QUALITY INFORMATION DOCUMENT North West
Transcription
QUALITY INFORMATION DOCUMENT North West
QUALITY INFORMATION DOCUMENT North West European Shelf Reanalysis NORTHWESTSHELF_REANALYSIS_PHYS_004_009 and NORTHWESTSHELF_REANALYSIS_BIO_004_011 Issue: 3.0 Contributors: Sarah Wakelin, James While, Robert King, Enda O’Dea, Jason Holt, Rachel Furner, John Siddorn, Matthew Martin, Robert McEwan ,Ed Blockley and Jonathan Tinker Approval Date by Quality Assurance Review Group : CHANGE RECORD When the quality of the products changes, the QuID is updated and a row is added to this table. The third column specifies which sections or sub-sections have been updated. The fourth column should mention the version of the product to which the change applies. Issue Date § Description of Change Author Validated By 1.0 January 2014 All Creation of the document Sarah Wakelin for V4.0 Ed Blockley 2.0 December 2014 All Addition of biological Robert McEwan product assessment (BIO_004_011) Ed Blockley 2.1 February 2015 All Revision after acceptance V5 Ed Blockley 2.2 13/03/2015 I.2, Warning remarks after II evidence of 004_011 interannual nutrient drifts 2.3 May 1 2015 all Change format to fit CMEMS graphical rules 3.0 21/01/2016 all Update for V2 (time series extension to 2014 and addition of MLD) Sarah Wakelin, Jon Tinker Marina Tonani 3.0 01/04/2016 all Revision after V2 AR Marina Tonani Marina Tonani Ed Blockley L. Crosnier TABLE OF CONTENTS I Executive summary ....................................................................................................................................... 4 I.1 Products covered by this document ........................................................................................................... 4 I.2 Summary of the results ............................................................................................................................... 4 I.3 Estimated Accuracy Numbers.................................................................................................................... 5 II Production Subsystem description ................................................................................................................ 6 III Validation framework ............................................................................................................................... 9 IV Validation results ......................................................................................................................................... 11 IV.1 Temperature ........................................................................................................................................... 11 IV.2 Salinity ..................................................................................................................................................... 22 IV.3 Currents .................................................................................................................................................. 31 IV.4 Tidal analysis .......................................................................................................................................... 33 IV.5 Transport ................................................................................................................................................ 35 IV.6 Mixed layer depth................................................................................................................................... 40 VII.1 Chlorophyll............................................................................................................................................ 47 VII.2 Nutrients ................................................................................................................................................ 50 VIII IX Quality changes since previous version .................................................................................................. 52 References.................................................................................................................................................... 53 I EXECUTIVE SUMMARY I.1 Products covered by this document NORTHWESTSHELF_REANALYSIS_PHYS_004_009 and NORTHWESTSHELF_REANALYSIS_BIO_004_011: The physical and biogeochemical parts of a North West European Shelf Reanalysis performed at ~7km resolution. For the physical product, comparisons are made to the North West European Shelf Hindcast at ~12km resolution POLCOMS_REANALYSIS. I.2 Summary of the results The quality of the NWS reanalysis simulation NORTHWESTSHELF_REANALYSIS_PHYS_004_009 from 01/01/1984 to 30/06/2014 has been assessed by comparison with observations. Results from the corresponding control run and the hindcast product. POLCOMS_REANALYSIS covering 01/01/1964 to 31/12/2004, are included for validation purposes. The results for the NWS reanalysis are summarised below. Temperature: Integrated over the whole domain, the bias in surface temperature is 0.4°C. Mean biases in the near surface layer (above 5m depth) have magnitude less than ~0.5°C while the largest errors ~-2°C are in the Bay of Biscay between 800m and 2000m deep. Salinity: Biases, on the practical salinity scale, are generally of magnitude less than 0.5. However, in the coastal regions of the Southern Bight of the North Sea, salinity is typically ~2 too fresh while the surface layer in the Irish Sea is ~0.5 too saline. Currents: Comparison with climatology at 15m depth shows that the reanalysis simulation reproduces major current systems in the region. Transport: For the transport of volume through sections there is fair agreement with the (limited) observational estimates available. Tidal elevations: The M2 constituent of tidal elevation has RMSE of 11.6 cm for amplitude and 15.6° for phase. The amplitude is underestimated in the Irish Sea and Southern Bight and overestimated in large tidal areas such as the Bristol Channel. The phase error is largest in the Southern Bight and German Bight. The quality of the NWS biological product NORTHWESTSHELF_REANALYSIS_BIO_004_011 has been assessed by comparison with observations during the period 03/01/1989 to 31/12/2003. The results for the NWS biogeochemical reanalysis are summarised below. Mixed layer depth: data for the assessment was generally sparse. Model errors were small during the spring/summer months and larger during winter when data was sparser. The model tends to underestimate the mixed layer depth in deep waters and overestimate it in shallower waters. Chlorophyll: For the period analysed the model is shown to have a negative bias in surface chlorophyll over the whole domain and all sub-regions with the mean error being -0.143. Compared to in-situ observations the inter-annual variability is captured well and there appears to be no appreciable drift in model bias. Nutrients: Few in-situ observations were available for comparison so model output has been qualitatively compared to World Ocean Atlas Climatology. Whilst summer depletion and winter recharge of nitrate, phosphate and silicate are captured there appears to be a reduction in total concentrations over the period assessed. Important notice: Further evaluation of product NORTHWESTSHELF_REANALYSIS_BIO_004_011 in March 2015 has revealed that there are substantial inter-annual drifts in the nutrient fields throughout the reanalysis period. Because the reanalysis was run in four sections [January 1984 -> March 1989; April 1989 -> December 2003; January 2004 -> June2014] and re-initialised at the start of each period, this also implies that there are temporal discontinuities at the boundaries between the four periods. These drifts and discontinuities are likely to have impacts on other parts of the biogeochemical system, which may be important in a number of user applications. Therefore users are strongly advised to contact the Service Desk for advice before embarking on projects using the biogeochemistry product NORTHWESTSHELF_REANALYSIS_BIO_004_011. It is intended that these issues will be resolved at the next release of this product. I.3 Estimated Accuracy Numbers Variable Metrics Units Decimal places temperature mean bias, standard deviation, RMS °C 2 salinity mean bias, standard deviation tidal elevation harmonic constants RMS cm / degree currents mean m s-1 transport mean, standard deviation Sv 2 heat flux mean W m-2 1 water flux mean kgm-2s-1 1 Mixed layer depth mean m 0 2 2/1 II PRODUCTION SUBSYSTEM DESCRIPTION Production centres name: Met Office, UK Production subsystem name: North West European Shelf Reanalysis physics and biogeochemistry (CMEMS names: NORTHWESTSHELF_REANALYSIS_PHYS_004_009 and NORTHWESTSHELF_REANALYSIS_BIO_004_011) Description The North West shelf reanalysis was produced using the Forecasting Ocean Assimilation Model 7km Atlantic Margin model (FOAM AMM7) which is comprised of version 3.4 of the Nucleus for European Modelling of the Ocean (NEMO) ocean model code (Madec, 2008) coupled to the European Regional Seas Ecosystem Model (ERSEM; Blackford et al. 2004). The model is located on the European North-West continental Shelf (NWS), from 40°N, 20°W to 65°N, 13°E, on a regular lat-lon grid with 1/15° latitudinal resolution and 1/9° longitudinal resolution (approximately 7km square). The model domain is shown in Figure 1 partitioned into shallow (on shelf) and deeper (off shelf) waters. Although the domain extends beyond the shelf to include some of the adjacent North-East Atlantic, the focus of this system is on the shelf itself and the deep water is primarily included to ensure there is appropriate cross-shelf exchange. Thus a hybrid s-sigma terrain following coordinate system (following Siddorn and Furner, 2013) with 51 levels is employed in order to retain vertical resolution on the shelf. However, in order to make analysis and visualization easier for users, the products are delivered on 24 geopotential (z-level) vertical levels based upon the ICES standard depths. Tidal forcing is included both on the open boundary conditions via a Flather radiation boundary condition (Flather, 1976) and through the inclusion of the equilibrium tide. The external elevation and depth mean velocity was determined from 15 tidal constituents taken from a tidal model of the north-east Atlantic (Flather, 1981). With the exception of the Baltic Sea, the model was forced at its open boundaries by temperature, salinity, sea surface height and depth integrated currents taken from global ocean analyses. For the early part of the reanalysis (January 1984 – March 1989) horizontal fluxes were taken from the National Oceanography Centre’s ORCA025 hindcast (Megann et al., 2014), before switching to GloSea data (MacLachlan et al., 2014) thereafter. For the period 07/2012-06-2014 and updated GLOSEA has been used since the previous run did not cover the new period (the Mean SSH for the NWS domain in GLOSAEA was adjust so that they both had the same mean on July 1st 2012 thus in the mean the SSH should not have a jump though of course spatially there will be difference). Baltic boundary conditions were taken from IOW-GETM model (see http://getm.eu/index.php?option=com_content&task=view&id=109&Itemid=42) and from June 2014 from CMEMS-BALTICSEA-ANALYSIS_FORECAST_PHYS_003_006. Freshwater inflow into the model from rivers and other sources was prescribed from the E-HYPE hydrological reanalysis (Donnelly, 2013) till 2012 and then from the updated version of E-HYPE (http://hypeweb.smhi.se). Surface forcing of 3 hourly precipitation, wind stress, pressure and shortwave fluxes were taken from the ECMWF ERA-Interim reanalysis (Dee et al., 2011). These fluxes were then processed through the CORE bulk forcing algorithms (Large and Yeager, 2004; Large and Yeager 2009) before being applied to the model. The inverse barometer effect of atmospheric pressure gradients on the sea surface height was also included. The light attenuation scheme is the RGB scheme within NEMO. A non-linear free surface was implemented using a variable volume layer method. The short time scales associated with tidal propagation and the free surface require a time splitting approach, splitting modes into barotropic and baroclinic components. The bottom boundary condition includes a log layer representation and a k-epsilon turbulence scheme is implemented with a generic length scale (Umlauf and Burchard, 2003). The model uses a non-linear free surface, and an energy and enstrophy conserving form of the momentum advection. It uses free slip boundary conditions. The tracer equations use a TVD advection scheme (Zalesak, 1979). For tracer diffusion a Laplacian diffusion scheme on geopotential surfaces is applied, whereas for momentum diffusion a mixed Laplacian/bilaplacian scheme is used, with the Laplacian operator applied on geopotential surfaces and the bilaplacian on model surfaces. For the reanalysis, assimilation of Sea Surface Temperature (SST) was performed using a 3DVar algorithm. Calculation of the assimilation increments was done using an adapted version of the NEMOVAR (Mogensen, 2012) system. Assimilation proceeded in three steps. Firstly a one day model forecast was performed, within which observations were compared to model output at the nearest time-step; this is a First Guess at Appropriate Time (FGAT) system. In the second stage, observation minus model differences were converted to SST increments by minimising a 3DVar cost function. In minimising this function seasonally varying estimates of the observation representativity error variance (assumed uncorrelated) and background error variance were used. The total observation error variance was obtained by adding an estimate of the measurement error variance of each observation to the representativity error variance. Information from observations was spread horizontally according to lengthscales that are inversely proportional to the potential vorticity gradient and have a maximum value of 130km. In the final stage the analysis was produced by rerunning the model for the same day with the increments added onto the SST field using the Incremental Analysis Update (IAU, Bloom et al., 1996) method. Increments were added into the model down to the base of the instantaneous mixed layer, where the mixed layer depth was determined by a 0.2°C temperature difference from the surface. From January 1984 until October 1995 the reanalysis assimilated NOAA-AVHRR data obtained from the Pathfinder Vn5.2 dataset (Casey et al., 2010). After October 1995 the NOAA-AVHRR feed was switched to use the ESA-CCI (Merchant et al.,2014) version of this data; this continued until February 2010 after which a GHRSST version (see http://www.ghrsst.org) of NOAA-AVHRR data, held internally at the UK Met-Office was used. ATSR data from the CCI project was assimilated from April 1991 until February 2010, with UKMO data used thereafter. Also from the CCI project, METOP-AVHRR data were assimilated from December 2006, switching to GHRSST feeds in February 2010. GHRSST data from the AMSRE and SEVIRI instruments were assimilated from August 2006, with the AMSRE instruments failing in November 2011. In-situ SST data were assimilated throughout the entire reanalysis run, using ICOADS (see http://icoads.noaa.gov/) data for the majority (January 1984 until December 2010) and data available through the Global Telecommunications System (GTS, see (http://www.wmo.int/pages/prog/www/TEM/GTS/index_en.html) thereafter. All data were quality controlled against the Ocean SST and sea Ice Analysis (OSTIA; Donlon et al., 2012) using the method of Ingleby & Huddleston (2007). The ecosystem model used World Ocean Atlas 2009 nutrients at the boundaries and river concentration climatology for nutrients and sediment concentration. The ecosystem model is forced by the physical model via an online coupling and is run at the same time step as the physical model, (300s). Both the reanalysis and control runs were completed in four segments, reflecting changes in forcing data. For the reanalysis the runs are divided into Run 1: 01/01/1984 to 31/03/1989 Run 2: 01/01/1989 to 31/12/2003 Run 3: 10/10/2003 to 30/06/2012 Run 4: 01/05/2012 to 30/06/2014 and for the control run Run 1: 01/01/1981 to 31/03/1989 Run 2: 02/01/1989 to 30/03/2003 Run 3: 01/01/2003 to 30/06/2012 Run 4: 01/05/2012 to 30/06/2014. In this document, the reanalysis product NORTHWESTSHELF_REANALYSIS_PHYS_004_009 (labelled as NEMO-reanalysis) is assessed for 01/01/1984 to 30/06/2014. In addition, the associated control run (labelled NEMO-control), for the same dates, and the previous hindcast, POLCOMS_REANALYSIS covering 01/01/1964 to 31/12/2004 from the Proudman Oceanographic Laboratory Coastal Ocean Modelling Scheme (POLCOMS), are also evaluated. Where the NEMO runs overlap, data are taken from the earlier run until the end of that run when the assessment switches to the next run in the series. The biogeochemical product NORTHWESTSHELF_REANALYSIS_BIO_004_011 has been assessed for the middle section only, 03/01/1989 to 31/12/2003. Users of the biogeochemistry product NORTHWESTSHELF_REANALYSIS_BIO_004_011 should be aware that, owing to inter-annual drifts in the model, there may be temporal inconsistencies at the join points between these sections as outlined in I.2 au-dessus. Bathymetry on the European NorthWest Shelf Bathymetry off of the European NorthWest Shelf Figure 1: FOAM AMM7 bathymetry (m) showing (left) the domain on the European NorthWest Shelf (defined here as total depth less than 200m) and (right) the domain off the shelf. III VALIDATION FRAMEWORK The products assessed are temperature, salinity, currents, tidal elevation, mixed layer depth, chlorophyll and nutrients. The model data are compared with climatological temperature fields from the World Ocean Atlas 2009 (WOA09, http://www.nodc.noaa.gov/OC5/WOA09/pr_woa09.html) to give maps at different vertical levels; with Pathfinder V5 Advanced Very High Resolution Radiometer (AVHRR) sea surface temperatures integrated over the model domain; with in-situ observations from the World Ocean Database 2013 (WOD13, http://www.nodc.noaa.gov/OC5/SELECT/dbsearch/dbsearch.html), averaged through vertical layers; and with mooring data time series at Cypris (Isle of Man), the Western Channel Observatory (L4, E1) and from the Harwich-Rotterdam ferry. Model salinities are compared to climatological fields from WOA09 to give maps at different vertical levels; with in-situ observations from WOD13, averaged through vertical levels; and with mooring data time series. Maps of time-averaged near surface (15m deep) model currents are compared to NOAA AOML drifter climatology (http://www.aoml.noaa.gov/phod/dac/drifter_climatology.html) and volume fluxes through the NOOS sections in the North Sea are compared to published values, where available. Harmonic constants of tidal elevations for the main tidal constituents are compared to observed values for NEMO-control. variable Description Observations Class Metrics temperature Time integrated means at depth levels; averages through depth layers; time series at mooring locations WOA09 climatology, WOD13 in-situ data, Pathfinder V5 AVHRR SST; moorings at Cypris, Western Channel Observatory (L4, E1) and Harwich-Rotterdam ferry 1, 2, 3 & 4 maps, time series, correlation, RMS errors salinity Time integrated means at depth levels; averages through depth layers; time series at mooring locations WOA09 climatology, WOD13 in-situ data, moorings at Cypris, Western Channel Observatory (L4, E1) and Harwich-Rotterdam ferry 1, 2, 3 & 4 maps, time series, correlation currents Near surface (15m) zonal and meridional current climatologies NOAA AOML drifter climatology; 1 maps transport Transports through NOOS sections in the North Sea Otto et al., 1990; Prandle et al., 1996 3 time series, mean values and standard deviations tidal constants Mixed depth Maps and statistics of errors layer Daily/monthly Kara MLD Tide gauge data from the British Oceanographic Data Centre 1 RMS errors MLD from EN4 temperature and salinity profiles. 1, 4 Maps, RMS and percent difference Chlorophyll Chlorophyll concentrations from top model layer Surface chlorophyll concentration from fluorometry at Western Channel Observatory (L4); GlobColour satellite product 1, 2, 3 & 4 Time series, RMS error, bias, maps. Nutrients Nitrate, silicate and phosphate concentrations from top model layer World Ocean Atlas 2009 climatology 1 maps Table 1: Summary of metrics and observations used in the assessment Sea level is not provided as part of this reanalysis product and therefore has not been validated in this document. The reason for excluding sea level is that it is not possible to produce a meaningful sea level trend product from this reanalysis because the model mean sea level is a product of the open boundary conditions. There are four distinct model data sources used for open boundary conditions each with their own reference sea level, internal model drifts and distinct long term trends. Thus it would be inappropriate to infer any long term sea level trend from the reanalysis as the signal will be dominated by the differences in the three distinct forcing datasets. The biogeochemical variables assessed are chlorophyll and nutrients. The model data are compared with satellite imagery for chlorophyll as well as in-situ measurements and climatological nutrient fields from the World Ocean Atlas 2009 (WOA09, http://www.nodc.noaa.gov/OC5/WOA09/pr_woa09.html). IV VALIDATION RESULTS IV.1 Temperature Monthly mean domain averaged sea surface temperatures (SSTs, Figure 2) show broad agreement between the three simulations considered and AVHRR SST. Mean values over the period shown are NEMO-control 11.66°C; NEMO-reanalysis 11.74°C; POLCOMS 12.11°C and AVHRR 12.11°C. The affect of the assimilation in NEMO is to reduce the annual range. Model SSTs are generally lower in the winter than the AVHRR values. Figure 2: Domain averaged SST time series for the model simulations and Pathfinder V5 AVHRR data (top); also shown is the difference between the NEMO-reanalysis and NEMO-control runs (bottom). The net downwards heat flux (Figure 3) integrated over the domain for 01/01/1984 to 30/06/2012 gives long term mean values of -23.0 Wm-2 for NEMO-control and -25.1 Wm-2 for NEMO-reanalysis. For comparison, net mean fluxes from ERA-40 reanalysis range from -10 to +10 Wm-2 in the southern North Sea, Irish and Celtic Seas and the Bay of Biscay to -70 to -50 Wm-2 south of Iceland (Kållberg et al., 2005). Figure 3: Domain averaged time series of net downwards heat flux (top) and the difference between the NEMO reanalysis and NEMO control runs (bottom). The modelled temperature fields at 0, 10, 30, 75, 300, 800 and 2000m are compared to World Ocean Atlas 2009 (WOA09) climatology (Figure 4-Figure 6). The model data are interpolated to the climatology vertical levels and time integrated for all whole years in the simulation (1984-2013 for NEMO and 1964-2004 for POLCOMS). Generally, differences between the models and the climatology are less than 2°C. All simulations underestimate the climate temperature in the Norwegian Sea; NEMO-reanalysis shows a marked improvement over NEMO-control in this region down to the 75m level and is also an improvement on the POLCOMS result. Except for the POLCOMS simulation at 800m, which is colder than the climatology in the south and west of the region, temperatures are too warm near the western boundary at all levels in all models. In-situ data from the World Ocean Database 2013 (WOD13) are used to give contemporary temperature comparisons (Figure 7-Figure 9). Daily mean model data for the days of the observations are interpolated to the location of the measurement. The resulting model – observation biases are averaged into layers for the whole of the dataset, including the first six months of 2014 for the NEMO simulations. Biases between the models and WOD13 tend to be smaller than between the models and climatology. Mean errors in the layers down to 300m for NEMO-reanalysis tend to be less than 0.5°C and are smaller than NEMO-control. POLCOMS tends to be too cold down to 300m. All models show a positive bias in the surface layer in the North Sea and are too cold in the Bay of Biscay between 300 and 2000m. Figure 4: Long term annual mean differences between NEMO-control temperature (1984-2011) and WOA09 at different depth levels (model-WOA09). Figure 5: Long term annual mean differences between NEMO-reanalysis temperature (1984-2011) and WOA09 at different depth levels (model-WOA09). Figure 6: Long term annual mean differences between POLCOMS temperature (19642004) and WOA09 at different depth levels (model-WOA09). Figure 7: Mean biases between co-located values of NEMO-control temperatures (1984-2014) and WOD13 observations in different layers (model-WOD13). Figure 8: Mean biases between co-located values of NEMO-reanalysis temperatures (1984-2014) and WOD13 observations in different layers (model-WOD13). Figure 9: Mean biases between co-located values of POLCOMS temperatures (1984-2004) and WOD13 observations in different layers (model-WOD13). Time series of the monthly mean biases between the models and WOD13, integrated over the whole domain into difference depth layers (Figure 10) show a general improvement in the model temperatures from POLCOMS to NEMO-control to NEMO-reanalysis. There is no evidence of an increase in the standard deviation of the biases (Figure 10) as time progresses. Figure 10: Monthly mean biases and standard deviations of the biases for model temperature compared to WOD13 observations, averaged over different depth layers. Time series of the mean temperatures integrated over the whole domain into different depth layers (Figure 11) show a general agreement between the model results in the surface layers down to 30m, although the annual range of NEMO-reanalysis is smaller than NEMO-control. Between 80 and 300m, NEMO-reanalysis temperatures exceed NEMO-control. Below 300m POLCOMS is ~1°C colder than the NEMO simulations, with the NEMO temperatures being in better agreement with climatology (Figure 4-Figure 6) and in-situ observations ( Figure 7-Figure 9). Figure 11: Domain averages of temperature monthly fields as a function of time in different layers. Using whole years of simulations only, there is a generally positive trend in SST (Figure 12) for the region. The main exception is in NEMO-control east of Iceland and in the Faroe-Shetland Channel where temperature decreases over 1984-2013. In the NEMO-reanalysis simulation this changes to a generally positive trend, which is more like the AVHRR data trend. NEMO-reanalysis trends are generally higher than for POLCOMS (these are over different time periods), but are still lower than those derived from AVHRR data. Figure 12: Maps of SST trend from the NEMO-control, NEMO-reanalysis and POLCOMS simulations and from Pathfinder V5 AVHRR data (note the different colour scale). The model temperatures are compared to four time series of observations: water column data from Cypris Station (4.833°W, 54.092°N) from the Isle of Man Long-term Environmental Time Series from 1984 to 2009, collected at between weekly and monthly intervals, downloaded from the British Oceanographic Data Centre (BODC), water column moorings E1 (4.367°W, 50.033°N) and L4 (4.217°W, 50.250°N) in the Western Channel Observatory collected by Plymouth Marine Laboratory at between weekly and monthly intervals and downloaded from http://www.pangaea.de; data are available for 1984, 1985 and 2002 to 2011 for E1 (Smyth et al., 2012) and 1988 to 2011 for L4 (Smyth et al., 2011), data for 2012-2014 were downloaded from http://www.westernchannelobservatory.org.uk, sea surface data from the Harwich to Rotterdam Ferry between 1.416°E, 52°N and 3.916°E, 52°N collected by Cefas at weekly frequency and published as monthly mean values (http://www.cefas.defra.gov.uk/) for 1984 to 2012. For comparison with Cypris, E1 and L4 data, model temperatures are extracted co-located with the date and location of each observation then monthly mean values are calculated and compared (Figure 13-Figure 15). For the Harwich-Rotterdam ferry, monthly mean model temperatures are used for comparison (Figure 15). At Cypris, the NEMO-reanalysis temperatures are ~0.15-0.2°C too cold down to 10m, while NEMO-control is ~0.1° too cold. At E1, there is good agreement between NEMO and observations above 20m but, below that depth, the model is ~0.6°C too warm. At L4, the NEMO results are ~ 0.15-0.45°C too warm. Compared to the Harwich-Rotterdam ferry data, NEMO-control is ~0.3-0.8°C too cold and NEMO-reanalysis is ~0.5-0.9°C too cold. Correlations between the mooring data and the model temperature are greater than 0.93 at all points. Figure 13: Mean values of temperature profile at Cypris Station, and correlations and standard deviations of monthly mean values. Figure 14: Mean values of temperature profile at station E1 in the English Channel, and correlations and standard deviations of monthly mean values. Figure 15: Mean values of temperature profile at station L4 in the English Channel, and correlations and standard deviations of monthly mean values. Figure 16: Mean values of temperature along the Harwich-Rotterdam ferry route at 52°N, and correlations and standard deviations of monthly mean values. The innovation bias and RMS for the models compared to the assimilated data (Table 2) show the improvement in the bias and RMS between the NEMO-control and NEMO-reanalysis simulations. period dataset NEMO-reanalysis Bias (°C) NEMO-control RMS (°C) Bias (°C) RMS (°C) 01/01/1984 – 31/03/1989 AVHRR 0.02 0.69 -0.13 0.86 in-situ 0.06 0.79 -0.03 0.91 01/01/1989 – 31/12/2003 AVHRR 0.04 0.56 0.06 0.80 AATSR 0.05 0.50 0.04 0.73 in-situ 0.05 0.55 0.12 0.81 AVHRR 0.05 0.50 0.17 0.79 AATSR 0.07 0.46 0.19 0.74 METOP 0.04 0.49 0.20 0.75 in-situ 0.03 0.46 0.25 0.84 AVHRR 0.08 0.46 NA NA AATSR NA NA NA NA METOP -0.00 0.57 NA NA In-situ -0.01 0.56 NA NA 10/10/2003 – 31/06/2012 01/05/201231/06/2014 Table 2: Innovation bias (observation – model) and RMS for surface temperature integrated over the whole domain, divided into the three legs of the NEMO-reanalysis simulation. IV.2 Salinity Monthly mean domain averaged sea surface salinities (SSS, Figure 17) show broad agreement between NEMO-control and NEMO-reanalysis, although NEMO-reanalysis results are fresher (mean is 34.91 compared to 34.94). POLCOMS is much fresher with a mean value of 34.77 over 1984-2004. Domain averaged net upwards water fluxes (Figure 18) for 01/01/1984 to 30/06/2012 are -9.6×10-6 kg m-2 s-1 for NEMO-control and -9.3×10-6 kg m-2 s-1 for NEMO-reanalysis. Figure 17: Domain averaged SSS time series for the model simulations (top); also shown is the difference between the NEMO-reanalysis and NEMO-control runs (bottom). Figure 18: Domain averaged time series of net upward water flux for the model simulations (top); also shown is the difference between the NEMO-reanalysis and NEMO-control runs (bottom). The modelled salinity fields at 0, 10, 30, 75, 300, 800 and 2000m are compared to WOA09 climatology (Figure 19-Figure 21). The model data are interpolated to the climatology vertical levels and time integrated for all whole years in the simulation (1984-2013 for NEMO and 1964-2004 for POLCOMS). Generally, differences between the models and the climatology are less than 0.5, with the models being too fresh. All models are too saline (up to 1 too salty) in the eastern North Sea in the surface layers and too fresh in the Bay of Biscay at 800m. In-situ data from WOD13 are used to give contemporary salinity comparisons (Figure 22-Figure 24). Daily mean model data for the days of the observations are interpolated to the location of the measurement. The resulting model – observation biases are averaged into layers for the whole of the dataset. In the NEMO simulations, salinity is too low in several coastal regions including west of Scotland and the Southern Bight of the North Sea and also in the Norwegian Trench; the surface layer is too saline in the Irish Sea and west and south of the Norwegian Trench, although differences here are smaller than in the model-climatology comparison indicating that the observed salinity over the averaging period differs from the long term climatology. The main feature of the salinity comparison for POLCOMS is that the modelled water around the UK coast, east of Denmark and in the surface layers to 30m in the Norwegian Trench are too salty. Figure 19: Long term annual mean differences between NEMO-control salinity (1984-2013) and WOA09 at different depth levels. Figure 20: Long term annual mean differences between NEMO-reanalysis salinity (1984-2013) and WOA09 at different depth levels. Figure 21: Long term annual mean differences between POLCOMS salinity (1964-2004) and WOA09 at different depth levels. Figure 22: Mean biases between co-located values of NEMO-control salinities (1984-2014) and WOD13 observations in different layers. Figure 23: Mean biases between co-located values of NEMO-reanalysis salinities (1984-2014) and WOD13 observations in different layers. Figure 24: Mean biases between co-located values of POLCOM salinities (1984-2004) and WOD13 observations in different layers. Time series of the mean biases between the models and WOD13, averaged into monthly means and integrated over the whole domain into different depth layers (Figure 25) show a general improvement in the model salinity from POLCOMS to NEMO. There is little difference between NEMO-control and NEMO-reanalysis. There is no evidence of an increase in the standard deviation of the biases (Figure 25) as time progresses. Figure 25: Mean biases and standard deviations of the biases of model salinity compared to WOD13 observations, averaged over different depth layers. Time series of the mean salinity integrated over the whole domain into different depth layers (Figure 26) show that generally NEMO-reanalysis is fresher than NEMO-control and POLCOMS is fresher than both NEMO runs. One exception is in the 300-800m layer from 1990 onwards when the salinity in the NEMO simulations reduces by ~0.1. The NEMO salinity values are in better agreement with observations (Figure 25) than the POLCOMS results. Figure 26: Domain averages of salinity monthly fields as a function of time in different layers. The trends in SSS in NEMO-reanalysis and NEMO-control are similar (Figure 27). Away from the coastal regions, the trend is towards decreasing salinity, which is especially large in the Norwegian Sea ~-0.02 yr-1. In POLCOMS the trend is an increase in salinity in the Norwegian Sea but a reduction elsewhere, including in coastal regions. Figure 27: Maps of SSS trend from the NEMO-control, NEMO-reanalysis and POLCOMS simulations. As for the temperature data, the model salinities were compared to four time series of observations at Cypris (surface salinity only), E1 and L4 in the English Channel and along 52°N on the Harwich to Rotterdam ferry. At Cypris, NEMO is more saline than the observations at the surface, with a mean value for NEMO-control of 34.68 and for NEMO-reanalysis of 34.55, while the observations have a mean of 34.17; for both NEMO-control and NEMO-reanalysis, correlations between the monthly mean model and observed values are 0.7 and standard deviations of the bias are 0.1. At E1 (Figure 28) below 30m, NEMO-control has good agreement with the observed mean profile; near the surface the model is too saline. Near the surface, NEMO-reanalysis is a better approximation to the observations but is too fresh throughout the profile. At L4 (Figure 29), except at the surface where both models are too saline compared to the mooring observation, the mean profile from NEMOcontrol is too saline whilst NEMO-reanalysis is close to the observed but too well mixed in the vertical. On the Harwich-Rotterdam ferry section (Figure 30), the models are on average too saline by ~0.4 in the west and too fresh by ~1.9 in the east. Correlations between monthly mean model and observed salinities are low at the English Channel stations E1 and L4, with values of up to 0.5 at E1 but negative correlations at L4. Correlation with the ferry data is higher with values ranging from 0.3 to 0.8. Figure 28: Mean values of salinity profile at station E1 in the English Channel, and correlations and standard deviations of monthly mean values. Figure 29: Mean values of salinity profile at station L4 in the English Channel, and correlations and standard deviations of monthly mean values. Figure 30: Mean values of salinity along the Harwich-Rotterdam ferry route at 52°N, and correlations and standard deviations of monthly mean values. IV.3 Currents Near surface climatologies for two components of currents averaged over all complete years of the simulations (Figure 31) show agreement between the model data sets, all models showing the clockwise circulation around the north of the Faroe Islands and southwards into the Faroe-Shetland Channel; the northwards flow on the Shetland side of the Faroe-Shetland Channel turning south into the North Sea; and the Norwegian Coastal Current. These features are in broad agreement with the NOAA AOML drifter climatology. The POLCOMS currents have stronger eddy fields in the Bay of Biscay compared to the NEMO simulations. NEMO-control (1984-2013) NEMO-reanalysis (1984-2013) POLCOMS (1964-2004) NOAA AOML drifter climatology Figure 31: Near surface mean currents for NEMO-control, NEMO-reanalysis and POLCOMS, also NOAA AOML drifter climatology. IV.4 Tidal analysis For sea level elevations, harmonic analysis of the dominant tidal constituents is compared against observations. Figure 32 show the amplitude and phase errors for M2. Figure 32 shows the comparison spatially. The amplitude is underestimated in the Irish Sea and Southern Bight and overestimated in large tidal areas such as the Bristol Channel, as also seen in Figure 33. The phase error is particularly large in the Southern Bight and German Bight. The RMSE for amplitude and phase for the constituents M2, S2, K2, N2, K1, P1 and O1 are shown in Table 3. Overall the tides are reasonably well represented but caution must be applied particularly with regards to tidal phase in areas such as the Southern and German Bights. Figure 32: M2 amplitude (left) and phase (right) error for NEMO-control run. Figure 33: M2 amplitude and phase versus observations Constituent RMSE Amp. (cm) RMSE Phase (Deg) M2 11.60 15.6 S2 4.50 15.8 K2 1.46 13.1 N2 3.39 20.7 K1 2.10 18.8 P1 0.99 19.7 O1 1.91 19.9 Table 3: RMSE amplitude and phase of control run tidal constituents against observations. IV.5 Transport Mean transports though the NOOS sections in the North Sea (Figure 34) are compared to available observations (Table 4). All models simulate the inflow to the North Sea through the Shetland North section in good agreement with the observation (Otto et al., 1990). The outflow from the North Sea through the Sojgnesjoen section is also simulated with the NEMO results being slightly smaller than POLCOMS and the observed value. In agreement with observations, the models show an inflow to the North Sea through the Shetland South section, with the POLCOMS transport being larger than the observed and the NEMO transports being smaller than observed. The NEMO transport into the North Sea through the Dover Strait is in good agreement with observations (Prandle et al., 1996), while POLCOMS has a weak outflow here. For eight of the NOOS sections, the net flux through each section (Figure 35 and Figure 37) and two components of transport in both directions normal to the section (Figure 36 and Figure 38) for POLCOMS and NEMO-reanalysis are shown; NEMO-control being similar to NEMO-reanalysis is not shown. Figure 34: Locations of NOOS sections for volume flux calculations. no section positive flux direction observed mean (Sv) NEMO-control mean (Sv) std (Sv) NEMO-reanalysis mean (Sv) std (Sv) POLCOMS mean (Sv) std (Sv) 1 Shetland North N -0.6 -0.52 0.21 -0.55 0.22 -0.55 0.21 2 Sojgnesjoen N 1.8 1.25 0.50 1.28 0.50 1.61 0.44 3 Shetland South NW -0.3 -0.11 0.04 -0.11 0.04 -0.48 0.22 4 Orkney N -0.93 0.37 -0.94 0.36 5 Utsira N 1.10 0.43 1.10 0.42 6 Lista NW 0.36 0.24 0.35 0.24 7 Aberdeen N -0.33 0.18 -0.33 0.18 8 Hantsholm N 0.33 0.22 0.32 0.22 9 Okso NE -0.01 0.01 -0.01 0.01 10 Flamborough Head N -0.01 0.06 -0.01 0.06 11 Terschelling NE 0.14 0.09 0.14 0.09 12 Noordwijk N 0.06 0.06 0.06 0.06 13 Dover Strait NE 0.08 0.06 0.07 0.06 -0.01 0.06 14 Cherbourgh E 0.08 0.06 0.07 0.06 15 Plymouth E -0.05 0.07 -0.05 0.07 -0.10 0.49 16 Rosslare N 0.05 0.06 0.05 0.06 -0.08 0.08 17 Dublin-Holyhead N 0.09 0.06 0.08 0.06 -0.01 0.08 18 Larne N 0.06 0.05 0.06 0.05 -0.00 0.08 19 Rottumerplaat E 0.10 0.07 0.10 0.07 20 Sylt N 0.08 0.06 0.08 0.06 21 Torungen-Hirtshals NE 0.00 0.01 0.00 0.01 22 Skagerrak E -0.01 0.01 -0.01 0.01 -0.20 0.08 23 Kategatt N -0.01 0.00 -0.01 0.00 0.094 Table 4: For the NOOS sections (Figure 34) means and standard deviations of monthly mean volume fluxes from 01/01/1984 to 30/06/2012 for the NEMO control and reanalysis simulations and 01/01/1964 to 31/12/2004 for the POLCOMS simulation (for available sections). Volume fluxes derived from observations are from Otto et al. (1990) for sections 1, 2 and 3 and from Prandle et al. (1996) for section 13. The directions of positive flux are denoted by N – northwards, E – eastwards, W – westwards, NE – northeastwards and NW – northwestwards. Figure 35: Monthly averaged net transports through selected NOOS sections (Figure 34) for the POLCOMS simulation, in Sverdrups. Figure 36: Two components of monthly averaged transports through selected NOOS sections (Figure 34) for the POLCOMS simulation, in Sverdrups. Figure 37: Monthly averaged net transports through selected NOOS sections (Figure 34) from the NEMO-reanalysis simulation, in Sverdrups. Figure 38: Two components of monthly averaged transports through selected NOOS sections (Figure 34) for the NEMO-reanalysis simulation, in Sverdrups. IV.6 Mixed layer depth V A note on the data processing The CMEMS NWS RAN output a modified version of the Kara MLD (Kara et al. 2000), using a with the reference depth at 3m instead of 10m. These were available as 25 hourly means (for each day 25 hourly instantaneous MLD estimates were averaged, including the midnight at the beginning and end of the period). This was done to approximately remove the tidal cycle. These are released as the daily product. This is a less appropriate strategy for when averaging over the month, when the tidal cycle is arguably less important, as it includes a “night-time” bias. However, 24 hourly mean files were not output for by this version of the RAN, and offline calculation of the MLD was not possible, and so the monthly mean files are simply the average of the 25 hour mean files for each month. We have limited number of hourly mean MLD and so have been able to quantify this “night-time” bias. Spatial maps of the bias, rms and percent difference between the (monthly and daily) means when calculated from 25 or 24 hour means are presented in Figure 39 for an exemplar year (1990). The time series of these differences for the NWS are presented in Figure 40. Figure 39 Bias (upper row), RMS (middle row) and percent difference (bias/24hour mean; lower row) from producing monthly (left column) and daily (right column) means from daily 25 hour means rather than daily 24 hour means. Figure 40 Time-series of the 25 and 24 hour mean MLD averaged over the shelf (left), the difference between these (middle), and the percent difference (right). The upper panel shows the daily mean files (from 1st January to 31st December 1990) , and the lower panel shows the monthly mean files (from January to December 1990). Given the small scale of these quantified differences, we consider this use of daily and monthly means as calculated from 25 hour means sufficient for the NWS. VI A note on the observations In order to validate the CMEMS RAN MLD we have use MLDs calculated from the quality controlled temperature and salinity profile database, EN4 (Good et al. 2013). These MLD are similar, but different from, the Kara MLD (referred to as EN4 MLD). In order to assess whether the EN4 MLD were fit for purpose to validate the CMEMS RAN Kara MLD, we have used the same EN4 MLD algorithm to the 25 hour mean 3d temperature and salinity fields of CMEMS RAN and compared them to the CMEMS RAN Kara MLD. This was done for each day of an exemplar year (2000). For each day, we compare the two MLD products on the shelf. We present a Taylor diagram (Figure 41) to show the spatial agreement, with time series of the shelf wide MLD, bias and the time-series of the relative standard deviation and correlation (as shown on the Taylor Diagram). We find that the correlation is generally above 0.95, and the shelf mean bias is generally < 1m, and there is greatest disagreement from May to October. We then look at the spatial pattern of the bias averaged over these months (Figure 42). When looking at these maps, we find there is a good agreement in the interior of the shelf, with the greatest differences in the Shetland Shelf (north of the North Sea), the Irish Shelf (west of Ireland) and in the outer Celtic Sea. Overall we find that there is a relatively good agreement between the two methods of calculating MLD on the shelf and conclude that the EN4 MLD’s are fit for the purpose of validating the CMEMS NWS MLDs, however recommend that for later versions, a consistent processing of the EN4 MLDs is undertaken. Figure 41 Comparison of simulated observed (EN4) MLD with modelled MLD. Taylor diagram showing the spatial agreement for each day of the (exemplar) year (2000), with time series of shelf-wide mean (modelled and observed) MLD, bias and Taylor statistics (relative standard deviation and correlation). Figure 42 Comparison of simulated observed (EN4) MLD with modelled MLD, averaged from May – October (inclusive), showing the mean modelled and observed MLD fields, the model-observed bias and the percent difference. VII MLD Evaluation Given the use of 25 hour means, and the EN4 MLD, and the relative paucity observations to evaluate the CMEMS NWS RAN MLD, we now make a careful evaluation. For each EN4 observed MLD, we extract the modelled MLD from the same grid-box, and same day (termed a data pair). The EN4 MLD data does not return a value when the water column is mixed, and so there are no observation-model MLD data pairs regions that are observed to be mixed (such as the southern North Sea and English Channel). From Figure 41 we identify the predominantly stratified months as being May – October, and so we focus our validation of this period. We initially compare the spatial patterns of the co-located modelled and observed MLD for the stratified months (May – October: Figure 43). It is clear from Figure 43 that there is much greater data density in the North Sea, and outer shelf regions, than in the Celtic and Irish Sea, and in the English Channel. There are no clear spatial patterns in the bias or rms. There is a slightly more positive modelobservations bias (model MLD too deep) in the eastern North Sea compared to the rest of the domain, and a negative (model MLD too shallow) in the sparsely sampled Celtic Sea. Figure 43 Spatial patterns of the co-located modelled and observed MLD for the stratified months (May – October). For each EN4 observed MLD, the modelled MLD for the same day and grid box is extracted. The average for these data (between May and August) for each grid box is shown here (upper left and right respectively). Below, we show the statistics of these data (for each grid box): the mean bias (model minus observation), rms and percentage difference (model minus observation bias divided by observation), lower row, from left to right respectively. We then look at all the model observation MLD data pairs (in terms of model minus observation bias), and plot as a function of time and seasonal cycle (Figure 44). The data appears to have stationarity (in terms of the mean and variance of the bias) over the duration of the CMEMS NWS RAN, with no apparent trend in either the mean or the variance. There is a clear seasonal cycle in the number of observations, although this may be accounted for the exclusion of observed mixed profiles. The biases appear to be positively skewed, and so the apparent mode of the data (e.g. ~-10 m - 0 m between day-number 150 – 250) does not necessarily reflect the mean. Figure 44 Timeseries of all model-observations bias pairs of data, plotted as a function of year, and day of year (upper left and right panels respectively). The lower panels show the same as the panels above, but zoomed in to give mode detail. The colouring gives a qualitative indication of data density. We now look at the model observation MLD data pairs (from the shelf) in a more quantitative way. For each month, we calculate the Taylor statistics (correlation, relative standard deviation, rms), the mean bias, mean observed and modelled MLD, and the number of data pairs (Figure 45). The Taylor diagram shows, with the exception of June – August, most correlations are greater than 0.6, and the model variance is generally greater than that of the observations (relative standard deviation > 1.). The months of June-August have much lower correlations (0.586, 0.442 and 0.361 respectively), but have similar modelled and observed variance, and much lower model minus observations biases. Figure 45 Quantitative analysis of the observed (EN4) and modelled (amm7) MLD data pairs. The Taylor diagram (left panel) shows agreement of the spatial patterns of modelled and observed MLD. The panels on the right hand side show the seasonal cycle of the mean MLD (from the sparse modelled and observed data pairs), the model minus observations bias, the correlation and relative standard deviation (repeating the data from the Taylor diagram) and the number of data pairs. When we look at individual validation regions (the North Sea, the Celtic seas (the Celtic and Irish Seas and the English Channel) and the outer shelf (the Irish and Shetland shelf)) for the stratified months (Figure 46), it can be seen that while the correlations are particularly low for the Celtic Seas relative to the North Sea, there are far fewer observations, and so the North Seas correlations are fairly similar to that of the whole shelf. Figure 46 Model (amm7) observation (EN4) data pairs, for May to October (from left to right), for the 3 main validation regions: the North Sea; the Celtic seas (the Celtic and Irish Seas and the English Channel); and the outer shelf (the Irish and Shetland shelf). The colour of the points show the data density. Each panel include the relative standard deviation (rsd), correlation (r), rms, correlation significance (p), the mean bias, and the number of data points (cnt). VII.1 Chlorophyll The model output was compared to observations from the Western Channel Observatory L4 buoy (Figure 47). This data consists of approximately weekly measurements from the surface of chlorophyll concentration derived from fluorometric techniques from 1992 until 2003. The model is capturing the timing of the seasonal variation in surface chlorophyll well and also the magnitude of winter minimums and spring maximums although the maxima in 1997 and 1999 are under estimated. This analysis will be extended for the final section of the reanalysis and fore earlier periods where data is available. Figure 47: Surface chlorophyll concentration (mg/m3) at L4 buoy (x) and model output (-). RMS error is 1.93 and model bias is 0.24. The model output was also compared to GlobColour satellite imagery products for chlorophyll concentration available from 1998 for case 1 waters which are summarised in Table 5. The biases are generally low with the best performance in the Irish Sea and the English Channel. For all regions the model is over estimating chlorophyll producing negative biases. Figure 48 shows a time-series of the errors for the whole domain and indicates that over the period analysed there is no significant increase in errors and the model is stable. Area Name Mean error (model-obs) RMS error Full Domain -0.143 0.667 Continental shelf -0.167 0.631 Off-shelf -0.135 0.679 Norwegian Trench -0.259 0.659 Northern North Sea -0.197 0.765 Southern North Sea -0.123 0.512 English Channel -0.073 0.419 Irish Sea -0.068 0.468 South Western Approaches -0.160 0.685 North Western Approaches -0.141 0.613 Table 5: Regional statistics for log10 of chlorophyll-a (mg/m3) concentration against the Globcolour product data for the period January 1998 – June 2002 Further analysis and comparison with satellite imagery will be carried out for the later parts of the reanalysis. This analysis began with the earliest available satellite data from the SeaWifs sensor and any analysis of earlier periods will depend on the quality and coverage of available observations. Figure 48: Log10 Chlorophyll concentration (log10(mg/m3)) mean error (dotted line), RMS error (solid line) for the full model domain from January 1998 to June 2002. Figure 49: Mean surface chlorophyll for June 1998-2003. GlobColour imagery left, model output right. Figure 49 shows surface chlorophyll concentration maps of GlobColour imagery (left) compared to model output (right) for the month of June highlighting the region of over production in the northwest of the model domain. This figure also gives an indication of the inter-annual variability of the model results. VII.2 Nutrients There are few available in-situ observations available for verification so an initial comparison to World Ocean Atlas climatology is made below. Figure 50, Figure 51 and Figure 52 show World Ocean Atlas nutrient data from February when nutrient concentrations are highest and August when they are lowest. These are shown next to monthly mean output from the model from 1990 at the beginning of the run and 2003 at the end. Figure 50 shows that the February (winter) and August (summer) surface nitrate compares well to the climatological data, however it is noted that there is a decline in winter nitrate concentration between 1990 and 2003. There is also an increase in the nitrate concentration in both February and August in coastal regions and the Irish Sea. This pattern is also seen in plots of phosphate in Figure 51. Figure 52 again shows a similar trend in decreasing surface concentrations with an even more marked increase in coastal concentrations. This section of the reanalysis is the longest and nutrient drifts will be compared to the other sections in future updates. Figure 50: Surface monthly nitrate concentration (mmol/m3) for February (top row) and August (bottom row) from World Ocean Atlas (left), model output from 1990 (middle) and model output from 2003 (right). Figure 51: Surface monthly phosphate concentration (mmol/m3) for February (top row) and August (bottom row) from World Ocean Atlas (left), model output from 1990 (middle) and model output from 2003 (right). Figure 52: Surface monthly silicate concentration (mmol/m3) for February (top row) and August (bottom row) from World Ocean Atlas (left), model output from 1990 (middle) and model output from 2003 (right). VIII QUALITY CHANGES SINCE PREVIOUS VERSION The NEMO reanalysis product NORTHWESTSHELF_REANALYSIS_PHYS_004_009 has been compared to the POLCOMS hindcast product NORTHWESTSHELF_REANALYSIS_PHYS_004_005 for the Northwest European Shelf. The main changes in quality are: Temperature biases have improved throughout the water column, with the exception of the 30-80m layer in the southern North Sea. Salinity biases have larger magnitudes in the coastal regions of the southern North Sea but have improved in the Norwegian Coastal Current and around the UK coast. The eddy climate has reduced in the Bay of Biscay. Volume transports through NOOS sections have changed but still compare well with the limited observations that are available. IX REFERENCES Blackford, J.C., Allen, J.I., Gilbert, F.J., (2004). Ecosystem dynamics at six contrasting sites: a generic modelling study. Journal of Marine Sciences, 52 (1-4), 191-215 Bloom S.C.,Takacs L.L.,Da Silver A.M., Ledvina, D., (1996) Data assimilation using incremental analysis updates, Monthly Weather Review, 124, 1256-1271. Casey, K.S., Brandon T.B.,Cornillon P., Evans R., (2010). "The Past, Present and Future of the AVHRR Pathfinder SST Program", in Oceanography from Space: Revisited, eds. V. Barale, J.F.R. Gower, and L. Alberotanza, Springer. DOI: 10.1007/978-90-481-8681-5_16 Dee D.P., Uppala S.M., Simmons A.J., Berrisford P., Poli P., Kobayashi S., Andrae U., Balmaseda M.A., Balsamo G., Bauer P., Bechtold P., Beljaars A.C.M., van de Berg L., Bidlot J., Bormann N., Delsol C., Dragani R., Fuentes M., Geer A.J., Haimberger L., Healy S. B., Hersbach H., Hólm E.V., Isaksen L., Kållberg P., Köhler M., Matricardi M., McNally A.P., Monge-Sanz B.M., Morcrette J.-J, Park B.-K, Peubey C., de Rosnay P., Tavolato C., Thépaut J.-N., Vitart F., (2011) The ERA-Interim reanalysis: configuration and performance of the data assimilation system, QJRMS, 137, 553-397 Donlon C. J., Martin M., Stark J., Roberts-Jones J., Fiedler E., Wimmer W., (2012) The Operational Sea Surface Temperature and Sea Ice Analysis (OSTIA) system, Remote sensing of Environment, 116,140158 Donnelly C., Arheimer B., Capell R., Dahne J. and Strömqvist J., (2013) Regional overview of nutrient load in Europe – challenges when using a large-scale model approach, E-HYPE. Understanding freshwater quality problems in a changing world. Proceedings of IAHS-IAPSO-IASPEI Assembly, Gothenburg, Sweden, July 2013 Flather R. A. (1981). "Results from a model of the north east Atlantic relating to the Norwegian Coastal Current". The Norwegian Coastal Current (Proceedings from the...symposium, Geilo, 9-12 September 1980). Bergen: Bergen University, 2, 427-458 Flather, R. A., (1976): A tidal model of the northwest European continental shelf. Memoires de la Societe Royale de Sciences de Liege, 6, 141-164. Good, S. A., Martin, M. and Rayner, N. A. (2013). "EN4: Quality controlled ocean temperature and salinity profiles and monthly objective analyses with uncertainty estimates." Journal of Geophysical Research: Oceans 118: 6704–6716 doi:10.1002/2013JC009067. Huthnance, J. M., (1984): Slope Currents and “JEBAR”. J. Phys. Oceanogr., 14, 795–810. Ingleby B., Huddleston M., (2007) Quality control of ocean temperature and salinity profiles Historical and real-time data, Journal of marine systems, 65, 158-175 Kållberg P., Berrisford P., Hoskins B., Simmons A., Uppala S., Lamy-Thépaut S., Hine R., (2005) ERA-40 Atlas. ECMWF ERA-Project Rep. Series 19, European Centre for Medium Range Weather Forecasts, Reading Kara, A. B., Rochford, P. A. and Hurlburt, H. E. (2000). "An optimal definition for ocean mixed layer depth." Journal of Geophysical Research-Oceans 105(C7): 16803-16821. Large W, Yeager S (2004) Diurnal to decadal global forcing for ocean and seaice models: the data sets and climatologies. Technical Report TN-460+STR, NCAR, 105 Large, W., Yeager, S., (2009): The global climatology of an interannually varying air-sea flux data set. Clim. Dynamics, 33:341–364, DOI 10.1007/s00382-008-0441-3 MacLachlan, C., Arribas, A., Peterson, D., Maidens, A., Fereday,D., Scaife, A.A., Gordon, M., Vellinga, M., Williams, A., Comer, R.E., Camp, J., and Xavier, P., (2014): Global Seasonal Forecast System 5 (GloSea5): a high resolution seasonal forecast system, Q. J. Roy. Meteor. Soc. Madec G., Delecluse P., Crépon M., Lott F., (1996) Large-Scale Preconditioning of Deep-Water Formation in the Northwestern Mediterranean Sea. Journal of Physical Oceanography, 26(8), 13931408. Madec,G. (2008) NEMO reference manual, ocean dynamics component, Institut Pierre-Simon Laplace, technical report Megann et al. (2014) GO5.0: The joint NERC-Met Office NEMO global ocean model for use in coupled and forced applications, Geosci. Model Dev. , 7, 1069-1092, doi:10.5194/gmd-7-1069-2014 Merchant C.J., Embury O., Roberts-Jones J., Fiedler E., Bulgin C.E., Corlet G.K., Good S., McLaren A., Rayner N., Morak-Bozzio S., Donlon C., (2014) Sea surface temperature datasets for climate applications from Phase 1 of the European Space Agency Climate Change Initiative (SST CCI), Geoscience Data Journal, Vol. 1 , Issue 2: 179-191. DOI: 10.1002/gdj3.20 Mogensen, K. S., Balmaseda, M. A., and Weaver, A.: The NEMOVAR ocean data assimilation system as implemented in the ECMWF ocean analysis for System 4, ECMWF Tech. Memo. 668., 2012. Otto L., Zimmerman J.T.F., Furnes G.K., Mork M., Saetre R., Becker G. (1990) Review of the physical oceanography of the North Sea. Neth J Sea Res 26:161–238 particulate metals through the Dover Strait. Cont Shelf Res 16 (2):237–257 Prandle D., Ballard G., Flatt D., Harrison A.J., Jones S.E., Knight P.J., Loch S., McManus J., Player R., Tappin A. (1996) Combining modeling and monitoring to determine fluxes of water, dissolved and Siddorn, J., Furner R. (2013): An analytical stretching function that combines the best attributes of geopotential and terrain-following vertical coordinates, Ocean Modelling, 66, 1-13 Smyth, T.J., Hardman-Mountford, N., Frost, M., Fishwick, J.R. (2011) Temperature and salinity from time series station L4 in the western English Channel from 1988 to 2011. Plymouth Marine Laboratory, doi:10.1594/PANGAEA.757204 Smyth, T.J., Hardman-Mountford, N., Frost, M., Fishwick, J.R. (2012) Temperature and salinity profiles at 10 m depth resolution from time series station E1 in the western English Channel from 1903 to 2011. doi:10.1594/PANGAEA.775761 Umlauf, L., Burchard, H., (2003) A generic length-scale equation for geophysical turbulence models. Journal of Marine Research 61, 235–265. Zalesak, S.T., (1979) Fully Multidimensional Flux-Corrected Transport Algorithms for Fluids, J. Comput. Phys., 31, 335–362.