**5. Discussion**

In the previous section, the main results were shown from the different SAMOA model set-up tests (SAM\_INI, SAM\_ADV, SAM\_H3D) performed at different coastal and port domains. All the SAMOA simulations were validated with observational data sources for sea level, surface currents, SST and SSS, and compared with its parent solution (IBI\_PHY).

The main aim of these SAMOA sensitivity tests is to assess the impacts that potential upgrades in the SAMOA set-up have on its solution at different domains. Once the quality control is performed (i.e., by comparing SAMOA solutions with IBI\_PHY solutions, joint with the available observations at each coast/port domain), it can be verified whether the proposed set-up may result in a new operational SAMOA release.

Prior to deciding a new operational SAMOA release, it would be needed to verify if the new solution improves/degrades the control one (and at what level). Furthermore, it is not enough to test the model set-up upgrade in a single SAMOA domain; it is needed to be assessed how the positive impacts extend across the different SAMOA port systems.

Table 6 summarizes the results of the SAM\_INI vs. SAM\_ADV and the IBI\_PHY parent solution (see SAMOA model set-up descriptions in Section 3.2. and the main validation results in the Section 4.1). In general terms, SAM\_ADV improves sea level significantly in respect to IBI\_PHY, but remains similar to the SAMOA control run (SAM\_INI). The SST simulated with the SAM\_ADV set-up improves the SAM\_INI one, but the solution is worse than the one from the Copernicus IBI\_PHY product. Surface salinity remains similar in all the model solutions, with SAM\_ADV outperforming in the Bilbao domain. With respect to surface current speed, a slightly better performance of IBI\_PHY is seen in the three cases where observations are available, but both SAMOA experiments outperform the surface current direction; the three solutions may be considered as similar for currents.

The proposed SAMOA SAM\_ADV set-up improves notably the Sea Level respect the IBI-MFC solution, especially at meso and macro-tidal environments. The higher spatial resolution of SAMOA systems allows to better capture of the harbour geometry and coastal surroundings, being a plausible explanation of such improvement in terms of sea level. The use of the new COARE bulk formula introduced in the SAM\_ADV test improves (significantly) the sea surface temperature metrics with respect to the SAM\_INI control one in 5 (4) out of the 7 cases analysed. This significant improvement led SAM\_ADV SAMOA set up to be transitioned into operations in October 2019, thus becoming the current (end of 2021) operational set-up.

**Table 6.** Summary of the impacts of the SAM\_ADV set-up. Performance of SAM\_ADV versus the control (SAM\_INI) one and the parent solution (IBI\_PHY), comparing in those domains where observations are available (for sea level, SST, SSS and surface currents). Legend: (SAM\_ADV relative performance with respect to SAM\_INI set-up/SAM\_ADV relative performance with respect to IBI\_PHY). Acronyms to define the relative performance of the proposed set-up: (I \*: relevant improvement/I: improvement/S: similar performance/W: mild worsening).


Improvements in operational systems tend to be incremental. The proposed SAM\_H3D configuration enhances the SAM\_ADV one in Surface Currents and SST; whereas Sea Level metrics remain similar (see summary of relative performances in Table 7). SAM\_H3D also outperforms IBI\_PHY at the following issues: (i) Sea Level in meso-tidal and macro-tidal environments; (ii) Surface current speed correlation and surface current direction; (iii) SST at intermediate and shallow waters (i.e., coastal buoys); (iv) Surface Salinity at Tarragona. The main reasons for this behaviour are discussed below.

**Table 7.** Summary of the impacts of the SAM\_H3D proposed set-up. Performance of SAM\_H3D versus the control (SAM\_ADV) one and the parent solution (IBI\_PHY), comparing in those domains where observations are available (for sea level, SST, SSS and surface currents). Legend: (SAM\_H3D relative performance with respect to SAM\_ADV set-up/SAM\_H3D relative performance with respect to IBI\_PHY). Acronyms to define the relative performance of the proposed set-up: (I \*: relevant improvement/I: improvement/S: similar performance/W: mild worsening).


First, better surface current correlations at SAM\_H3D (respect to SAM\_ADV) are related to three factors: (i) tidal-induced currents are correctly nested with the parent solution; (ii) wind-induced currents are better modelled with the COARE 3.6 parametrization; and (iii) SAMOA solutions exhibit more spatial heterogeneity than IBI\_PHY. This last issue (heterogeneity) can be partly explained due to SAMOA higher spatial resolution. For instance, the model grid is able to reproduce sharp bathymetric pressure gradients; joint with the atmospheric forcing data that features local orographic constraints.

In this sense, atmospheric forcings are an important cause of the mismatch. SAMOA uses HARMONIE that has a systematically higher wind speed bias than ECMWF: around 1.5 m/s yearly-averaged difference along the coastal and deep waters of Iberian Peninsula, Balearic and Canary Islands. At the inland stations, biases between the two forcings are similar, mostly due to more stations to be assimilated. Moreover, HARMONIE overestimation can increase up to 7 m/s during extreme events, such as the recent Storm Gloria (January 2020) (Sotillo et al., 2021a [47]); whilst ECMWF tend to lower overestimations

(close to 3 m/s). Such HARMONIE overestimation is one of the main reasons for the higher bias and RMS at some stations in the SAMOA surface currents, with respect to IBI\_PHY.

Wind direction, though, provides a trade-off: whereas ECMWF biases can reach close to 30◦ at coastal locations, HARMONIE remains more accurate (biases around 10–20◦). The higher horizontal resolution of HARMONIE (2.5 km vs. 10 km) is able to solve the orographic constraints, island boundaries and intraday processes (such as sea-breezes) more accurately. For instance, the Tarragona system is prone to wind jets (Grifoll et al., 2016 [48]), mainly constrained by the mountain-ranges close to the coastal fringe. As can be seen in the Tarragona case (Figure 6a), IBI\_PHY forced with ECMWF-IFS cannot reproduce the wind jet. ECMWF-IFS coarse resolution (10 km) is not enough for modelling the narrow sharp topographic gradients between the mountain ranges near the Ebro Delta, but HARMONIE does it (see the SAMOA solutions in Figure 6d,g).

Hence, proper current direction is an important remark, especially when tidal-currents act concomitant to wind-induced currents. Errors in the wind direction can lead to windinduced stresses; that albeit accurate in modulus, incorrectly reinforce/weakens surface currents driven by other physical processes.

IBI\_PHY uses CORE bulk formula (Large and Yaeger, 2004 [49]), SAM\_ADV the COARE 3.0 (Fairall et al., 2003 [33]), and SAM\_H3D the COARE 3.5 (Edson et al., 2013 [39]). Charnock coefficient in COARE 3.5 (and its parametrized surface-drag coefficient) is lower for moderate winds than in COARE 3.0 (Brodeau et al., 2017 [50]). In COARE 3.0, Charnock coefficient is set to constant (0.011) at wind speeds from 0 to 10 m/s; whereas in COARE 3.5, Charnock is set to 0.006 until 6 m/s. Due to the systematic overestimation of the HAR-MONIE wind fields, the reduction of the wind-stresses can be a factor for the enhancement of SAM\_H3D surface currents metrics, especially in correlation and general bias, when moderate wind conditions prevail.

Another difference between SAM\_ADV and SAM\_H3D is that the gustiness parameter is lower in COARE 3.5 than in the former release; and that may boost surface stresses at extreme wind regimes. Gustiness effects are not included in CORE, that may be a secondary factor for better IBI\_PHY RMS metrics (ECMWF atmospheric forcings influence, being a primary one). Nonetheless, CORE drag coefficients tend to be higher than COARE at moderate wind speeds; and vice versa, from moderate to high winds (Pelletier et al., 2018 [51]). Further research in air-sea interaction would shed some light on this key issue.

SAM\_H3D is the solution that shows more resemblance with the spatial signature from surface HF radar (Figures 6 and 7). Note however, that both SAMOA solutions are heavily constrained by the IBI\_PHY solution. For instance, the spurious SW fluxes at the Eastern corner at the Tarragona model (Figure 6) are mainly due to the IBI parent solution. Eddies in the boundaries and corners, such as the clockwise one close to the Southern corner can be also problematic, especially when IBI patterns are unrealistic. The SAM\_H3D nesting ensures mass and momentum conservation from the parent to child grids, but errors in the former can be easily propagated.

This latter phenomenon may partly explain the low predictive skill in the Almería system, both with IBI\_PHY and SAMOA. A possible reason may be transient spurious mesoscale structures in IBI\_PHY at Cabo de Gata (Figure 2). High surface current direction bias is found at the deep-water EXT buoy (almost 30◦, in comparison with the 3◦ bias found when compared at the Bilbao or Tarragona EXT buoys). Note that these errors in direction are inherited from the parent solution (IBI\_PHY), suggesting that the transport direction is incorrect. This phenomenon also affects both heat and salinity fluxes. Hence, a plausible solution to minimize undesired influences related to spurious dynamic features identified in the IBI\_PHY parent solution may be to expand the SAMOA coastal domain, fixing the SAMOA boundaries outside of the areas where such spurious structures are favoured in the IBI\_PHY solution; then generating a SAMOA intrinsic solution, less constrained by the limited-area domain and the IBI\_PHY data imposed as boundary condition.

Regarding SST, the enhancement of SAM\_H3D solution may be due to three factors: (i) better characterization of the heat fluxes with COARE 3.5; (ii) consistent circulation

fields and (iii) the use of HSIMT-TVD advection scheme for passive tracers. Despite that SAMOA does not have data assimilation, the one-year metrics are close, or even better (at some cases) than IBI\_PHY. Note that the Copernicus Regional System counts with a data assimilation scheme. Then, satellite SST fields and in-situ profiles data are assimilated in the IBI\_PHY analyses.

In this sense, Gran Canaria highlights the role of coastal circulation in shallow-depths heat fluxes, as the enhancements of the circulation fields are consistent with SST ones. The RMS is lower in IBI\_PHY than SAMOA, though; most probably due to the radiation forcings (be it short or long-wave), that have higher variability in HARMONIE than ECMWF.

SAMOA significantly improves the sea level, especially in those areas with meso and macro-tides. There is no important difference between SAM\_ADV and SAM\_H3D, suggesting that the main improvements come from the increase of horizontal resolution, that lead to better capture of the harbour geometry and coastal surroundings. Water mass fluxes and piling-up effects are better reproduced.

However, the performance is substantially lower at the Barcelona and Tarragona cases (as shown in the Table 3 metrics), in which all the three models require further assessment on two aspects:


Wave-driven currents are another point in which SAMOA will devote further efforts. At present (late 2021), the Copernicus IBI\_PHY solution is one-way coupled with the IBI-WAV system (based on the WAM (WAMDI Group, 1988 [53]) spectral wave model) by adding (i) the Stokes drift, (ii) wave-induced mixing and (iii) wind drag coefficients formulas based on the sea-state. This addition may be particularly important during wave storms, as suggested in Sotillo et al., 2021a [47] and Lorente et al., (2021) [54]. Current developments of the SAMOA system, developed within the EuroSea Project, are going in the same direction, with strong focus on the local processes beneath intermediate and shallow waters.

Same focus on local coastal processes, that has become the leitmotif of the SAMOA forecasting system, will also be taken on improving the land-sea connection. Proper modelling of salinity fluxes require operational run-off discharge products, such as the one analysed in this issue (Sotillo et al., 2021b [55]), rather than climatological values. All these improvements, albeit ambitious, represent a clear SAMOA roadmap for the coming years. Moreover, as suggested in the case of Gran Canaria (i.e., SST improves when surface currents do it), the synergistic gain from addressing all of them, would enhance the predictive skill of SAMOA operational products at an integrated level.
