From a 7 km model to AMM15, and back to Copernicus
The Met Office's ocean forecasts for the north-west European shelf go back a long way: by 2019 the system had been providing products for more than 15 years, built on FOAM, its Forecasting Ocean Assimilation Models. These seas are mostly shallow and highly tidal, busy with fishing, oil, gas and renewable energy, and their coasts are exposed to flooding from high tides, waves and storm surges. For 8 years the forecasts delivered to Copernicus Marine and the projects before it came from AMM7, a model with a 7 km grid. It reproduced the large-scale circulation of the shelf, but not its small eddies, frontal jets or internal tides.
AMM15, the Atlantic Margin Model at 1.5 km, was developed by the Met Office and the National Oceanography Centre through their Joint Weather and Climate Research Programme, and presented by Graham and colleagues in February 2018. The step was costly: about 15 times as many wet cells as AMM7, and about 400 node hours a day against 20. A 30-year run without data assimilation improved the mean state overall, more off the shelf than on it, and left the North Sea's biases largely unchanged. The authors warned that a finer horizontal grid alone would not bring improvements across the whole domain.
Two years of trials, 2016 and 2017, run beside AMM7, served to decide whether to go ahead, and real-time AMM15 forecasts started at the beginning of November 2018. Tonani and colleagues described that first system in 2019: a 6-day forecast, forced by ECMWF's IFS at the request of Copernicus Marine, computed in about 4 hours on 48 nodes of the Met Office's Cray XC40. The product's quality document records the coupling to a wave model as a science update in September 2020.
Then came a break. The consortium the Met Office led, with Germany's Federal Maritime and Hydrographic Agency (BSH), the National Oceanography Centre and Plymouth Marine Laboratory, left the Copernicus service, and from November 2023 the product of the same name was an extraction of the IBI production centre's system, while the Met Office offered AMM15 through its own UK Marine and Climate Advisory Service. In November 2025 the Met Office's system returned under a new consortium, with a 7-day forecast; Copernicus now lists BSH in the lead, with the Met Office, Plymouth Marine Laboratory and Hereon. The quality document calls it similar to the system available before November 2023, and reports that a Copernicus-funded internal study found that one better on tides and sea temperature than the IBI extraction, and otherwise comparable.
The science: the tide in the equations
AMM15 is a regional configuration of NEMO v3.6 (Nucleus for European Modelling of the Ocean), the ocean code behind SMOC's circulation too, set up here for a sea the tide dominates. Tides are the main source of variability over most of the shelf, so the model has a non-linear free surface: its layers stretch and squeeze as the water rises and falls. The short time scales of a tide require the computation to be split in two, the depth-averaged flow stepping about every 3.5 seconds and the rest every 60 seconds. Friction on the sea floor follows a logarithmic layer, and vertical mixing a generic length scale turbulence scheme.
The tide is computed in the model, not added to it. It enters in two ways: along the open boundaries, as 11 constituents (M2, S2, N2, K2, K1, O1, P1, Q1, M4, MS4, MN4) taken from the TPXO7.2 solution at 1/12°, and as the equilibrium tide. From there it travels across the shelf by the model's own physics, over EMODnet's 2015 bathymetry, moved from lowest astronomical tide to mean sea level. SMOC works another way: it adds a depth-averaged tide from FES2014 to a circulation computed without one. In AMM15, tide and circulation come out of the same equations as one current.
The model computes on a rotated grid, its pole moved so that a false equator crosses the domain and the cells keep about the same size from south to north. The papers and the quality document give about 1.5 km in both directions; the product manual gives 1/60°. The product delivers it interpolated onto a regular grid of 1/33° in longitude by 1/74° in latitude, about 2 km by 1.5 km at 50°N. That is finer than the internal Rossby radius on the shelf, of the order of 4 km, so eddies and fronts can be computed with only a little added viscosity. In the vertical it has 51 levels that follow the sea floor, from about 0.3 m thick in shallow water to 99 m in the deepest, interpolated onto 33 standard depths.
The ocean is coupled both ways with the wave model WAVEWATCH III v4.18, every hour. Waves act on the current through three terms: the stress on the water, changed by wave growth and dissipation; the Stokes-Coriolis force in the momentum equation; and a sea surface roughness that depends on wave height. The wave model, in turn, feels the surface current. The surface current itself is the first model level, not interpolated: 1 m thick where the sea is deeper than 50 m, thinner elsewhere. No Stokes drift is added to it, unlike SMOC's total current, which sums the circulation, the tide and the Stokes drift. It does not hold the small net forward motion of the water under passing waves.
How a run is computed
One run is made each day, from 00 UTC. It starts with two analysis cycles over the two days before, from 48 to 24 hours back and from 24 hours back to the start, then runs the forecast 168 hours ahead. In each cycle the model first runs for a day and is compared with the observations at the time they were made; NEMOVAR, a 3D-Var scheme, then works out corrections to temperature, salinity, velocity and sea surface height, kept dynamically balanced; a last run feeds them in gradually. The analysis 24 hours back starts the next day's run, and the wide window lets in observations that arrive late.
The system assimilates sea surface temperature from ships, drifters and moored buoys and from satellites (AVHRR, SEVIRI, VIIRS, AMSR2); temperature and salinity profiles from Argo floats, gliders and other instruments; and along-track sea level anomaly from the altimeters of Cryosat, AltiKa, Jason-3, Sentinel-3 and Sentinel-6. The altimetry and the model are detided before the corrections are worked out, and the tide is then added back, so the tide stays the model's own. No measured current is assimilated: the velocities move only through their balance with the other fields, and the HF radars used in validation stay out of the analysis.
Above the sea, the Met Office's global Unified Model drives the system, its hourly and three-hourly fields applied directly as fluxes. AMM15 is nested one way: at its Atlantic edge, the Met Office's global 1/12° ORCA12 model gives temperature, salinity, sea surface height and depth-integrated currents, and at its Baltic edge the Copernicus Baltic Sea model gives temperature, salinity and depth-integrated currents. River flows come from a daily climatology of gauge data averaged over 1980–2014, with an older one for major rivers missing from it, along the French and Norwegian coasts for instance. They are typical flows for the date, not those of the day.
Copernicus Marine has a target of 12:00 UTC for the daily publication; the hourly surface current has been seen published about 08:50 to 10:15 UTC, the two days of analysis rewritten and the previous forecast replaced. Daily means over 25 hours, which remove the tide, and quarter-hourly surface fields sit beside the hourly ones, with a rolling archive of two years. The model computes from about 45°N to 63°N and out to about 20°W; the distributed rectangle is slightly smaller, 46°N to 62.74°N and 16°W to 13°E, on 958 × 1240 points, about 724,500 of them water. All the grid points in the Baltic and in the Kattegat south of 57.25°N are masked.
At sea
AMM15 covers the English Channel and Dover Strait, the North Sea up past Shetland, the Irish Sea and St George's Channel, the Celtic Sea, the waters around Brittany, the west of Ireland and Scotland, the Pentland Firth, the Skagerrak, and the Bay of Biscay only as far south as 46°N: Iberia and the Mediterranean are outside it. Its grid draws more of these waters than a global one: in a square of ±0.08° around the Raz Blanchard, also called the Alderney Race, SMOC's 1/12° grid has 4 water points and AMM15's has 60, as Saily WR measured in November 2026. That is more detail, and a stronger stream may show up in it; neither proves the stream is right.
The producers' validation, over 2022 and 2023, is firmest on the tide as a height of water. At 146 tide gauges on open coasts, the main semi-diurnal constituent, M2, has a median amplitude difference of 0.22% (0.26 cm, with a root mean square difference of 9.54 cm for a median amplitude of 128.77 cm) and a median timing difference of 8.2 minutes, with a root mean square difference of about 21 minutes. They find the largest timing errors in the German Bight, where the shallows and an amphidromic point nearby make the phase hard to place, and in the Skagerrak and Kattegat, where tides are small. These figures tell of the rise and fall of the sea, not of its stream.
The stream itself is harder to check: the producers call modelled currents very difficult to evaluate, for lack of data, and leave them out of the summary of their results. They compare it with three HF radar systems. In the German Bight, the correlation is about 0.9 for tidal currents and 0.7 for the residual flow; in the Skagerrak the residual flow does better than the small tide; off Bergen, in the Norwegian Trench, the correlation is low for both. Tonani and colleagues found AMM15's tidal current closer than AMM7's to the German Bight's radars in March 2017, a smaller root mean square difference everywhere the radars reach. None of these radars stands in the Channel, the Irish Sea or off Brittany, so these figures say little about the stream in any given race.
Near the coast the model has limits of its own. NEMO 3.6 cannot wet and dry, so a minimum depth of 10 m is imposed: there is water in the model where the seabed really dries out, and the tide's propagation is wrong in such depths. The product manual advises against using the fields directly where large areas are shallower than 10 m, naming the Wadden Sea; the configuration paper adds the East Anglian coast, the Bristol Channel, Morecambe Bay and the Solway Firth. A river in flood is not in the forecast, and a grid of about 2 km cannot draw the eddy behind a rock: Tonani and colleagues write that coastal processes are not fully resolved. In a tidal gate or close to land, tidal atlases and pilot books remain the reference.
AMM15 in Saily WR
Saily WR takes AMM15's hourly surface current from Copernicus Marine's public store, on its own grid, from H+0 to H+168 of each day's run. A server downloads each new run once it is published, and the Data page shows live which AMM15 run the app is on, its horizon and the rectangle it covers. As the Copernicus Marine licence requires, the data are credited wherever they are shown: "E.U. Copernicus Marine Service Information (CMEMS)".
There is nothing to choose: inside its waters a route sails in AMM15 by itself, over SMOC. Past H+168, or where none of the four AMM15 points around the route has water, the route takes SMOC. A grid cell touching land keeps AMM15, drawn from its points with water alone: the route switches to SMOC, on a much coarser grid, only in a harbour or a pass AMM15 does not see.
At the domain's edge, at 46°N notably, the route passes from one current to the other with no blending, so the current it meets can jump there. The route's summary names each current it crossed, in order, for instance "AMM15, then SMOC"; on a route sailed in several, the route log, on screen and printed, names the source of each leg, and the CSV has a "Current source" column. On the routing map, the Current layer shows AMM15 inside its waters and the global current around it.
If Copernicus stops publishing, the last run stays in use while its steps cover the time, even days old. On a long trip, whose time step lasts hours, a leg keeps one value of the current for its whole length although the tide turns about every 6 hours, so the time gained on a strong stream can be overstated. In very light wind against the stream, a route can also stop short of its destination. And Saily WR has not compared AMM15's current with observations: nothing here says that a route sailed in it is closer to the truth than one sailed in SMOC.
Sources
- AMM15: a new high-resolution NEMO configuration for operational simulation of the European north-west shelf, Graham et al., Geoscientific Model Development (2018)
- The impact of a new high-resolution ocean model on the Met Office North-West European Shelf forecasting system, Tonani et al., Ocean Science (2019)
- Product User Manual for NWSHELF_ANALYSISFORECAST_PHY_004_013, issue 3.0, Copernicus Marine Service (2025)
- Quality Information Document for NWSHELF_ANALYSISFORECAST_PHY_004_013, issue 3.0, Copernicus Marine Service (2025)
- Atlantic - European North West Shelf - Ocean Physics Analysis and Forecast, Copernicus Marine Service (2026)
- Currents, 2D, hourly: the dataset's catalogue entry, Copernicus Marine Service (2026)
- North West Shelf – Monitoring Forecasting Centre (NWS MFC), Copernicus Marine Service (2026)
- NWS MFC poster: the NOWMAPS consortium, Copernicus Marine Service (2023)
- Met Office marine data service, Met Office
- Product User Manual for GLOBAL_ANALYSISFORECAST_PHY_001_024, issue 2.4, Copernicus Marine Service (2026)
- Service commitments and licence, Copernicus Marine Service (2026)