QUALITY INFORMATION DOCUMENT

(1)

QUALITY INFORMATION DOCUMENT

For Global Ocean-Delayed Mode in-situ Observations of

surface (drifters and HFR) and sub-surface

(vessel-mounted ADCPs) water velocity.

INSITU_GLO_UV_L2_REP_OBSERVATIONS_013_044

http://doi.org/10.13155/41256

Issue 2.1

Contributors: Hélène Etienne, Nathalie Verbrugge, Christine Boone, Anna Rubio, Lohitzune Solabarrieta,

Lorenzo Corgnati, Carlo Mantovani, Emma Reyes, Marina Chifflet, Julien Mader, Thierry Carval

CMEMS version scope: December 2020 release

(2)

QUID for Global Ocean-Delayed Mode in-situ Observations of surface (drifters and HFR) and

sub-surface (vessel-mounted ADCPs) water velocity INS_GLO_UV_L2_REP_OBSERVATIONS_013_044

Ref : CMEMS-INS-QUID-013-044 Date : 28 October 2020

Version : 2.1

CHANGE

RECORD

Issue Date § Description of

Change Author Checked By

1.0 14/12/2015 all First version of document

H. Etienne

1.1 17/08/2016 I.1, II.2, IV, almost all figures

Time series extended with the year 2015

H. Etienne

1.2 24/04/2017 Almost all figures

Update with new data

H.Etienne L. Petit de la Villéon 1.3 23/05/2017 all Multiple changes

under the V3 review

F. Hernandez and H. Etienne

F. Hernandez

1.4 30/10/2017 all Time series extended with the year 2016

H.Etienne

1.5 29/11/2018 all Time series extended with the year 2017 and part of 2018

H.Etienne

2.0 18.03.2020 all Addition of HF radar REP surface currents dataset A Rubio, M. Chifflet, L. Corgnati, C. Mantovani, E. Reyes, J. Mader Addition of ADCP velocity profiles time series

Update of the entire drifters time series

Thierry Carval H. Etienne 2.1 31.08.2020 §1.2.2 Update of the document concerning the radar_total dataset A. Rubio L. Solabarrieta 2.1 28.10.2020 Update of the document concerning the radar_total dataset and changes requested in the SQO

A. Rubio L. Solabarrieta

(3)

sub-surface (vessel-mounted ADCPs) water velocity INS_GLO_UV_L2_REP_OBSERVATIONS_013_044 Ref : CMEMS-INS-QUID-013-044 Date : 28 October 2020 Version : 2.1

TABLE

OF

GLOSSARY

AND

ABBREVIATIONS

ADCP Acoustic Doppler Current Profiler

AOML Atlantic Oceanographic and Meteorological Laboratory CF Climate Forecast (convention for NetCDF)

CMEMS Copernicus Marine Environment Monitoring Service CORIOLIS In situ data system for operational oceanography DAC Drifter data Assembly Center

DBCP Data Buoy Cooperation Panel DOC Drifter Operations Center DT Delayed Time

DUACS Data Unification and Altimeter Combination System

ECMWF European Center for Medium-range Weather Forecasts model EU European Union

FTP File Transfer Protocol GDAC Global Data Archiving Centre GDP Global Drifter Program

GLO Global

GMES Global Monitoring for Environmental and Security GTS Global Telecommunication System

HF High Frequency HFR High Frequency Radar INSITU In situ

INS-TAC In situ Thematic Assembly Centre ISAS In Situ Analysis System

MDT Mean Dynamic Topography MFC Monitoring and Forecasting Centre NetCDF Network Common Data Form NRT Near Real Time

PUM Product User Manual RAN ReANalysis

RDAC Regional Data Archiving Centre REP Reprocessed

ROOS Regional Ocean Observing Systems R&D Research and Development

RT Real Time

S Sea Salinity

SDN SeaDataNet

SDC SeaDataCloud

SVP Surface Velocity Program T Sea Temperature

TAC Thematic Assembly Centre URN Uniform Resource Name

UV Horizontal components of velocity VHF Very High Frequency

(6)

Version : 2.1

I

EXECUTIVE SUMMARY

I.1 Products covered by this document

This document aims to give a detailed picture of the processes and tools used to validate the datasets INSITU_GLO_UV_L2_REP_OBSERVATIONS_013_044.

Short Description Product code Area Delivery Time

Global REP INSITU_GLO_UV_L2_REP_OBSERVATIONS_013_044 Global Biannual

This product consists of 3 datasets each dedicated to near-surface currents measurements:

● drifters: the NOAA Atlantic Oceanographic and Meteorological Laboratory (AOML) Surface Velocity Program (SVP) Drifter’s retreatment from 1990 to August 2019. It provides the drifter's position, velocity and includes temperature measurements. In addition a wind slippage

correction from 1993 to August 2019 is provided by the INS TAC (CLS).

● radar_total: near-surface zonal and meridional velocities measured by High Frequency radars (HF radars, as acronym HFR), standard deviation of near-surface zonal and meridional velocities, Geometrical Dilution of Precision (GDOP), quality flags and metadata. These surface observations are part of the European HF radar Network (see Mader et al, 2017 and Corgnati et al., 2018). ● adcp: three-dimensional water current velocities over a depth range along a vessel underway

trajectory. The initial REP product contains current velocities measured on research vessels in the global ocean. Additional ADCP platforms will be considered in next versions: moorings in deep water, coastal stations that are affected by local bathymetry and coastal processes.

(7)

Version : 2.1

Product code INSITU_GLO_UV_L2_REP_OBSERVTIONS_013_044

For Global Ocean- Delayed Mode in-situ Observations of surface (drifters and High Frequency radars) and sub-surface (vessel-mounted ADCPs) water velocity

Dataset[1] drifter radar_total adcp

Geographical coverage

global European and US coasts from coast to up >200 km (depending on the operating frequency) global Horizontal resolution

Discrete Gridded (Typically ranges from a few hundred meters to 5-6 km, depending on HF Radar operating central frequency and bandwidth)

discrete

Temporal resolution

6-hourly Hourly (exceptions with 15’ or 30’)

2 minutes

Vertical resolution - - 0 to 100 meter deep

Available time series

01/01/1990 to 07/08/2019

01/2009 to 07/2020 (but not for all the systems)

04/01/2001 to 27/08/2017

Variables Zonal and Meridional Velocities at 15-m depth, Zonal and Meridional wind-slippage correction, Surface Temperature if available. QC variables and metadata (global attributes)

Zonal and Meridional Velocities at the surface (actual depth depending on the operating frequency), standard deviation of zonal and meridional velocities at the surface Geometrical Dilution of Precision (GDOP). QC variables and metadata (global attributes) Vertical profiles of eastward_sea_water_velocity, northward_sea_water_velocity, upward_sea_water_velocity + depth, QC variables, metadata

Delivery time biannual biannual biannual

File Format netCDF-4 NetCDF4 classic model NetCDF4 classic model Delivery

mechanism

CMEMS Information Service – multiyear ftp (ftp://my.cmems-du.eu)

Table 1: List of INSTAC datasets for which this document applies

(8)

Version : 2.1

I.2 Summary of the results

I.2.1 Dataset: drifter

The SVP drifter’s data are collected, quality controlled and distributed by the AOML Data Assembly Center (DAC). Then a wind slippage correction is computed by the INS TAC (CLS).

Most of the quality control procedures are performed by AOML. Quality of drifter’s measurements depends on the platform and on its physical integrity, in particular the loss of the drogue. If the drifter’s measurement is in nominal mode, it is considered that its velocity corresponds to the 15m depth velocity. If not, it is considered that the drifter’s velocity provides an estimation of the surface current. The drogue loss information is provided in the data set and defines the way the wind slippage correction has to be used.

I. A validation control is performed on the velocity and wind slippage correction by comparison to the geostrophic current derived from altimetry products from THE CMEMS (SEALEVEL_GLO_PHY_L4_REP_OBSERVATIONS_008_047).

In some regions and time periods, the number of measurements can be critically low due to the drifter launch time schedule (Figure 1) and their geographical locations (as in high latitudes.) Moreover, by construction, the wind slippage correction is not available in some small basins (like the Mediterranean Sea) and before 1993.

To compute the wind slippage correction, the new release of the geostrophic current and a new

Ekman/Stokes model have been used. Hence, the entire

INSITU_GLO_UV_L2_REP_OBSERVATIONS_013_044 has been reprocessed.

Figure 1: Count of transmitting drifters per month from 1990 to August 2019. Thick line is the total number of

(9)

Version : 2.1

I.2.2 Dataset radar_total

The last inventory shows that there are 72 HFRs currently deployed and active in various coastal areas of the European seas (see live map in http://eurogoos.eu/high-frequency-radar-task-team/). This number is growing with seven new HFRs installed per year (Figure 3). The EU HFR node delivers in near real-time and at hourly basis, maps of surface current velocities from the HF radars that are actively processing and/or delivering their (formatted or raw) data to the node. The HFR node also collects and process near real time HFR for advanced QA/QC and aggregation of files to build the REP dataset of historical surface current velocity data from those operators connected to the node. In the European framework, the EU HFR Node is now managing data from 11 European HFR networks (built by 28 radar sites) and from 2019 is also receiving data from 5 US HFR networks (Figure 3).

Figure 2- Distribution of the HFR systems contained in the radar_total dataset (zoom in europe in Europe. The

operational systems are plotted in green, future installations in yellow and past deployments in magenta. Source: http://eurogoos.eu/high-frequency-radar-task-team/).

From the 11 EU networks, 10 are sending data in NRT since 2019, and 7 have provided time series of historical data before 2019. EU HF radars are distributed amongst the different Regional Ocean Observing Systems (ROOS) areas coordinated by the European Global Ocean Observing System (EuroGOOS): 56% in MONGOOS (Mediterranean Operational Network for the Global Ocean Observing

(10)

Version : 2.1

System), 32% in IBIROOS (Ireland-Biscay-Iberia Regional Operational Oceanographic System) and 5% in NOOS (north West European Shelf Operational Oceanographic System) (Last update of the inventory, March 2020 as shown in Figure 3).

Figure 3.- Number of HFR stations reporting to the EuroGOOS HFR Task Team from 2004 to 2020. Source: updated version from Roarty et al,. 2019. The operational systems are plotted in green, future installations in yellow and past deployments in red.

I.2.3 Dataset: adcp

I.2.3.1 Coverage in time of the ADCP product

The following figures give an overview of the number of ADCP measurements of 11 complete years of observations (one year = 365 days for one platform). A total of 6 research vessels reported 310 million current observations.

(11)

Version : 2.1

Figure 4: distribution of ADCP measurements per year counted in platform-day (one platform one day = +1)

I.2.3.2 Coverage in space of the ADCPs product

The following figure provides a synoptic view of the coverage in space of the ADCP product within “in situ TAC”. Table 2 and Figure 6 present the number of observations.

(12)

Version : 2.1

PLATFORM_CODE NAME NB_DAY_OBS

FNCM L'ATALANTE 1271

FZVN LE SUROIT 924

FABB BEAUTEMPS-BEAUPRE 907

FMCY POURQUOI PAS? 781

FNFP THALASSA 445

FNUR ANTEA 49

Table 2: ADCP platforms and number of days observed

Figure 6. Distribution per Research Vessel of ADCP days of observations

I.2.3.3 Information on the quality of the data

These diagrams give a synoptic view on the quality of the data available in the historical product per region. The output data quality information about ADCP provided by the GLOBAL INSTAC is displayed in the Figure 7 and Table 3, providing the percentage of observations flagged with a good data quality flag. After quality control, 95% of ADCP observations are flagged as ‘good data’; 5% of ADCP observations are flagged ‘bad data’.

QC flag QC label nb observation

4 bad data 15 217 270 1 good data 295 290 446

nb observations 310 507 716

(13)

Version : 2.1

Figure 7_{. ADCP observations flags: 95% ´good data’, 5% of ‘bad data’}

I.3 Estimated Accuracy Numbers

Table 4 and Table 5 summarizes the accuracy of the measurements that can be expected depending on the platforms and sensors. This is the best accuracy that a user can expect for the data.

Dataset Reference Current (m/s)

drifter Niiler et al, 1995 0.01

radar_total Liu et al., 2010; Ohlmann et al.; 2007, Molcard et al., 2009; Kalampokis et al., 2016; Lana et al. (2015)

0.03–0.12

adcp See Table 5 -

Table 4: Accuracy of the measurements expected from the three datasets Production Subsystem description

The following table summarizes the accuracy of the measurements for the ADCPs that can be expected depending on the sensors. This is the best accuracy then a user can expect for the in situ data to which a quality flag “Good data” (see Table 5) has been applied after validation process. The definition of the reference values is obtained from different sources. The specific reference is given in the tables below and the values are given for the different parameters.

(14)

Version : 2.1

Table 5. Accuracy numbers for measured time series and ADCP estimated parameters for different ADCP sensors.

(15)

Version : 2.1

II

PRODUCTION SUBSYSTEM DESCRIPTION

II.1 Where do measurements come from?

II.1.1 Dataset: drifters

The AOML Drifter Operations Centre manages the deployment of drifting buoys type “SVP” around the world in the frame of the Global Drifter Program (http://www.aoml.noaa.gov/phod/dac/index.php). Once verified operationally, they are reported to AOML's DAC. Incoming data from the drifters are then placed on the Global Telecommunications System (GTS) for distribution to meteorological services everywhere.

A uniform design for the modern SVP drifter was proposed in 1992 (Sybrandy and Niiler, 1992). The SVP drifter has a spherical surface buoy and a semi-rigid drogue that maintains its shape in high-shear flows. The modern data set of SVP drifters (Figure 8) has a holey-sock drogue centred at a depth of 15 meters beneath the sea surface. Different sensors are set on the drifters (Table 6 and Figure 9). Drifter locations are estimated from satellite fixes. Initially transmitted through the Argos satellite system, SVP satellite transmission is now transitioning to the Iridium systems that allows more flexibility, overview with higher rate and capacity of data transmission. Hence, Iridium allows to estimate hourly buoy’s positions.

AOML's DAC collects the raw data, applies quality control and editing procedures, and interpolates positions and temperature to regular 6 hours intervals by Krigging techniques (Hansen and Poulain, 1996). Drifter velocity is then estimated from the interpolated positions.

The wind slippage correction is estimated afterward by the INS TAC from a method derived from Rio (2012), using the drogue loss detection given by the AOML DAC (Lumpkin et al., 2013).

(16)

Version : 2.1

(17)

Version : 2.1

Name Description Measurement

SVP Standard Surface Velocity Program drifter SST

SVPB SVP with Barometer SST, air pressure

SVPC SVP with seabird Conductivity SST, SSS

SVPW SVP with Wind sensor SST, wind speed and direction

SVPBS SVP with Barometer and Salinity SST, Pressure

SVPBW SVP with Barometer and Wind sensor SST, wind speed and direction, air pressure.

Table 6: Description of the instrument types.

Figure 9: Instrument type distribution during the overall period.

II.1.2 Dataset: radar_total

HFR is a land-based remote sensing technology (see Figure 10) that has shown to be a cost-efficient tool

to monitor coastal regions at a range of up to 200 km. Oceanographic HF-radars are mainly utilized to measure ocean surface current fields for various applications such as search and rescue, oil spill monitoring, marine traffic information or improvement as well as data assimilation of numerical circulation models [Paduan and Rosenfeld, 1996; Gurgel et al., 1999; Rubio et al. 2018; Roarty et al., 2019].

HF-radar relies on resonant backscatter resulting from coherent reflection of the transmitted wave by the ocean waves whose wavelength is half of the transmitted one. This is the Bragg scattering phenomenon and it results in the first order peak of the received (backscattered) spectrum (Paduan and Graber, 1997). The difference between the theoretical speed of the waves and the velocity observed,

(18)

Version : 2.1

resulting in the Doppler shift in the observed Bragg peaks, is due to the velocity of the radial component of the current (the current in the same direction of the signal), that can be therefore estimated.

HF radars belong to the remote sensing instruments family. Based on the analysis of the electromagnetic waves (in the high frequency band) back scattered by the ocean surface and the associated doppler shift, each radar station is able to determine the radial component of the surface current velocity, i.e. the component of the velocity along the radial direction away from -negative- or toward -positive- the radar antenna itself, for each cell of a given polar grid, determined by the angular/radial resolution and max range of the instrument. Each HF radar station produces then a two-dimensional map on a polar grid containing partial information on the radial surface current velocity (Figure 11- left panel). To obtain complete information, i.e. total surface current vectors (radar_total), data from at least two HFR stations with overlapping coverage must be combined. Total velocities are calculated using unweighted least square fit that maps radial velocities measured from individual sites onto a cartesian grid. The final product is a map of the horizontal components of the ocean surface currents on a regular grid in the area of overlap of two or more radar stations (Figure 11- right panel).

HFRs provide current data only relative to the surface within an integration depth ranging from tens of centimetres to 1-2 meters depending on the operating central frequency. These data are first collected by the European HF Radar Node and then verified and (when necessary) post-processed before being transferred to INS-TAC GLO CMEMS.

Figure 10- HF radar's principle of operation. Source: SOCIB ICTS – MEDCLIC project (http://medclic.es/en/).

©ICTS

SOCIB

(19)

Version : 2.1

Figure 11.-Maps of HF radar radial velocities for two radial sites, showing the radial coverage (left) and total

surface velocities in the overlapping area (right).

The European HFR Node acts as the focal point for the European HFR data providers and implements the HFR data stream from the data providers to the CMEMS-INSTAC Global Production Unit (PU).

In the case of the REP data the European HFR Node collects both the historical datasets sent by the providers (periods before April 2019) and the data delivered (since April 2019) in NRT by the connected providers.

For the historical data the collection dataflow is structured as follows:

● if the data provider can generate HFR historical data series according to the defined data format and QC standards, the node only collects and analyses the data, reprocess the time series if needed (always in exchange with the providers) and pushes them to the Global PU. ● If the data provider cannot generate HFR data according to the defined data format and QC

standards, the HFR Node harvests the raw data from the provider, harmonizes, quality-controls, formats the data, re-processes the time series if needed (always in exchange with the providers), and pushes them to the Global PU.

The strength and flexibility of this solution reside in the architecture of the European HFR node, that is based on a centralized database, fed both by the operators via a webform and by the software routines running on the node, containing updated metadata of the HFR networks and the needed information for processing, Quality Controlling and archiving the data (schematized in Figure 12). A set of shared software tools uses all that information for processing native HFR data for quality controlling and converting them to the standard format for distribution. This strategy guarantees that, whatever the workflow, the NRT and REP data are processed by the same software tools.

(20)

Version : 2.1

Figure 12: Schematic view of the production system description for HF radar data

II.1.3 Dataset: adcp

The French research oceanographic vessels are equipped with Acoustic Doppler Current Profilers (ADCP) that measure ocean current profiles along the vessel tracks.

● Four vessels are equipped with an ADCP Ocean Surveyor 150 kHz (OS150) and an ADCP Surveyor 38 kHz (OS38) : Pourquoi Pas?, Atalante, Beautemps-Beaupré and Thalassa.

● A fifth vessel, Suroît, is equipped with an ADCP Broad Band 150 kHz (BB150). More on “The current profilers of the French Oceanographic Fleet”.

Observations from Research vessel mounted ADCP are processed in delayed mode by “Cascade data processing chain” operated by Ifremer/Sismer specialists.

The 150kHz ADCP is built by the American company R.D.I (http://www.rdinstruments.com). With each acoustic impulse, the current in the entire water column is measured down to depths of 200m. The 38kHz ADCP is an Acoustic Doppler Current Profiler built by the American company R.D.I

(http://www.rdinstruments.com). With each acoustic impulse, the current in the entire water column is measured down to depths of 850m. Both are attached to the hull of the vessel. They simultaneously emit four acoustic beams, the main frequency of which is around 38 kHz or 150kHz. These beams are oriented at 30° from the vertical of the ship. The acoustic wave emitted is reflected on the substance in

suspension and is shifted proportionately to the speed of the reflectors (the Doppler effect). These particles are assumed to move at the speed of the water mass. A current is thus calculated in the direction of each beam. As the orientation of the sensor in relation to the ship, the orientation of the ship in relation to a point of reference on land, and the speed of the vessel are known, an absolute current can be calculated.

(21)

Version : 2.1

Figure 13: The four beams of an ADCP installed on the hull of a ship to observe ocean currents along the ship track

The data is processed with Cascade (Automated chain for following embedded acoustic Doppler current meters) software. Cascade is a piece of validation and visualization software for ADCP hull

measurements initiated in 1998 at LPO (Laboratory of Ocean Physics - Ifremer) to meet its research needs for processing and analyzing ADCP measurements. It is regularly updated. It is now implemented operationally by Ifremer/Sismer data centre. It is comprised of a set of matlab programmes.

After Cascade processing, ADCP data files are available and downloadable in NetCDF format.

Copernicus Marine in situ TAC reformat the NetCDF files extracting the current EOVs (sea water current components).

II.2 Description of the datasets

II.2.1 Dataset: drifter

Initially, this dataset is an extraction of the AOML Global Drifter Program (GDP) interpolated database (http://www.aoml.noaa.gov/phod/dac/dacdata.php) over the 1990 to August 2019 period. Then a wind slippage correction is added for the period 1993 to August 2019 by the INS TAC. The time range covered by this correction is shorter because the wind slippage needs the geostrophic current and Mixed Layer Depth (MLD) to be computed. Geostrophic currents are available since 1993. These geostrophic currents and MLD are delivered in CMEMS through the products SEALEVEL_GLO_PHY_L4_REP_OBSERVATIONS _008_047 and MULTIOBS_GLO_PHY_REP_015_002.

As can be seen in Figure 14, the number of SVP drifters has continuously increased from the early 90’s. But with the use of new drifter’s design during the mid-2000s and the displacement of the buoys in the Antarctic Circumpolar Current (Figure 15), the number of undrogued instruments has raised (Lumpkin et

(22)

Version : 2.1

al, 2013). The number of temperature, velocity and wind slippage correction data follow the same trend (Figure 14).

Note that the detection of the drogue loss is important to correctly infer surface current velocities from this dataset (also reminded in the CMEMS-INS-PUM-013-044).

Figure 14: Monthly distribution of the temperature (black), velocity (red) and wind slippage correction (blue)

(23)

Version : 2.1

Figure 15: 2°x2° bin mean number of “drogued” (upper) and “undrogued” (lower) drifter measurements over the 1990 to 2019 period.

II.2.1.1 Velocity

Drifter velocities are computed by AOML: after the QC and editing of drifter positions, trajectory positions are re-interpolated regularly by the Krigging technique on 6 hours interval (Hansen and Herman, 1989; Hansen and Poulain, 1996). Then velocities are derived from finite differences of their position fixes using a 12-h centred scheme.

SVP drifter’s velocity is not the perfect measurement of the water column averaged over the drogue depth. The water can sink, or the drifter can slip due to wind influence on the surface float. Hence the resulting drifter velocity is the addition of the 15 meters depth large-scale current, the upper-ocean wind-driven flow, the influence of tides and Stokes Drift and other forces on the drogue and the surface

(24)

Version : 2.1

body of the drifter, and the slip. The slip depends on the drogue loss status of the buoy and the windage is provided within the dataset. Drogue loss detection performed by AOML is detailed below (§II.2.1.2). The spatial repartition of the measurement is sparse or null in high latitudes and coastal areas. Less data is also available in the Mediterranean Sea (Figure 16). The repartition of the velocity data over the period (Figure 14) follows the drifter’s deployment schedule.

Figure 16: 2° x 2° bin mean of the number of velocity measurements from 1990 to August 2019.

II.2.1.2 Wind slippage correction by the INS TAC

As long as the drogue remains attached to the drifter, the downwind slip is estimated at 0.7 cm.s-1 per 10 m.s-1 of wind speed (Niiler and Paduan, 1995). If a SVP drifter loses its drogue, it will slip downwind at a speed of 20 cm/s per 10 m/s of wind (Pazan and Niiler, 2001).

A direct wind slippage correction, also called “windage”, of zonal and meridional velocity is estimated by the INS TAC following Rio (2012) method. The total drifter velocity Ud is decomposed into different

contributions:

- geostrophic current Ug

- Ekman/Stokes current Ue (wind-driven current)

- ageostrophic current Ua, including tides and other high frequency signals

- the slippage from the wind Us, estimated here.

When the drogue is lost, the drifter is directly under the influence of the wind stress. Using geostrophic current from altimetry (as mentioned earlier) and an estimation of the wind-driven velocities from a new

(25)

Version : 2.1

Ur = (Ud - Ug - Ue)f ∝ Us Equation 1

where f stands for a low pass filter to remove inertial oscillation, tidal and high frequencies signal impacting the drifter’s velocity. The filter cutoff length is taken as the maximum between the 24 hours tidal period and the inertial period corresponding to the more equatorward latitude of the single drifter trajectory.

Then the wind slippage correction Us is computed as:

Us = α W Equation 2

where α is a coefficient which minimizes the correlation between the wind speed W and the residual drifter velocity Ur.

For this purpose, a 0.25° resolution 3h wind stress τ from the ECMWF ERA5 reanalysis is interpolated onto the drifter positions and used to compute the slip. The wind speed is computed using a bulk formula:

W=(τ/ρaCD)1/2 Equation 3

where ρa=1.3 kg.m-3 is the air density and Cd = 1,4.10-3 is a non dimensional drag coefficient.

Using the same wind stresses, the Ue component is computed, based on an empirical Ekman/Stokes

model. This model has been recently updated. In all previous work (Rio et al, 2014), empirical formulations of what we called the Ekman response to wind stress have been calculated based on the information from ocean current velocities derived from the trajectory of surface floats and 15m depth drogued SVP buoys. Both measurements are of lagrangian type (they follow the water parcels) and therefore include both the Ekman wind driven component and the Stokes drift. But we expect the Stokes drift to substantially modify the classical Ekman formulation in the first 2-4 meters (depending on the wind speed). At 15m depth however, we expect the wind-driven current to be mostly due to Ekman drift. We therefore speak about wind-driven current or Ekman/Stokes current. The new wind-driven current uses a more complete formulation including a priori knowledge of the mixed layer depth as a proxy for the Ekman depth.

6 . 0 0 0

)

0 (









i w

z

e

u



₁₅ 15 0.7

)

15 (









i w

z

e

u



Equation 4

β and θ parameters are estimated in 0.25° latitude x 1 m MLD grid at the surface and 15 m. Both surface (undrogued buoy) and 15m depth (drogued buoy) wind slippage are computed and available in the data set.

The wind slippage correction is computed as follows: each buoy trajectory is divided into its drogue and undrogue time series. For each of them:

- when the trajectory is longer than 60 days, an adaptative moving correlation window is used to

(26)

Version : 2.1

length. The first/last days of the trajectory are then completed using the mean value over the buoy time series (drogued or undrogued)

- For trajectory shorter than 60 days, a climatological α (drogued and undrogued drifters

separately) is used to compute the wind slippage correction. These 2 climatologies are computed with all the values from trajectories longer than 60 days of drogued and undrogued drifters respectively.

As the wind-driven model is not a fully global product, no wind slippage correction is available in the Mediterranean Sea (Figure 17). Moreover, no correction is provided ±5° from the equator for the drogued drifters (see IV.1.1)

Elsewhere, this new data base provides a wind slippage correction of the entire drifter’s trajectories from 1993 to August 2019.

Figure 17: 2° by 2° bin mean of the number of wind slip correction data from 1990 to August 2019

II.2.1.3 Temperature

All standard SVP drifters measure temperature 20-30 cm beneath the sea surface. AOML re-interpolate along the 6h-sampled trajectories the temperature measurements using the same Krigging technique that for position fixes. These temperature observations are flagged following INS TAC standards. Spatial and time repartition of the data are shown on Figure 18 and Figure 14.

(27)

Version : 2.1

Figure 18: 2° by 2° bin mean of the number of temperature data from 1990 to August 2019.

II.2.2 Dataset: radar_total

The dataset radar_total REP consists in 11 time series of maps of near-surface zonal and meridional velocities and their corresponding standard deviation, Geometrical Dilution of Precision (GDOP), quality flags and metadata. The data series available have different lengths, depending on the availability of data from the providers. Details on the HFR radar systems providing REP data for this release are summarized in Table 5. The spatial coverage of the different systems is showcased by the corresponding maps of mean currents provided in Figure 19.

(28)

Version : 2.1

NETWORK STATION Lat Lon Tf (MHz) Dates of available REP

data series ROOS INSTITUTION

HFR-COSYNA WANG 53,79 7,92 12,50 12/2019-07/2020 NOOS Helmholtz-Zentrum Geesthacht BUES 54,12 8,86 10,80 SYLT 54,79 8,28 12,50

HFR-MATROOS MON 52,03 4,17 16,17 01/2020-07/2020 Rijkswaterstaat

OUD 51,82 3,88 16,17 HFR-EUSKOOS MATX 43,45 -2,75 4,53 01/2009-72/2020 IBIROOS EUSKALMET-AZTI HIGE 43,38 -1,78 4,53 HFR-Galicia SILL 42,10 -8,90 4,46 06/2015-72/2020

Puertos del Estado

FIST 42,88 -9,27 4,46 VILA 43,16 -9,21 4,46

INTECMAR

PRIO 43,57 -8,31 4,46

HFR-Lisboa 590_IHOC_PL016 38,67 -9,33 12,43 04/2012-72/2020 IBIROOS

Instituto Hidrografico 590_IHOC_PL015 _38,41 _-9,21 _12,92 _IBIROOS HFR-South 590_IHOC_PL017 _36,99 _-8,95 _13,50 2/2016-72/2020 IBIROOS 590_IHOC_PL018 _37,09 _-8,44 _13,50 590_IHOC_PL019 _37,18 _-7,44 _12,47

MAZA 37,13 -6,83 13,50 Puertos del Estado HFR-Gibraltar

CEUT 35,90 -5,31 26,28

12/2019-07/2020 IBIROOS MONGOOS

Puertos del Estado

CARN 36,08 -5,43 26,28 TARI 36,00 -5,61 26,28 CAMA 36,09 -5,81 26,28

HFR-Ibiza FORM 38,67 1,39 13,50 06/2012-07/2020 MONGOOS SOCIB

GALF 38,95 1,22 13,50

HFR-DeltaEbro

SALO 41,06 1,17 13,50

12/2019-72/2020 MONGOOS Puertos del Estado

ALFA 40,67 0,83 13,50 VINA 40,46 0,48 13,50 HFR-TirLig MONT 44,15 9,65 26,28 08/2016-72/2020 MONGOOS CNR-ISMAR TINO 44,03 9,85 26,28 PCOR 44,14 9,66 26,28 VIAR 43,86 10,24 26,28 PFIN 44,30 9,22 13,50

HFR-Vigo SUBR 42,25 -8,86 46.5 02/2010 - 05/2016 IBIROOS Universidad de Vigo

TORA 42.20 -8-80 46.5

HFR-US-Alaska * * * * 01/2019 - 07/2020 Global US HFR National

Network

HFR-US-EastGulfCoast * * * * 01/2019 - 07/2020

Global US HFR National Network HFR-US-Hawaii * * * * 01/2019- 07/2020 Global US HFR National

Network HFR-US-PuertoRicoVirginIs lands * * * * 01/2019 - 07/2020 Global US HFR National Network HFR-US-WestCoast * * * * 01/2019 - 07/2020 Global US HFR National Network

Table 7- Details of the HFR radar systems providing REP data and time availability of corresponding total REP datasets (Tf: Transmit frequency). *The total surface current data from the US systems are directly harvested

from https://hfrnet-tds.ucsd.edu/thredds/HFRADAR_USWC.html and translated to the EU standard nc format by the EU-Node. They contain less information than the EU files, namely the details of the radial sites used for

(29)

Version : 2.1

Figure 19: Mean surface current maps for the indicated HFR systems and periods. The means are computed in the area of 80% temporal coverage as recommended by Roarty et al., 2012 for the corresponding periods as specified in the title of the subplots (continues in next page).

(30)

Version : 2.1

Figure 19: (continued) Mean surface current maps for the indicated HFR systems and periods. The means are computed in the area of 80% temporal coverage as recommended by Roarty et al., 2012 for the corresponding periods as specified in the title of the subplots.

II.2.3 Dataset: adcp

The ADCP files are produced by a data processing chain such as Cascade (see II.1.3). The processed and quality-controlled files are pushed in Copernicus in situ TAC NRT multiparameter product (INSITU_GLO_NRT_OBSERVATIONS_013_030).

To publish the ADCP REP dataset, these files are copied in the REP workplace to be homogeneously assessed with a visual inspection by an ADCP specialist.

The files with a valid assessment and visual inspection are pushed in the REP ADCP product.

The questionable data files are signaled to their regional originator in the QC feedback loop. The originator either modifies the QC (file update) or confirms the data quality (no file update).

(31)

Version : 2.1

(32)

Version : 2.1

III VALIDATION FRAMEWORK

III.1 Dataset: drifter

III.1.1 AOML QC steps

Drifter data are daily downloaded from Argos/Iridium and also received at AOML once a month. Deployment time and position of first good transmission from the water are determined.

Dead buoys are identified:

● Transmission from the same location after a successful deployment (grounded) ● No new data after last update (quit)

Bad locations from ARGOS/GPS raw data are checked, based on speed between consecutive locations. Bad points are deleted.

Deviant SST values are removed by applying a temperature change criterion relative to the recent temperatures measured by the buoy. SST from each drifter are compared with Reynold’s climatology to determine temperature sensor failure.

Buoys that possibly lost their drogues are identified and drogue loss date is determined.

To eliminate the more obvious errors in raw position, the DAC applies a two-step quality control scheme (Hansen and Poulain, 1996). The velocity is calculated by finite differencing the raw position both forward and backward in time. A position is flagged as "bad" if it produces a velocity greater than four standard deviations from the mean for both forward and backward passes.

This two-way differencing scheme is used because a forward-only calculation may fail to identify a bad fix if it comes immediately after a gap in data acquisition.

III.1.2 Drogue loss detection by AOML

A drogue presence automatic detection has been implemented by the GDP (Lumpkin et al., 2013) and reanalysis of the historical data has been performed. The methodology is based on the Rio (2012) study combined with the information’s from the submergence and transmission frequency variations of each drifter.

First, a wind-driven motion model is computed using wind stresses deduced from CCMP and COARE3.0 formulation and applied only on drifters that are flagged as drogued and the first 90 days of the drifters after 2000. A residual velocity U’= Ud - Ug is computed and low pass filtered with a cut off frequency of 5

days (removing the inertial, diurnal and tidal frequencies). U’ is used to evaluate a wind-driven model of the form:

u’ = a. τ1/2 v’= b. τ1/2

(33)

Version : 2.1

Then the difference between the low-pass filtered residual velocity of each drifter and the collocated u’ is computed and written as in equation 2. The time series of α for W > 1.5 m.s-1 is used. A value of α = 0.015 is chosen as a detection of the drogue loss (see Lumpkin et al., 2013 for more details.)

This control is then combined with the submergence record information and from the radio frequency of the drifter-satellite communication (there is an increase of the diurnal variation when drogue is lost.)

III.1.3 Assessment of the Wind slippage correction

The wind Slippage correction proposed here as been tested. The mean Us for undrogued and drogued

buoys are computed in order to check if the surface trajectories are effectively more impacted by the wind slip than the 15m measurements.

We then compare the satellite altimetry derived geostrophic current Ugeo to equivalent Ug signal from

drifters:

Ug ≈ (Ud – Ue)f Equation 5

When undrogued, the wind slippage correction is also removed:

Ug ≈ (Ud – Ue -Us)f Equation 6

We compute this value for drogued and undrogued drifters, in a way to compare the resulting statistics. We also check the impact of Us on the results.

III.2 Dataset: radar total

The EU HFR Node is in charge of checking the validity of the HF radar data files and of applying the NRT QC in compliance with the EU common data and metadata Standard for HF radar surface current data. Table 8 summarizes the mandatory tests for application according to the European common QC standard for NRT HFR current data. The tests are subject to change in some details.

Short description QC tests

Syntax

X Data Density Threshold

X Velocity Threshold X Variance Threshold X Temporal Derivative X GDOP Threshold X

(34)

Version : 2.1

A more detailed description of the mandatory NRT QC tests performed on HFR total current data are: ● Syntax check: this test will ensure the proper formatting and the existence of all the necessary

fields within the total NetCDF file. This test is performed on the NetCDF files and it assesses the presence and correctness of all data and attribute fields and the correct syntax throughout the file. This test is performed by the European HFR Node before pushing data to the distribution platforms.

● Data Density Threshold: this test labels total velocity vectors with a number of contributing

radials bigger than the threshold with a “good data” flag and total velocity vectors with a number of contributing radials smaller than the threshold with a “bad data” flag.

● Velocity Threshold: this test labels total velocity vectors whose module is bigger than a

maximum velocity threshold with a “bad data” flag and total vectors whose module is smaller than the threshold with a “good data” flag.

● Variance Threshold: this test labels total vectors whose temporal variance is bigger than a

maximum threshold with a “bad data” flag and total vectors whose temporal variance is smaller than the threshold with a “good data” flag. This test is applicable only to Beam Forming (BF) systems. Data files from Direction Finding (DF) systems will apply instead the “Temporal Derivative” test reporting the explanation “Test not applicable to Direction Finding systems. The Temporal Derivative test is applied.” in the comment attribute.

● Temporal Derivative: for each total bin, the current hour velocity vector is compared with the

previous and next hour ones. If the differences are bigger than a threshold (specific for each grid cell and evaluated on the basis of the analysis of one-year-long time series), the present vector is flagged as “bad data”, otherwise it is labelled with a “good data” flag. Since this method implies a one-hour delay in the data provision, the current hour file should have the related QC flag set to 0 (no QC performed) until it is updated to the proper values when the next hour file is generated.

● GDOP Threshold: this test labels total velocity vectors whose GDOP (Geometrical Dilution Of

Precision) is bigger than a maximum threshold with a “bad data” flag and the vectors whose GDOP is smaller than the threshold with a “good data” flag.

These mandatory QC tests are manufacturer-independent, i.e. they do not rely on particular variables or information provided only by a specific device and they are required for labeling the HFR total velocity data as Level 3B (Level 3A data that have been processed with a minimum set of QC). Please refer to Appendix A of the JERICO-NEXT deliverable D5.14 “Recommendation Report 2 on improved common procedures for HFR QC analysis” for the processing level definition.

For some of these tests, HFR operators will need to select the best thresholds. Since a successful QC effort is highly dependent upon selection of the proper thresholds, this choice is not straightforward, and requires trial and error before final selections are made. These thresholds are not determined arbitrarily but based on historical knowledge or statistics derived from historical data.

For the REP data the processing of the data consists in the following steps:

1 – Harvesting of standardized/native data: Historical data (only periods prior to 04/2019 for the connected systems) are obtained directly from the providers upon invitation sent by emails with

(35)

Version : 2.1

data of the systems connected for NRT operations after 04/2019 are directly retrieved from the node’s THREDDS and stored in the same HIST repositories.

2- Reprocessing of native data: Native data are reprocessed using the 'HFR_Node__Historical_Data_Processing software in matlab.

3- Screening of the data and plots for the Advanced Quality Control: Data series are organized in a standard system/time folder tree. The following plots by year and system are produced:

(i) Temporal series of the spatial average of the current velocity module, its standard deviation and the total coverage (see example in Figure 20)

(ii) Temporal series of the QC flags for all the grid nodes with data (see example in Figure 21) (iii) Maps of the mean value of QC flags for the target year and Maps of mean velocity module and its standard deviation for the target year (see example in Figure 22)

(iv) Spatial (x-axis) vs. temporal (y-axis) coverage 80/80 annual metric (already showed in NRT for each system in the dashboard). It allows to check it the system has reached the goal of providing surface currents over the 80% of the area during 80% of the time (see examples in Figure 23 and Figure 24)

(v) Map of the mean velocity field in the area of 80% temporal coverage (see examples in Figure 19).

4- Report on the HIST data quality. Based on (3) a report is produced by system where the performance is analysed year by year and periods for reflagging (expert but subjective analysis) are proposed.

● Two new possible flags are proposed in addition to those already in use (i.e. 0: no QC performed; 1: good data; 4 bad data):

Qc flag Meaning

2 Probably good data

3 Bad data that are potentially correctable

● In addition, possible changes in the processing of the data (namely in the thresholding strategy) are proposed.

This step is divided in two stages in close collaboration with the data providers:

● In a first step, the report is sent to the provider for its validation and agreement or feedback on the comments and the reflagging/reprocessing proposed.

● After the providers feedback changes in the original data series (reflagging or reprocessing) are indicated. A final version of the report is produced, to be inserted in the QUID.

5- Reflagging: It is only performed if the provider validates the reflagging proposition. If no reflagging is performed, the REP data contain the same information of the NRT files.

(36)

Version : 2.1

Figure 20: Temporal series of the spatial average of the current velocity module (top panel), its standard deviation (middle panel) and the grid points of the total coverage (bottom panel). Example for HFR-EUSKOOS system in 2018. Black dots are the values obtained considering all the data in the domain, in green those considering only data with QC flag =1 (good data).

Figure 21: Temporal series of the QC flags for all the grid nodes with data. Example for HFR-EUSKOOS system in 2018.

(37)

Version : 2.1

Figure 22: Maps of the mean velocity module and the mean value of QC flags for the target year (left column) and their standard deviations (right column) for the target year. Example for HFR-EUSKOOS system in 2018.

Figure 23: Spatial (x-axis) vs. temporal (y-axis) coverage 80/80 annual metric recommended by Roarty et al., 2012. Allows to check if the system has reached the goal of providing surface currents over the 80% of the area during 80% of the time (two figures). Example for HFR-EUSKOOS system in 2018.

(38)

Version : 2.1

Figure 24: Map of the % of availability of data in each grid point and contour showing the area of temporal availability >80%. Example for HFR-EUSKOOS system in 2018.

III.3 Dataset: adcp

The research vessels ADCP data are validated under the authority of the scientists in charge of the oceanographic cruises. For each cruise, an ADCP data scientist operates a data processing chain such as Cascade (as described in §II.1.3).

Each cruise data is aggregated into one file per vessel by Copernicus in situ TAC in the multiparameter product (INSITU_GLO_NRT_OBSERVATIONS_013_030).

Each aggregated data file is assessed and visually inspected every 6 months by an in situ TAC ADCP specialist.

(39)

Version : 2.1

IV VALIDATION RESULTS

IV.1 Dataset: drifter

IV.1.1 Ocean current and wind slippage correction

Validation of the wind slippage correction has been performed from 1993 to end of 2018.

The spatial patterns of the zonal velocity measured by drifters are similar in the drogued (15m depth) and corrected undrogued (surface) drifters (Figure 25). The main currents are consistent with the general ocean circulation. Zonal velocity is stronger at the surface than at 15m depth in all the basins and main ocean circulation structures. This is expected, as the velocity measured by the drifters is a total signal with a stronger component of the wind induced circulation at the surface than at 15m (Rio et al; 2014). As expected, the time-mean wind slippage correction is smaller for drogued buoys (Figure 26). Moreover, the latitudinal repartition of the zonal correction (Figure 27) is similar to the zonal wind distribution (see Fig 1b in Lumpkin et al; 2013). Wind slippage correction is stronger in area of important wind speed, with a maximum south of 40°S and near 50°N.

In the previous version, we find a pattern of high value of the α coefficient in equation 1 on the drogued buoys in the equatorial area. Improvement of the geostrophic current derived from altimetry with use of CNES-CLS18 MDT and the wind-driven formulation in the equatorial band has greatly reduced the residual current Ur and the resulting estimated correction at 15m. But the accuracy of the wind slippage

method is still not sufficient to provide a correction for the drogued buoys at the equator. Moreover, if the correction is estimated superior to ±0.03 m.s-1, it is recommended to remove it from the 15m depth velocity.

As the windage is stronger at the surface, this correction is relevant even at the equator for the undrogued drifters.

The geostrophic signal extracted from the drogued drifters (equation 5) gives some good overall statistical results (Table 9) compared to altimetry products. Correlation between the 2 signals ranges between 0.81 and 0.85.

For undrogued drifters, the introduction of the wind slippage correction improves the comparison to geostrophy (Table 10). Correlation with Ugeo is increased (near +0.01) and the percent of RMS difference

is reduced by more than 7% in the zonal direction and 5% in the meridional direction.

Compared to the previous version, the correlation with altimeter geostrophy is quite similar in both zonal direction and meridional direction. But the percent of the RMS difference is reduced of 1.3 (zonal velocity) and 3 (meridional velocity.)

(40)

Version : 2.1

Figure 25: 2°x2° bin drogue on (upper) and drogue off (lower) mean zonal drifter’s velocity in m/s over the 1993 to end of 2018.

(41)

Version : 2.1

Drogue on Correlation RMS difference

(m/s)

RMS difference/RMS geost (%)

Ug vs Ugeo 0.85 (0.84) 0.09 (0.11) 36.5 (37.8)

Vg vs Vgeo 0.81 (0.80) 0.09 (0.11) 48.2 (51.1)

Table 9: Statistical validation results of the drogued drifter’s velocity Ug/Vg versus satellite altimeter derived

geostrophy Ugeo/Vgeo. Statistics have been computed on 13 885 668 data (±5° equatorial band excluded).

Previous version statistics are in brackets (referenced to DUACS18 geostrophy).

Drogue off Correlation RMS difference (m/s) RMS difference/RMS

geost (%)

Ug vs Ugeo 0.80 (0.79) 0.11 (0.11) 56.9 (57.7)

Vg vs Vgeo 0.76 (0.75) 0.10 (0.11) 64.9 (67.9)

Ug-Us vs Ugeo 0.81 (0.81) 0.10 (0.10) 48.2 (49.5)

Vg-Vs vs Vgeo 0.77 (0.76) 0.10 (0.10) 59 (62.2)

Table 10: Statistical validation results of the undrogued drifter’s velocity Ug/Vg versus satellite altimeter

derived geostrophy Ugeo/Vgeo. Statistics have been computed on 19 354 979 data. Previous version statistics

(42)

Version : 2.1

Figure 26: 2°x2° bins drogue on (upper) and drogue off (lower) mean zonal wind slippage correction in m/s over the 1993 to end of 2018 period.

(43)

Version : 2.1

Figure 27: Time series (monthly mean) of longitude mean of zonal wind slippage correction for drogued drifters (upper) and undrogued drifters (lower).

QUALITY INFORMATION DOCUMENT