Surface heat budget over the Weddell Sea: Buoy results and model comparisons

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 17, NO. C2, 313, 1.129/2JC372, 22 Surface heat budget over the Weddell Sea: Buoy results and model comparisons Timo Vihma, Juha Uotila, Bin Cheng, and Jouko Launiainen Finnish Institute of Marine Research, Helsinki, Finland Received 6 April 2; revised 24 August 21; accepted 26 September 21; published 26 February 22. [1] The surface heat budget over the Weddell Sea ice cover in 1996 was studied on the basis of data from Argos buoys equipped with meteorological sensors. In addition, a thermodynamic sea ice model, satellite-based data on the sea ice concentration, sonar results on ice thickness distribution, and output from large-scale meteorological models were all utilized. Applying the buoy data, the sensible heat flux over sea ice was calculated by Monin-Obukhov theory using the gradient method, and the latent heat flux was obtained by the bulk method. A second estimate for the surface fluxes was obtained from the thermodynamic sea ice model, which was forced by the buoy observations. The results showed a reasonable agreement. The dominating component in the heat budget over ice was the net longwave radiation, which had a mean annual cooling effect of 28 W m 2. This was balanced by the net shortwave radiation (annual mean 13 W m 2 ), the sensible (13 W m 2 ) and latent ( 3 Wm 2 ) heat fluxes, and the conductive heat flux through the ice (5 W m 2 ). The regional surface fluxes over the fractured ice cover were estimated using the buoy data and Special Sensor Microwave Imager (SSMI)-derived ice concentrations. In winter the regional surface sensible heat flux was sensitive to the ice concentration and thickness distribution. The estimate for the area-averaged formation rate of new ice in leads in winter varies from.5 to.21 m per month depending on the SSMI processing algorithm applied. Countergradient fluxes occurred 8 1% of the time. The buoy observations were compared with the operational analyses of the European Centre for Medium-Range Weather Forecasts (ECMWF) and the reanalyses of the National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR). The 2 m air temperature and surface temperature were 3.5 and 4.4 C too high, respectively, in the ECMWF and 3.2 and 3. C too low in the NCEP/NCAR fields, but the models reproduced the synoptic-scale temperature variations well. The errors seem to be related to the cloud cover and the surface boundary conditions. Neither of the models recognizes leads in the ice pack, and the ice and snow thicknesses are often far from reality. The distribution of the cloud cover in the both models differed a lot from observations. INDEX TERMS: 3339 Meteorology and Atmospheric Dynamics: Ocean/atmosphere interactions (312, 454); 3349 Meteorology and Atmospheric Dynamics: Polar meteorology; 454 Oceanography: Physical: Ice mechanics and air/sea/ ice exchange processes; KEYWORDS: surface fluxes, ice thermodynamics, Weddell Sea, sea ice, leads 1. Introduction [2] The surface heat budget over ice-covered seas depends on the radiative, turbulent, and conductive heat fluxes. In summer the thermal differences at the air-ice-sea interfaces are small, but during most of the year in the polar oceans the sea ice and its snow cover act as insulators between the relatively warm ocean and the cold atmosphere. Most of the observations of the surface heat budget of sea ice have been made over Arctic and sub-arctic regions. On the basis of the Arctic data, in winter the atmospheric surface layer over thick sea ice is typically stably stratified because of large heat losses via longwave radiation. The turbulent heat flux is therefore generally directed downward [Vowinckel and Orvig, 1973; Untersteiner, 1986; Serreze et al., 1992]. The heat flux through the ice and snow strongly depends on the thickness of these layers [Makshtas, 1991]. [3] Antarctic sea ice is generally thinner than that in the Arctic, and the ice concentration is typically lower. It is therefore not so clear that a stable stratification also prevails in the wintertime atmospheric surface layer over Antarctic sea ice. Results from various regions and seasons have indicated both stable and Copyright 22 by the American Geophysical Union /2/2JC372 unstable stratification [Andreas and Makshtas, 1985; Kottmeier and Engelbart, 1992; Wamser and Martinson, 1993; Launiainen and Vihma, 1994; Vihma et al., 1996]. In the Weddell Sea the surface layer air temperature over sea ice in winter is typically from 1 to 25 C [Kottmeier et al., 1997]. Synoptic-scale variations in the air temperature and turbulent fluxes are strong and depend especially on the cloudiness and wind direction. Large spatial variations in the snow thickness [Massom et al., 1997; Worby et al., 1996] also have a strong effect on the stratification. [4] The Weddell Sea is covered by ice during the austral winter, but leads are generated by tidal and larger-scale ice drift divergence [Padman and Kottmeier, 2], and wider polynyas are frequent in the southeast and south near the ice shelves and the coast of the Antarctic continent [Zwally et al., 1985]. Over ice-free areas, cracks, leads, and polynyas the heat flux from the sea to the air may reach values of several hundred watts per square meter [Andreas et al., 1979]. The sensible heat flux is usually the largest component, but latent heat flux and net longwave radiation are also important. Even if open water or thin, new ice only compose a few percent of the surface area, the large upward fluxes may make a major contribution to the regional heat budget. [5] Reliable estimates of the regional surface fluxes over fractured sea ice are still rare, although their importance is well demonstrated for the sea ice dynamics [Stössel and Claussen, 5-1

2 5-2 VIHMA ET AL.: SURFACE HEAT BUDGET OVER THE WEDDELL SEA 4W 2W 6W W 6S 2 km 6S 7S 7S 6W 4W 2W W Figure 1. Trajectories of the two meteorological buoys in the Weddell Sea (buoy 5895 as a solid line and buoy as a dotted line). 1993], the sea ice heat and mass balance [Eisen and Kottmeier, 2], the atmospheric boundary layer [Vihma, 1995], the largescale atmospheric circulation [Simmons and Budd, 1991], and the ocean convection [Chapman, 1999]. Our objective is to provide such estimates in this paper. Hence we analyze data obtained by automatic marine meteorological buoys and compute estimates for turbulent and radiative surface fluxes over the sea ice and leads. The buoy data and an application of a thermodynamic ice model provide local heat fluxes; information on the ice concentration and thickness distribution is used to estimate regional fluxes. Data from analyses and forecasts of large-scale meteorological models are then compared to the buoy observations. The key concepts we try to address are (1) the contribution of the various flux components in the local surface energy balance over sea ice, (2) the importance of the ice concentration and thickness distribution on the regional surface energy balance, and (3) the accuracy of the large-scale atmospheric models in estimating the near-surface air temperature, the wind speed, and the surface fluxes over the Antarctic sea ice. The last aspect is particularly important for the forcing of sea ice and ocean models. 2. Buoy Observations [6] During the FINNARP-95/96 expedition of the Finnish Antarctic Research Program, seven Argos buoys were deployed from R/V Aranda on ice floes in the central Weddell Sea in January February Two of the buoys, numbers 5895 and 25161, had a meteorological mast with instruments for the following quantities: atmospheric pressure, air temperature at two height levels, air humidity, wind speed, and direction. In addition, the buoy hull temperature was measured at a depth of.7 m below the snow surface. The rest of the buoys were used in studies of ice dynamics [Uotila et al., 2]. The trajectories of the two meteorological buoys are presented in Figure 1. The buoys were located and the data were transmitted by the Argos satellite survey, yielding an average observation interval of 1 hour 1 min. Buoy saved and transmitted hourly data, however. The buoy data were transmitted to the Global Telecommunication System (GTS) of the World Meteorological Organization. [7] To ensure data quality and to measure the temperature gradient in the atmospheric surface layer, the buoys had air temperature sensors at two different levels. The measurement heights, accuracy of sensors, and functioning periods of the buoys are presented in Table 1. A more detailed description of the sensor configuration, types, and calibrations is given in a technical data report [Vihma et al., 1997]. The anemometers occasionally stopped during the austral winter, which was most probably due to ice or snow accretion. The wind speed for these periods (16 days in all) was estimated from the European Centre for Medium-Range Weather Forecasts (ECMWF) geostrophic winds by assuming that the surface wind is 45% of the geostrophic wind [Uotila et al., 2]. It should be noted that an accurate measurement of the surface temperature over sea ice, with snow accumulating and melting, is very difficult to make using an automatic buoy with a restricted electrical power capacity (e.g., not allowing the use of an infrared thermometer). In order to estimate the sensible heat flux therefore measurements were made of the temperature gradient in the air. On the other hand, a buoy measurement of the humidity gradient was not accurate enough to derive the latent heat flux. Instead, it can be derived by utilizing the data of the air temperature gradient and a humidity measurement at a single level. 3. Flux Calculations 3.1. Diagnostic Method Turbulent fluxes. [8] The surface sensible heat flux H over ice was computed from the measurements of wind speed and the air temperature gradient applying the Monin-Obukhov similarity theory: rc p k 2 ðq 2 q 1 ÞV 2 H ¼ ½lnðz 2 =z Þ c M ðz 2 =LÞŠ½lnðz 2 =z 1 Þ c HE ðz 2 =LÞþc HE ðz 1 =LÞŠ ; where r is the air density, c p is the specific heat, and k is the von Karman constant (=.4). q is the potential temperature of the air, V is the wind speed, and z is the aerodynamic roughness length. L is the Obukhov length (in meters), which depends on the sensible heat flux. The functions c M and c HE describe the stability effect. (1)

3 VIHMA ET AL.: SURFACE HEAT BUDGET OVER THE WEDDELL SEA 5-3 Table 1. Buoy Observations a Observation Heights, m Accuracy Buoy Observation Period b V T a RH V, m s 1 T a, C RH, % Feb to 11 Jan , , , c 5 d Jan. to 29 May , c 5 d a V is the wind (speed and direction), T a is the air temperature, and RH is the relative humidity. b Some of the buoy sensors failed earlier. c The accuracy of the mutual difference between the two temperature sensors was.2.5 C. d The accuracy decrease in temperatures below about 1 C. We calculate z over ice according to Banke et al. [198] using 1 cm as the geometric roughness, which yields 1 mm for z. For c M and c HE we apply the method of Holtslag and de Bruin [1988] for the stable region (1 m/l ) and that of Högström [1988] for the unstable region (1 m/l < ). The subscripts 1 and 2 refer to the lower and higher measurement heights on the buoys, respectively. However, the temperature difference q 2 q 1 is often small, and thus its measurement is liable to errors. To avoid the worst errors, we checked the results by calculating the surface temperature q s corresponding to the heat flux: where h = 46.5(e/q 2 ), LWR ¼ e LWR # est 4 S ; ð5bþ where T S is the surface temperature, s is the Stefan-Boltzmann constant, N is the cloud fraction from to 1, e is the surface emissivity (.97), and e is the water vapor pressure (in hp. The net shortwave radiation at the surface (SWR) was estimated by the formula of Shine [1984], with the cloud effect according to Bennett [1982]: q s ¼ q ZT ¼ q 2 H ½ ln ð z 2=z Þ c M ðz 2 =LÞŠ½lnðz 2 =z T Þ c HE ðz 2 =LÞŠ rc p k 2 ; V 2 ð2þ S cos 2 Z SWR ¼ ðcos Z þ 1:Þe1 3 þ 1:2 cos Z þ :455 ð1 :52NÞð1 aþ ; ð6þ where q s is defined as equivalent to q ZT ; the temperature at the height of the thermal roughness length z T, which we calculate according to Andreas [1987]. Cases with the calculated q s exceeding C or q s q 2 4 C were considered as suspect and were not included in the analyses (for the basis of the threshold, see, e.g., Overland et al. [2]). This eliminates unrealistic spikes from the results. Such spikes could not be easily distinguished from the directly observed values of q 1 and q 2. Both (1) and (2) were solved iteratively (separately) using the method of Launiainen and Vihma [199]. In calculating the fluxes over leads the bulk method (3) was applied, assuming a surface temperature at the freezing point of 1.8 C. Over narrow leads, constant values were used for z [Andreas and Murphy, 1986] and z T [Launiainen, 1983]. rc p k 2 ðq 2 q S ÞV 2 H ¼ ½lnðz 2 =z Þ c M ðz 2 =LÞŠ½lnðz 2 =z T Þ c HE ðz 2 =LÞŠ : ð3þ [9] The latent heat flux LE was calculated applying the bulk method both over the ice and leads: rlk 2 ðq 2 q S ÞV 2 LE ¼ ½lnðz 2 =z Þ c M ðz 2 =LÞŠ½lnðz 2 =z T Þ c HE ðz 2 =LÞŠ ; ð4þ where q s is the saturation specific humidity at the surface, calculated from q s from (2), q 2 is the specific humidity of the air, and l is the enthalpy of vaporization. The roughness length for humidity is assumed to equal z T Radiative fluxes. [1] The downwelling longwave radiation at the surface (LWR#) was estimated from the formula of Prata [1996], with the cloud effect according to Jacobs [1978], finally yielding for the net longwave radiation LWR: h p LWR #¼ 1 ð1 þ hþexp ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi i 1:2 þ 3:h sq 4 2 ð1 þ :26NÞ; ð5aþ where S is the solar constant and Z is the solar zenith angle. We used the following values for albedo:.85 for dry snow,.77 for melting snow,.7 for bare ice, and.4 for melting ice [Perovich, 1998; Allison et al., 1993]. The state of the surface was deduced from the results of the thermodynamic model (section 3.2). Over a lead surface, albedo was calculated as dependent on the solar altitude [Payne, 1972]. [11] The most uncertain factor in the calculation of the radiative fluxes is the cloud cover. We compared three data sets of the total cloud cover: (1) the operational analyses of the ECMWF [1997], (2) the reanalyses of the National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) [Kalnay et al., 1996], and (3) estimates we calculated on the basis of the buoy air temperatures applying the method of Makshtas et al. [1999]. This method is applicable for a polar winter, when the air temperature is mostly determined by the cloud cover. The algorithm produces a distribution of the total cloud cover that is U-shaped, in agreement with a large observational data set [Makshtas et al., 1999]. Applied to our air temperature data from March through October, it produced a mean cloud cover of.59. The corresponding ECMWF mean cloud cover was.83, and that of the NCEP/NCAR was.5. The ECMWF distribution of the cloud cover was close to a gamma distribution, while that of the NCEP/NCAR was close to uniform; that is, they both strongly disagreed with the Makshtas et al. [1999] results. Hence we used the Makshtas et al. results in our calculations from March through October, while during austral summer we applied the ECMWF cloud cover. This was because it had a better agreement with our (limited period) ship observations than the NCEP/NCAR cloud cover and because the ECMWF relative humidities agreed better with the buoy data. The insufficient cloud cover of the NCEP/NCAR in the polar regions has also been noted by World Climage Research Program (WCRP)/Scientific Committee on Oceanic Research (SCOR) [2] Conductive fluxes. [12] Measurements of the temperature at a depth of.7 m below the snow surface allowed us to estimate roughly the heat flux from the ocean through the ice and snow. In a stationary situation with homogeneous layers of snow

4 5-4 VIHMA ET AL.: SURFACE HEAT BUDGET OVER THE WEDDELL SEA Ice thickness (m) March Month September c) 1 1 % 5 % ice thickness (m) ice thickness (m) Figure 2. ( Mean monthly ice thickness in the central Weddell Sea according to Strass and Fahrbach [1998] and the ice thickness distribution for ( March and (c) September based on a combination of the mean monthly ice thicknesses and the annual thickness distribution. and ice and in the absence of phase transitions the heat flux (EH) from ice and snow can be presented as [Makshtas, 1991] q s T i EH ¼ l i ; ð7þ h i þ ðl i =l s Þh s where h s is the snow thickness and h i is the ice thickness between the surface and the ice temperature measurement depth. T i is the ice temperature, and l i and l s are the heat conduction coefficients of ice and snow, for which we used values l i =2.1Wm 1 K 1 and l s =.3 Wm 1 K 1, respectively [Makshtas, 1991]. The temperature at the snow surface was calculated by the gradient method, as described in section Unfortunately, the snow thickness could not be measured, except at the time of buoy deployments. We therefore developed a simple relation for the snow thickness guided by the observations of Massom et al. [1998]. It yields an average h S of.22 m and assumes a dependence on the ice thickness H i above the thinner floes: h S = min(.2h i,.25 m). For H i we used the mean monthly ice thicknesses based on the sonar data of Strass and Fahrbach [1998] from the central Weddell Sea (Figure 2). With the T i sensor being installed.7 m below the surface, h i was accordingly.7 m h S Thermodynamic Sea Ice Model [13] We used a thermodynamic sea ice model [Launiainen and Cheng, 1998] as a second approach to calculating the heat budget over the sea ice. In an application over the Baltic Sea ice cover the model results for the surface temperature and fluxes have agreed well with direct observations [Cheng et al., 21; Launiainen et al., 21]. The model includes calculation of the heat conduction through the snow and ice and calculates the air-snow interface temperature q S, surface fluxes, and penetration of shortwave radiation into the snow and ice. The iterative surface temperature serves as the key parameter interacting with the surface heat balance. The calculation scheme of Cheng and Launiainen [1998] is used to solve the heat conduction equation for an ice and snow column consisting, for example, of 1 vertical layers in the ice and 5 in the snow. The thermal conductivities of the snow and ice (l i 2.1 Wm 1 K 1 and l s.3 Wm 1 K 1 ) depend on temperature and salinity. The model uses the wind speed, cloudiness, and air temperature and humidity as input data. The shortwave and longwave radiations are calculated as in (5) and (6). Observations on the oceanic heat flux at the ice bottom in the Weddell Sea have a wide range: from <2 W m 2 in the western Weddell Sea [Robertson et al., 1995] to 27 W m 2 around the zero meridian [McPhee et al., 1999], and on the basis of the drift region of our buoys we applied a moderate value of 5 W m 2. In the model runs we fixed the snow and ice thickness according to the observations (Figure 2. This is because a thermodynamic model cannot reproduce a realistic annual cycle of the ice thickness in the Weddell Sea, where much of the increase in the ice thickness is due to rafting and ridging of the floes. [14] In applying the model the fundamental difference from the calculations in section 3.1 is that we no longer use the measurement of the air temperature gradient for calculating the surface temperature. The wind speed V 2, air temperature q 2, and air humidity were given as measured by buoy The model was run using a 1 min time step, and the buoy data (with an average observation interval of 1 hour) were linearly interpolated to 1 min intervals. Being independent of the measurement of the air temperature gradient, the model results will be compared with the surface temperature and flux results based on (1) (7). Later, the simulations will be extended to cover a distribution of ice and snow thicknesses (section 4.3). 4. Results 4.1. Heat Budget Over Sea Ice General characteristics. [15] The components of the ice surface energy budget are shown in Figure 3 as a time series from 3 February 1996 to 11 January 1997, and the mean annual and seasonal turbulent and conductive fluxes are given in Table 2. The results obtained by the thermodynamic sea-ice model were in agreement, with respect to the sign and order of magnitude, with

5 VIHMA ET AL.: SURFACE HEAT BUDGET OVER THE WEDDELL SEA H (-) LE (...) SWR (- - -) LWR (- - -) EH (-.-.-) positive to the surface / L c) 2 15 count < > / L Figure 3. ( Time series of the ice surface energy budget components (1 day running means) based on buoy 5895 data from 3 February 1996 to 11 January 1997: sensible heat flux (solid line), latent heat flux (dotted line), net shortwave radiation (upper dashed line), net longwave radiation (lower dashed line), and conductive heat flux (dot-dashed line). The fluxes are positive toward the surface. ( Time series and (c) distribution of the stability parameter 1 L 1. In Figure 3b, only the cases with the wind speed exceeding 3 m s 1 are shown. those based on the measurement of the air temperature gradient. Comparisons are shown in Figure 4. (As the radiative fluxes were parameterized with the same equations in both the buoy data analyses and the model, these are not shown in the figure.) The turbulent fluxes were calculated from individual buoy observations, after which the diurnal means (shown in most of the following figures) were averaged. In Figure 3 we present 1 day running means to make the graphs easier to read. The large dayto-day variations become apparent, for example, from Figure 4. [16] The results presented in Figures 3 and 4 and in Table 2 may be summarized as follows: stable stratification prevails over the ice, and the net longwave radiation is the dominating component in the surface energy budget throughout most of the year, except in summer. Together with the small and almost permanent contribution from the latent heat flux and an occasional contribution from the sensible heat flux, the net longwave radiation acts as the heat sink for the surface. There are three smaller heat sources: the sensible heat flux, which is most frequently (always in the 1 day running means) downward; the conductive heat flux through the ice; and the net shortwave radiation. Occasionally, the latent heat flux is also toward the surface. The net heat flux at the surface was around 3 to 5 W m 2 in winter, turning positive in October.

6 5-6 VIHMA ET AL.: SURFACE HEAT BUDGET OVER THE WEDDELL SEA Table 2. Mean Seasonal Values and Standard Deviations for the Sensible Heat Flux H, the Latent Heat Flux LE, and the Heat Flux Through the Ice and Snow (EH) a Flux Season b Sea Ice Leads Buoy Data c Ice Model d Buoy Data c H annual 12 ± ± 8 95 ± 15 summer 16 ± 1 9 ± 5 ± 18 autumn 9 ± 7 11 ± ± 81 winter 7 ± 9 2 ± ± 92 spring 19 ± ± 6 66 ± 97 LE annual 3 ±6 3 ±5 53 ± 42 summer 1 ±8 5±7 17 ± 15 autumn 4 ±3 2 ±3 59 ± 35 winter 5 ±4 1 ±2 83 ± 38 spring 2 ±8 6 ±5 48 ± 42 EH annual 5 ± 4 5 ± 4 summer 2 ± 2 ± 1 autumn 7 ± 4 9 ± 5 winter 5 ± 5 8 ± 4 spring 4 ± 4 2 ± 2 a All fluxes are in W m 2 and positive to the surface. b Summer is defined as the season from December to February, autumn from March to May, winter from June to August, and spring from September to November. c The results are based solely on the measurements of buoy d The results are based on the measurements of buoy 5895 with an application of the thermodynamic sea ice model. [17] The overall accuracy of the results depends a lot on the meteorological conditions, but comparison of results of the two methods gives a first estimate for the accuracy. The results typically agree to within 5 W m 2, but in certain cases the discrepancy is larger. In addition to the accuracy, we have to consider how representative are the local fluxes over ice. Measurements of the air temperature gradient at buoy were not accurate enough to derive the turbulent fluxes, but the air temperature, humidity, and wind speed data for the period 3 February to 28 March 1998 were used to drive the thermodynamic ice model, as was done with the data from buoy The results for the sensible heat flux and surface temperature are shown in Figure 5. The mean values agreed within 3 W m 2 and 1. C, and the maximum differences were 1 W m 2 and 4.1 C, respectively. These occurred in cases with higher air temperatures and stronger winds measured by buoy 25161, which was located closer to the open ocean Radiative and conductive fluxes. [18] The net longwave radiation at the surface is a quantity very difficult to parameterize accurately on the basis of surface observations only, and the results are sensitive to the method used. This is because the downward longwave radiation depends on the temperature and emissivity profiles through the atmosphere, and especially in cold temperatures, these profiles may deviate a lot from the conditions for which most parameterization schemes have been developed. These problems and the various parameterization schemes are discussed further by Launiainen and Cheng [1998]. The Prata [1996] equation, however, has a solid theoretical and empirical basis, and results comparable to within a few watts per square meter were obtained using the formula of Guest [1998], developed on the basis of Weddell Sea data. In any case the results are fairly sensitive to cloudiness, as was indicated by the large (15 W m 2 ) standard deviation of diurnal means. The errors in the daily net longwave radiation in midwinter are therefore assumed to be up to ±2 W m 2. [19] The net shortwave radiation has an important contribution to the heat budget in summer, but for a winter period of 15 days it is <1 Wm 2. The contribution of the solar radiation to the surface heat budget is somewhat difficult to define. The solar radiation penetrates through the surface but is mostly absorbed in the first few centimeters of snow, and part of the heat returns to the surface in the form of conductive heat flux from below. To distinguish between the contributions of solar heating and the conductive heat flux originating from the ocean, we have included all the solar radiation absorbed in the snow and ice body in the term net shortwave radiation. Accordingly, we have calculated the conductive flux on the basis of the temperature difference between the air-snow interface and the depth of.7 m (for the results based solely on the buoy dat and between the air-snow and ice-ocean interfaces (for the results based on the ice model). The flux shows small magnitudes in summer and spring, but some larger values in autumn and winter (Table 2 and Figure 4d) Turbulent fluxes. [2] The sensible heat flux has a downward mean value in all seasons (Table 2). This may, however, result from different physical processes. In winter the effective radiative cooling of the surface results in a downward sensible heat flux, but the buoy data show even larger sensible heat fluxes in spring and summer. These are related to stronger winds and warm air advection from the open sea. Consistent with the downward sensible heat flux, the surface temperature is on average 1.1 C lower than the 3.5 m air temperature. The buoy and model results show good agreement (Figure 4c). In summer the buoy results for the sensible heat flux (mean 16 W m 2 ) are larger than the model results (mean 9 W m 2 ), while the situation is reversed in winter (7 and 2 W m 2, respectively). The estimates for the latent heat flux indicate weak evaporation most of the time. A few peaks of condensation are seen during spring and summer (Figure 4. [21] Several factors caused inaccuracy in the estimates of the turbulent fluxes. These include the errors in the measurements of air temperature gradient (in particular due to solar heating in summer), air humidity, and wind speed; errors in the estimation of the surface roughness of ice and leads; changes in the observation heights due to snow accumulation and melt (an effect of 1 2Wm 2 only); and horizontal inhomogeneity and nonstationarity, which may affect the validity of the Monin-Obukhov theory. The sources of inaccuracy are discussed in more detail by Launiainen and Vihma [1994]. In the present study, the accuracy may be slightly better because we rejected the cases of q S > C and q S q 2 4 C, and unrealistically large fluxes are therefore avoided. Considering the inaccuracy of the air temperature gradient, an error of.5 C approximately corresponds to an error of 1 W m 2 in the sensible heat flux. An error of 1. m s 1 in the wind speeds typically causes an error of 2 5 W m 2 in the sensible heat flux, and the effects related to the uncertainty of surface roughness are of the same order of magnitude. The q S is calculated on the basis of the q S, and its errors are further reflected into the latent heat flux. Assuming an error of 3% in the calculated sensible heat flux, the error in q S is 2 8% in the conditions met in this study, and the corresponding error in the latent heat flux is 2 3% Heat Budget Over Leads [22] Although we do not have quantitative information on the existence of narrow leads in the vicinity of the buoys (which were continuously on ice floes), we estimate the surface heat budget for such hypothetical leads. The calculation of the heat budget over leads is based on the assumption that the properties of the air mass observed by the buoy are also representative of those over nearby leads. This is reasonable for narrow leads, which only cause a small modification to the air mass flowing over them [Andreas et al., 1979; Alam and Curry, 1997]. In cases of large winter polynyas the wind would accelerate, and the air mass would be heated during its traverse, but considering the surface fluxes, these two effects tend roughly to balance each other [Vihma, 1995; Dare and Atkinson, 1999]. We further

7 VIHMA ET AL.: SURFACE HEAT BUDGET OVER THE WEDDELL SEA c) o C d) Figure 4. Results based solely on the buoy data (solid lines) compared with those based on application of the thermodynamic ice model (dotted lines) for ( sensible heat flux, ( latent heat flux, (c) surface temperature, and (d) conductive heat flux (diurnal means). assume that the lead surface temperature is at the freezing point of 1.8 C, with the exception that in summer it depends on the ice concentration rising linearly to C with the ice concentration decreasing from.8 to.5. The components of the surface energy budget over leads are shown as time series in Figure 6. The turbulent sensible heat flux is the dominating component in winter, and the net shortwave radiation is the dominating component in summer. Most of the time the surface also releases heat by evaporation and net longwave radiation, components which are of comparable magnitude. The net heat flux over leads is positive ( 27 W m 2 ) from early October to February. From March through September it has a mean value of 31 W m 2, and the standard deviation of diurnal means is as large as 17 W m 2. It mostly arises from the variability in the wind speed and air temperature. The seasonal values for the turbulent fluxes are shown in Table 2 and demonstrate large differences from the fluxes over thick sea ice. With a typical 4 m wind speed of 6 m s 1 the winter mean sensible heat flux ( 182 W m 2 ) corresponds to a surface-air temperature difference of 17 C Regional Heat Budget [23] The components of the regional heat budget were calculated on the basis of the results obtained for the sea ice and leads, applying the Special Sensor Microwave Imager (SSMI) ice concentration data. Two data sets were used: one processed using the so-called Bootstrap algorithm (obtained from J. Comiso, NASA Goddard Space Flight Center, Greenbelt, Maryland) and another processed using the NASA-Team algorithm including a weather correction for ice concentrations <5% (Heygster et al. [1996], obtained from the University of Bremen, Bremen, Germany). First, we assumed only two surface categories: open leads

8 5-8 VIHMA ET AL.: SURFACE HEAT BUDGET OVER THE WEDDELL SEA 3 Sensible heat flux W / m Julian day in Surface temperature deg C Julian day in 1996 Figure 5. Comparison of ( surface sensible heat flux and ( surface temperature as calculated with the thermodynamic ice model forced by buoy 5895 data (solid line) and buoy data (dashed line). and (thick) ice. The area-averaged net heat flux was simply calculated as hqi ¼fQ w þ ð1 f ÞQ i ; where f is the open water fraction, Q w and Q i are the net heat fluxes over leads and ice, respectively, and the angle brackets denote an ð8þ area average. For Q i we took the model-based values to make the results comparable with the further analyses presented in this section. The various components of hqi were calculated analogously to (8), and the results are shown in Table 3. In accordance with the spatial resolution of the SSMI Bootstrap data the area averages represent a scale of km 2 around buoy The area-averaged net flux was typically negative in winter and positive in summer, particularly in the northern Weddell Sea in 3 H - LE... SWR LWR -.-. positive to the surface Figure 6. Time series of the surface energy budget components (1 day running means) over leads based on buoy 5895 data. Sensible heat flux (solid line), latent heat flux (dotted line), net shortwave radiation (dashed line), and net longwave radiation (dot-dashed line).

9 VIHMA ET AL.: SURFACE HEAT BUDGET OVER THE WEDDELL SEA 5-9 Table 3. Regional Surface Fluxes of Sensible, Latent, and Conductive Heat Over the Weddell Sea, Calculated As Area Averages of the Fluxes Over Leads, Open Water, and Various Classes of Ice Thickness a Flux Season Thick Ice and Open Water Realistic Ice Thickness Distribution Bootstrap fr NASA-Team fr Bootstrap fr NASA-Team fr hhi annual summer autumn winter spring hlei annual summer autumn winter spring hehi b annual summer 3 3 autumn winter spring a The annual and seasonal means in W m 2 are given. See the text for explanation for the Bootstrap and NASA-Team open water fraction (fr). b The area-average is calculated for the ice-covered area, and thus does not depend on the ice concentration. November and December 1996, when the ice melted. The regional sensible heat flux typically varied in the range from 2 to +3 W m 2, with an annual mean of 6 13 W m 2 (Table 3). [24] The assumption of the presence of only thick ice and leads is, of course, a simplification. We studied the importance of the ice thickness distribution by modeling the year-round surface heat budget over a wide range of ice and snow thicknesses. The same atmospheric forcing data based on the buoy observations were used. An example of the sensitivity of the sensible heat flux to the ice and snow thicknesses is given in Figure 7. The thicker the ice and snow, the smaller the conductive heat flux through them, which is reflected in a lower surface temperature and a higher turbulent heat Mean annual sensible heat flux Snow thickness (cm) Ice thickness (cm) Figure 7. Dependency of the mean annual surface sensible heat flux on the ice and snow thicknesses (note the logarithmic scales) as calculated by the thermodynamic ice model forced by the buoy data.

10 5-1 VIHMA ET AL.: SURFACE HEAT BUDGET OVER THE WEDDELL SEA f (%) c) Figure 8. Diurnal means of regional surface fluxes in the Weddell Sea based on the realistic ice thickness distributions. ( Net heat flux at the surface as calculated with the SSMI Bootstrap (lower solid line) and NASA- Team data (lower dotted line). The diurnal mean open water fraction at the buoy site is shown in the upper part: SSMI Bootstrap results as a solid line, NASA-Team results as a dotted line, and NCEP/NCAR values as a dashed line. ( The sensible heat flux with the SSMI Bootstrap data (solid line) and with the NASA-Team data (dotted line), and (c) the fluxes of latent heat (dashed line), net longwave radiation (solid line), and net shortwave radiation (dotted line). In Figure 8c the ice concentration is based on the SSMI Bootstrap data. flux from the air. The sensitivity is large for a thin ice and snow cover but decreases rapidly with increasing thicknesses. [25] The components of the area-averaged heat budget were then calculated using the model results and monthly ice thickness distributions derived from the observations by Strass and Fahrbach [1998] for the central Weddell Sea (their station 283). We estimated the monthly distributions by combining their mean monthly ice thicknesses and annual thickness distribution. The results for March and September, the extreme months, are shown in Figure 2. For the snow thickness h S we applied the same relation as in section The results are summarized in Table 3 and Figure 8. The results calculated on the basis of this more realistic ice thickness distribution differ from those based on the assumption of thick ice and leads only. The presence of thin ice is important for the area-averaged sensible heat flux in winter. The seasonal mean decreases by 5 7 W m 2 when the ice thickness distribution is accounted for in the calculations (Table 3). Over a compact ice cover (1% Bootstrap concentration) the maximum diurnal difference of 26 W m 2 occurred on 2 July 1996, with a sensible heat flux of 9 W m 2, as calculated with the realistic ice thickness distribution, and 35 W m 2, as calculated for thick (1.9 m) ice. [26] Next we look at the effect of the ice concentration based on the two processing techniques. The concentrations differ mostly during winter with a compact ice cover, when the Bootstrap concentration was typically 1% while the NASA-Team concen-

11 VIHMA ET AL.: SURFACE HEAT BUDGET OVER THE WEDDELL SEA 5-11 Air temperature C o Wind speed 1 m s Figure 9. Time series of ( 2 m air temperature and ( 1 m wind speed according to the buoy (solid line), ECMWF model (dashed line), and NCEP/NCAR model (dotted line). For clarity the wind speed is drawn as 1 day running means. tration was 9 95%. These differences were reflected particularly in the regional sensible heat flux in winter. The seasonal mean is 13 Wm 2 when calculated from the Bootstrap concentrations but only 3 W m 2 when calculated from the lower NASA-Team concentrations. The maximum differences in the diurnal hhi and the area-averaged net flux were 43 and 71 W m 2, respectively. These occurred in a case with strong wind (11.5 m s 1 ), cold air ( 16.9 C), moderate downward H over ice (24 W m 2 ), and large upward H from leads ( 32 W m 2 ). The Bootstrap ice concentration was 1%, while the NASA-Team value was 88%. In comparable conditions, differences in hhi exceeding 2 W m 2 occurred on 33 days. During the austral summer the two estimates for hhi correlated well (Figure 8b and Table 3). The different ice concentrations have a large effect on the estimate of the potential amount of new ice formation in leads. For the period from March through September, using the mean net flux over leads ( 31 W m 2 ) and the mean Bootstrap open water fraction (.19), the areaaveraged formation rate of new ice would be.5 m per month, while the value would be.21 m per month using the mean open water fraction from the NASA-Team algorithm (.78). The estimates assume open leads and no oceanic heat flux from below. The latter value fits well to the results of Eisen and Kottmeier [2] (.1.15 m per month in the eastern and up to.3 m per month in the western Weddell Se, also based on the NASA-Team ice concentrations. [27] From the point of view of parameterizations in large-scale models the occurrence of so-called countergradient fluxes form an interesting problem. In most cases a downward hhi coincided with a positive hq a i hq S i (hq S i was calculated analogously to (8), and the air temperature observed over the ice was used for hq a i). Occasionally, however, an upward hh i was associated with a positive hq a i hq S i; that is, a countergradient flux occurred. In these cases the open water fraction was typically 2 5%. The wind speed was weak (4. ± 3.2 m s 1 for the mean and standard deviation), yielding small downward fluxes over the ice, but the air temperature was low ( 16. ±7.7 C), allowing moderate upward fluxes over the leads. Taking the ice concentrations from the Bootstrap results, such countergradient fluxes prevailed for 8% of the time; for the NASA-Team ice concentrations the corresponding figure was 1%. 5. Comparisons With Large-Scale Models [28] The buoy results were compared with the operational analyses of the ECMWF [ECMWF, 1997] and the reanalyses of the NCEP/NCAR [Kalnay et al., 1996]. From the ECMWF we used 6 hourly analyses on a regular latitude-longitude grid of size , corresponding to 62 km 21 km at the latitude of 7 S. The NCEP/NCAR values were the diurnal means of the 6 hourly values (forecasts, averages, or instantaneous values, depending on the model variable) on a grid ( km at 7 S). The model values were cubically interpolated to the locations of buoys 5895 and on the basis of the nearest grid points. The nearly year-round results from buoy 5895 served as primary references for the comparisons, but the shorter-period wind and temperature data from buoy were also examined. The ECMWF and NCEP/NCAR cloud covers were already shortly evaluated in section [29] Comparisons of the air temperature and the wind speed measured by the buoys and modeled by the ECMWF and NCEP/ NCAR are shown in Figures 9 and 1 and in Table 4. Regarding the buoy air temperatures as references, we see that the NCEP/NCAR 2 m air temperatures have a cold bias of 3.2 C in the annual mean, while the ECMWF values have a 3.5 C warm bias. The deviations are large, particularly in winter over a compact ice field, but they are somewhat smaller for the NCEP/NCAR temperatures. The air temperatures of the both models, however, respond positively to weather variations, which is seen as a good correlation with the buoy values (correlation coefficient r.9; Figure 1). The surface temperatures of both models compared to the buoy results similarly to the air temperatures (Table 4). The ECMWF 2 m relative humidities (with respect to water saturation) were on average comparable to the buoy data (annual means of 84 and 8%, respectively). The NCEP/NCAR 2 m relative humidities were somewhat higher, with an annual mean of 93%. Considering the

12 5-12 VIHMA ET AL.: SURFACE HEAT BUDGET OVER THE WEDDELL SEA r =.87 r =.92 T(ECMWF) T(NCEP) T(buoys) T(buoys) c) 15 r =.57 d) 15 r =.69 V(ECMWF) 1 5 V(NCEP) V(buoys) V(buoys) Figure 1. Comparisons of the observed (buoy 5895 as dots and buoy as circles) and model values for ( buoy and ECMWF 2 m air temperature (in C), ( buoy and NCEP/NCAR 2 m air temperature, (c) buoy and ECMWF 1 m wind speed (in m s 1 ), and (d) buoy and NCEP/NCAR 1 m wind speed. The buoy 1 m winds are calculated from measurements at a height of 3.5 m. relative humidity with respect to ice saturation, the annual means were 87% for the buoy data, 9% for the ECMWF, and 17% for the NCEP/NCAR. (According to Vowinckel and Orvig [197, p. 26], supersaturation with respect to ice may be common over polar sea ice.) Because of the too warm air temperatures, the ECMWF specific humidities were too high, with an annual mean of 2.1 g kg 1 compared to the buoy value of 1.8 g kg 1. Related to the cold bias in the air temperature, the NCEP/NCAR specific humidities were in better agreement, with an annual mean of 1.7 g kg 1. [3] For the wind speed comparisons the buoy wind speeds observed at a height of 3.5 m were converted to a height of 1 m taking into account the effects of roughness and stability [Launiainen and Vihma, 199]. The model wind speeds typically deviate.5 2 m s 1 from the buoy observations (Figure 9). The NCEP/ NCAR 1 m wind speed correlates with the buoy (5895) wind somewhat better than does the ECMWF wind (r =.69 and.57, respectively), but the magnitude of the ECMWF 1 m wind speed is on average closer to the buoy observations. The mean wind speeds both for the buoy and ECMWF 1 m wind are 6.3 m s 1 but are only 5.1 m s 1 for the NCEP/NCAR 1 m wind. The quantity having the best overall correspondence with the buoy observations was, in fact, the NCEP/NCAR wind speed at the lowest atmospheric model level, having a mean value of 6.2 m s 1 and a correlation coefficient of.66. Table 4. Mean Values of the Air and Surface Temperature, Wind Speed, Relative Humidity With Respect to Water Saturation, and Surface Sensible and Latent Heat Fluxes As Observed Over the Ice by Buoy 5895 and Modeled by the ECMWF and NCEP/NCAR Quantity Period / Conditions a Buoy ECMWF NCEP/NCAR T a (2 m) annual compact ice field fractured ice field T S annual compact ice field fractured ice field V (1 m) annual compact ice field fractured ice field RH (2 m) annual compact ice field fractured ice field hhi annual 9 15 compact ice field 9 18 fractured ice field 1 2 hlei annual compact ice field fractured ice field a The annual values are calculated on the basis of data from Julian days 34 to 366, the values for a compact ice field from days 5 to 31, and the values for a fractured ice field from days 34 to 49 and 311 to 366.

13 VIHMA ET AL.: SURFACE HEAT BUDGET OVER THE WEDDELL SEA Julian day in Julian day in 1996 Figure 11. Time series (1 day running means) of ( the surface sensible heat flux and ( the surface latent heat flux, based on the ice model forced by the buoy data (solid line) and the NCEP/NCAR (dashed line) and ECMWF (dotted line) models. [31] The turbulent surface fluxes are compared in Figure 11 and Table 4. The area-averaged buoy results calculated with the Bootstrap ice concentrations and sonar-based ice thickness distributions are used as a reference. During winter the reference sensible heat fluxes are typically between the values from the ECMWF and NCEP/NCAR models, and the differences are large. Only in February and early November 1996 were all three values in good agreement (Figure 11). All the estimates for the latent heat flux provided small-flux magnitudes in winter, but the scatter was largest during summer and spring. [32] An analysis of the reasons for the differences between the buoy and model values is not straightforward because the model values are affected by numerous factors. Firm conclusions would require sensitivity runs using the ECMWF and NCEP/NCAR models. It seems, however, that the warm bias in the ECMWF surface and 2 m air temperatures can be at least partly due to the excessive cloud cover in winter, and the cold bias in the NCEP/ NCAR can be due to the inadequate cloud cover. Certain boundary conditions can also provide possible explanations for some of our findings. First, we have to stress that the 2 m air temperature is interpolated from the surface skin temperature (a prognostic variable over sea ice) and the temperature at the lowest atmospheric model level (at the height of about 3 m in the ECMWF and 5 m in the NCEP/NCAR model). Neither of the models applies a fractured ice cover, but the ice concentration is either or 1. The NCEP/NCAR sets the model ice concentration to 1 if the SSMI data indicate a value exceeding.55. This is an important restriction in the model and can partly explain the cold bias in the NCEP/ NCAR surface and 2 m air temperatures. In the NCEP/NCAR the ice thickness was a constant of 3 m, while the water equivalent of the snow thickness (a model variable) was usually about.1 m. This corresponds to an unrealistically thick snow cover of the order of.3 1 m, which can also contribute to the cold bias, particularly in spring (Figure 9, when the real ice and snow thicknesses were much smaller. In the ECMWF model the ice concentration was set to 1 if the surface temperature dropped below 1.7 C. The model started to use satellite data for the ice coverage on 23 April 1996 (C. Larsson, personal communication, 1997), but we cannot detect from Figures 9 or 11 any clear improvement since this date. Further, in the ECMWF model the ice thickness was a constant of 1 m without any snow cover. This may also contribute to the warm bias, which was particularly large in winter (Figure 9. [33] The differences in the surface sensible heat flux are mostly related to the erroneous surface temperature. The near-zero H in the ECMWF is in agreement with the too warm surface, and the excessive downward H in the NCEP/NCAR is in agreement with the too cold surface. Hence the flux comparisons also suggest that the biases found are due to the surface boundary conditions rather than to any major bias in the air temperature or wind speed at the lowest atmospheric model level. Because of the lack of leads, no countergradient fluxes occurred in the model results. The coarse horizontal resolution in the NCEP/NCAR model prevents it from detecting the mesoscale cyclones common in the Weddell Sea [Heinemann, 199], and this may well explain the low bias in the mean wind speed. The worse wind speed correlation of the ECMWF model can be due to the large warm bias in the air temperature, which affects the near-surface wind via an erroneous stratification effect. [34] Most of the verification work of the reanalyses products has so far addressed low and middle latitudes [WCRP/SCOR, 2]. Renfrew et al. [22], however, studied the ice edge zone in the Labrador Sea. Despite a cold bias of 1.3 C in the ECMWF 2 m air temperature they concluded that the turbulent fluxes of the ECMWF model are suitable for driving ocean models. Unfortunately, the same seems not to be valid for the ice-covered Weddell Sea. 6. Conclusions [35] This study was a continuation of our previous work [Launiainen and Vihma, 1994] with a wider utilization of additional sources of information, such as the thermodynamic sea ice model, large-scale model (ECMWF and NCEP/NCAR) fields, data on ice thickness distribution, and SSMI ice concentration data. The surface sensible heat flux over the sea ice was calculated by the gradient method based on the Monin-Obukhov theory. A check on the reliability of the flux was made by calculating the corresponding surface temperature. The latent heat flux was calculated by the bulk method applying the calculated surface temperature. The radiative fluxes were parameterized applying Prata [1996] (almost