The mean seasonal model SST from 1985 to 2004 is compared with remotely sensed products. These include the National Oceanic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) product and the European Space Agency (ESA) Climate Change Initiative (CCI) product . The period 1985–2004 is chosen for two reasons. First, it allows for the CO5 hindcast to be spun up from rest in 1981. Secondly, it presents a common period with which to compare the POLCOMS hindcast that ends in 2005.

Figure  compares the CO5 control and POLCOMS hindcast SST bias against the AVHRR data. The largest bias in CO5 SST is the cold bias extending from eastern Iceland south-eastwards to the Faroe–Shetland Channel (FSC) and from the FSC north-westwards to the northern boundary of the domain. This SST bias is less apparent in summer as seen in Fig. c. The reduction in the bias might be caused by over-stratification in summer. The regions immediately surrounding the cold bias area appear to be warm biased in summer. This suggests the cold bias may be of a remote origin such as the boundary condition. Elsewhere off shelf there is a smaller cold bias in winter, spring and autumn. Along the Celtic shelf break, there is a slight warm bias. The model is probably underestimating the cold water surface signal associated with enhanced vertical mixing at the shelf break. In summer, off shelf southward of 50∘ N, CO5 appears to be too warm. On shelf, CO5 SST is slightly cold biased in most regions for most seasons.

However, there are some warm biases, particularly in summer. The Southern Bight, the Western Isles of Scotland and the western Irish Sea all have summertime warm biases. The English Channel also has a warm SST bias in autumn.

Figure e–h show the equivalent seasonal SST bias for the POLCOMS hindcast. POLCOMS also has a large cold bias from Iceland to the FSC and from the FSC to the northern boundary in winter, spring and to some extent in autumn. However, the POLCOMS SST cold bias appears to be more extensive. It also extends south-westwards from the FSC to roughly the Porcupine Bank. Near the western boundary, there is also a significant warm SST bias in POLCOMS north of 55∘ N in winter. Off shelf in summer, there is a large warm bias in POLCOMS across much of the domain. However, there is also a large summertime SST cold bias in the Norwegian Trench, the Skagerrak and the Kattegat.

In summary, the CO5 control hindcast appears to have a much smaller SST bias than the preceding POLCOMS hindcast. One particularly large bias in CO5 is the large cold bias in the northern part of the domain which is also present in POLCOMS. This bias is explored further with comparisons against temperature and salinity profiles, as well as climatology. CO5 does appear to be too warm off shelf in summer but much less so than POLCOMS. On shelf, CO5 is generally slightly cold biased, whereas POLCOMS alternates from a large wintertime cold bias to a large summertime warm bias. POLCOMS is too cold in the Norwegian Trench during summer, while CO5 appears to do reasonably well there.

The mean sea surface salinity (SSS) of CO5 for 1985–2004 and the POLCOMS hindcast are compared against the World Ocean Atlas 2013 (WOA13) global climatology , the KLIWAS North Sea climatology (KLIWAS) and EN4 profile data in Fig. . A similar pattern in negative SSS bias as SST bias from Iceland to the FSC and to the northern boundary is present in CO5. With the exception of this northern region, CO5 off shelf is in reasonably good agreement with both the climatology and the mean profiles. However, POLCOMS appears too fresh off shelf except along the western French coast, indicating an offset in surface salinity between CO5 and POLCOMS. On the shelf, CO5 is in general slightly too saline. In particular, the Irish Sea is saline biased in CO5 and indicates the E-HYPE freshwater flux may be too small there.

Both CO5 and POLCOMS have large SSS biases compared to the climatologies and profiles in the Norwegian Trench. POLCOMS is too saline in the Norwegian Trench, while the salinity bias in CO5 is a dipole: near the Norwegian coast it is too fresh and near the western limb of the Norwegian Trench it is too saline. In POLCOMS, not only are there fewer vertical levels but the vertical resolution near the surface is proportional to the ocean depth as in CO4. Consequently, compared to CO5, the surface resolution in POLCOMS in the Norwegian Trench is much reduced. The surface resolution in POLCOMS over the Norwegian Trench is typically 4 to 5 m compared to the uniform 1 m resolution for CO5. This may account for the much more saline SSS in POLCOMS there. The Baltic boundary in POLCOMS is also more similar to CO4 than CO5 when using climatological river points to represent the exchange with the Baltic. The sensitivity experiments below investigate the effect of both these changes within the NEMO framework. With regards to the dipole in CO5, the resolution at 7 km is not sufficient to resolve the intense mixing processes in the trench where northward-flowing freshwater of Baltic origin along the Norwegian coast mixes laterally with adjacent incoming southward-flowing saline Atlantic water. It is anticipated that with increased horizontal resolution, better representation of eddy-induced mixing may reduce the dipole there.

POLCOMS and CO5 have biases of opposite signs in the German Bight; CO5 is too fresh and POLCOMS is too saline. POLCOMS uses the climatological rivers as in CO4 in contrast to the E-HYPE rivers used in CO5. Thus, the sensitivity experiments S30_1, S30_2 and S30_3 that compare the different river sources should help to understand the difference in this bias. POLCOMS also appears to be too fresh in the Southern Bight, and this may be contributing to the saline bias in the German Bight. POLCOMS may not be advecting the Rhine outflow to the east close enough to the coast. CO5 in contrast appears to be too fresh in the vicinity of the Rhine outflow.

To assess how CO5 and POLCOMS behave throughout the water column off shelf, they are compared against WOA13 data. Figure  displays both zonal transects and depth level temperature biases for 1985–2004 compared to WOA13. Both CO5 and POLCOMS temperature biases are included in Fig. . Figure  is the equivalent salinity plot. As the mean is for the entire period, seasonal biases such as in the SST plots of Fig.  are not discernible. The location of the transects are chosen to intersect regions of particularly large bias. Note that these comparisons use the CMEMS POLCOMS dataset, which was interpolated onto standard depth levels from the native POLCOMS grid which uses 40 s levels in the vertical . The interpolated POLCOMS data are particularly coarse at depth which is reflected in the step-like nature of the POLCOMS bias plots at depth. This accounts for some of the differences seen towards the bottom of profiles in Fig. .

In Fig. , the first two columns are zonal transects of difference in the mean temperature from the WOA13 climatology over the period 1985–2005. The first column is for CO5 and the second POLCOMS. The geographical extent of the biases highlighted in the transects is shown for four depths in the last two columns of Fig. . Both CO5 and POLCOMS have a cold water bias centred around roughly 1000 m that originates near the southern boundary away from the relaxation zone. A warm temperature bias surrounds the cold temperature bias away from the coast. A similar pattern in salinity bias is shown in Fig. . It appears the models are diffusing both horizontally and vertically the warm and saline waters of Mediterranean origin entering the domain from the southern boundary. The extra diffusion in the relaxation zone and the relatively coarse vertical resolution of about 100 m at a depth of 1000 m may be contributing to the loss in identity of the Mediterranean waters. The anomaly is also present in the Bay of Biscay but is much reduced in CO5 further north.

In the seasonal SST anomalies, a large cold bias was shown in both CO5 and POLCOMS in winter. This cold bias is also present with respect to the WOA13 climatology. In CO5 and POLCOMS, it extends down to around 500 m. There is a warm bias in CO5 along the sea bed of the Iceland–Faroe ridge at around 500 m and at a similar depth on the Shetland side of the FSC. The vertical resolution of POLCOMS is quite coarse at this depth. However, it suggests that at depths greater than 500 m POLCOMS is warm biased in the FSC and Norwegian Sea, and CO5 appears to be close to the climatology below 500 m. There is a similar pattern in the salinity bias with both CO5 and POLCOMS relatively fresh near the surface in this region. On the other hand, POLCOMS appears to be slightly fresher than WOA13 off shelf right through depth for most of the domain. Off shelf, away from Biscay and the northern boundary, CO5 salinity is quite similar to WOA13.

The KLIWAS climatology for the NWS in combination with EN4 provides an alternative to WOA13 for evaluation of the models on the shelf itself. Figures  and compare CO5 with both the KLIWAS climatology and the EN4 data over the period 1985–2004. A comparison of POLCOMS against KLIWAS is also included as a reference. Figure  focuses on the summer months when there is seasonal thermal stratification, while Fig.  is the salinity mean for all seasons. Including all seasons allows for a larger number of in situ profiles to compare against. In addition to biases at depth levels of 10, 30 and 40 m, transects are taken through areas of significant bias to give an overview of the vertical structure in the model bias.

Generally, the structure of the temperature bias between CO5 and EN4 is in reasonable agreement with the structure of the bias between CO5 and the KLIWAS climatology. In the seasonally stratified areas of the North Sea, CO5 compares favourably near the surface compared to POLCOMS. POLCOMS there is significantly warm biased. Immediately below the thermocline, both CO5 and POLCOMS are cold biased with the cold bias in POLCOMS being somewhat larger than CO5. In CO5, the cold bias does not extend to the bed and in fact reverses sign to be warm biased near the bed, whilst in POLCOMS the cold bias reduces towards the bed with only a small bias remaining at the sea floor. The light attenuation schemes in CO5 and POLCOMS are quite different and may partially explain why POLCOMS is more warm biased at the surface and more cold biased at depth. The light scheme used in POLCOMS (PDWL) is also implemented in CO4 and is included in the sensitivity experiments to enable its impact to be assessed.

The CO5 salinity bias against EN4 is also broadly in agreement with the bias against the KLIWAS climatology. As in the surface plots of Fig. , over most of the North Sea, CO5 is slightly too saline through depth. Along the coasts of Holland, Germany and Denmark, CO5 is clearly too fresh, suggesting too much riverine input as discussed earlier. Away from the coasts, POLCOMS is in fairly good agreement with EN4 and KLIWAS while just slightly fresher at depth. The transects in Fig.  are taken to go through the Norwegian Trench and the Rhine plume. CO5 is shown to be roughly 0.5 too fresh above 20 m in the Norwegian Trench near the coast of Norway, while below 20 m CO5 is slightly too saline. The warmer and more saline water from the Atlantic appears to make CO5 too saline along the rim of the Norwegian Trench. In contrast, POLCOMS is shown to be typically greater than 1.1 too saline above 20 m in the Norwegian Trench, while below 40 m, POLCOMS switches from the large saline bias to a significantly fresh bias. It appears that CO5 is representing the haline stratification in the Norwegian Trench with greater fidelity than POLCOMS. Both the vertical resolution and the Baltic boundary condition may play some role in this and are included in the sensitivity experiments that follow.

The effect of the changes of surface vertical resolution between CO4 and CO5 is shown in Fig. . Sensitivity experiment S30_1 is exactly the same as the control (CNTL) except it uses 33 SH vertical levels instead of 51 SF levels. Although there are some small changes in summertime stratification off the shelf, the most dramatic change concerns the surface salinity in the Norwegian Trench. Figure  is the difference in salinity at 5 m between the control experiment CNTL (SF51) and sensitivity experiment S30_1 (SH33). The extra vertical resolution in the control run results in less diffusion of the surface fresh layer with depth. The POLCOMS hindcast also has much less vertical resolution at the surface than CO5, which may be one factor underlying its saline bias in the surface waters of the Norwegian Trench.

Both the river forcing and Baltic boundary condition are changed in CO5 from climatological inputs to E-HYPE riverine inputs and IoW Baltic boundary data. The 30-year sensitivity experiments S30_2 and S30_1 are compared in Fig. . S30_1 is a 33-level version of the CO5 control but with exactly the same E-HYPE rivers and IoW Baltic boundary. S30_2 is exactly the same as S30_1 except that it uses the older climatological inputs for rivers and Baltic boundary as used in CO4. Figure a shows the surface salinity bias against EN4 data for S30_1. Figure b is the same but for S30_2 and shows a large reduction in the freshwater bias in the German Bight. Figure c compares experiment S30_3 with S30_1 to show differences created by the change in rivers alone. For most of continental Europe, the E-HYPE rivers clearly have a greater discharge as shown in Fig. . The difference is pronounced in the German Bight and along the Norwegian coast and is likely contributing to the fresh bias in CO5 compared to EN4 data there. Another possible source of salinity bias could be incorrect transports and representation of the North Sea circulation. Figure shows a background field of mean transport in the North Sea and also the mean transport across selected NOOS transects into and out of the North Sea. The calculated transports are similar to reported values from observational estimates as shown in Table .

Around parts of the coast of the UK, the E-HYPE river discharge in some regions is actually slightly less than the climatology or only slightly greater in others. This is also reflected in the difference of salinity shown in Fig. c. Combining the correlation between areas of larger E-HYPE river discharge than climatological rivers and the larger salinity biases in CO5 with what appears reasonable transports in CO5, it seems likely that the E-HYPE rivers are the first-order source of the salinity biases observed in CO5.

Figure d compares experiment S30_4 with S30_1 to show the impact resulting from the different Baltic boundaries. The IoW boundary results in a slightly more saline SSS over in the Norwegian Trench. The effect of the Baltic boundary condition is much smaller than the freshening due to the E-HYPE rivers resulting in an overall freshening compared to the climatologies.

The summertime biases in temperature were shown to be significantly different between POLCOMS and CO5. Sensitivity experiment S30_5 explores replacing the light attenuation scheme in CO5 with the PDWL scheme. Figure  compares the control experiment and experiment S30_5 over the summer. Figure a compares each run on shelf in regions of seasonal stratification. Using the PDWL light scheme has three effects: it increases the warm surface temperature bias, it increases the mid-depth cold bias and it reduces the bias from 40 m to the bed. The partition of solar radiation into a penetrating part and a non-penetrating part is dealt with differently in each scheme and influences the degree of bias at the surface. In the PDWL scheme, all of the non-penetrating part is added to the surface layer, while in the RGB scheme there is still a slight penetration of the quickly attenuating light. The cold bias in both models indicates that the depth of the thermocline is too shallow, which could be either be due to the light not penetrating far enough in both schemes or insufficient vertical mixing. At depth, the PDWL scheme results in less heat being mixed down, resulting in a better agreement with the bed temperature as the RGB scheme is biased warm there. However, in mixed areas on shelf, both models appear to be too warm, which may indicate a bias in the surface flux forcing.

Figure b compares each scheme against the mean of the profiles off shelf south of 60∘ N. Figure b does not show depths below 140 m as the differences due to light below this depth are negligible. The large cold bias in the upper layers of the ocean north of 60∘ N biases the whole field cold. Thus, to obtain a better representation of the effect of the light scheme in the absence of large underlying biases, we restrict the mean to south of 60∘ N. As the light penetrates more deeply off shelf in the PDWL light scheme, the warm bias at the surface is less than the RGB scheme and the cold bias below 20 m is also reduced. Both schemes are similarly cold biased below 60 m where the direct effect of light penetration is small.

The 5-year sensitivity experiments show that the most significant differences between CO4 and CO5 are related to the lateral boundary conditions. A bug in NEMO v3.2 prevented the application of the inverse barometer effect on the open ocean lateral boundaries. Thus, two sensitivity experiments with NEMO v3.4 were conducted: S5_1 with this bug deliberately included and S5_2 without. An additional experiment, S5_3, is an equivalent experiment with exactly the same forcing but with NEMO v3.2 as the base model.

The resulting 5-year mean SST difference between S5_1 and S5_2 is shown in Fig. a. Clearly the switching on or off of the inverse barometer on the open boundary has a large impact on the SST in the north of the domain. The difference between the SST RMSE of S5_1 and S5_2 shown in Fig. d. The much larger RMSE of S5_1 indicates that the inclusion of the inverse barometer effect on the boundary considerably reduces the SST skill there. However, if the inverse barometer is not included on the boundaries, anomalous northward-flowing boundary jet currents result. Figure b and e are the equivalent mean and RMSE differences between S5_2 and S5_3, which are very similar to that of Fig. a and d. The difference (not shown) between S5_1 and S5_3 is much smaller. Thus, a large component of the difference between CO4 and CO5 is the difference in the application of the inverse barometer effect on the lateral boundary.

Another significant difference was the open ocean source data interpolated onto the open boundaries of CO5 and CO4. The 30-year sensitivity experiments of CO5 used data from the 1/4∘ global ocean domain (ORCA025). However, CO4 was forced using a 1/12∘ model of the North Atlantic (NATL12). In operational implementation of CO5, the higher resolution NATL12 model is also used to derive open boundaries. Sensitivity experiment S5_4 is exactly the same as S5_1 but replaces the ORCA025-derived boundaries with boundaries derived from the NATL12 model. The mean and RMSE SST differences between S5_1 and S5_4 are shown in Fig. c and f. The NATL12 data result in a warmer SST from Iceland to the Faroe Islands and a reduced RMSE compared to the ORCA025 data.

The GLOSEA5 surface data are compared against WOA13 in Fig. . Compared to WOA13, GLOSEA5 is anomalously cold and fresh along the CO5 boundary of 65∘ N. This suggests that a significant component of the northern bias in CO5 originates in the global model that provides its northern boundary condition.

Another important difference between CO4 and CO5 is the surface fluxes. In operational implementation, the surface fluxes are taken from the Met Office NWP model. The experiments in were also forced with NWP directly prescribed fluxes. However, comparable NWP model data were not available from before 2006, and thus the longer runs, as in the CO5 control, use ERAI surface fluxes. Furthermore, the NWP fluxes are directly prescribed in contrast to the CORE bulk formulae used with ERAI. A Haney correction must also be applied when using direct fluxes with a prescribed reference SST as used by the NWP model itself.

Sensitivity experiment S5_5 is exactly as S5_1 but with ERAI-derived fluxes instead of NWP fluxes. Figure a and b compare the mean SST and SSS between S5_5 and S5_1. The SST is almost uniformly warmer with NWP fluxes than ERAI-derived fluxes, particularly in the Bay of Biscay, around the coast of the UK and into the Skagerrak and southern Norwegian Trench.

However, it should also be noted that because direct fluxes use the Haney correction, the resulting model SST is indirectly relaxed to the prescribed SST in a hindcast simulation. Furthermore, in NEMO version 3.2, the surface stress from direct fluxes was based on absolute wind velocity rather than wind velocity relative to the moving ocean surface. This has important localised effects in the vicinity of persistently strong surface currents, such as the Skagerrak. This sensitivity of the model to relative winds versus absolute winds using ERAI-derived forcing is also investigated. Figure  is the difference in the mean for 1 year of surface stress, salinity, current and temperature between using the relative wind velocity to the ocean surface and the absolute wind velocity.

Furthermore, the details of the fluxes near coastlines, and particularly the wind stress in the Skagerrak and southern Norwegian Trench, are quite different between the lower-resolution ERAI and higher-resolution NWP fluxes. The differing resolution of the surface forcing and the use of absolute instead of relative wind stress is thus likely to play a role in the different sensitivity experiments.

In Fig. b, it is shown that the experiment with ERAI-derived forcing is slightly more saline on shelf but significantly fresher in the Skagerrak and the Norwegian Trench mirroring the SST differences there. The difference in the shear stress modifies both the transport of the surface fresh layer out of the Skagerrak and the transport of relatively saline water from the North Sea into the Skagerrak. The difference in relative and absolute winds is significant also along the shelf break from the Shetland Islands northwards. In this case, the effect of using the absolute wind velocity is to reduce the transport of North Atlantic water northwards, which results in locally lower mean SST. With respect to the difference between ERAI-derived forcing and NWP-forced experiments, the difference in the SST in this local region is reduced. The reduction in the mean difference of SST is due to the countervailing effects of general domain-wide cold bias between ERAI-derived forcing and NWP fluxes, and the local cooling due to using absolute winds with the direct NWP fluxes.