main latex file Click here to download Manuscript: model_comp.tex
Click here to view linked References
Validation of three global ocean models in the Weddell Sea
Angelika H. H. Renner a,b,∗, Karen J. Heywood b Sally E. Thorpe a a British
Antarctic Survey, Natural Environment Research Council, High Cross, Madingley Road, Cambridge, CB3 0ET, UK
b School
of Environmental Sciences, University of East Anglia, Norwich, NR4 7TJ, UK
Abstract We present a validation of three global z-level eddy permitting/resolving ocean general circulation models against hydrographic observations and remote sensing sea ice data from the Weddell Sea. The Weddell Sea is a region in which complex processes such as water mass formation, sea ice formation and melt, and circulation under ice shelves take place. The representation of these processes is challenging even for the current generation of eddy-permitting ocean models. Simulating the hydrographic structure of this basin is a stringent test for models, notably so when considering the global influence of the regional processes. The performance of OCCAM (at two resolutions), ORCA025 and TPAC are tested regarding water mass properties, sea ice seasonality (OCCAM and ORCA025 only) and volume transport. OCCAM simulates the deep water masses reasonably well in both resolutions. The eddy resolving run is not significantly better than the eddy permitting simulation. ORCA025 and TPAC represent the surface layers and the Weddell Gyre circulation
better but are generally too warm throughout the water column. All models underestimate Weddell Sea Bottom Water formation. Both OCCAM and ORCA025 struggle to correctly model the sea ice cover: OCCAM overestimates the summer ice extent while little multi-year sea ice remains in ORCA025. TPAC exhibits considerable drift in potential temperature during the model run. The choice of a model for a study has to be made carefully taking into account the model’s performance in the specific area, application and variable of interest. We identify several starting points for improving the models, namely model numerics and parameterisations of subgridscale processes, ensuring more accurate forcing datasets, correct initialisation, and ice-ocean interactions, all of which are likely to have larger benefits than simply increasing the horizontal resolution beyond eddy permitting. Key words: global ocean model, Weddell Sea, hydrography, sea ice, deep water masses, Southern Ocean
1
1
Introduction
2
Various complex processes such as deep and bottom water formation, sea ice formation
3
and melt, circulation under ice shelves and mixing make the simulation of the Weddell Sea
4
a challenge for ocean models. A correct representation of the Weddell Sea is, however, of im-
5
portance for the simulation of the global oceans (e.g. Hellmer et al., 2005). The dense water
6
masses exported from the Weddell Sea are important drivers of the thermohaline circulation
7
and contribute significantly to the ventilation of the deep oceans (Hellmer and Beckmann,
8
2001; Jacobs, 2004). They are formed from saline Warm Deep Water (WDW), the Weddell
9
Sea variation of Circumpolar Deep Water, and fresh and very cold shelf waters (Foster and
10
Carmack, 1976; Foldvik et al., 1985; Weppernig et al., 1996). The shelf waters interact with
11
the atmosphere in coastal polynyas, where heat loss to the atmosphere and brine rejection
12
during sea ice formation lead to the formation of High Salinity Shelf Water (HSSW). A second ∗ British Antarctic Survey, Natural Environment Research Council, High Cross, Madingley Road, Cambridge, CB3 0ET, UK Email address:
[email protected] (Angelika H. H. Renner).
13
form of shelf water modification occurs underneath ice shelves, where supercooling and melt of
14
the ice shelf base produces the extremely cold and rather fresh Ice Shelf Water (ISW). At the
15
continental slope, the shelf waters mix with WDW and form Weddell Sea Deep and Bottom
16
Water (WSDW and WSBW, respectively). The downslope flow of these cold and dense water
17
masses has been observed along the southern and western Weddell Sea (e.g. Fahrbach et al.,
18
2004). The sea ice cover in the Weddell Sea is highly seasonal. While the release of freshwater
19
during the ice melt in summer is important for the modification of Antarctic Surface Waters
20
(AASW) in the central and northern Weddell Sea, brine rejection during ice formation aids the
21
formation of dense water on the shelves (Gill, 1973). Ice formation rates are particularly high
22
in coastal polynyas in front of the ice shelves in the southern and southwestern Weddell Sea
23
(Renfrew et al., 2002).
24
Teleconnections between the Southern Ocean and the North Atlantic were previously re-
25
ported by Hellmer et al. (2005). They used the adjoint method to include high resolution model
26
results for the Weddell Sea in a global ocean circulation model. They found that improving the
27
simulation of processes in the Weddell Sea has a major impact on the export of WSDW and
28
the location and transport of the Antarctic Circumpolar Current (ACC). Colder and relatively
29
fresh water spreads into the South Atlantic and changes in the North Atlantic are reported.
30
This reinforces the importance of correctly simulating the Weddell Sea in global ocean and
31
climate models.
32
Stand-alone ocean general circulation models (OGCM) are valuable tools to investigate
33
smaller scales and improve parameterisations, numerics and the overall representation of oceanic
34
processes. Climate models also benefit from better, higher resolution ocean models (Roberts
35
et al., 2008). With better codes, parameterisations, topography and forcing datasets, and higher
36
resolution, OGCMs have greatly improved. However, detailed and correct representation espe-
37
cially of polar processes which are crucial on a global scale remains difficult. To improve the
38
simulation of water mass formation and modification in polar regions, sea ice models are cou-
39
pled to OGCMs. Sea ice models still struggle to realistically represent the ice cover, both in
40
the Arctic and the Southern Ocean and especially with respect to ice thickness (e.g. Timmer-
41
mann et al., 2004; Holland and Raphael, 2006; Martin and Gerdes, 2007; Renner and Lytle,
42
2007). These areas are also important for the heat flux between atmosphere and ocean and the
43
overturning circulation (Marsland et al., 2007) as the ocean loses heat to the atmosphere and
44
shelf waters become cooler. St¨ossel et al. (1998, 2002) demonstrated effects of Antarctic sea
45
ice on the simulation of a global OGCM. Amongst other aspects, they show how tuning of a
46
parameter such as sea ice salinity changes Antarctic Bottom Water formation. In addition to
47
modelling specific difficulties, validation of the models in polar regions suffers from the scarcity
48
of observational data. Therefore, one of the aims of the World Ocean Circulation Experiment
49
(WOCE) was to obtain a baseline dataset, including hydrographic sections across the Weddell
50
Sea. The recent developments in modelling and observations make this an opportune time to
51
revisit the performance of ocean models in polar regions and in particular in the Weddell Sea.
52
In this paper, we test the performance of three global ocean general circulation models in the
53
Weddell Sea and suggest possible reasons for the different model behaviours. First, we present
54
the models used in the comparison, followed by a description of the observational datasets. The
55
Weddell Sea hydrography and the sea ice cover in the models are compared between models
56
and with observations in the fourth section. The discussion brings together the similarities and
57
differences of the models and we propose mechanisms leading to the model characteristics. At
58
the end of the discussion, we present key points for improving OGCMs. The conclusion gives
59
an overview of the benefits of each model and suggests the application where each model is
60
most suitable.
61
2
The Models
62
For this study we use output of three global ocean models, one of them at two horizontal
63
resolutions. All are z-level models with partial bottom cells, and two include sea ice. This section
64
gives a brief overview of their characteristics, with the main details summarised in Table 1.
65
2.1
OCCAM
66
The Ocean Circulation and Climate Advanced Modelling (OCCAM) project model run by
67
the National Oceanography Centre, Southampton, is a global ocean general circulation model
68
(OGCM) of the Bryan-Cox-Semtner type (Bryan, 1969; Semtner, 1974; Cox, 1984) coupled with
69
a dynamic-thermodynamic sea ice model (Aksenov, 2002). The sea ice dynamics are based on an
70
elastic-viscous-plastic formalism. The thermodynamics use one vertical layer for snow and one
71
or two layers for thin and thick ice, respectively. The ice albedo is set to 0.66. Details are given
72
by Webb et al. (1998), Webb (2000) and Coward and de Cuevas (2005). The model exists in
73
different resolutions, both in the horizontal and the vertical. We use the eddy-permitting 1/4◦ -
74
and the eddy-resolving 1/12◦ -versions with 66 z-levels with partial step bottom layer (OCCAM
75
1/4 and OCCAM 1/12 respectively). This corresponds to a grid spacing of approximately 6
76
x 9.3 km at 50◦ S and 2.1 x 9.3 km at 77◦ S for OCCAM 1/12, 17.9 x 27.8 km at 50◦ S
77
and 6.2 x 27.8 km at 77◦ S for OCCAM 1/4, and layer thicknesses of 5 m at the surface
78
to 208 m for the deepest layer for both models. The model runs were forced using 6-hourly
79
winds and heat fluxes from the NCEP/NCAR reanalysis data (Kalnay et al., 1996; Large et al.,
80
1997), and monthly cloud fraction and solar radiation from the International Satellite Cloud
81
Climatology Project. Monthly precipitation is based on Microwave Sounding Unit data and
82
blended with Xie and Arkin (1997) observational data. The bathymetry is constructed from
83
Smith and Sandwell (1997) and DBDB5 (DBDB5, 1983). Sill depths of important straits are
84
checked and adjusted manually (Coward and de Cuevas, 2005). The runs for both resolutions
85
were initialised using WOCE Hydrographic Program Special Analysis Centre (SAC) climatology
86
(Gouretski and Jancke, 1996) for potential temperature (θ) and salinity (S) with additional data
87
for the Arctic Ocean. Sea ice in the Southern Ocean was set to 1.50 m thickness of sea ice and
88
0.15 m snow in all cells south of 65.25◦S with an ice concentration of 99% in each affected grid
89
cell. Both OCCAM 1/12 and OCCAM 1/4 were run for 20 years corresponding to January
90
1985 to December 2004 (runs 401 and 103, respectively).
91
2.2
ORCA025
92
ORCA025 is a global configuration of the ocean model OPA (Oc´ean PArall´elis´e; Madec
93
(2006)) coupled with the Louvain la Neuve sea ice model LIM2 (Fichefet and Morales Maqueda,
94
1997). LIM2 is a dynamic-thermodynamic sea ice model with three layers (one for snow and
95
two for ice) and a viscous-plastic rheology. Ice albedo is parameterised following Shine and
96
Henderson-Sellers (1985) with modifications (Greenfell and Perovich, 1984). The model run
97
used for this study (run G70) was performed at the Laboratoire des Ecoulements G´eophysiques
98
et Industriels (LEGI) in Grenoble. For ORCA025, the global, tripolar Mercator grid has a
99
nominal horizontal resolution of 1/4◦ at the equator. In our study region, the grid spacing
100
therefore varies from 17.9 × 17.9 km at 50◦ S to 6.2 × 6.2 km at 77◦ S. The 46 vertical levels
101
have a thickness of 6 m at the surface increasing to 250 m for the lowest level. The model
102
uses partial bottom cells. The setup follows Barnier et al. (2006) except for the vertical mixing
103
where a modified version of the turbulent kinetic energy (TKE) parameterisation for the vertical
104
mixing coefficient was used that includes the effects of long waves, Langmuir cells and the
105
penetration of TKE in depth (Molines et al., 2006, updated 2007).
106
Run G70 was forced by the Drakkar Forcing Set No. 3 (DFS3; Molines et al. (2006, updated
107
2007); Brodeau et al. (submitted)). DFS3 is compiled from various sources. Long- and shortwave
108
radiation fields are from CORE (Coordinated Ocean-Ice Reference Experiments) forcing (Large
109
and Yeager, 2004). Precipitation south of 30◦ N is from the GXGXS (uncorrected CORE)
110
dataset and CORE north of 30◦ N. Air temperature, wind and air humidity are 6-hourly ERA40-
111
reanalysis fields (1958-2001) and ECMWF(2002-2004). River runoff is implemented from Dai
112
and Trenberth (2002) using a new dataset including coastal runoff. For a comparison of G70
113
with runs using the standard CORE forcing see Barnier et al. (2007). The model bathymetry is
114
regridded from ETOPO2 merged with the Bedrock Mapping Project data (Lythe and Vaughan,
115
2001) south of 72◦ S and GEBCO on the continental shelves (i.e. in depths shallower than 200
116
m). Additional smoothing and manual editing in key regions has been performed (Molines et al.,
117
2006, updated 2007). θ and S were initialised from Levitus et al. (1998) in mid to low latitudes
118
and the Polar Science Center Hydrographic Climatology (PHC) v2.1 (Steele et al., 2001) in
119
high latitudes. The initial sea ice state was taken from a previous run after 10 years of spin up
120
(Molines et al., 2006, updated 2007). The model is run for 47 years corresponding to the years
121
1958 to 2004.
122
2.3
TPAC
123
The OGCM of the Tasmanian Partnership for Advanced Computing (TPAC) is based on
124
the Geophysical Fluid Dynamics Laboratory’s (GFDL) Modular Ocean Model (MOM), version
125
3.1. It is a primitive equation, ocean only model extending from 80◦ N to 80◦ S. Horizontal
126
resolution is 1/8◦ × 1/8◦ corresponding to 8.9 x 13.9 km at 50◦ S and 3.1 x 13.9 km at 77◦
127
S. In the vertical, the model has 24 z-levels with minimum layer thickness of 22 m at the
128
surface and 489 m at the bottom, with partial bottom cells. Unusual for an eddy-resolving
129
model (and different from OCCAM 1/12), TPAC includes isopycnal mixing by using the Gent
130
and McWilliams (1990) parameterisation. Sea ice is not considered. More details are given by
131
Meijers et al. (2007).
132
For the run we use in this study, the model is forced by applying hemispheric wintertime
133
boundary conditions to which NCEP-R2 variability and trends (Kanamitsu et al., 2002) were
134
added. Sea surface temperature is constrained to the NOAA Optimum Interpolation Sea Surface
135
Temperature V2 (Reynolds et al., 2002). Surface salinity is adjusted by a salt flux calculated
136
using the NCEP-R2 latent heat flux and precipitation rate. The time scale for both sea sur-
137
face temperature and salinity restoring is 30 days. The model is initialised from the WOCE
138
Hydrographic Programme SAC atlas (Gouretski and Jancke, 1998) and Levitus (1982) data
139
in the Arctic. The model bathymetry is based on the TerrainBase dataset (Row et al., 1995).
140
Integration time is 20 years, corresponding to 1982-2001, using asynchronous time stepping.
141
3
142
3.1
The observational datasets and extracted model output Hydrographic sections
143
θ and S data were selected from hydrographic transects made in the Weddell Sea between
144
Joinville Island and Kapp Norvegia onboard RS Polarstern between 1989 and 1998 (Table 2; Fig.
145
1, Fahrbach et al. (2007)). They were designated section SR04 of the WOCE Repeat Sections
146
Program. Fahrbach et al. (2004) present details of the measurements and instruments as well
147
as a description of the data. Using monthly mean fields, which correspond in time with the
148
months the CTD data were collected, model θ and S were bilinearly interpolated from the four
149
nearest grid points onto the location of the CTD casts. As each model has a different vertical
150
resolution, we then linearly interpolated the model data to fit the CTD depth grid (converted
151
to depth from a 2 dbar pressure grid), thereby ensuring consistency between the models when
152
calculating model to CTD differences and statistics.
153
Following Fahrbach et al. (2004) we calculate the area occupied on the sections by WDW,
154
WSDW and WSBW, as well as the average θ, S and depth for each water mass. We use the
155
same criteria as Fahrbach et al. (2004): WDW is the water below the mixed layer with θ > 0◦
156
C, WSDW lies below that and has −0.7◦ C < θ ≤ 0◦ C, and WSBW is at the bottom of the
157
water column with θ ≤ −0.7◦ C.
158
In addition to the sections across the Weddell Sea Fahrbach et al. (2004) use for their
159
analysis, we include the section SR04e (April 1998) to maximise the available validation data.
160
Since the section does not extend all the way onto the shelf in the eastern Weddell Sea, the
161
derived water mass properties (average θ, S, depth and occupied area) are biased and cannot
162
be compared directly with the results from the previous sections. For example, θ and S will
163
be lower than in the other sections as water is transformed while flowing on and along the
164
continental shelf and slope by interaction with ice shelves and sea ice formation.
165
Observations in regions such as the Weddell Sea are often compromised by circumstances
166
encountered during the cruise and by time limitations. Station spacing therefore varies between
167
the different occupations of the SR04 section. When directly comparing a CTD section to the
168
model section at the respective time, we extract and interpolate the model data onto exactly
169
the set of CTD station positions of each of the five sections. For the timeseries of water mass
170
properties we show the range of values for the four sets of stations excluding the station set of
171
section SR04e. Representative of the five occupations of the SR04 transect, we present the full
172
depth section and the θ − S diagram of the spring section SR04b in Figs. 2 and 3.
173
3.2
Sea ice data
174
Sea ice has been observed from satellites using passive microwave radiometry since the
175
1970s (e.g. Parkinson, 2004). For this study, we use monthly sea ice concentrations derived
176
using the Bootstrap algorithm from brightness temperatures measured by Special Sensor Mi-
177
crowave/Imagers (SSM/I) (Comiso, 1990, updated 2007), available from the National Snow
178
and Ice Data Center. We restrict the region considered for extent and area calculations to the
179
Weddell Sea, namely from the east coast of the Antarctic Peninsula and east of 58.5◦ W to
180
10◦ W (Fig. 1a)). Ice extent is defined as the area of all grid boxes covered by ice whereas ice
181
area is the area of the grid boxes covered by ice multiplied by the respective ice concentration.
182
As satellites struggle with recognising very low concentrations, we use a threshold value of 15%,
183
below which concentrations are not considered for calculations of ice extent and area.
184
4
185
4.1
Results OCCAM 1/12
186
OCCAM 1/12 captures the general features of the Weddell Sea hydrography (Figs. 2 and 3):
187
The AASW lies above WDW below which θ decreases again with depth through WSDW and
188
WSBW. There are, however, several deviations from the observations. In general, the depth of
189
the pycnocline and thereby the mixed layer is too deep and too cold in summer. In the CTD
190
sections, the surface mixed layer (ML; defined as the layer from the surface to the depth where
191
the gradient in potential density is greater than 0.02 kg m−3 per 10 db) extends down to 50 to
192
100 m in the central Weddell Sea (Fahrbach et al., 1992). In OCCAM 1/12, the average ML
193
depth in the deep Weddell Sea (i.e. bathymetry deeper than 4000 m) ranges from approximately
194
140 m at the start of the run to around 230 m after 15 model years and extends to 300 m in
195
the western part of the Weddell Sea.
196
Beneath the mixed layer on the eastern side of the section, the modelled θ is colder then
197
the observations to a depth of 2500 m (Fig. 2). This signal can be attributed to the opening
198
of a polynya in the Maud Rise region shortly after the beginning of the model run. Following
199
the opening of the polynya in the OCCAM 1/12 sea ice cover in July 1985, a patch of water
200
characterised by a low WDW θmax appears in the eastern Weddell Sea in August 1985. The
201
polynya persists until early summer when the sea ice retreats. It recurs in 1986 and the WDW
202
θmax decreases further. In the following model years, the cold water signal slowly moves around
203
the Weddell Gyre.
204
Averaged over the entire WDW layer along the section, θ is higher than observed (Fig. 4).
205
As the mixed layer is too thick, the area occupied by WDW is smaller than observed. Above
206
the continental slope, both in the western and the eastern Weddell Sea, the downward slope
207
of the thermocline is less pronounced than in the observations leading to a core of water up to
208
1◦ C warmer situated above the 1000–3000 m isobaths. The cold water plume of newly formed
209
WSBW on the western continental slope which is visible in the CTD sections (Fahrbach et al.,
210
2001) is not resolved in the model (Fig. 2).
211
The area covered by WDW is underestimated in OCCAM 1/12 in all sections. After a
212
decrease in WDW area percentage during the first five years of the model run, the amount
213
of WDW increases again after 1990 but does not reach the initial level (Fig. 4). Average θ
214
and S increase slightly during the run. The increase of the average depth is more pronounced.
215
Contrary to WDW, the area percentage of WSDW increases in 1985-1990 and decreases after
216
that (Fig. 5). WSDW is overestimated at the cost of WDW. The area percentage of WSDW
217
remains greater than observed. Average θ and depth are fairly constant but S increases over
218
the model run. The area occupied by WSBW in OCCAM 1/12 is very close to observations,
219
as is the average depth. θ is on average about 0.03◦ C warmer in the model, excluding the
220
last section SR04e (April 1998). For this section, the CTD data show much lower average θ
221
and S as the warmer and saltier eastern Weddell Sea is not included completely. Bottom water
222
production in the southern and western Weddell Sea leads to an asymmetric distribution of
223
WSBW with more WSBW in the western part. Although this pattern is present in OCCAM
224
1/12 as well, it is rather weak, which indicates underestimated bottom water formation. The
225
total amount of WSBW along the section line fits the observations well though. It decreases
226
during the model run (Fig. 6). As for WDW and WSDW, the average WSBW S increases.
227
WSBW also becomes warmer and deeper.
228
The seasonal cycle is most visible in the sea ice cover (Figs. 7 and 8). In OCCAM 1/12,
229
the climatological minimum ice extent occurs in February, agreeing with observations, and the
230
maximum is in August, a month earlier than observed in the SSM/I climatology. Although the
231
annual cycle in the model agrees well temporally with the observations, the extent of the ice
232
cover is far overestimated by OCCAM 1/12 (Fig. 7 a). The ice extent at its minimum reaches
233
2.2 × 1012 m2 , almost 70% more than observed. Instead of a small area of high concentration of
234
sea ice in the southwestern Weddell Sea and near the Antarctic Peninsula, the ice spreads over
235
the entire Weddell Sea (Fig. 8). However, the summer ice cover is different in OCCAM 1/12
236
before and after 1994. In the first part of the run the ice extent reaches a summer minimum
237
of 5.5 × 1010 m2 in 1990 (Fig. 7 b). After 1994, the seasonal variations are small and ice
238
concentration and extent remain high even in summer. During winter, ice concentrations are
239
at 99%, the maximum value allowed in the model, over most of the Weddell Sea. Only at the
240
ice edge, which is too far north, and close to the coast at the tip of the Antarctic Peninsula
241
and the South Shetland Islands, is the ice less dense (Fig. 8). Concentrations remain too high
242
near the coast in the central Weddell Sea and prevent the formation of coastal polynyas such
243
as that observed in front of the Ronne-Filchner ice shelf (e.g. Renfrew et al., 2002).
244
4.2
OCCAM 1/4
245
The representation of the central Weddell Sea in OCCAM 1/4 resembles that of OCCAM
246
1/12. The main water masses are present with similar properties to those of OCCAM 1/12
247
(Figs. 2–6) and their properties fit the observations well. As in OCCAM 1/12, the mixed layer
248
is too deep (Fig. 2) and gets deeper during the model run. During the first nine years of the
249
model run (1985-1993), OCCAM 1/4 is more saline in the mixed layer than OCCAM 1/12.
250
Also, while in OCCAM 1/12 the upper two layers experience a warming over the entire section
251
in summer during that period, OCCAM 1/4 does not show this behaviour. After 1994 the
252
surface layers warm significantly only twice (in summer 1999/2000 and 2001/2002), and on
253
these occasions the warming is visible in both resolutions but only over the continental shelf
254
and slope. In OCCAM 1/4, the surface layer is at or close to the freezing point at the times of
255
all sections.
256
The amounts of WDW (WSDW) in OCCAM 1/4 show only a small increase (decrease)
257
from 30 to 35% (58 to 57%) during the model run. Only WSBW significantly decreases: the
258
occupied area drops from 11% to below 7%, a reduction by more than a third of the initial
259
amount. The WSDW is slightly deeper in the lower resolution version, occupies less area along
260
the section than in OCCAM 1/12 and is warmer. WSBW is also deeper in OCCAM 1/4, but
261
θ, S and the amount of WSBW are similar to OCCAM 1/12.
262
The different behaviour of the two runs in the surface layer coincides with differences in
263
the ice cover. The seasonal cycle in the sea ice of OCCAM 1/4, although present, is far less
264
pronounced than in the satellite observations (Fig. 7 a). The climatological minimum ice extent
265
in February is 3.2 × 1012 m2 , a reduction of less than 30% of the August maximum. The
266
observations show a far larger seasonal cycle where in summer the ice cover is reduced by about
267
70% of the winter extent. As with the higher resolution version, OCCAM 1/4 produces dense
268
pack ice with maximum ice concentrations in winter (Fig. 8). In contrast to OCCAM 1/12, the
269
sea ice concentration remains high even in summer. As in the surface water properties, the two
270
versions of OCCAM become more similar in the sea ice evolution after 1994 (Figs. 7 a and 7 b).
271
The seasonal sea ice cycle in OCCAM 1/4 is consistent throughout the model run; the abrupt
272
change observed in OCCAM 1/12 does not appear. The polynya in the OCCAM 1/12 ice cover
273
in 1985 and 1986 does not appear in OCCAM 1/4 and the propagation of colder water around
274
the Weddell Sea as observed in OCCAM 1/12 is not visible in OCCAM 1/4.
275
4.3
ORCA025
276
ORCA025, the other 1/4◦ -model in our comparison, is generally too warm and, except in
277
the uppermost layer and above the shelves, slightly too saline (Figs. 2 and 3). In summer, the
278
surface layer is up to 2.8◦ C warmer than measured. At the shelf break, both temperature and
279
salinity are lower than in the CTD data. The modelled descent of the isolines starts further
280
away from the shelf over a deeper part of the continental slope and is less steep in the model.
281
Towards the end of summer (sections S04a and SR04e, not shown), the surface layer becomes
282
too saline. Overall, the salinity gradient from the surface to the salinity maximum in the WDW
283
is less steep than observed. The mixed layer depth is stable throughout the run at around 100
284
m. However, the average depths of the deeper water masses increase (Figs. 4-6). In ORCA025,
285
the average temperatures of WDW are 0.3◦ C too high. WDW warms substantially during the
286
first ten model years. The warming is slower during the rest of the run (Fig. 4). In ORCA025,
287
WDW occupies a larger area along the sections (on average 43.9% in the sections) than in the
288
CTD measurements (28.6%) and the average depth is 285 m deeper. Consequently, WSDW and
289
WSBW are deeper as well (Fig. 5,6). The WSDW salinity is overestimated by 0.005 (Fig. 5).
290
Average θ for WSDW matches the observations quite well, but the coldest deep water in the
291
model only just qualifies for WSBW and reaches a minimum of −0.719◦ C in the austral spring
292
section of 1989 (SR02). As in the OCCAM runs, the plume of newly formed WSBW along the
293
western continental shelf is not visible.
294
The Weddell Sea in ORCA025 becomes warmer during the model run. As the temperature
295
increases, the area occupied by WDW becomes larger and spreads deeper while the average
296
depth of WSDW and WSBW increases and the area occupied by WSBW decreases (Fig. 4-6).
297
At the same time, salinity increases too. Eventually, the entire water column is so warm that
298
from 1994 onwards no water fulfills Fahrbach et al. (2004)’s criterion for WSBW. The average
299
area, temperature and salinity of WSDW vary only a little.
300
The sea ice in ORCA025 is very different from the simulations by OCCAM (Fig. 8,7).
301
The timing of minimum and maximum ice coverage and the annual mean ice extent agree
302
with SSM/I data (Fig. 7). In winter, the ice concentrations are closer to the observations than
303
in both OCCAM versions (Fig. 8). However, although ORCA025 manages to simulate lower
304
concentrations close to the coast, they are still too high. The climatological maximum extent
305
is slightly overestimated but still below the values for OCCAM 1/12 and OCCAM 1/4. In
306
summer, almost all sea ice disappears and the ice extent decreases to 6.8 × 1010 m2 , an order
307
of magnitude less than observed. Only a very small amount of low concentration multi-year ice
308
remains in the southern Weddell Sea.
309
4.4
TPAC
310
TPAC is, like ORCA025, too warm (Figs. 2 and 3). At the surface, θ drops to below −1.4◦
311
C during September to December while during the rest of the year and particularly in late
312
summer and early autumn θ reaches values of more than −0.4◦ C. In the CTD data, θ remains
313
close to the freezing point at −1.8◦ C in the upper 10 m in all sections except for the summer
314
section SR04d where θ reaches 0.5◦ C near the Antarctic Peninsula (not shown). In the model,
315
cold and fresh Antarctic Surface Water occupies only the uppermost model layer and the mixed
316
layer is far too shallow at just over 30 m, which corresponds to the upper two model layers.
317
The surface layer and the waters below down to 150 m are too saline in all sections. Only over
318
the continental slope both in the eastern and western Weddell Sea is the model colder and
319
fresher than observed, indicating that the downward sloping of the isopycnals is too far away
320
from the coast. The plume of cold fresh water flowing down the western continental slope is
321
not reproduced.
322
TPAC is the only model in our comparison that is overall too saline in all of the sections.
323
While the maximum salinity (Smax ) in the CTD section is between 34.693 and 34.706 in the
324
WDW, even the average salinity in the WDW exceeds these values reaching 34.714 to 34.738
325
and Smax is 34.793 in section SR04b (Figs. 3 and 4). The WDW is on average more than 0.2◦
326
C warmer than observed. As the entire water column is too warm in TPAC, the area occupied
327
by WDW is also overestimated. The water warms further during the model run and the WDW
328
area percentage increases while WSDW decreases. WDW gets warmer and saltier but remains
329
slightly colder than in ORCA025. As the area occupied by WDW increases, so does the average
330
depth. The WSDW area decreases during the run (Fig. 5). The salinity of the deep water
331
corresponds to the observations, but θ is too high from the beginning of the run. The coldest
332
bottom water in TPAC is far too warm at −0.45◦ C (Fig. 3). Using Fahrbach et al. (2004)’s
333
criteria, TPAC does not simulate any WSBW at all (Fig. 6). As even in the first month of the
334
model run (January 1982) no WSBW is present, this suggests that the initialisation dataset is
335
too warm. TPAC shows a clear drift in both θ and S. The increase during the run is similar to
336
ORCA025 but occurs over a far shorter period of time (20 years cf. 47 years).
337
4.5
Error statistics
338
In order to quantify the model performance in reproducing the observed hydrography and
339
sea ice cover in the Weddell Sea, we calculated the basic statistical properties of observed
340
and modelled variability, correlation, and root mean square error (RMSE) for observed and
341
modelled θ and S along the sections, and sea ice extent and area in the Weddell Sea (Table 3).
342
The equations for the statistical variables are given in Table 3.
343
Figure 9 combines the various measures in a single diagram (Taylor, 2001). We use the CTD
344
measurements and the remote sensing sea ice observations as the reference datasets, marked
345
by the point labelled “reference” at standard deviation (σ) = 1 and correlation coefficient (R)
346
= 1. To include θ and S from all sections and the sea ice data, we use the normalised standard
347
deviation σnorm , that is σnorm = σdata /σobs for both model and observational data. The better
348
the correlation of the modelled data to the observations, the closer the corresponding model
349
marker will be to the x-axis. Similarly, the better the standard deviation (i.e. the variability)
350
fits the observed value, the closer the marker will be to the dashed black line marking the unit
351
circle. The red dotted semi circles mark how good the fit is taking into account equally both
352
correlation and agreement of variability. This “goodness of fit” corresponds to a centred pattern
353
root mean square error (RMSE), where a general bias in the model data is not considered.
354
The models are rather spread out on the diagram, i.e. they simulate some sections well
355
and some badly (Fig. 9). Generally, the models simulate θ along the sections better than S.
356
In OCCAM 1/12, the spatial variability of θ along the sections is close to the observations
357
and σnorm is close to 1. However, the patterns observed are not represented very well and R
358
is below 0.8 for all sections except S04a. The spatial distribution of S correlates better to the
359
observations but σnorm is higher and the RMSE larger. The seasonal cycle of the sea ice extent
360
is more poorly simulated than in OCCAM 1/4 and ORCA025 due to the offset of the annual
361
cycle.
362
OCCAM 1/4 matches the patterns and the distribution of the observed θ better than the
363
high resolution OCCAM 1/12. The RMSE of S varies between the sections from as low as 0.036
364
to a maximum of 0.61 (centred RMSE = 0.60). The temporal development of the sea ice extent
365
is closer to observations than in OCCAM 1/12, but the persistent summer ice cover and the
366
strong underestimation of the seasonality lead to very low σnorm (0.45).
367
Both for θ and S, the correlation is better in ORCA025 than in the other models: R varies
368
between 0.80 and 0.95. σnorm is larger though which suggests an overestimation of the spatial
369
variability in both variables. The centred pattern RMSE for θ is much lower than the standard
370
RMSE (difference of more then 0.1 in all sections) which confirms that ORCA025 is biased
371
towards high θ. The correlation of the sea ice extent in the model to the observations is very
372
high at R=0.98, but σnorm is well above 1 as well which is caused by the strong seasonality and
373
the underestimation of the summer sea ice in ORCA025.
374
TPAC shows the lowest correlations for θ and S of all the models. It also has the highest
375
standard RMSE for both variables. As in ORCA025, the centred RMSE for θ is much lower
376
than the standard RMSE, a sign of the warm bias of the model.
377
5
Discussion
378
Although all three models are global primitive equation z-level models, they behave very
379
differently. There are no systematic biases common to all the models in the region. Two res-
380
olutions of the same model (OCCAM) also produce different results. Nevertheless, OCCAM
381
1/4 and OCCAM 1/12 have common issues. Both model versions display a too deep and too
382
cold mixed layer (Fig. 2). This is linked to an extensive sea ice cover (Fig. 8). Very high ice
383
concentrations prevent heat uptake through solar radiation during the summer months, thus
384
preventing a warming of the ocean surface. In winter though, the lack of open water reduces
385
heat loss from the ocean to the atmosphere. In both OCCAM 1/12 and OCCAM 1/4, the first
386
effect seems to be stronger and helps to produce a very cold, deep mixed layer. The high ice
387
concentrations also reduce ice production. In the real oceans, ice formation in particular in
388
coastal polynyas (that do not appear in OCCAM at all) is an important factor as the process
389
is closely linked to salinification, HSSW formation (Renfrew et al., 2002; Foldvik et al., 2004)
390
and hence WSBW formation. The formation of coastal polynyas in OCCAM is compromised
391
by the weak wind forcing applied to both resolutions, and the missing offshore winds (Fig. 10).
392
A cold anomaly in the WDW θmax moving cyclonically around the Weddell Sea appears
393
only in OCCAM 1/12 and is not seen in the observations. The cold water is produced in a large
394
polynya in the Maud Rise region which is present in the model in the austral winter in 1985
395
and 1986, when the sea ice cover has diverged from the initial state. Travel times for anomalies
396
associated with the Antarctic Coastal Current around the entire perimeter of the Weddell Sea
397
are in the range of a few years (Beckmann et al., 1999; Schodlok et al., 2001). The cold water
398
signal which appears on the inner Weddell Sea side of the Antarctic Coastal Current travels
399
much slower in OCCAM 1/12. It takes two years for it to move from the Greenwich Meridian
400
to the hydrographic section at Kapp Norvegia. By the time corresponding to the last section
401
(SR04e, April 1998), the cold anomaly in OCCAM 1/12 has reached the western end of the
402
section at the Antarctic Peninsula. An indication of the weak Antarctic Coastal Current and
403
the slow circulation of the Weddell Gyre in both OCCAM versions is the small difference of sea
404
surface height (SSH) between the Antarctic Peninsula and the minimum SSH in the middle of
405
the gyre. Fig. 11 (upper panel) shows that OCCAM 1/12 and OCCAM 1/4 have a much weaker
406
gyre than ORCA025 and TPAC. This is due to the weak wind forcing in OCCAM compared
407
with the much stronger winds in ORCA025 (Fig. 10). The OCCAM wind stress is higher than
408
ORCA025’s in the central Weddell Sea while ORCA025 winds are much stronger north of 65◦
409
S and along the Antarctic coast The weak Weddell Gyre in OCCAM may contribute to the
410
extensive sea ice cover: the slow currents lead to long residence times of water in cold regions and
411
prevent faster heat transport into the Weddell Sea by inflow of warmer ACC water. Although
412
still weak, the Antarctic Coastal Current is faster in OCCAM 1/12 than in OCCAM 1/4 even
413
though the forcing is the same. The difference between the two resolutions could be explained by
414
the Gent and McWilliams (1990) parameterisation (GM90) used in OCCAM 1/4. GM90 is used
415
to remove baroclinic instabilities for numerical stability and, although the Antarctic Coastal
416
Current is mostly stable, GM90 reduces the horizontal density gradients across the current and
417
weakens the Coastal Current (Beckmann, pers. comm.). Also, the very dense and mostly land-
418
fast sea ice pack in OCCAM 1/4 with concentrations of > 95% reduces momentum transfer
419
from the wind to the upper ocean. The lower sea ice concentrations in the higher resolution
420
version allow for stronger wind forced currents.
421
OCCAM 1/12 also does not seem to have reached equilibrium in the Weddell Sea. While
422
the ACC transport through Drake Passage seems to be stable soon after the start of the model
423
run (Fig. 11b), the mixed layer depth in the central Weddell Sea continues to increase as does
424
the depth of θmax in the WDW. This indicates problems with the surface fluxes. OCCAM 1/4
425
on the other hand does appear to have reached equilibrium.
426
In ORCA025, the absence of multi-year ice facilitates too strong warming of the mixed
427
layer, the exact opposite to OCCAM. The warming is enhanced by a low baroclinic ACC
428
transport (Fig. 11 b). In the form of a feedback loop, the strength of the baroclinic ACC
429
transport is linked to the bottom water formation in the Weddell Sea: a weak ACC relates
430
to a weak meridional surface buoyancy gradient. This means that warmer surface water can
431
penetrate further south and thereby lateral heat transport into the Weddell Sea is enhanced
432
which reduces bottom water formation. This in turn reduces the stratification which weakens
433
the thermohaline forcing component of the ACC (Olbers and W¨ ubber, 1991; Cai and Baines,
434
1996). The initialisation of the model is a major reason for the missing summer sea ice: the final
435
state of a ten year spin up run has been used which had almost no sea ice left in the Weddell
436
Sea (J.-M. Molines, pers. comm.). The ice cover did not recover during the model run. One
437
likely explanation for this is the high sea surface temperature observed throughout the Weddell
438
Sea during the first month of the model run. In January 1958, the first month of the model run,
439
the surface temperature is only at the freezing point beneath existing sea ice; elsewhere in the
440
Weddell Sea it is up to 2◦ C. The almost complete loss of the highly reflective ice cover during
441
summer allows for large heat uptake by the ocean through solar radiation and therefore strong
442
warming. In winter, strong winds in the ECMWF forcing (Fig. 10) can quickly advect any
443
sea ice out of the Weddell Sea and thereby promote a faster decrease to the very low summer
444
concentrations. This export of freshwater would also explain the increase in salinity over the
445
ORCA025 model run.
446
In our comparison, TPAC is the only ocean model that is not coupled to a sea ice model.
447
Combined with the loose constraining of sea surface temperature and salinity to hemispheric
448
wintertime conditions, this can explain the high salinities. Excluding sea ice means that the
449
influence of sea ice formation, melt and drift on the freshwater budget is missing. The constraint
450
to wintertime conditions probably leads to overestimation of the surface salinities because the
451
input of freshwater due to sea ice melt in summer is ignored. The missing sea ice cycle also
452
affects the modification of shelf waters, so that dense water formation is heavily restricted and
453
bottom water missing. The baroclinic ACC transport decreases during the model run. This
454
allows increasing entrainment of warm water into the Weddell Sea, leading to the increasing
455
water temperatures, the increase of WDW and the decrease of WSDW. TPAC shows a large
456
drift which demonstrates the need for suitable forcing (possibly with stronger relaxation of sea
457
surface temperature and salinity to climatology) or coupling to a sea ice model.
458
Model validations using observations always struggle with the snapshot-like nature of most
459
observational data. Often, model results are compared with either observations from a single
460
point in time or a very short period (e.g. a hydrographic section), or long-term climatologies.
461
While the first approach neglects the fact that variability in the real world and in the model do
462
not necessarily coincide, the latter neglects the spread caused by exactly that variability. Using
463
repeat sections provides information on the temporal variability. Sampling the models at the
464
different sets of stations demonstrates variations due to station spacing. Hydrographic sections
465
are measured over several weeks and include short-term features such as eddies. Ocean models
466
cannot exactly reproduce the observations but aim to capture the level of natural variability
467
over the course of the model run. The models considered in this study manage to simulate a
468
certain level of variability which covers some of the changes seen in the observations, e.g. area
469
percentage, θ, and S of WDW in OCCAM 1/4, θ of WSDW in OCCAM 1/12, OCCAM 1/4 and
470
ORCA025, and WSBW area percentage and depth in OCCAM 1/12 and OCCAM 1/4 (Figs. 4,
471
5, and 6, respectively). They are, however, often in disagreement with the observed mean values.
472
This is most pronounced for the deep water masses and in ORCA025 and TPAC where both
473
models fail to reproduce the observed amounts and the characteristics of WSDW and WSBW.
474
During the entire model runs, the average depth of the WSDW layer was 200 to 400 metres
475
too deep, θ of WSDW in TPAC 0.1◦ C too warm, and WSBW completely missing in TPAC.
476
The WSBW layer in ORCA025 disappears after 38 years of the model run. Less drastically,
477
both OCCAM versions struggle as well and the WSBW in both resolutions is too fresh and too
478
warm. They do however simulate all major features of the Weddell Sea hydrography. Using the
479
repeat sections in relation to the entire model runs shows that the model biases are consistent
480
for each model, e.g. for all sections, the area percentage of WDW is too low in OCCAM 1/12
481
and too high in ORCA025, but there are exceptions, notably for section S04a.
482
483
Although the models are set up differently and use different forcings, general conclusions emerge by comparing their performances:
484
1. Increasing resolution is not a panacea. As noted by Barnier et al. (2006), model nu-
485
merics thoroughly tested at coarse resolutions may require reformulation for higher resolution.
486
Moving to higher resolution, both horizontally and vertically, can improve model simulations
487
significantly (Lee et al., 2002; Thoma et al., 2006). However at the level of resolution in eddy-
488
permitting and eddy-resolving models, better parameterisations and model numerics are just as
489
important. Examples are better advection schemes, improved and suitably chosen subgridscale
490
parameterisations, and use of partial bottom cells (Lee et al., 2002; Legg et al., 2008; Penduff
491
et al., 2007). It seems that the step from coarse 1◦ or 2◦ models to eddy-permitting models was a
492
much more vital one in terms of improving model hydrography and large-scale circulation than
493
going from eddy-permitting to eddy-resolving. The absence of major differences between our
494
OCCAM 1/12 and OCCAM 1/4 results demonstrate this. Within the range of resolutions in the
495
models presented in this paper, we do not find one to perform better than the others. Very high
496
resolution, such as OCCAM’s 1/12◦ or POP 1/10 as used by Maltrud and McClean (2005),
497
becomes important when examining processes that are strongly influenced by eddies, where
498
currently used subgrid parameterisations in eddy-permitting models are still insufficient (e.g.
499
Legg et al., 2006). In high latitudes however, model grids of nominally eddy-permitting models
500
such as the ORCA025 grid (Madec and Imbard, 1996) can provide high enough resolution to
501
include the effects of eddies.
502
2. The choice of the forcing datasets is important. Barnier et al. (2007) use various forcings
503
for the same model and show the large impact of the different datasets on the model simula-
504
tions. Their experiments led to the development of a coherent set of forcing variables (Brodeau
505
et al., submitted). In the Weddell Sea, errors in the freshwater and heat flux forcings can have
506
considerable impact on the ocean by altering sea ice production and melt and the density of
507
shelf waters. Demonstrating the sensitivity of the ocean to wind forcing, Saenko and Weaver
508
(2001) and Saenko et al. (2002) found that wind induced sea ice motion is important for the
509
formation of Antarctic Intermediate and Bottom Water and ocean ventilation. The differences
510
in our model simulations support the importance of realistic wind forcings. The weak NCEP
511
wind forcing is at least partly responsible for the build-up of the very dense sea ice pack and
512
the weak Weddell Gyre in the OCCAM runs. Gnanadesikan et al. (2004) demonstrated how
513
observations such as Levitus et al. (1998) used for restoring models did not allow deep water
514
masses to outcrop at the surface and hence restricted the formation of deep and bottom waters
515
in the Ross and Weddell Seas. Surface restoring of temperature and salinity cannot therefore
516
replace the accurate simulation of the underlying processes in order to produce the right water
517
masses.
518
3. Sea ice models need to be improved. The lack of a sea ice model in TPAC makes a
519
realistic simulation of the processes in the Weddell Sea difficult. The missing bottom water, the
520
warming of the entire water column, and the low salinities above the continental shelf and slope
521
show the necessity of the inclusion of sea ice formation and melt processes for a correct model
522
hydrography. The relaxation using salt flux derived from latent heat and precipitation is not
523
sufficient to ensure adequate bottom water formation without the sea ice processes. From the
524
ORCA025 run, we learn that the initialisation of the ice cover can impact the entire model run.
525
Kerr et al. (2009) analysed ocean model runs with and without a coupled sea ice model. They
526
found that the effect of the sea ice model on the ocean is complex. Having used a 1◦ model, they
527
conclude that both refined resolution and improved parameterisations of physical processes are
528
necessary to simulate ice-ocean interactions and the hydrography in the Weddell Sea. Hellmer
529
(2004) and Wang and Beckmann (2007) showed that including ice shelves in a coupled sea
530
ice-ocean model improves the simulation of the sea ice cover and alters the hydrography in the
531
Weddell Sea with global effects. None of the models discussed in this paper include ice shelves.
532
A previous model intercomparison study suggested biases specific to z-level models. As part
533
of the DYNAMO (Dynamics of the North Atlantic circulation) project, Willebrand et al. (2001)
534
compare three models at 1/3◦ horizontal resolution with different vertical coordinate schemes
535
(z-level, sigma and isopycnic). They find that the z-level model suffers from the steplike repre-
536
sentation of the topography combined with insufficient resolution (30 vertical levels). Combined
537
with inadequate numerical algorithms for advection and mixing, this leads to excessive diapyc-
538
nal mixing and poor representation of downslope flows. Similarly, investigating the performance
539
of two z-level models including an earlier version of OCCAM 1/4 in the Scotia Sea, Thorpe
540
et al. (2005) suggest that low vertical resolution, poor representation of the bathymetry, and
541
issues with the vertical mixing schemes lead to errors in the circulation and loss of Antarctic
542
Bottom Water. In this paper, we have shown that the new version of OCCAM 1/4 with the
543
higher vertical resolution and partial bottom cells is much improved and capable of simulating
544
Weddell Sea deep water masses. In ORCA025 and TPAC, however, with fewer vertical levels
545
and different vertical mixing schemes, WSBW is lost rapidly during the model run or is not
546
present at all, suggesting both the parameterisation and the vertical resolution are not suf-
547
ficient. All our models are too warm and too saline in the bottom layer of the Weddell Sea
548
suggesting that further improvements can be made with the representation of bottom water
549
formation processes including the simulation of cold overflows.
550
6
Conclusions
551
Comparing the output of three global OGCMs with observational data and with each other,
552
we find significant differences. The results of the same model with different horizontal resolution
553
(OCCAM 1/12 and OCCAM 1/4) are very similar but the higher resolution in OCCAM 1/12
554
and the different approach to vertical mixing lead to different behaviour in the upper ocean and
555
the sea ice model. Qualitatively, the results of ORCA025 and TPAC are close to each other but
556
very different from OCCAM. The mismatches between models and observations are genuine
557
and not an artefact of the snapshotlike nature of the hydrographic sections. The models are at
558
high enough resolution to reproduce some variability due to station location, but struggle with
559
the average values or fail to simulate the deep water masses correctly.
560
Ocean models are often used and developed to study specific areas or regions, for example
561
the North Atlantic overturning circulation or the Agulhas retroflection. They are tuned to
562
represent those regions well, while other regions might be less well simulated, as long as on a
563
global scale, results seem reasonable. As the models in this comparison produce very different
564
results for different parameters, it is not possible to choose “the best model”. Instead, the choice
565
should be made according to the application. We find that for analyses regarding deep water
566
mass properties, ORCA025 and TPAC perform poorly and OCCAM does best. However, the
567
seasonality of the sea ice cover and the surface of the ocean is very limited in OCCAM. The high
568
resolution of OCCAM 1/12 allows for much more detail as eddies are explicitly resolved and
569
topography is more precise. Thompson and Heywood (2008) demonstrate how higher resolution
570
in observations changes our understanding of processes in regions with narrow jets and the same
571
can apply to model analyses. For processes which are closely linked to the growth and retreat of
572
sea ice, for example assessing the spring phytoplankton bloom or air-sea CO2 fluxes, ORCA025
573
is more suitable. Problems can arise where the existence of multi-year ice is of importance.
574
Reasons for the different model performances include the use of different forcing datasets,
575
different initialisations and different parameterisations of vertical mixing and other subgridscale
576
processes. When using a global model for a regional study, the choice has to be made carefully;
577
our study has shown how models can be suitable for specific applications such as outflow of
578
bottom water, ocean-topography interaction, or ecological studies. It is difficult to say, which
579
improvements to OGCMs are most likely to be the most effective. Comparisons to previous
580
studies have shown the effect of better mixing and advection parameteristions and the benefit
581
of vertical resolution and partial bottom cells. Eddy-permitting OGCMs are being tested to
582
be used in the new generation of coupled climate models (e.g. Shaffrey et al., 2008, early
583
online release; WCRP, 2009). Keeping this in mind, we suggest that at least for global models
584
the focus should be on improving vertical mixing and horizontal advection schemes, better
585
parameterisations of processes at the ice-ocean interface and in the sea ice models, and on more
586
self-consistent and realistic forcing datasets, and not on increasingly higher resolutions.
587
Acknowledgements
588
We thank the modelling groups for their help and providing the model data: Andrew Coward
589
and the OCCAM team at the National Oceanography Centre, Southampton, the DRAKKAR
590
project team, esp. Jean-Marc Molines and Bernard Barnier at LEGI-Grenoble, and Jason
591
Roberts at ACECRC. Eberhard Fahrbach at the Alfred Wegener Institute made the hydro-
592
graphic data available. Nuno Nunes provided the ACC transports and Aike Beckmann valuable
593
advice. Comments from Robert Hallberg and two anonymous reviewers helped to improve the
594
manuscript. This work was funded by a EU Marie Curie Early Stage Training Fellowship in
595
Antarctic Air-Sea-Ice Science (FAASIS) by the European Commission Marie-Curie Actions at
596
the School of Environmental Sciences, University of East Anglia, and the British Antarctic
597
Survey, contract number MEST-CT-2004-514159.
598
References
599
Aksenov, Y., 2002. The sea ice-ocean global coupled ARCICE project report part 1; description of dynamical-thermodynamical sea
600
ice model. SOC Research & Consultancy Report 103, 83pp., Southampton Oceanography Centre.
601
Barnier, B., Brodeau, L., Le Sommer, J., Molines, J.-M., Penduff, T., Theetten, S., Treguier, A.-M., Madec, G., Biastoch, A.,
602
B¨ oning, C., Dengg, J., Gulev, S., Bourdall´ e Badie, R., Chanut, J., Garric, G., Alderson, S., Coward, A., de Cuevas, B., New, A.,
603
Haines, K., Smith, G., Drijfhout, S., Hazeleger, W., Severijns, C., Myers, P., 2007. Eddy-permitting ocean circulation hindcasts
604
of past decades. CLIVAR Exchanges vol. 12 (3), p. 8–10.
605
Barnier, B., Madec, G., Penduff, T., Molines, J.-M., Treguier, A.-M., Le Sommer, J., Beckmann, A., Biastoch, A., B¨ oning, C.,
606
Dengg, J., Derval, C., Durand, E., Gulev, S., Remy, E., Talandier, C., Theetten, S., Maltrud, M., McClean, J., De Cuevas,
607
B., 2006. Impact of partial steps and momentum advection schemes in a global ocean circulation model at eddy-permitting
608
resolution. Ocean Dynamics 56(5-6), 543–567.
609 610 611 612 613 614
Beckmann, A., Hellmer, H. H., Timmermann, R., 1999. A numerical model of the Weddell Sea: Large-scale circulation and water mass distribution. Journal of Geophysical Research 104(C10), 23,375–23,391. Brodeau, L., Barnier, B., Treguier, A.-M., Penduff, T., Gulev, S., submitted. An ERA40-based atmospheric forcing for global ocean circulation models. Ocean Modelling. Bryan, K., 1969. A Numerical Method for the Study of the Circulation of the World Ocean. Journal of Computational Physics 135, 154–169.
615
Cai, W., Baines, P. G., 1996. Interactions between thermohaline- and wind-driven circulations and their relevance to the dynamics
616
of the Antarctic Circumpolar Current, in a coarse-resolution global ocean general circulation model. Journal of Geophysical
617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638
Research 101 (C6), 14,073–14,093. Comiso, J., 1990, updated 2007. DMSP SSM/I daily polar gridded sea ice concentrations, June to September 2001. Boulder, CO, USA: National Snow and Ice Data Center. Digital media. Coward, A. C., de Cuevas, B. A., 2005. The OCCAM 66 Level Model: physics, initial conditions and external forcing. Tech. Rep. SOC Internal Report No. 99, Southampton Oceanography Centre. Cox, M. D., 1984. A primitive equation 3-dimensional model of the ocean. GFDL Ocean Group Technical Report No 1, Geophysical Fluid Dynamics Laboratory/NOAA, Princeton University, Princeton, N.J. 08542. Dai, A., Trenberth, K. E., 2002. Estimates of freshwater discharge from continents: Latitudinal and seasonal variations. Journal of Hydrometeorology 3, 660–687. DBDB5, 1983. U.S. Naval Oceanographic Office, and the U.S. Naval Ocean Research and Development Activity, Digital Bathymetric Data Base-5 minute grid. Fahrbach, E., Harms, S., Rohardt, G., Schr¨ oder, M., Woodgate, R. A., 2001. Flow of bottom water in the northwestern Weddell Sea. Journal of Geophysical Research 106(C2), 2761–2778. Fahrbach, E., Hoppema, M., Rohardt, G., Schr¨ oder, M., Wisotzki, A., 2004. Decadal-scale variations of water mass properties in the deep Weddell Sea. Ocean Dynamics 54, 77–91, doi:10.1007/s10236-003-0082-3. Fahrbach, E., Rohardt, G., Krause, G., 1992. The Antarctic Coastal Current in the southeastern Weddell Sea. Polar Biology 12, 171–182. Fahrbach, E., Rohardt, G., Sieger, R., 2007. 25 years of polarstern hydrography (1982-2007). WDC-MARE Reports, vol. 5, 88pp, incl. CD-ROM. Fichefet, T., Morales Maqueda, M. A., 1997. Sensitivity of a global sea ice model to the treatment of ice thermodynamics and dynamics. Journal of Geophysical Research 102(C6), 12,609–12,646. Foldvik, A., Gammelsrød, T., Tørresen, T., 1985. Circulation and water masses on the southern Weddell Sea shelf. In: Jacobs, S. S.
639
(Ed.), Oceanology of the Antarctic Continental Shelf. Antarctic Research Series. Vol. 43. AGU, Washington D.C., pp. 5–20.
640
Foldvik, A., Gammelsrød, T., Østerhus, S., Fahrbach, E., Rohardt, G., Schr¨ oder, M., Nicholls, K. W., Padman, L., Woodgate,
641
R. A., 2004. Ice shelf water overflow and bottom water formation in the southern Weddell Sea. Journal of Geophysical Research
642
109, C02015, doi:10.1029/2003JC002008.
643 644
Foster, T. D., Carmack, E. C., 1976. Temperature and salinity structure in the Weddell Sea. Journal of Physical Oceanography 6, 36–44.
645 646
Gent, P. R., McWilliams, J. C., 1990. Isopycnal mixing in ocean circulation models. Journal of Physical Oceanography 20(1), 150–155.
647
Gill, A. E., 1973. Circulation and bottom water production in the weddell sea. Deep-Sea Research 20(2), 111–140.
648
Gnanadesikan, A., Dunne, J. P., Key, R. M., Matsumoto, K., Sarmiento, J. L., Slater, R. D., Swathi, P. S., 2004. Oceanic ven-
649
tilation and biogeochemical cycling:Understanding the physical mechanisms that produce realistic distributions of tracers and
650 651 652 653 654 655 656
productivity. Global Biogeochemical Cycles 18, GB4010, doi:10.1029/2003GB002097. Gouretski, V. V., Jancke, K., 1996. A new hydrographic data set for the South Pacific: synthesis of WOCE and historical data. WHP SAC Technical Report No. 2, WOCE Report No. 143/96 (unpublished manuscript), WHP Special Analysis Centre. Gouretski, V. V., Jancke, K., 1998. A new world ocean climatology: Optimal interpolation of historical and WOCE hydrographic data on neutral surfaces. WOCE Technical Report No. 162/89, WHP Special Analysis Centre. Greenfell, T. C., Perovich, D. K., 1984. Spectral albedos of sea ice and incident solar irradiance in the southern Beaufort Sea. Journal of Geophysical Research 89, 3573-3580.
657
Griffies, S. M., 1998. The Gent-McWilliams skew flux. Journal of Physical Oceanography 28(5), 831–841.
658
Hellmer, H. H., 2004. Impact of Antarctic ice shelf basal melting on sea ice and deep ocean properties. Geophysical Research Letters
659 660 661 662 663
31, L10307, doi:10.1029/2004GL019506. Hellmer, H. H., Beckmann, A., 2001. The Southern Ocean: a ventilation contributor with multiple sources. Geophysical Research Letters 28(15), 2927–2930. Hellmer, H. H., Schodlok, M. P., Wenzel, M., Schr¨ oter, J. G., 2005. On the influence of adequate Weddell Sea characteristics in a large-scale global ocean circulation model. Ocean Dynamics 55, 88–99.
664
Hibler, W. D., 1979. A dynamic thermodynamic sea ice model. Journal of Physical Oceanography 9, 815-846.
665
Holland, M. M., Raphael, M. N., 2006. Twentieth century simulation of the southern hemisphere climate in coupled models. Part
666 667 668 669 670
II: sea ice conditions and variability. Climate Dynamics 26, 229–245. Hunke, E. C., Dukowicz, J. K., 1997. An elastic-viscous-plastic model for sea ice dynamics. Journal of Physical Oceanography 27, 1849-1867. Jacobs, S. S., 2004. Bottom water production and its links with the thermohaline circulation. Antarctic Science 16(4), 427–437, doi:10.1017/S095410200400224X.
671
Kalnay, E., Kanamitsu, M., Kistler, R., Collins, W., Deaven, D., Gandin, L., Iredell, M., Saha, S., White, G., Woollen, J., Zhu,
672
Y., Chelliah, M., Ebisuzaki, W., Higgins, W., Janowiak, J., Mo, K. C., Ropelewski, C., Wang, J., Leetmaa, A., Reynolds, R.,
673
Jenne, R., Joseph, D., 1996. The NCEP/NCAR 40-year reanalysis project. Bulletin of the American Meteorological Society
674
77(3), 437–471.
675 676 677 678 679 680 681 682 683 684 685 686 687 688
Kanamitsu, M., Ebisuzaki, W., Woollen, J., Yang, S. K., Hnilo, J. J., Fiorino, M., Potter, G. L., 2002. NCEP-DOE AMIP-II reanalysis (R-2). Bulletin of the American Meteorological Society 83(11), 1631–1643. Kerr, R., Wainer, I., Mata, M. M., 2009. Representation of the Weddell Sea deep water masses in the ocean component of the NCAR-CCSM model. Antarctic Science doi: 10.1017/S0954102009001825. Large, W. G., Danabasoglu, G., Doney, S. C., McWilliams, J. C., 1997. Sensitivity to surface forcing and boundary layer mixing in a global ocean model: annual-mean climatology. Journal of Physical Oceanography 27(11), 2418–2447. Large, W. G., Yeager, S. G., 2004. Diurnal to decadal global forcing for ocean and sea-ice models: the data sets and flux climatologies. Tech. rep., CGD Division of the National Center for Atmospheric Research. Lee, M.-M., Coward, A. C., Nurser, A. J. G., 2002. Spurious diapycnal mixing of the deep waters in an eddy-permitting global ocean model. Journal of Physical Oceanography 32, 1522–1535. Legg, S., Hallberg, R. W., Girton, J. B., 2006. Comparison of entrainment in overflows simulated by z-coordinate, isopycnal and non-hydrostatic models. Ocean Modelling 11, 69–97. Legg, S., Jackson, L., Hallberg, R. W., 2008. Eddy-resolving modeling of overflows. In: Hecht, M. W., Hasumi, H. (Eds.), Ocean Modeling in an Eddying Regime. Geophysical Monograph 177. AGU, Washington D.C., pp. 63–82.
689
Levitus, S., 1982. Climatological Atlas of the World Ocean. Noaa prof. paper, GFDL, Princeton University, NJ.
690
Levitus, S., Boyer, T. P., Conkright, M. E., O’Brian, T., Antonov, J., Stephens, C., Stathopolos, L., Johnson, D., Gelfeld, R., 1998.
691 692 693 694 695
World ocean database 1998. NOAA Atlas NESDID 18, US Government Printing Office, Washington, DC. Lythe, M. B., Vaughan, D. G., 2001. BEDMAP: a new ice thickness and subglacial model of Antarctica. Journal of Geophysical Research 106, 11335–11351. Madec, G., 2006. NEMO reference manual, ocean dynamics component : NEMO-OPA. Preliminary version. Note du pole de mod’elisation, nstitut Pierre-Simon Laplace, France.
696
Madec, G., Imbard, M., 1996. A global ocean mesh to overcome the North Pole singularity. Climate Dynamics 12, 381–388.
697
Maltrud, M. E., McClean, J. L., 2005. An eddy resolving global 1/10◦ ocean simulation. Ocean Modelling 8(1-2), 31–54.
698
Marsland, S. J., Church, J. A., Bindoff, N. L., Williams, G. D., 2007. Antarctic coastal polynya response to climate change. Journal
699 700 701
of Geophysical Research 112, C07009, doi:10.1029/2005JC003291. Martin, T., Gerdes, R., 2007. Sea ice drift variability in Arctic Ocean Model Intercomparison Project models and observations. Journal of Geophysical Research 112, C04S10, doi:10.1029/2006JC003617.
702
Meijers, A. J., Bindoff, N. L., Roberts, J. L., 2007. On the total, mean and eddy heat and freshwater transports in the southern
703
hemisphere of a 1/8◦ × 1/8◦ global ocean model. Journal of Physical Oceanography 37(2), 277–295, doi:10.1175/JPO3012.1.
704
Molines, J. M., Barnier, B., Penduff, T., Brodeau, L., Treguier, A. M., Theetten, S., Madec, G., 2006, updated 2007. Definition of
705 706 707 708 709 710
the interannual experiment ORCA025-G70 1958-2004. LEGI Technical note. Olbers, D., W¨ ubber, C., 1991. The role of wind and buoyancy forcing of the Antarctic Circumpolar Current. In: Latif, M. (Ed.), Strategies for Future Climate Research. Max-Planck-Institut f¨ ur Meteorologie, Hamburg, pp. 161–192. Parkinson, C. L., 2004. Southern Ocean sea ice and its wider linkages: insights revealed from models and observations. Antarctic Science 16 (4), 387–400. Penduff, T., Le Sommer, J., Barnier, B., Treguier, A.-M., Molines, J.-M., Madec, G., 2007. Influence of numerical schemes on
711
current-topography interactions in 1/4◦ global ocean simulations. Ocean Science 3 (4), 509–524.
712
URL http://www.ocean-sci.net/3/509/2007/
713 714 715 716 717 718 719 720
Renfrew, I. A., Moore, G. W. K., Guest, P. S., Bumke, K., 2002. A comparison of surface layer and surface turbulent flux observations over the Labrador Sea with ECMWF analyses and NCEP reanalyses. Journal of Physical Oceanography 32, 383–400. Renner, A. H. H., Lytle, V., 2007. Sea-ice thickness in the Weddell Sea, Antarctica: a comparison of model and upward-looking sonar data. Annals of Glaciology 46, 419–427. Reynolds, R. W., Rayner, N. A., Smith, T. M., Stokes, D. C., Wang, W., 2002. An improved in situ and satellite sst analysis for climate. Journal of Climate 15, 1609–1625. Roberts, M. J., Donners, J., Harle, J., Stevens, D., 2008. Impact of relative atmosphere-ocean resolution on coupled climate models. CLIVAR Exchanges 13(1), 8–11.
721
Row, L. W., Hastings, D. A., Dunbar, P. K., 1995. Terrainbase worldwide digital terrain data documentation manual.
722
Saenko, O. A., Schmittner, A., Weaver, A. J., 2002. On the role of wind-driven sea ice motion on ocean ventilation. 2002 32,
723 724 725 726 727 728 729 730 731 732
3376–3395. Saenko, O. A., Weaver, A. J., Nov. 2001. Importance of wind-driven sea ice motion for the formation of Antarctic Intermediate Water in a global climate model. Geophysical Research Letters 28, 4147–4150. Schodlok, M. P., Rodehacke, C. B., Hellmer, H. H., Beckmann, A., 2001. On the origin of the deep CFC maximum in the eastern Weddell Sea - numerical model results. Geophysical Research Letters 28 (14), 2859–2862. Semtner, A. J., 1974. A general circulation model for the World Ocean. Technical Report No 9, Department of Meteorology, University of California, Los Angeles. Semtner, A. J., 1976. A model for the thermodynamic growth of sea ice in numerical investigations of climate. Journal of Physical Oceanography 6, 379-389. Shaffrey, L. C., Stevens, I., Norton, W. A., Roberts, M. J., Vidale, P. L., Harle, J. D., Jrrar, A., Stevens, D. P., Woodage, M. J.,
733
Demory, M. E., Donners, J., Clark, D. B., Clayton, A., Cole, J. W., Wilson, S. S., Connolley, W. M., Davies, T. M., Iwi, A. M.,
734
Johns, T. C., King, J. C., New, A. L., Slingo, J. M., Slingo, A., Steenman-Clark, L., Martin, G. M., 2008, early online release.
735
UK-HiGEM: The new UK high resolution global environment model. Model description and basic evaluation. Journal of Climate
736
doi:10.1175/2008JCLI2508.1.
737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758
Shine, K. P., Henderson-Sellers, A., 1985. The sensitivity of a thermodynamic sea ice model to changes in surface albedo parameterization. Journal of Geophysical Research 90, 2243-2250. Smith, W. H. F., Sandwell, D. T., 1997. Global seafloor topography from satellite altimetry and ship depth soundings. Science 277, p. 1957–1962. Steele, M., Morley, R., Ermold, W., 2001. PHC: a global ocean hydrography with a high quality Arctic Ocean. Journal of Climate 14, 2079–2087. St¨ ossel, A., Kim, S. J., Drijfhout, S. S., 1998. The impact of Southern Ocean sea ice in a global ocean model. Journal of Physical Oceanography 28, 1999–2018. St¨ ossel, A., Yang, K., Kim, S. J., 2002. On the role of sea ice and convection in a global ocean model. Journal of Physical Oceanography 32, 1194–1208. Taylor, K. E., 2001. Summarizing multiple aspects of model performance in a single diagram. Journal of Geophysical Research 106(D7), 7183–7192. Thoma, M., Grosfeld, K., Lange, M. A., 2006. The impact of mixing parameterisation and bathymetry filtering on the simulated hydrography along steep continental shelf regions in terrain following ocean models. FRISP Report 17. Thompson, A. F., Heywood, K. J., 2008. Frontal structure and transport in the northwestern Weddell Sea. Deep-Sea Research I 55, 1229–1251. Thorpe, S. E., Stevens, D. P., Heywood, K. J., 2005. Comparison of two time-variant forced eddy-permitting global ocean circulation models with hydrography of the Scotia Sea. Ocean Modelling 9, 105–132. Timmermann, R., Worby, A., Goosse, H., Fichefet, T., 2004. Utilizing the ASPeCt sea ice thickness data set to evaluate a global coupled sea ice-ocean model. Journal of Geophysical Research 109, C07017, doi:10.1029/2003JC002242. Wang, C., Beckmann, A., 2007. Investigation of the impact of Antarctic ice-shelf melting in a global ice-ocean model (ORCA2-LIM). Annals of Glaciology 46, 78–82.
759
WCRP, 2009. Workshop Report: World Modelling Summit for Climate Prediction. WCRP Report, vol. 131, 31pp.
760
Webb, D. J., 2000. Evidence for shallow zonal jets in the South Equatorial Current region of the southwest Pacific. Journal of
761 762 763 764 765
Physical Oceanography 30(4), 706–720. Webb, D. J., de Cuevas, B. A., Coward, A. C., 1998. The first main run of the OCCAM global ocean model. Tech. Rep. Internal Document, No. 34, 44pp, Southampton Oceanography Centre. Weppernig, R., Schlosser, P., Khatiwala, S., Fairbanks, R. G., Nov. 1996. Isotope data from Ice Station Weddell: Implications for deep water formation in the Weddell Sea. Journal of Geophysical Research 101, 25723–25740.
766
Willebrand, J., Barnier, B., B”oning, C., Dieterich, C., Killworth, P. D., Le Provost, C., Jia, Y., Molines, J.-M., New, A. L., 2001.
767
Circulation characteristics in three eddy-permitting models of the North Atlantic. Progress in Oceanography 48(2-3), 123–161.
768
Xie, P. P., Arkin, P. A., 1997. Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates,
769
770
771
772
and numerical model outputs. Bulletin of the American Meteorological Society 78(11), 2539–2558.
List of Tables 1
Summary of the main characteristics of the models under validation. References for the bathymetry datasets are given in the text.
29
773
2
Overview of WOCE hydrographic sections used for the model validation
774
3
Error statistics for ocean model - observation comparison following Taylor
775
(2001). RMSEs for θ, and sea ice extent are given in
◦
C and m2 , respectively.
30
31
Table 1 Summary of the main characteristics of the models under validation. References for the bathymetry datasets are given in the text. Model
Ocean component
Vertical mixing scheme
Surface
salinity
Sea ice component
Horizontal and ver-
restoring OCCAM 1/12
Bryan-Cox-Semtner
(Coward and de
with free surface; eddy-
Cuevas, 2005)
type
No
isopycnic
mixing
scheme
resolving
Bathymetry source
tical Resolution
Relaxation of surface
Aksenov
(2002);
grid box salinity to
thermodynamic
with
dynamicelastic-
1/12◦ × 1/12◦ ;
Smith & Sandwell v6.2
66 vertical levels
with 72◦
DBDB5
south
Levitus et al. (1998)
viscous-plastic rheology (Semtner,
of
S; sill depths
at timescale of 30
1976; Hunke and Dukowicz, 1997)
checked manually
days OCCAM 1/4
As OCCAM 1/12; eddy-
Isopycnal mixing scheme
(Coward and de
permitting
(Griffies, 1998) with Gent
Cuevas, 2005)
As OCCAM 1/12
1/4◦ × 1/4◦ ;
As OCCAM 1/12
As for OCCAM 1/12
66 vertical levels
and McWilliams (1990) parameterisation
ORCA025 (Barnier
et
al.,
2006)
OPA9 (Madec, 2006): hy-
Modified TKE parameteri-
Relaxation to Levitus
LIM2
drostatic primitive equa-
sation
et al. (1998)/PHC at
Maqueda,
tion model with free sur-
timescale of 60 days
thermodynamic,
face; eddy permitting
for the upper two ver-
rheology (Hibler, 1979)
(Fichefet
and
1997):
Morales dynamic-
1/4◦ Mercator grid;
ETOPO2 with Bedrock
46 vertical levels
south of 72◦
viscous-plastic
GEBCO on the continental shelves; addi-
tical levels
tional smoothing, editing in key areas
TPAC (Meijers 2007)
et
al.,
MOM 3.1: primitive equa-
Isoneutral
tion model; eddy-resolving
form of
the
McWilliams
mixing
in
Gent
and
NCEP-R2 latent heat
(1990)
flux and precipitation
subgrid-scale mixing
Relaxation
rate, days
S and
using
timescale
30
No sea ice included
1/8◦ × 1/8◦ ; 24 vertical levels
TerrainBase
Table 2 Overview of WOCE hydrographic sections used for the model validation Section name (Cruise name)
Date
Colour in Fig. 1a
SR02 (ANT VIII/2)
12 Sep - 08 Oct 1989
dark blue
SR04b (ANT IX/2)
22 Nov - 15 Dec 1990
light blue
SR04d (ANT X/7)
18 Dec 1992 - 12 Jan 1993
dark green
S04a (ANT XIII/4)
20 Mar - 12 Apr 1996
red
SR04e (ANT XV/4)
01 Apr - 11 Apr 1998
light green
Table 3 Error statistics for ocean model - observation comparison following Taylor (2001). RMSEs for θ, and sea ice extent are given in ◦ C and m2 , respectively. Statistical parameter
θ
Model
S
Sea ice extent
SR02
SR04b
SR04d
S04a
SR04e
SR02
SR04b
SR04d
S04a
SR04e
OCCAM 1/12
0.71
0.79
0.77
0.82
0.63
0.83
0.87
0.84
0.89
0.83
0.75
OCCAM 1/4
0.74
0.81
0.74
0.83
0.82
0.79
0.81
0.77
0.82
0.84
0.88
ORCA025
0.84
0.93
0.88
0.89
0.80
0.95
0.93
0.92
0.88
0.90
0.98
TPAC
0.75
0.77
0.75
0.72
0.70
0.70
0.72
0.72
0.70
0.65
-
OCCAM 1/12
0.40
0.34
0.36
0.30
0.48
0.054
0.047
0.049
0.048
0.056
8.71 × 1011
RMSE =
OCCAM 1/4
0.39
0.33
0.39
0.30
0.33
0.061
0.041
0.057
0.036
0.056
9.76 × 1011
p1
ORCA025
0.44
0.41
0.422
0.43
0.54
0.046
0.049
0.045
0.056
0.053
7.12 × 1011
TPAC
0.50
0.50
0.50
0.48
0.59
0.068
0.073
0.061
0.082
0.089
-
OCCAM 1/12
0.39
0.33
0.35
0.30
0.47
0.053
0.047
0.049
0.047
0.056
7.39 × 1011
centred pattern RMSE =
OCCAM 1/4
0.39
0.33
0.39
0.30
0.33
0.060
0.041
0.057
0.036
0.055
6.96 × 1011
p1
ORCA025
0.32
0.28
0.30
0.32
0.39
0.046
0.049
0.044
0.056
0.052
6.07 × 1011
TPAC
0.35
0.34
0.35
0.37
0.42
0.068
0.073
0.060
0.082
0.088
-
OCCAM 1/12
1.02
1.05
1.12
0.98
1.03
1.17
1.65
1.25
1.88
0.89
0.92
OCCAM 1/4
1.07
1.10
1.17
1.03
1.00
1.17
1.31
1.23
1.39
0.89
0.45
ORCA025
1.11
1.33
1.27
1.32
1.17
1.38
1.76
1.40
1.98
1.13
1.50
TPAC
0.99
1.05
1.07
0.99
1.02
1.21
2.00
1.35
2.45
1.15
-
R= 1 1 N σx ×σy
Σ((x − x ¯)(y − y¯))
(x = obs.; y = model)
N
Σ(y − x)2
(x = obs.; y = model)
N
Σ((y − y¯) − (x − x ¯))2
(x = obs.; y = model)
normalised σ = σy /σx (σx =
p1
N
Σ(x − x ¯)2 )
(x = obs.; y = model)
776
777
List of Figures
1
a) CTD stations. Colour key to individual transects is given in Table 2. The
778
bathymetry shown is from Smith and Sandwell V10.1. The white dashed line
779
marks the western limit north of the Antarctic Peninsula for sea ice extent
780
calculations. b-d) model bathymetries: b) OCCAM 1/12 (dashed line: OCCAM
781
1/4 coastline in the Weddell Sea), c) ORCA025, d) TPAC.
782
2
Section SR04b, Nov-Dec 1990. Uppermost panel: CTD observations for θ (left)
783
and S (right). Below: differences between model and CTD for θ (left) and
784
S (right), from second panel to bottom panel: OCCAM 1/12, OCCAM 1/4,
785
ORCA025, and TPAC. The upper 500 m are expanded to allow for more detail.
786
The white lines on the x-axis indicate CTD station positions. The black lines
787
on the right hand side of the difference plots indicate the model depth levels.
788
3
790
4
35
θ - S diagram for section SR04b (Nov/Dec 1990). The dotted lines are contours of constant potential density anomaly (kg m−3 ).
789
34
36
Properties of WDW in OCCAM 1/12, OCCAM 1/4, ORCA025 and TPAC
791
(left to right) averaged along the SR04 section line. From top to bottom:
792
Percentage of area along section covered by WDW; average WDW θ; average
793
WDW S; average WDW depth. The grey shading indicates the variation due
794
to the different station spacing of the hydrographic sections (excluding the set
795
of stations of SR04e). The triangles indicate the values derived from the CTD
796
sections, the circles the values from the corresponding model sections. Filled
797
markers denote full sections, the open markers section SR04e which does not
798
extend over the entire eastern margin of the Weddell Sea. Note the extended
799
x-axis for ORCA025.
37
800
5
Fig. 4 but for WSDW.
38
801
6
Fig. 4 but for WSBW.
38
802
7
Sea ice extent in the Weddell Sea: a) climatological annual cycle (1988-2002),
803
and b) monthly means.
39
804
8
Climatological sea ice concentrations for the years 1988-2002 in February (upper
805
panel) and September (lower panel) for SSM/I observations, OCCAM 1/12,
806
OCCAM 1/4 and ORCA025 model results (TPAC does not include sea ice).
807
9
Taylor diagram for θ (+), S (⋄), and sea ice (∗) statistics. Dark blue markers
808
represent OCCAM 1/12, red markers OCCAM 1/4, grey markers ORCA025
809
and cyan markers TPAC.
810
10
from NCEP Reanalysis 2 for TPAC (from left to right). The NCEP Reanalysis
812
2 wind field is on the original 2◦ resolution grid instead of the TPAC model grid
813
because only snapshots of the forcing were stored during the model run. 11
40
Annual mean wind stress magnitude (N m−2 ) for OCCAM, ORCA025, and
811
814
39
40
a) Difference of sea surface height (SSH) between the tip of the Antarctic
815
Peninsula (63.45◦ S, 55◦ W) and the SSH minimum in the Weddell Sea. b)
816
Monthly mean baroclinic ACC transports through Drake Passage (referenced
817
to the bottom).
41
Fig. 1. a) CTD stations. Colour key to individual transects is given in Table 2. The bathymetry shown is from Smith and Sandwell V10.1. The white dashed line marks the western limit north of the Antarctic Peninsula for sea ice extent calculations. b-d) model bathymetries: b) OCCAM 1/12 (dashed line: OCCAM 1/4 coastline in the Weddell Sea), c) ORCA025, d) TPAC.
Fig. 2. Section SR04b, Nov-Dec 1990. Uppermost panel: CTD observations for θ (left) and S (right). Below: differences between model and CTD for θ (left) and S (right), from second panel to bottom panel: OCCAM 1/12, OCCAM 1/4, ORCA025, and TPAC. The upper 500 m are expanded to allow for more detail. The white lines on the x-axis indicate CTD station positions. The black lines on the right hand side of the difference plots indicate the model depth levels.
Fig. 3. θ - S diagram for section SR04b (Nov/Dec 1990). The dotted lines are contours of constant potential density anomaly (kg m−3 ).
Fig. 4. Properties of WDW in OCCAM 1/12, OCCAM 1/4, ORCA025 and TPAC (left to right) averaged along the SR04 section line. From top to bottom: Percentage of area along section covered by WDW; average WDW θ; average WDW S; average WDW depth. The grey shading indicates the variation due to the different station spacing of the hydrographic sections (excluding the set of stations of SR04e). The triangles indicate the values derived from the CTD sections, the circles the values from the corresponding model sections. Filled markers denote full sections, the open markers section SR04e which does not extend over the entire eastern margin of the Weddell Sea. Note the extended x-axis for ORCA025.
Fig. 5. Fig. 4 but for WSDW.
Fig. 6. Fig. 4 but for WSBW.
Fig. 7. Sea ice extent in the Weddell Sea: a) climatological annual cycle (1988-2002), and b) monthly means.
Fig. 8. Climatological sea ice concentrations for the years 1988-2002 in February (upper panel) and September (lower panel) for SSM/I observations, OCCAM 1/12, OCCAM 1/4 and ORCA025 model results (TPAC does not include sea ice).
Fig. 9. Taylor diagram for θ (+), S (⋄), and sea ice (∗) statistics. Dark blue markers represent OCCAM 1/12, red markers OCCAM 1/4, grey markers ORCA025 and cyan markers TPAC.
Fig. 10. Annual mean wind stress magnitude (N m−2 ) for OCCAM, ORCA025, and from NCEP Reanalysis 2 for TPAC (from left to right). The NCEP Reanalysis 2 wind field is on the original 2◦ resolution grid instead of the TPAC model grid because only snapshots of the forcing were stored during the model run.
Fig. 11. a) Difference of sea surface height (SSH) between the tip of the Antarctic Peninsula (63.45◦ S, 55◦ W) and the SSH minimum in the Weddell Sea. b) Monthly mean baroclinic ACC transports through Drake Passage (referenced to the bottom).