Unstructured grid modelling of o ﬀ shore wind farm impacts on seasonally stratiﬁed shelf seas

Shelf seas comprise approximately 7% of the world’s oceans and host enormous economic activity. Development of energy installations (e.g. O ﬀ shore Wind Farms (OWFs), tidal turbines) in response to increased demand for renewable energy requires a careful analysis of potential impacts. Recent remote sensing observations have identiﬁed kilometre-scale impacts from OWFs. Existing modelling evaluating monopile impacts has fallen into two camps: small-scale models with individually resolved turbines looking at local e ﬀ ects; and large-scale analyses but with sub-grid scale turbine parameterisations. This work straddles both scales through a 3D unstructured grid model (FVCOM): wind turbine monopiles in the eastern Irish Sea are explicitly described in the grid whilst the overall grid domain covers the south-western UK shelf. Localised regions of decreased velocity extend up to 250 times the monopile diameter away from the monopile. Shelf-wide, the amplitude of the M 2 tidal constituent increases by up to 7%. The turbines enhance localised vertical mixing which decreases seasonal stratiﬁcation. The spatial extent of this extends well beyond the turbines into the surrounding seas. With signiﬁcant expansion of OWFs on continental shelves, this work highlights the importance of how OWFs may impact coastal (e.g. increased ﬂooding risk) and o ﬀ shore (e.g. stratiﬁcation and nutrient cycling) areas.


Introduction
Shelf seas comprise approximately 7% of the world's oceans and are the focus of an enormous amount of economic activity. They have been the subject of intensive study for the last century during which time 5 their dynamics have been well documented (e.g. Proudman and Doodson, 1924;Doodson and Corkan, 1933;Cloet, 1954;Simpson and Hunter, 1974;Pingree and Maddock, 1977;Pingree et al., 1982). Shelf seas are dynamic environments subject seasonal heating, atmo-10 spheric fluxes, tides, river inputs and open ocean exchange (Holt et al., 2012). They are characterised by seasonal temperature stratification in which summer heating is able to stratify the water column and overcome strong tidal mixing (Holt and Umlauf, 2008). 15 Stratified regions are characterised by a sharp seasonal thermocline (developed with the onset of positive net heat flux into the water (Smyth et al., 2014)) whilst nonstratified regions are dominated by tidal mixing and remain vertically homogeneous year round. Tidal mixing 20 fronts form at the interface between these stratified regions and shallower non-stratified waters (Simpson and Hunter, 1974).
Shelf seas are critical in maintaining a complex ecosystem (Proctor et al., 2003;Holt et al., 2012;Wake-25 lin et al., 2012) which has been shown to modulate the impacts of climatic variability (Barange et al., 2011;. The behaviour of the shelf sea ecosystem is predominantly controlled by the timing and rate of water transport from low light and nutrient rich deep 30 waters into high light nutrient poor surface waters (Pingree et al., 1982;Richardson et al., 2000). Tidal mixing fronts separate the nutrient-rich from nutrient-poor waters, thus their location, stability and strength are critical in the evolution of many ecosystem processes (e.g. 35 spring bloom phenology and magnitude). Furthermore, the interplay between density gradients and tidal currents generates temporally varying Strain-Induced Periodic Stratification (SIPS) which in turn feeds back and modifies the water column structure and tidal ellipses 40 (Simpson et al., 1990;Palmer, 2010). Therefore processes which act to modify either density or currents have the potential to modify the magnitude and timing of SIPS (Souza and Simpson, 1996;Palmer, 2010).
Within the north-western European shelf seas, the 45 UK government's commitment to Offshore Wind Farms (OWFs) is considerable: as of 2008, the UK has installed OWFs with a combined 3.7GW capacity; the UK could be committed to delivering 22.4% (29GW) of its total electricity generation (129GW) from OWFs under 50 the EU 2020 Renewable Energy Targets (Carbon Trust, 2008). This massive UK investment in OWFs is under way with 1,183 offshore wind turbines already installed.
However, despite the rapid introduction of large-scale OWFs to the UK's shelf seas, the potential impacts these 55 devices have on shelf sea hydrodynamics remain relatively unknown. This is due largely to the cost and complexity of observational campaigns capable of capturing the potential impacts. Recent analyses of remote sensing data have observed impacts derived from the 60 introduction of offshore wind turbines several kilometres from their siting (Li et al., 2014;Vanhellemont and Ruddick, 2014), thus, large-scale work to quantify these impacts is needed.
The majority of the work to date has been of i) small-65 scale models with explicit individual turbines (Roulund et al., 2005;Jensen et al., 2006;Okorie, 2011) and ii) large-scale model domains with turbine impacts parameterised as sub-grid scale processes through increased bed roughness (Lambkin et al., 2009), water column 70 velocity (Shapiro, 2011), turbulence models (Rennau et al., 2012) or Linear Momentum Actuator Disk Theory (LMADT) (Serhadlıoglu et al., 2013). Parameter-isation is computationally efficient, however, it omits small-scale turbulent processes which can have impor-75 tant impacts for horizontal and vertical water structure (Christie et al., 2012). van Rennau et al. (2012) 90 found a small effect from turbines represented as 25 and 50m dry elements (results were scaled linearly by factors of 5 and 10 to represent turbines of 5m diameter, an approach which is likely to yield overestimates of the impacts). A number of models have assessed im-95 pacts from tidal turbines (i.e. submerged structures) using momentum sink parameterisations (e.g. Yang et al., 2013Yang et al., , 2014 and have shown that, in macrotidal estuaries (e.g. South Puget Sound), such as are common on the UK continental shelf, tidal stream turbine wakes 100 extend approximately 1.5km (Yang et al., 2014). A 2D TELEMAC model was used to investigate a proposed wind farm off the east coast of Ireland (the Dublin Array) (MRG Consulting Engineers Limited, 2013).
The model domain covered 2,800km 2 and contained 105 100,000 elements with 145 turbine monopiles represented as 5m diameter hexagonal islands. The results indicated that horizontal impacts from the wind turbine monopiles reduced surface current speeds by 5% of the maximum at distances of up to thirty times the monopile 110 diameter (MRG Consulting Engineers Limited, 2013).
Given the 2D nature of the model, no assessment of the change in vertical velocity, temperature and salinity structure was possible.
The two scales adopted in the approaches to date 115 (spatially limited and high resolution and vice versa) suffer from limitations of scaling: high resolution models are unable to determine whether impacts propagate to distances beyond each model domain; coarse models cannot resolve the small-scale impacts and their po-120 tentially important role. One approach to resolving this conundrum is to nest model grids. This, however, often means a one way exchange of information which limits the export of impacts from the small grids to the larger ones, a process which can have important effects 125 on the model results (Zhou et al., 2014). To resolve this, a model must seamlessly resolve and communicate processes across a range of scales, and an unstructured model provides this ability (Jones and Davies, 2007a).
The work presented here expands on existing work by 130 bridging the required scales and extends the approach into the vertical to address implications for stratified shelf seas. The modelling focuses on the impacts generated by the addition of turbine monopiles to the UK shelf, with an emphasis on the eastern Irish Sea (Fig. 1).

135
The UK shelf is tidally dominated (tidal ranges reach 14m in the Bristol Channel), seasonally stratified (onset in April, dissipation in August) shelf sea. Typical surface temperatures are between 3 and 20 • C and salinities vary strongly with proximity to river mouths (where 140 they approach zero), but are typically 35.5PSU further offshore. Current speeds rarely exceed 2m s −1 , although localised current speeds can exceed 3m s −1 e.g. in the Pentland Firth, Scotland (Martin-Short et al., 2015).
To assess the performance of the Finite Volume Com-

Model set up
The Finite Volume Community Ocean Model (FVCOM) is a prognostic, unstructured-grid, finitevolume, free-surface, 3D primitive equation coastal 165 ocean circulation model (Chen et al., 2003). FVCOM solves the 3D momentum, continuity, temperature, salinity and density equations by computing fluxes between unstructured triangular elements. Vertical turbulent mixing is modelled with the General Ocean Turbu-170 lence Model (GOTM) using a k-formulation (Umlauf and Burchard, 2005) whilst horizontal mixing is parameterised through the Smagorinsky scheme (Smagorinsky, 1963). The vertical grid in FVCOM is described in terrain following (sigma) coordinates where shallower 175 areas resolve vertical structure with finer detail.
FVCOM has been widely used in shelf and coastal domains for a range of problems where a strong need exists to resolve varying horizontal scales, including: physical modelling of temperature and salinity stratifi-180 cation (Chen et al., 2007;Yang and Khangaonkar, 2008;Huang, 2011;Zheng and Weisberg, 2012); tracer evolution in complex estuaries (Torres and Uncles, 2011); the relationship between hydrodynamics and pursuit diving bird behaviour (Waggitt et al., 2016a,b); the be-185 haviour of sequestered CO 2 leak plumes (Blackford et al., 2013); and tracking the dispersal of lice (Adams et al., 2012(Adams et al., , 2014.

Flume grids 190
The inclusion of wind turbine monopiles must be analysed in terms of the impact the model grid resolution has on the calculated results. To that end, three

195
The flume is 3×1km in size and all three configurations have a 5m diameter turbine 1km from the left boundary. The turbine is represented as 5m diameter hexagonal island (i.e. infinitely high walls) to best capture the shape of the monopile whilst minimising the number of ele-200 ments required to represent it. The grid resolutions decrease from 2.5m at the monopile and change linearly with distance to 20m, 10m and 2.5m over a radius of 150m for the low (3,683 nodes), medium (6,869 nodes) and high (27,541 nodes) resolution grids, respectively.

205
The flume configurations all feature identical forcing, with a constant bed roughness (z 0 ) of 0.03m, a si-  (Chen et al., 2013). The water column temperature and salinity are warmer and fresher (12.5 • C and 33.5PSU, respectively) in the top 5m of water than the remaining 25m, where temperature and salinity are 11 • C and 33.8PSU, respectively. There is 220 no wind, heating, precipitation or air pressure forcing.     the sub-grid scale turbulence parameterisation; and the potential energy anomaly as a measure of stratification.

290
The potential energy anomaly (φ) in J m −3 is defined as: where g is the acceleration due to gravity (9.81m s −2 ), h is the water depth (m), ρ is the density (kg m −3 ) andρ is the depth-averaged density: (2) The grid sensitivity variables in Table 1 show that the low and medium grids have a very similar response to the monopile across all three measured parameters. The high resolution grid, in contrast, shows differences from the other two grids: in the potential energy anomaly In the speed anomaly results, the low grid is overestimating the horizontal velocity signature within the wake region which, again, indicates the wake is not as 315 well resolved (the region of low velocity is less well defined in the lower resolution grids). The TKE is similar in all grids, indicating the sub-grid scale parameterisations for turbulence are capturing the processes in the lower resolution grids.

Grid configuration
The model domain is defined by the initial coast-400 line, derived from the Global Self-consistent, Hierarchical, High-resolution Shoreline (GSHHS) (Wessel and Smith, 1996) and sampled at resolutions between 200 and 1200m. The model grid is constructed such that the resolution is controlled by a size function based on 405 coastline curvature, water depth (h), bathymetry gradient and gravity wave propagation speed ( gh, where g is the acceleration due to gravity (9.81m s −2 )) ( Fig. 1).
This ensures that areas with complex coastlines, high seabed gradients and shallow water depths have smaller 410 elements to ensure tidal wave propagation is well resolved (Legrand et al., 2007). The gradient control is depth-limited (50m threshold) so only the shallowest parts of the domain are adjusted by both the water depth and its gradient (otherwise water depth only). From an 415 initial grid, the unstructured grid is iteratively adjusted such that the element sizes fit the size function. Final manual adjustment of the grid ensures the quality criteria in the FVCOM manual (Chen et al., 2013) are met.
The model has 20 vertical layers distributed in the ver-420 tical with a quadratic function (the resolution of the surface and near-bed layers is higher than those in the midwater column).
Due to the large computation requirements associated with running a grid in which each monopile is explic- at the wind turbine monopiles (Fig. 1). Reanalysis-2 heat flux data (Kanamitsu et al., 2002) interpolated to the model grid. Heat flux is prescribed at the surface and added to the water column using the COARE2.6 bulk air-sea flux algorithm (Fairall et al., 2003) in the vertical diffusion terms as layer fluxes. Pre- The model is run for two periods: January and May.

Shelf model set up
The north-west European continental shelf is season-ally stratified, with stratification typically beginning in April. By modelling January and May, we are able to 515 include both fully mixed and stratified waters to investigate how the turbines affect stratification.

Tidal harmonics
The computational expense of running the model  The amplitudes and phases for M 2 , S 2 and M 4 in Fig.   4 show a high similarity with existing model analysis (e.g. Jones and Davies, 2007a,b;Jones et al., 2009), with observation derived maps (Proudman and Doodson, 1924;Doodson and Corkan, 1933) as well as with 550 remote sensing analyses (Egbert et al., 1994;Andersen, 1995

Fine-scale coastal circulation
The fine grid modelled surface velocities were compared with high frequency radar data to assess spatial model skill at reproducing coastal currents. The Liv-

Liverpool Bay Coastal Observatory radar
The time series analysis of the data from Liverpool 630 Bay CObs yields a correlation over the majority of the domain in excess of 0.8 (Fig. 7a). In areas where the number of samples (Fig. 7b) is lower (i.e. close to the radar sources), the coefficient decreases. This, however,

PRIMaRE radar
The model-PRIMaRE radar statistical analysis results are shown in Fig. 9. Compared with the CObs data, the 670 PRIMaRE data are noisier with artefacts from the radar visible in the statistical analyses (the large arcs in Fig.   9a). Although the spatial resolution is higher here than in the CObs data, the time sampling is hourly compared with every 20 minutes. The correlation coefficient in

Validation summary
Tidal analysis of the model results against a range of observed data, including HF radar data and coastal and 695 offshore tide gauges, has shown the model accurately reproduces the physical environment of the model domain. The ranges in Bartlett (1998)

Results
The anticipated impacts from the introduction of a large number of monopiles into an energetic shelf sea are that locally generated increased turbulence dissipates rapidly and the impacts are limited to the near-710 field. These local impacts would be evident in the velocity fields, mainly as reductions in the wakes, but also increases around the monopile sides. Since the shelf sea is a large dynamic system, alterations in the energy balance at OWF-scale would be accommodated by altering 715 the flows within a relatively short distance (a few tens of kilometres) and beyond that, little to no impact would be evident in the circulation.
In the vertical structure of the water column, the local effects would be relatively marked, with potentially 720 increased mixing due to turbulence from the monopiles.
This would be evident only in areas which experience seasonal stratification and in which the maximum current speed is high enough to create turbulent fields able to induce vertical mixing. Beyond the OWFs, the im-725 pact is likely to be relatively minor.
Analysis of the spatial and temporal changes in the The wakes observed from remotely sensed imagery (Li et al., 2014;Vanhellemont and Ruddick, 2014) indicate they extend up to 1.3km from the wind turbine  difference in the latter is instead a change in distribution around the amphidrome rather than a shift in its position following the introduction of the turbine monopiles.

Vertical water column structure
The turbine monopile impacts on the vertical struc-845 ture of the water column are illustrated in vertical profiles through six wind turbines in Fig. 12. Four parameters are shown: horizontal current speed (a), vertical velocity (b), temperature (c) and salinity (d).
With the tide flowing from left to right in Fig. 12a, fication. Stratified water will experience smaller vertical velocities than their fully mixed counterparts due to the increased work required to overcome the density gradient.
As the water column stratifies due to increased sur-870 face heat in the spring, the impact of the monopiles is evident ( Fig. 12c and 12d).

Seasonal stratification
In seasonally stratified seas (such as those around the UK), the stratification can be measured through the po-885 tential energy anomaly. The potential energy anomaly represents the amount of energy required to fully mix the water column, thus, a value greater than zero indicates a stratified water column, less than zero means the column is convectively unstable (i.e. less dense water 890 below high density) and zero a fully mixed water column. The climatic conditions of the north-west European continental shelf mean that the water column stratification onset is controlled by decreased wind stress and freshwater inputs and increased summer insolation 895 creating a warm and slightly more saline surface layer which is able to persist despite strong tidal mixing, inducing a stratification feedback (Holt and Umlauf, 2008). Tidal mixing during the winter is able to overcome these stratifying factors to produce a fully mixed 900 water column.    (Simpson and Hunter, 1974;Bowers and Simpson, 1987;Holt and Umlauf, 2008

Discussion
The modelling presented here quantifies how the introduction of 242 offshore wind turbines impacts a tidally dominated, seasonally stratified shelf sea. The turbines are explicitly represented in the model grid 945 rather than parameterised through roughness or momentum effects (Melville and Sutherland, 1989;Yang et al., 2013). The model is compared against a range of in situ observations (NTSLF tide gauges, BODC offshore pressure sensors, CObs and PRIMaRE HF radar) both as 1D 950 and 2D time series and through harmonic analysis. The results of the model validation indicate that the model performs well within the expectations required for accurate coastal and estuarine modelling put forward in Bartlett (1998).

955
The approaches taken to quantify the impacts of  Analysis of the spatial characteristics of wakes expressed in the surface current speed has shown good agreement with the existing remote sensing outputs.
Since the modelled wakes generated in the lee of the monopiles remain at all states of the tide, each tidal 1015 turbine has a radius over which its impact is felt. For the turbines in the modelling performed here, the radius for each turbine is approximately 1km, inside of which the reduction in speed is 5% of the regional maximum current speed. Therefore, the spacing of wind turbines within a wind farm has consequences on the total horizontal area which is affected by their introduction. Farms which are more closely spaced therefore and "Hornsea") would fall within areas of significant seasonal stratification (Holt and Umlauf, 2008;Souza, 2013). The introduction of these structures within re-1105 gions which are critical to the shelf ecosystem (Pingree et al., 1982;Richardson et al., 2000) may have impacts on resources within those areas e.g. fisheries, bird habitats (Miller and Christodoulou, 2014). Recent analyses of remotely sensed sea surface temperature shows the 1110 development of fronts is often associated with "charismatic megafauna" (Miller and Christodoulou, 2014).
Thus, adding turbine monopiles can alter the structure of the fronts and impact on the migration of these megafauna. The knock on impacts from OWFs will be 1115 felt not only in the lower trophic levels (via stratification, mixing and nutrient cycling) but all the way to large marine vertebrates.

Conclusions
The use of a 3D unstructured hydrodynamic model of  lar, but also offshore. In areas where seasonal stratification occurs, increased vertical transport induces greater mixing leading to a decrease in stratification. The horizontal extent of this disturbance is significantly larger than the sum of the footprint of the monopiles.

1155
Whilst the modelling outlined here describes the first attempt at accurately scaled monopiles in a large-scale domain, it is not without its limitations. The performances of the sub-grid scale mixing parameterisations (Smagorinsky (1963) in the horizontal and GOTM with 1160 a k-formulation from Umlauf and Burchard (2005)) are pushed to their limits in the vicinity of the turbine monopiles. Whilst the grid sensitivity analysis has shown that the resolution used in the shelf model is an underestimate of the impacts generated from higher res-1165 olution grids, a non-hydrostatic FVCOM model configuration could overcome some of these limitations. However, given the already onerous computational requirements, it may not yet be practical to perform this work.
There is also proof-of-concept work in which FVCOM 1170 is coupled with a CFD model (Wu and Tang, 2010), which would be an alternative route of investigation, although it is also not without significant challenges.
When considering future offshore wind farms, this work highlights the importance of their placement on 1175 impacts onshore (e.g. flooding, visual, noise), in intertidal areas (e.g. bird habitats) and offshore (e.g. nutrient cycling through seasonal mixing), particularly when candidate sites are subject to seasonal stratification. This is particularly important given the future expansion of the UK's offshore wind farms and the proposed locations of the next round of construction, which includes some large offshore wind farms in areas of significant stratification close to the coast.