**4. Results**

This Section shows the main results from the comparison of the different SAMOA model tests (SAM\_INI, SAM\_ADV and SAM\_H3D; please refer to Table 1 and Figure 3b) together with their parent solution: IBI\_PHY. Results are split in two Subsections: Section 4.1 gives information on the SAM\_INI, SAM\_ADV (and IBI\_PHY) performances for the June 2019 period, at the 9 original SAMOA-1 domains (green dots at Figure 2): Barcelona (BCN), Tarragona (TAR), Bilbao (BIL), Almería (ALM), Ferrol (FER), La Gomera (GOM), Gran Canaria (GCA), Santa Cruz de la Palma (PAL), Tenerife (TEN); whereas Section 4.2. assesses SAM\_ADV, SAM\_H3D and IBI\_PHY from October 2019 to October 2020 at 5 SAMOA domains (i.e., BCN, TAR, BIL, ALM and GCA).

#### *4.1. Impacts in SAMOA Solutions Related to Changes in the OBC Treatment and the Use of a Bulk Formula to Deal with Atmospheric Forcing (SAM\_ADV vs. SAM\_INI)*

This first SAM\_INI configuration was carried out based on the SAMOA-1 model set-up after the update of the HARMONIE atmospheric forcing (see Section 3.1). The upgraded SAMOA set-up (SAM\_ADV test) related to changes in the OBC treatment and the use of a Bulk formula is further described in Section 3.2.

The error metrics from both SAMOA tests and the IBI\_PHY solution are summarized in Table 3. In general, it is seen that SAM\_ADV improves considerably the SAM\_INI Sea Surface Temperature (SST):


At a more detailed level, Figure 4 presents the SST at Gran Canaria coastal buoy. Despite that the three models are able to capture the intradaily cycles, SAM\_ADV (green line) and IBI\_PHY (black line) show similar trends. Additionally, for the first part of the month, SAM\_ADV shows more agreemen<sup>t</sup> with the observations than SAM\_INI (blue line). However, also note that at the second half of the month, there was a sudden rise in temperatures that only SAM\_INI captured (though overestimating it).

**Figure 4.** Observed and modelled Sea Surface Temperature (SST) at Gran Canaria PdE EXT mooring buoy (see location in Table 2) in June 2019. Units in Celsius degrees. The dotted red line represents the observations; the solid black line is the parent solution (CMEMS IBI\_PHY); the solid blue line is the initial SAMOA set-up (SAM\_INI); and the solid green line represents the SAMOA solution obtained with the advanced set-up (SAM\_ADV).

Regarding mean sea-level, SAM\_INI and SAM\_ADV show similar performance, improving considerably IBI\_PHY, especially at meso-tidal (GOM, GCA and PAL domains) and macro-tidal environments (BIL, FER, VIG). The three models show very high correlations (close to 0.96); but the RMS is lower (close to 0.05 m) for the SAMOA set-ups, whereas IBI\_PHY reaches 0.14 m. The same pattern is also found for the COE: both SAMOA have values close to 0.77, but IBI\_PHY does not surpass 0.66.

Sea Surface Salinity (SSS) remains with similar metrics between SAMOAs, not improving the IBI\_PHY performance. RMS are identical and the COE is negative at all three. Hence, SAMOA does not add new information on this variable.

Surface currents also exhibit similar metrics amongs<sup>t</sup> the different models. Note though, that at a qualitative assessment level, SAM\_ADV has shown more consistency with the IBI\_PHY patterns than SAM\_INI. These phenomena especially happened at the SAMOA boundaries, where the presence of spurious rim-currents was partially alleviated. The consistency of the SAM\_ADV OBC scheme was confirmed with the SAM\_H3D test (next Subsection).

#### *4.2. Impact in the SAMOA Solutions Related to Changes in the Physics and in the Temporal Frequency of the Imposed Data along Boundaries (SAM\_H3D vs. SAM\_ADV)*

As pointed in the previous section, SAM\_ADV seemed to enhance coherency between the SAMOA systems and their parent solutions (IBI\_PHY). However, differences were still high in some cases, affecting the whole domain and specially along the boundaries. The availability of new hourly-3D CMEMS IBI-MFC products, joint with the upgrade of some physical parametrizations, allowed building the SAM\_H3D set-up at 5 systems that were compared with the SAM\_ADV and the IBI\_PHY parent solution. Their results are summarized in Tables 4 and 5 and Figures 5–7.

**Figure 5.** Observed and modelled Sea Surface Temperature (SST) at 3 coastal buoys. (**a**): Tarragona coast. (**b**): Bilbao coast. (**c**): Gran Canaria coast. Period shown: October 2019–October 2020. Units in Celsius degrees. The dotted red line represents the observations; the solid black line is the CMEMS IBI-MFC; the solid blue line is the current operational SAMOA set-up (SAM\_ADV); the solid green line is the proposed set-up nested into hourly 3D IBI\_PHY data (SAM\_H3D). Further information about the buoy stations (location and mooring depth) can be found in Table 2.

**Figure 6.** HF Radar vs. model solutions (SAM\_H3D and SAM\_ADV experiments, together with the Copernicus IBI parent solution) at Tarragona during March 2020. Monthly-averaged surface current field modelled and observed, depicted at subpanels (**a**) (IBI\_PHY)-(**d**) (SAM\_ADV)-(**g**) (SAM\_H3D), joint with (**c**) (HF Radar), respectively. However, the error metrics subpanels (**b**,**e**,**f**,**h**,**i**) are computed from hourly-averaged fields. Model-observation monthly biases for IBI\_PHY, SAM\_ADV and SAM\_H3D are depicted in the central column (subpanels **b**,**e**,**h**); SAM\_ADV and SAM\_H3D model correlation fields with HF Radar observations are shown in subpanels (**f**,**i**). Units in m/s. The red rectangle in subpanels (**<sup>a</sup>**–**<sup>c</sup>**) the SAMOA coastal domain. Red dots at subpanel (**c**) represent locations where time series metrics are computed (summarized in Table 5).

**Figure 7.** HF Radar vs. model solutions (SAM\_H3D and SAM\_ADV experiments, together with the Copernicus IBI\_PHY parent solution) at Gran Canaria during September 2020. Monthly-averaged surface current field modelled and observed, depicted at subpanels (**a**) (IBI\_PHY)-(**d**) (SAM\_ADV)- (**g**) (SAM\_H3D), joint with (**c**) (HF Radar), respectively. However, the error metrics subpanels (**b**,**e**,**f**,**h**,**i**) are computed from hourly-averaged fields. Model-observation monthly biases for IBI\_PHY, SAM\_ADV and SAM\_H3D are depicted in the central column (subpanels **b**,**e**,**h**); SAM\_ADV and SAM\_H3D model correlation fields with HF Radar observations are shown in subpanels (**f**,**i**). Units in m/s. The red rectangle in subpanels (**<sup>a</sup>**–**<sup>c</sup>**) the SAMOA coastal domain. Red dots at subpanel (**c**) represent locations where time series metrics are computed (summarized in Table 5).

Regarding SST (Table 4), lower biases are found in the SAM\_H3D (deep water TAR buoy, ALM and GCA) and IBI\_PHY (BCN, BIL and coastal TAR buoy). Correlation is close to 0.98 at IBI\_PHY and SAM\_H3D, slightly higher than for the SAM\_ADV case (0.96); pointing the correct capture of the intraday variability. IBI\_PHY presents lower RMS (0.65 ◦C) than SAM\_H3D (0.72 ◦C) and SAM\_ADV (0.97 ◦C). The COE shows good agreemen<sup>t</sup> between SAM\_H3D (0.83) and IBI\_PHY (0.84), whilst SAM\_ADV is significantly lower (0.76).

Gran Canaria (GCA) is the system in which SAM\_H3D clearly outperforms IBI\_PHY and SAM\_ADV: (i) the bias is lower ( −0.02 ◦C) than in IBI\_PHY ( −0.15 ◦C) and SAM\_ADV (−0.31 ◦C) cases; (ii) SAM\_H3D RMS (0.4 ◦C) is in the same line than IBI\_PHY (0.38 ◦C), and lower than SAM\_ADV (0.59 ◦C).

Error metrics are consistent with the time-series plots (Figure 5), as the model performance between IBI\_PHY (black line) and SAM\_H3D (green line) are fairly similar at the coastal buoys. Both systems capture the main trends and seasonal changes. However, SAM\_ADV (blue line) presented a consistent negative/positive bias in the Winter-Spring/Summer, most probably due to heat-fluxes mismatches at the model domain. Note also that the error metrics between SAM\_ADV and SAM\_H3D present more agreemen<sup>t</sup> at deep waters (closer to the SAMOA domain boundary) than at the coastal locations (far away from the boundaries, heavily influenced by coastal circulation processes and a proper advection scheme).

Surface salinity is improved in Tarragona with the SAMOA solution, though SAM\_ADV shows slightly better metrics ( −0.13 PSU bias, 0.54 correlation) than SAM\_H3D ( −0.14 PSU, 0.49) and IBI\_PHY ( −0.18 PSU, 0.43). At Bilbao and Almería, the error metrics are quite similar for the three models, although SAM\_ADV and IBI\_PHY moderately outperform SAM\_H3D.

As in Section 4.1, SAMOA Sea Level solution outperforms notably IBI\_PHY at mesotidal and macro-tidal environments: the RMS at these two places is significantly lower with SAMOA (0.13 m Bilbao, 0.08 m Gran Canaria) than IBI\_PHY (0.27 m Bilbao, 0.12 m Gran Canaria). COE presents important differences (close to 0.9 in SAMOA, whilst IBI\_PHY has 0.74 in Bilbao and 0.8 at Gran Canaria).

Sea Level at Barcelona and Tarragona exhibit similar behaviour on the three models. However, at these two domains, IBI\_PHY solution exhibits a higher COE (around 0.6) and the long-term metrics underperform both at SAM\_ADV and SAM\_H3D.

Apart from SST, the most prominent differences at the SAM\_H3D are found in the sea surface currents, both at current speed and direction (Table 4). Current direction metrics from SAMOA outperform at deep-water buoys: (i) SAM\_H3D ( −1.4◦) bias is lower than IBI\_PHY ( −4.6◦); (ii) correlation and RMS are also better at SAM\_H3D (0.3, 74.2◦) than at IBI\_PHY (0.27, 84.2◦).

Regarding surface current speed at Tarragona deep-water buoy, significant improvement in the correlation can be found (0.35 (SAM\_ADV) vs. 0.36 (SAM\_H3D) vs. 0.25 (IBI\_PHY)). RMS is similar for all three datasets (0.16 vs. 0.18 vs. 0.17, respectively); but the bias is lower in IBI\_PHY (0.01 m/s), and higher in SAM\_H3D (0.05 m/s) than in SAM\_ADV (0.03 m/s).

Tarragona and Gran Canaria are the two SAMOA that have coverage from Radar HF. Hence, special emphasis would be given. Error metrics have been computed at five specific points (Table 5), representative of the area (see their position at subpanels (c) in Figures 6 and 7).

Figure 6 shows the monthly-averaged HF-radar measurements, model results and error metrics in Tarragona during March 2020. In that month, there were four clustered NW inland wind-jets in the first week of the month, reaching hourly wind-speeds up to 20 m/s. From 10th March until the end of the month, the atmospheric conditions were fairly stable, though: wind speed had a mean average close to 3–4 m/s. Note however, that at Figure 6c, even at a monthly-averaged scale, these inland wind-jets are the most remarkable signature.

Both SAMOA systems reproduce the origin and the propagation of wind-jet. But SAM\_H3D is the solution that shows more resemblance with the surface HF radar (Figure 6g–i), exhibiting correlations close to 0.5 at the wind-jet influence area. IBI\_PHY does not reproduce the wind-jet and it tends to overestimate the Northern Current and General circulation, remaining some influence from circulation features associated to extreme storm events occurred in previous months (i.e., Mediterranean Storm Gloria).

At the Gran Canaria case (Figure 7) there were persistent, but moderate, NE wind speed (lower than 8 m/s, with monthly-averaged values close to 3 m/s). The monthlyaveraged HF radar exhibits consistency (Figure 7c) with these wind field conditions. Surface circulation is mainly SW, exhibiting higher values at the South (close to 0.30 m/s) than in the North (around 0.1 m/s). IBI\_PHY (Figure 7a) exhibits a clockwise gyre near the Gran Canaria harbour. This eddy is also reproduced in SAM\_ADV set-up (Figure 7d–f), but it is not in SAM\_H3D (Figure 7g–i).

SAM\_ADV overestimates currents at the Southern part of the domain (Figure 7d). Biases at the SAM\_H3D boundary are lower than SAM\_ADV and even IBI\_PHY. Additionally, SAM\_H3D exhibit higher correlation at the Southern part of the domain (close to 0.7). Note that in SAM\_ADV, the correlation is negative at the central part of the Eastern boundary. These findings are consistent regardless of the measurement devices, as the analysis performed with RadarHF and deep-water buoy observational data denote similar patterns in terms of error metrics.
