Introduction
One of the major problems in sea-ice modelling is the high sensitivity of the model results to the prescribed forcing, in particular to oceanic heat flux and wind. The quality of the model results is, therefore, highly dependent on the quality of the forcing fields. Since the Southern Ocean area is probably the one with the coarsest observation net on Earth, one can expect that errors in the forcing fields may be at least as important as model errors.
In order to avoid the specification of an oceanic heat flux, the sea-ice (SI) model of Hibler and Ackley (Reference Hibler, III. and Ackley1983) has been coupled to a one-dimensional oceanic mixed-layer (OML) model (Lemke and others, Reference Lemke, Owens and Hibler, III.1990; Owens and Lemke, Reference Lemke, Owens and Hibler, III.1990).
Similarly, the atmospheric forcing was determined interactively by coupling the same SI model to a one-dimensional atmospheric boundary-layer (ABL) model, where the forcing has been raised to the geostrophic level (Koch, Reference Koch1988).
In this paper, the influence of the coupling of the SI model alternatively to the OML and ABL models will be investigated and is, in the following, presented in a hierarchy of coupling schemes. The influence of the wind forcing from a different level is shown in a final sensitivity experiment.
The Si-Oml Model
The coupled SI-OML model from Lemke and others (Reference Lemke, Owens and Hibler, III.1990) and Owens and Lemke (Reference Lemke, Owens and Hibler, III.1990) has been applied to the entire Southern Ocean region (Stössel and others, Reference Stössel, Johannessen, Haugan and Owens1990). The SI model is based on Hibler (Reference Hibler, III., Hansen and Takahashi1984), with an additional continuity equation for the snow thickness (Owens and Lemke, Reference Lemke, Owens and Hibler, III.1990). The OML model is one-dimensional and incorporates an exponentially shaped pycnocline (Lemke, Reference Lemke1987). The model is run on a spherical, circumpolar grid with a latitudinal resolution of 2.5° and a longitudinal resolution of 5°, extending from 50°−80°S, using a daily time step.
The forcing is prescribed in the lowest layer of the atmosphere and at a depth of 500 m in the ocean.The forcing with respect to the intermediate ocean temperature and salinity, the geostrophic current, the precipitation and the cloudiness is derived from climatologies (Stössel and others, Reference Stössel, Johannessen, Haugan and Owens1990). The atmospheric forcing with respect to temperature, humidity and wind, on the other hand, is determined from daily (real-time) computations derived from the global analyses of the ECMWF (Trenberth and Olson, Reference Trenberth and Olson1988), where the lowest level is described at 1000hPa. To drive the SI-OML model, the wind is taken directly from that level, following Trenberth and others (Reference Trenberth and Large1989); whereas the temperature and humidity fields above the surface are computed by interpolation between the 850hPa and the 1000hPa fields or, if the 1000hPa level is above the surface, by extrapolation and lapse rate formulation, respectively.
Cyclo-stationarity of the response of ice volume and ice extent is achieved after five years of simulation. The forcing for the first five years of model integration is specified from data analyses for the year 1985, whereas for the sixth year it is transiently given by the data for 1986.
Fig. 1. Ice thickness contours, as simulated by the SI-OML model for the approximate date of (a) minimum (1 March 1986) and (b) maximum (27 September 1986) ice extent.
This final year will be used in the following analyses of the model results.
With this new kind of forcing, the SI-OML model is run without any changes in the model parameters (i.e. ice strength and lead-closing parameter) as well as drag and transfer coefficients relative to the standard version described in Stössel and others (Reference Stössel, Johannessen, Haugan and Owens1990). The resulting ice thickness distribution is shown in Figure 1 for the approximate time of minimum (a) and maximum (b) ice extent. Day 60 corresponds to 1 March 1986 and day 270 to 27 September 1986.
Compared to the standard results in Stössel and others (Reference Stössel, Johannessen, Haugan and Owens1990), where monthly forcing values from the Taljaard and others (Reference Taljaard, Crutcher and Jenne1969) climatology have been employed, the new results with the daily forcing are similar in the Weddell Sea, where the observed characteristic ice thickness pattern is captured by the model. In other
Fig. 2. Ice thickness contours, as simulated by the SI-OML- ABL model for the same dates as in Figure 1.
regions, for example east of Victoria Land and in the Bellingshausen Sea, slight improvements are achieved compared to the previously published results. The main deficiencies compared to observations are a slightly overestimated winter ice extent and a general over-estimation of the compressions in convergent ice areas.
The Si-Oml-Abl Model
Since the network of meteorological observing stations in the Southern Ocean area is rather sparse, the fields of the global analyses consist mainly of model forecast results from the numerical weather prediction (NWP) model of the ECMWF. The NWP model results of the lowest level (especially the air temperature), however, are highly dependent on the specified lower boundary conditions (especially the sea-surface temperature patterns and the sea-ice boundaries). In order to reduce the influence of those conditions (which are primarily climatological) on the sea-ice model results, the forcing level is raised to the next higher level which, in the case of the ECMWF analyses, is the 850hPa level. This is done by coupling the SI-OML model to the ABL model used by Koch (Reference Koch1988). This model resolves the Prandtl layer with the Monin-Obukhov theory and the barotropic Ekman layer with the Rossby number similarity theory, both in a vertically integrated form.
The increase of the forcing level concerns the variables temperature, humidity and wind. The sea-level pressure is derived from the same data set in order to determine the potential temperature.
Initially, the SI-OML-ABL model is run without any specific tuning. The resulting ice thickness distribution is shown in Figure 2. The differences compared to the previous results (Fig. 1) are striking: the ice extent is reduced significantly and the ice thickness distribution resembles that of a pure thermodynamic sea-ice model (Stössel and others, Reference Stössel, Johannessen, Haugan and Owens1990). Compared to the results with surface wind forcing (Stössel, Reference Stössel, Johannessen, Haugan and Owens1990) the mean ice velocities are significantly reduced, especially in the Ross Sea (up to 8cms−1; Fig. 3).
Fig. 3. Mean ice velocities for September 1986, as simulated by the SI-OML-ABL model.
Although the thinner ice thickness compares better with observations (Wadhams and others, Reference Wadhams, Lange and Ackley1987), the results are rather unrealistic. Since the ice extent could be described rather well by Koch (Reference Koch1988), with the Hibler and Ackley (Reference Hibler, III. and Ackley1983) sea-ice model for the Weddell Sea, the present model hierarchy will be degraded to that model in the following.
Reduction of the Model Physics
The essential differences between the present SI-OML model and the sea-ice model of Hibler and Ackley (Reference Hibler, III. and Ackley1983) are the coupling to the OML model and the prognostic simulation of the snow thickness (Owens and Lemke, Reference Lemke, Owens and Hibler, III.1990). Figure 4 demonstrates the effect of neglecting those two processes on the seasonal cycles of ice extent and ice volume. The lower curves (1 in Fig. 4) show the results of the previous run with the complete SI-OML-ABL model. Neglecting the prognostic determination of the snow cover raises the ice volume by about 2 x 103 km3 in summer and 4 x 103 km3 in winter, whereas the ice extent increases only slightly by about 1 x 106 km2 in Fig. 4). This corresponds qualitatively to the results of a similar experiment in Owens and Lemke (Reference Lemke, Owens and Hibler, III.1990) and Stössel and others (Reference Stössel, Johannessen, Haugan and Owens1990). The quantitative differences can be attributed primarily to the different transfer coefficients of the turbulent heat fluxes, which are determined diagnostically by the friction velocity when coupled to the ABL model.
If, in addition, the OML model is decoupled and, consistent with Hibler and Ackley (Reference Hibler, III. and Ackley1983), a constant
Fig. 4. Seasonal cycles of ice extent and ice volume for the year 1986, as simulated by: (1) the SJ-OML-ABL model; (2) the same as (1) without snow; (3) the same as (2) with the OML model decoupled (SI- ABL model); and (4) the same as (3) with higher lead-closing parameter.
mixed-layer thickness of 30 m and a constant oceanic heat flux of 2 W m−2 is assumed, one achieves the seasonal cycles marked by 3 in Figure 4. With this modification, the winter maximum ice volume increases by 4 x 103 km3 and the maximum ice extent by 5.5 X 106 km2, whereas the summer values remain almost unchanged. The influence of the prognostic mixed layer is relatively large and resembles the corresponding influence on the pure thermodynamic results (Stössel and others, Reference Stössel, Johannessen, Haugan and Owens1990). This indicates that the overall ice drift is also underestimated in the present experiment.
An increase in summer ice extent of about 1 x 106 km2, and a further increase of the maximum ice volume of about 4 x 103 km3, can finally be achieved by raising the lead-closing parameter in the sea-ice model
Fig. 5. Ice thickness contours, as simulated by the SI-ABL model neglecting snow, using a higher lead-closing parameter and employing the 1000 hPa wind field instead of the 850 hPa one, for the same dates as in Figure 1.
from 0.5 to 1.0 m, as was done by Koch (Reference Koch1988) and Hibler and Ackley (Reference Hibler, III. and Ackley1983). This parameter is empirical, and determines the rate of lateral freezing, with the higher value leading to slower lead closing. As compared to the observed seasonal cycle of ice extent for 1986 (Gloersen and Campbell, Reference Gloersen and Campbell1988), where the minimum is about 3.4 x 106 km2 and the maximum about 17.7 x 106 km2, the results with the above modifications have significantly improved (curves labelled 4 in Fig. 4).
Sensitivity to Wind Forcing
Whereas the latter improvements are valid for the seasonal cycles, this is not the case for the areal ice thickness distribution, which is the most sensitive indicator for incorrect sea-ice simulations. Particular strong ridging of ice occurs at the west coast of the Antarctic Peninsula. This resembles the results with the alternative climatological wind forcing from Hellerman and Rosenstein (Reference Hellerman and Rosenstein.1983), shown in Stössel and others (Reference Stössel, Johannessen, Haugan and Owens1990), where the wind stress vectors are mostly zonally (eastward) distributed over the entire domain, especially in winter. As can be expected, there is also a more zonal pattern in the 850 hPa wind field of the ECMWF analyses than in the corresponding 1000hPa field, where the wind field is influenced by the orography.
The ice thickness distribution improves significantly by replacing the 850 hPa winds by those of the 1000 hPa level, keeping the wind turning angle at zero, neglecting snow and mixed-layer effects, and using the higher lead-closing parameter (Fig. 5). Compared to Figure 1, the new results are more realistic in the way that the winter ice extent is lower and the ridging in the convergent ice regimes is much less. Compared to Figure 2, on the other hand, the overall ice extent and ice thickness are largely increased (towards observations) and the pure thermodynamic character of the thickness distribution is replaced by a more dynamic one.
In spite of these improvements, the observed characteristic ice thickness distribution in the Weddell Sea (which might be considered as some kind of reference area for the entire Southern Ocean) is still under-developed in this experiment, indicating that the surface drag, which is calculated via the stability dependent friction velocity, is too low.
Conclusions
Without any tuning, the SI-OML model forced with daily (real-time) atmospheric variables (temperature, humidity and wind) from the global analyses of the ECMWF instead of monthly (climatological) variables of the Taljaard and others (Reference Taljaard, Crutcher and Jenne1969) data set, yields reasonable simulation results. This supports the quality of the 1000 hPa values of the ECMWF analyses. The main deficiency, namely the over-estimated ridging in convergent ice areas, can be improved by either tuning the ice strength parameter of the sea-ice model and simultaneously the drag coefficient for the air/ice stress or by applying an adequate Prandtl-layer formulation similar to that in the ECWMF-AGCM, both of which will be demonstrated in a subsequent paper.
The simulation using the SI-OML-ABL model with the above-mentioned forcing variables raised to the 850 hPa level is less successful, unless the snow cover is neglected, the OML model decoupled and the lead-closing parameter increased. Even then the overall ice thickness distribution remains unrealistic, unless the 1000hPa wind field is applied instead of the higher level one. The main reason for this feature is the more zonal pattern of the wind velocity in the 850 hPa field, where orographic influences are rather small. Those influences, in terms of katabatic and barrier winds, however, seem to be decisive for realistic ice thickness distributions (see also Schwerdtfeger, Reference Schwerdtfeger1979), making it necessary to employ the lower-level wind fields. Since the wind field is less predetermined by the specified surface boundary conditions of the NWP model than the temperature and humidity fields, this measure should be less critical.
The model results indicate that the parameterizations in the coupled SI-OML-ABL model have to be read-justed in order to make them reasonable. Assuming that the parameters of the SI-OML model are adequate, the sensitivities of several ABL model parameters were analyzed. It will be shown in a subsequent publication that the results can be improved, for example, by raising the roughness lengths over ice and water, and by increasing the thermally induced turbulence (buoyancy).
Acknowledgements
W.D. Hibler III, P. Lemke, W.B. Owens and C. Koch are thanked for placing their models at my disposal, P. Lemke, M. Heimann and K. Herterich for comments on an earlier version of the paper, H. Schriever for typing the manuscript and M. Grunert for drawing the figures. Thanks are also due to the Deutscher Wetterdienst (DWD), Offenbach, for providing the analyses of the ECWMF, and to R. Schnur for preprocessing these data. This work was partially supported by the Sonderforschungsbereich (SFB) 318 and the Max-Planck-Institut für Meteorologie, Hamburg.
