„5 
(a) ROAM-NBS vs, OBS 
CA 
u 60 
7 
i 
45a 
u ® 
s ® 
B m 60 
0 £ 
5 5 | 
Br > 
S 
10 258 
56 
A 
ia 
2 25 
'ongitude T°E7 
AT 
20 25 
'angitude 1°E7 
V. Maurer et al.: Evaluation of coupled and uncoupled simulations 
(b) ROAM-NBS vs. NEMO-NBS 
MAM 
E 
£ 
u? 
wu 
0.0 
© 
u 
u 
Q.1.8 
Lu} 
ww 
5 
0.2 
— be 
= 
x 
U 
us 
zz 
= 
CE] 
56 
+ 
A 
= 
5 
20 © 
— Be 
= 
z 
CN 
& 60 
3 
x 
CE: 
5 
S 
a) 
Fo.a 
n.1 8 
w 
0.1.0 
EC 
dw 
u 
—-.2 
f 
_;An 
2 Lu 
longitude [°E7 
20 25 
longitude [*°E* 
Figure 9. Differences in the mean sea ice concentration for ROAM-NBS vs. Copernicus observations (a) and ROAM-NBS vs. NEMO- 
NBS (b) for winter and spring for September 1981-November 2020. 
3.3.2 Ocean temperature 
in addition to the SST evaluation in Sect. 3.1, the evolution 
of temperature in different depths over time and mean tem- 
perature profiles are compared against observational profiles 
for the ROAM-NBS and NEMO-NBS simulations to eval- 
Jate the stratification of the Baltic. The chosen stations, re- 
sembling those of Meier (2007), cover the main basins of 
che Baltic Sea and are displayed in Fig. 10a. The observa- 
tional data are in-situ profile data from Copernicus. The sea- 
sonal mean is calculated over all time instances between Jan- 
uary 1979 and December 2020 where the observational data 
exist. 
At the monitoring stations Bornholm Deep (SMHIBY5) 
and Gotland Deep (SMHIBY15), both simulations tend to 
fit in-situ observations in the upper layers and underestimate 
;emperatures in the deeper layers over the entire evaluation 
period (see Fig. A5). The seasonal cycle is captured by both 
simulations. 
The mean temperature profiles in the Arkona Basin 
(Arkona and FINO2, Fig. 10b) generally match the obser- 
vational data for both simulations. In winter, a cold bias 
of 1-2 °C can be quantified in the Arkona Basin, which is 
slightly stronger for ROAM-NBS than for NEMO-NBS. At 
FINO2, the mean profile for summer also reveals a cold bias 
of about 1.0°C, here wich NEMO-NBS being cooler than 
ROAM-NBS. At station SMHIBY5, which is located in the 
Bornholm basin, both model runs coincide well with obser- 
vational data at the sea surface and in layers above a depth of 
50 m; differences between the coupled and uncoupled simu- 
lations are small (Fig. 10b). In the bottom layer, a cold bias 
can be observed. This cold bias at station SMHIBY5 is larger 
in summer than in winter, whereas the intermediate layer is 
accurately captured in summer. Similarly, within the Gotland 
Deep (SMHIBY15, Fig. 10), both model runs underestimate 
che mean temperature at depths below 100 m, mostly due to 
a weak salinity stratification (see Fig. 12). The overall cold 
Dias at the bottom of the Gotland Deep will also be shown in 
Sect. 4.1. As in the Arkona Basin, the coupled model ROAM- 
NBS has a larger cold bias in winter than the NEMO-NBS 
Geosci. Model Dev... 19. 543578, 2026 
stand-alone run. The intermediate layer is well captured at 
station SMHIBY15 during the summer months. 
The last two stations, SMHIBY31 and SMHISR5C4, lie 
in the Landsort Deep and Gulf of Bothnia, respectively. The 
shallow depth of the Landsort Deep in the simulations arises 
{rom Laplacian smoothing of the EMODNET bathymetry 
and the use of the nearest grid cell for station SMHIBY31, 
so that the deepest smoothed cell (370 m) does not align with 
the station’s grid point. For the available model depth, both 
simulations’ seasonal mean temperature results agree well 
with the observational data. In the Gulf of Bothnia in sum- 
mer, both simulations exhibit a warm bias near the surface, 
but a small bias below 40 m. In winter, the sign of the surface 
and near-surface bias is reversed compared to summer. How- 
ever, for the calculation of mean profiles, the model and ob- 
servational datasets were not masked for ice concentrations, 
which contributes to larger discrepancies in surface and near- 
surface temperatures at station SMHISR5C4 during the win- 
ter months. 
Overall, both simulations exhibit smaller temperature bi- 
ases in summer than in winter. At most stations, the simu- 
lated temperatures align more closely with observations near 
the surface than in the deeper layers for both seasons. In 
summer, the temperature profiles also display an intermedi- 
ate layer, although its magnitude is underestimated in both 
model runs. 
3.3.3 Salinity 
The model sea surface salinity is validated against an interpo- 
lated level-4 analysis of the sea surface salinity based on in- 
situ and satellite observations from Copernicus. The winter 
and summer differences between the simulated mean sea sur- 
face salinity of ROAM-NBS and observations for the period 
December 1993-November 2020 are shown in Fig. 11. These 
years were chosen as the observational dataset is only avail- 
able for this period (Table 2). The sea surface salinity tends to 
be underestimated at the Norwegian and German coasts and 
the Baltic Sea and tends to be overestimated at the passage 
from the Baltic Sea to the North Sea. However. both mod- 
https://doi.ore/10.5194/smd-19-543-2026