by cloud-top cooling. An additional contributing factor could be if the interaction between the clouds and the short-wave radiation is parameterized more like in mid-latitude stratocumulus, that are often opti-cally denser than Arctic stratocumulus in the pres-ence of more condensation nuclei.

Surface Fluxes

Figures 4 and 5 show the simulated atmospheric heat fluxes in all the three simulations, as com-pared to the SHEBA ASFG flux measurements; statistics for these results are also summarized in Table 1. While the errors in the model simu-lations are roughly constant during the Arctic winter, when the sun does not rise, there are ob-vious errors in both magnitude and phase in the models during summer. The EC energy balance peaks much too early in the summer, while the two COAMPSTM simulations are similar, and are more in phase with the measurements, although also peaking somewhat early. The annually aver-aged error in the EC simulation results is relative-ly close to zero, while the C0 and C1 results have negative biases of about 20–25 W m-2. However, the near-zero bias in the EC total heat flux seems to be the result of two compensating errors, dur-ing different parts of the season. There is an al-most constant modest negative bias in winter and a large localized positive bias in spring and early summer. C0 and C1 also have a negative bias, slightly smaller than in EC during winter, which remains negative through the whole year. In fact, these biases are due to different processes and do not become clear until the components of the heat flux are examined. Figure 5 examines the annual cycle of the indi-vidual components of the heat flux and their error. The turbulent fluxes are generally relatively small both in the model simulations and in the measure-ments, ± ~ 20 W m-2 for the weekly averages; larger one-hour values are present in both (not shown). The turbulent heat flux in C0 generally has a positive bias during the whole year, consis-tent with the warm bias. Both sensible and latent heat fluxes have absolute peaks around day 140, probably due to a combination of a slight tem-perature error in combination with a large posi-tive peak in the windspeed error. The C1 sensible heat flux is relatively close to the measurements, which is expected since the surface temperature is prescribed. The latent heat flux is too large dur-ing summer, which is consistent with the too high near-surface humidity. The EC sensible heat flux mostly has a negative bias that is larger during the winter, while the latent heat flux has a positive bias from spring through autumn. Overall, although the relative errors in the turbulent fluxes are sometimes very large, this may actually be a smaller problem as the fluxes

Figure 4. Plots of the total atmospheric heat flux (Wm-2), the sum of the turbulent sensible and latent heat flux and the net long- and shortwave radiation, showing (a) its annual cycle from models and observations, and (b) the model error as a function of time. The gray shaded region in the error plot outlines an error estimate defined as ± 3 standard deviations, calculated from nine model grid points surrounding the SHEBA site.

Figure 5. In (a–d) the annual-cycle of the components of the modeled and observed atmospheric surface heat-flux (Wm-2), and (e–h) the model errors. The gray shaded region in the error plot outlines an error estimate defined as ± 3 standard deviations, calculated from nine model grid points surrounding the SHEBA site.


226 Ambio Vol. 33 No. 4–5, June 2004

themselves are quite small. Certainly, the large temperature bias during the winter, in particular in C0, can not be explained by heat-flux errors of this magnitude. It appears more likely that the turbulent fluxes, even in C0 and EC where the surface tem-peratures are allowed to respond to erroneous fluxes, are more or less responding to errors in other parameters. One may therefore speculate if the importance of the Arctic boundary layer fluxes may be more indirect than direct. For example, an error in these heat fluxes can set the stage for cloud formation in the boundary layer. This can much more effectively alter the energy balance at the surface, through the cloud’s interaction with both solar and terrestrial radiation, than errors in the turbulent fluxes them-selves. The net radiative fluxes reveal biases of a larger magnitude in all three model-simulations, in particular during summer. It is clear that the net shortwave radiation in the EC simulation is too large. This error can be due either to a too small albedo or to a too small cloud cover, more likely the latter. The C0 and C1 results are similar to each other, and opposite and out of phase to the EC error. In fact, results from C0 and C1 during spring, early summer and late autumn are in good agreement with the observations. It is during the second half of summer that the COAMPSTM results develop a large negative bias, roughly of the same magnitude but out of phase with the positive EC bias. This is partly due to a too high surface albedo, not taking ac-count of the development of melt ponds. However, the timing also corresponds to the period when the near-surface humidity is too large and the vertical profile of the temperature bias is indicative of a too deep well-mixed layer. The conclusion is that this error is due both to errors in the cloudiness and to an erroneous surface albedo. As the observations show a large cloud fraction during this time, the cloudiness error is likely due to too thick or too optically dense boundary-layer clouds, rather than a too high cloud fraction. The longwave radiation errors have different structure. Dur-ing summer, when the net radiation is small, the error is also small, presumably because the low-level clouds saturate and become black bodies at a relatively small liquid-water path compared to their effect in shortwave radiation. Too thick or too dense clouds then have no additional effect on the error in the longwave radiation. During the winter, the net long-wave radiation is crucial, as there is no solar radiation. Contrary to the almost intuitive conclusion, that the near-surface warm bias should be due to an abundance of low-level clouds in the model, the net longwave radiation in all 3 simulations is more negative than in the observations. This must mean that there are actually less clouds in all the simulations than in the ob-servations, and that the low-level warm bias begs another ex-planation. The EC and C0 results are similar, with a somewhat larger error in the EC results, while the C1 is closer to the measurements, at least during some periods. The larger error in both the C0 and EC results are at least partly explainable by their too high surface temperatures.

DISCUSSION AND CONCLUSIONS Tentative results from first attempts to simulate the SHEBA-year with the COAMPSTM model are presented. The results are evaluated against observations from the SHEBA site, from the ASFG mast and from radio soundings, and are compared to ECMWF simulation results. With a few exceptions, the regional-model results are prom-ising and are, all in all, as good as those from the ECMWF forecast model. The latter was run in 24-hr increments, where each run was initialized with an analysis, using a data assimila-

tion cycle. The standard deviations of the simulated results for most variables analyzed here are of the same order of magni-tude as the standard deviation of the corresponding observa-tion. The root-mean-square error is also of the same magni-tude as the standard deviation, or smaller, with the exception of the atmospheric heat fluxes. Here the errors are slightly larger than the standard deviation for the sensible heat flux and the net longwave radiation. For latent heat flux and net short-wave radiation, the error is significantly larger, about a factor of two. The correlation coefficients are high for temperature and humidity and lower for the wind speed. With the excep-tion of shortwave radiation, the correlation coefficients are low or very low for the heat fluxes. It is worth noting, however, that for all the variables where the correlation is lower, the IoA is significantly higher than the correlation. As all the error statistics were evaluated from on hourly data, this indicates that there are timing problems on that time-scale, that may not necessarily invalidate the results on a longer times scale. The physics behind these calculations may thus be better than what would appear, even given such low correlation. The largest problem is the sometimes quite large systematic biases. For temperature and humidity during summer, these are due to an inadequate treatment of the melting of snow and ice on the surface. Instead of raising the temperature above the melting point, increased heat flux should result in increased melting; this is not treated properly in COAMPSTM. The tem-perature errors in winter is larger and more challenging to ex-plain, as it seems unlikely that they are due to overestimating low clouds. Even with the first model level, as low as 15 m, observations show that the actual surface layer is sometimes much thinner. This renders the assumptions on the surface-layer heat-fluxes, governing the temperature profile close to the surface, invalid. This by itself can prevent the model from generating the very stable conditions that are often observed. A second candidate for causing this bias is the heat flux through the ice and snow. Measurements indicate that the most impor-tant factor regulating this heat flux is the snow cover on top of the ice, even if this is much thinner than the ice. With a more insulating snow and ice cover, the surface temperatures would drop faster in response to a negative heat balance at the surface. Thus, a more successful model must treat this explic-itly and not as part of one homogeneous slab of ice and snow, which is currently the case in COAMPSTM. The windspeed bias is roughly constant and persists in spite of the fact that the mo-mentum flux bias is close to zero (not shown). More analysis is required to resolve this issue. The roughness length specified for ice surfaces in COAMPSTM is quite small, which may be one reason for this. The biases in the radiation fluxes dominate the atmospheric heat flux error. They are sometimes quite large, but can be un-derstood in terms of lack or abundance of low-level clouds. The latter are presumably, at least sometimes, a consequence of errors in the surface humidity, which is related to the error in surface temperature. The turbulent fluxes are small and seem to respond quickly to whatever forcing is applied. In conclusion, it seems that many of these errors would be much reduced by correcting the temperature error. This requires a closer look at the clouds and the energy balance at the surface.

References and Notes

1. IPCC 2001. Climate Change 2001. The Scientific Basis. Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change, Houghton, J.T., Ding, Y., Griggs, D.J., Noguer, M., van der Linden, P.J., Dai, X., Maskell, K. and Johnson, C.A. (eds). Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 881pp.

2. SEARCH 2001. SEARCH: Study of Environmental Arctic Change, Science Plan. Polar Science Center, Applied Physics Laboratory, University of Washington, Seattle, 91pp.


Ambio Vol. 33 No. 4–5, June 2004 227

3. Räisänen, J. 2001. CO2-induced climate change in the Arctic area in the CMIP2 experi-ments. SWECLIM Newsletter 11, 23-28.

4. Walsh, J.E., Kattsov, W.M., Chapman, W.L., Govorkova, V. and Pavlova, T. 2002. Com-parison of Arctic climate by uncoupled and coupled global models. J. Clim. 15, 1429-1446.

5. Serreze, M.C., Walsh, J.E., Chapin III, F. S., Osterkamp, T., Dyurgerov, M., Ro-manovsky, V., Oechel, W.C., Morison, J., Zhang, T. and Barry, R.G. 2000. Observa-tional evidence of recent change in the northern high-latitude environment. Climate Change 46, 159-207.

6. Mahrt, L. 1998. Stratified atmospheric boundary layers and breakdown of models. The-oret. Comput. Fluid Dynamics 11, 263-279.

7. Zilitinkevich, S. 2002. Third-order transport due to internal waves and non-local turbu-lence in the stably stratified surface layer. Quart. J. Roy. Met. Soc. 128, 913-926.

8. Intrieri, J., Fairall, C.W., Shupe, M.D., Persson, P.O.G. andreas, E.L., Guest, P.S. and Moritz, R.E. 2002. Annual cycle of Arctic surface cloud forcing at SHEBA. J. Geophys. Res. 107 (C10), 8039, doi:10.1029/2000JC000439.

9. Beesley, J.A., Bretherton, C.S., Jacob, C. andreas, E.I., Intrieri, J.M. and Uttal, T.A. 2000. A comparison of cloud and boundary layer variables in the ECMWF forecast model with observations at the Surface Heat and Energy of the Arctic (SHEBA) ice camp. J. Geophys. Res. 105, No. D10, 12337-12349.

10. Vihma, T, Hartman, J. and Lu" pkes, C. 2003. A case study of an on-ice flow over the Arctic marginal sea-ice zone. Bound.-Layer Meteorol. 107, 189-217.

11. Rinke, A., Lynch, A.H. and Dethloff, K. 2000. Intercomparison of Arctic regional cli-mate simulations: Case studies of January and June 1990. J. Geophys. Res. 105, 29669-29683.

12. Pinto, O.J., Curry, J.A. and Lynch, A.H. 1999. Modeling clouds and radiation for the November 1997 period of SHEBA using a column climate model. J. Geophys. Res. 104, 6661-6678.

13. Curry J.A. and Lynch, A.H. 2002. Comparing arctic regional climate models. EOS Trans. Amer. Geophys. Union. 83, 87.

14. Jones, C.G., Wyser, K., Ullerstig, A. and Willén, U. 2004. The Rossby Centre Regional Atmospheric Climate Model (RCA). Part II: Application to the Arctic Climate. Ambio 33, 211-220.

15. Perovich, D.K. andreas, E.L., Curry, J.A., Eiken, H., Fairall, C.W., Grenfell, T.C., Guest, P.S., Intrieri, J., Kadko, D., Lindsay, R.W., McPhee, M.G., Morison, J., Moritz, R.E., Paulson, C.A., Pegau, W.S., Persson, P.O.G., Pinkel, R., Richter-Menge, J.A., Stanton, T., Stern, H., Sturm, M., Tucker, W.B. and Uttal, T. 1999. Year on ice gives climate in-sights. Eos Trans. 80, 483-486.

16. Hodur, R.M. 1997. The Naval Research Laboratory’s coupled ocean/atmosphere meso-scale prediction system. Month Wea. Rev. 125, 1414-1430.

17. Persson, P.O.G., Fairall, C.W. andreas, E.L., Guest, P.S. and Perovich, D.K. 2002. Measurements near the atmospheric surface flux group tower at SHEBA: Near-surface conditions and surface energy budget. J. Geophys. Res. 107, 8045, doi:10.1029/2000JC000705.

18 The I0A is defined as

(P – O)2 IoA = 1 –

( P – O + O – O )2

1

N

1

N

where P and O are predicted and observed values, respectively, N is the number of ob-servations and an over-bar indicates a time average.

19. Hanna, S.R. 1994. Mesoscale meteorological model evaluation techniques with empha-sis on needs of air quality models. In: Mesoscale Modeling of the Atmosphere. Pielke, R.A. and Pearce, R.P. (eds), Meteorological Monographs, published by the American Meteorological Society, pp. 47-62.

20. Andreas, E.L., Guest, P.S., Persson, P.O.G., Fairall, C.W., Horst, T.W., Moritz, R.E. and Semmer, S.R. 2002. Near-surface water vapor over polar sea ice is always near ice satu-ration, J. Geophys. Res. 107, 1-17

21. This work was carried out partly within the SWECLIM program and Financial support from the Foundation for Strategic Environmental Research (MISTRA) and the Swedish Meteorological and Hydrological Institute (SMHI) is gratefully acknowledged. Addi-tional support from the Swedish Science Council (Vetenskapsrådet) is also acknowl-edged. The authors are grateful for the support from the ARCMIP community in setting up the simulations and to the SHEBA Atmospheric Flux Group for access to the obser-vations. We thank Christian Jakob and his coworkers at ECMWF for producing the ECMWF column data set for SHEBA.

Michael Tjernström completed his BSc in me-teorology at Stockholm University in 1980, as a part of his Air Force Officer Training, and worked as a forecaster for a few years before commenc-ing graduate studies at Uppsala University in 1983. His PhD thesis on boundary-layer clouds was presented in 1988, and he remained in Up-psala as researcher and lecturer until 1998. He then received a senior scientist position from the Swedish Natural Sciences Research Council and returned to Stockholm University. He became professor in [email protected]

Mark Zagar completed an undergraduate educa-tion in physics/meteorology at the University of Ljubljana and presented a PhD on small-scale adaptation at the universities of Ljubljana and Toulouse, France, in 2000. He worked at the Slo-venian Meteorological Service from 1993 until 2000, and later as a post-doctoral fellow at the Stockholm [email protected]

Gunilla Svensson completed an undergraduate education at Uppsala University in 1988 and pre-sented a PhD thesis on mesoscale modeling of photochemical air pollution in 1995 in Uppsala. After a stay as a post-doctoral fellow at California Institute of Technology in Pasadena, USA, she has been an assistant professor at Stockholm University since 1997. She was appointed asso-ciate professor in [email protected]

Their address: Department of Meteorology, Stock-holm University, SE-106 91 Stockholm, Sweden

model simulations of the arctic atmospheric boundary layer from the sheba year

Documents