Improving Data Quality for the Australian High Frequency Ocean Radar Network Through Real-Time and Delayed-Mode Quality-Control Procedures

remote sensing

Article Improving Data Quality for the Australian High Frequency Ocean Radar Network through Real-Time and Delayed-Mode Quality-Control Procedures

Simone Cosoli * , Badema Grcic, Stuart de Vos and Yasha Hetzel

Ocean Graduate School and the UWA Oceans Institute, The University of Western Australia, 35 Stirling Highway, Crawley, WA 6009, Australia; [email protected] (B.G.); [email protected] (S.d.V.); [email protected] (Y.H.) * Correspondence: [email protected]; Tel.: +61-8-6488-7314

Received: 31 July 2018; Accepted: 14 September 2018; Published: 16 September 2018

Abstract: Quality-control procedures and their impact on data quality are described for the High-Frequency Ocean Radar (HFR) network in Australia, in particular for the commercial phased-array (WERA) HFR type. Threshold-based quality-control procedures were used to obtain radial velocity and signal-to-noise ratio (SNR), however, values were set through quantitative analyses with independent measurements available within the HFR coverage, when available, or from long-term data statistics. An artifact removal procedure was also applied to the spatial distribution of SNR for the first-order Bragg peaks, under the assumption the SNR is a valid proxy for radial velocity quality and that SNR decays with range from the receiver. The proposed iterative procedure was specially designed to remove anomalous observations associated with strong SNR peaks caused by the 50 Hz sources. The procedure iteratively fits a polynomial along the radial beam (1-D case) or a surface (2-D case) to the SNR associated with the radial velocity. Observations that exceed a detection threshold were then identified and flagged. After removing suspect data, new iterations were run with updated detection thresholds until no additional spikes were found or a maximum number of iterations was reached.

Keywords: HF ocean radar systems; quality control; remote-sensing

1. Introduction The Australian Ocean Radar facility at the University of Western Australia is part of the Integrated Marine Observing System (IMOS), a national collaborative research infrastructure tasked with collection and dissemination of ocean data. The radar facility measures met-ocean parameters from an array of High-Frequency Radar (HFR) systems deployed around the coastline of Australia. Commercial direction-finding (SeaSonde) and phased-array (WERA) HFR systems, provided respectively by Codar Ocean Sensors (COS) and Helzel MessTechnik, are in operation in the Rottnest Shelf and Turquoise Coast (Western Australia, WA; Figure1), Coffs Harbour and Newcastle (New South Wales, NSW, Australia), and the South Australian Gulf regions (SAG). Each HFR node is configured primarily to sample ocean currents with a maximum range of over 200 km; however, at selected locations where the phased-arrays systems are installed, waves and winds can also be obtained. Radar data, freely available from the IMOS portal (https://portal.aodn.org.au/), are used for scientific research, operational modeling, coastal monitoring, fisheries, and other applications [1–7]. This paper focuses on the quality-control (QC) procedures that have been implemented for the phased array systems in Australia and are applied operationally on both the near real-time (NRT) and delayed-mode (DM) products. The main focus is on radial currents and surface current maps,

Remote Sens. 2018, 10, 1476; doi:10.3390/rs10091476 www.mdpi.com/journal/remotesensing Remote Sens. 2018, 10, 1476 2 of 15 but some of the proposed methodologies have the potential to be applied to wind and wave data available from the phased-array systems. Routinely, QC tests for HFR data are applied at different processing levels, including Doppler spectra, radial velocity maps, and surface current vectors [8]. Most of the tests on Doppler spectra (e.g., noise floor and computation, first-order Bragg peak detection, and the detection and removal of radio-frequency interference and ship echo [9,10]) are embedded within the manufacturer software. Other tests (for example, on radial velocity speed or signal-to-noise ratio (SNR)) rely on the definition of threshold values which should be set on the basis of historical knowledge or statistics derived from more recently acquired data [8]. Examples of operational procedures applied in RT mode for HFR in general and WERA systems in particular can be found in [11,12], while advanced methodologies capable of handling complex cases of external radio frequency interferences are described in [13]. The paper is organized as follows. Section2 describes the network set up and some of the present limitations. Section3 provides details on the choice of the data quality proxy and the thresholds, and documents the accuracy levels of the HFR systems deployed across the country. Additional details on the NRT and DM QC procedures are also provided in this section. Section3 focuses mostly on the HFR systems deployed in WA, due to the large number of mooring deployments in the region that can be used for validation of the HFR data. However, results and methodologies are applied operationally to the HFR systems across Australia and can be easily extended to the global HFR network. Discussions and conclusions are presented in Sections4 and5, respectively.

2. Materials and Methods The Australian Ocean Radar network includes the compact cross-loop systems direction-ﬁnding HFR system provided by Codar Ocean Sensors (US) and the WERA phased-array HFR systems manufactured by Helzel MessTechnik (Germany). Operational settings (frequency, bandwidth, integration times, and output rates) are detailed in Table1. Since 2014, all systems have operated within the ITU frequency bands in compliance with the 2012 World Radiocommunication Conference, ITU Resolution 612, which designated speciﬁc bands between 3 and 50 MHz to support oceanographic radar operations. The WERA systems operate at 9.335 MHz with 33.4 kHz bandwidth in the Rottnest shelf (WA) and the South Australia Gulfs region (SAG), and at 13.5 MHz with 100 kHz bandwidth at Coffs Harbor (COF, NSW). Within these frequency bands, a general increase in the average noise levels was documented [11] and a more pronounced diurnal variability was observed with respect to the frequency bands previously utilized by the Australian HF radar network. In general, operating range decreased after the frequency change due to stronger radio frequency interference.

Table 1. Operational settings for the High-Frequency Radar (HFR) systems across Australia; locations of the HFR systems for the Western Australia (WA) node are given in Figure1.

Operating Frequency HFR Node HFR Type Bandwidth (kHz) Integration Time Output Rate (Center, MHz) TURQ SeaSonde 4.463 25 17 min 1 h (radials) 1 h (vectors) ROT WERA 9.335 33.4 5 min 5 min (radials) 1 h (vectors) SAG WERA 9.335 33.4 5 min 5 min (radials) 1 h (vectors) NEWC SeaSonde 5.2625 14 30 min 30 min (radials) 1 h (vectors) COF WERA 13.5 100 5 min 5 min (radials) 1 h (vectors)

At each WERA node, the two systems operate in alternate mode with short 5 min acquisition cycles to avoid mutual interference and for monitoring purposes. The receive array is composed of equally spaced 16 element linear arrays, and the standard 4 element conﬁguration is used for the transmit array. When operated in real-time mode, radial current maps are transmitted over a 3 G connection to a central server for QC and upload to the data portal after conversion to netCDF format. Every hour, radial maps from the two stations are averaged and converted to NRT-QC vector maps on a regular grid with 4 km (ROT and SAG) and 1.5 km (COF) horizontal resolutions. Vector maps in netCDF format are also uploaded every hour to the data portal and made available in near real Remote Sens. 2018, 10, x FOR PEER REVIEW 3 of 15

equally spaced 16 element linear arrays, and the standard 4 element configuration is used for the transmit array. When operated in real-time mode, radial current maps are transmitted over a 3 G connection to a central server for QC and upload to the data portal after conversion to netCDF format. Every hour, radial maps from the two stations are averaged and converted to NRT-QC Remote Sens. 2018, 10, 1476 3 of 15 vector maps on a regular grid with 4 km (ROT and SAG) and 1.5 km (COF) horizontal resolutions. Vector maps in netCDF format are also uploaded every hour to the data portal and made available time.in near Raw real data time. are Raw stored data on are site stored and on retrieved site and on retrieved a quarterly on a basis quarterly (every basis three (every months) three for months) offline processing.for offline processing. Proprietary Proprietary software provided software by provided the manufacturer by the manufacturer is used at site is and used in at offline site and mode in tooffline process mode the to raw process data; the real-time raw data; and real-time offline conversion, and offline quality conversion, control, quality and vectorcontrol, mapping and vector are performedmapping are through performed processing through tools processing developed tools in Pythondeveloped and in MATLAB Python and languages. MATLAB languages. Quality-controlled subsurface subsurface mooring mooring data data for for the the Rottnest Rottnest Shelf Shelf region region (Figure (Figure 1) 1are) are used used for forvalidation validation and and fine-tuning fine-tuning purposes, purposes, primarily primarily for forthe thelarge large number number of sustained of sustained moorings moorings and and the thewide wide range range of oceanographic of oceanographic processes processes in the in area the [5–7]. area [ 5The–7]. data The set data used set here used spans here a spans time aperiod time periodbetween between 01 February 01 February 2015, 2015, 00:00 00:00 UTC, UTC, and and 30 June30 June 2015, 2015, 23:59 23:59 UTC. UTC. Subsurface Subsurface current current data, collected by the IMOSIMOS AustralianAustralian National Mooring Network (ANMN) Facility, are availableavailable for download through through the the data data portal portal [14 – [14–22],22], and include and include Nortek Nortek 190 KHz 190 Continental KHz Continental current profilers current atprofilers WATR20 at and WATR20 WACA20 and moorings, WACA20 a moorings, 600 KHz RDI a 600 Workhorse KHz RDI ADCP Workhorse at WATR04 ADCP mooring, at WATR04 a RDI Workhorsemooring, a SentinelRDI Workhorse ADCP at Sentinel NRSROT, ADCP a RDI at Long NRSROT, Ranger a ADCPRDI Long at WATR50, Ranger andADCP a Nortek at WATR50, Aquadopp and Proa Nortek at WATR10 Aquadopp mooring Pro sites. at WATR10 Details on mooring the mooring sites. Details design andon the the mooring QC processing design can and be the found QC inprocessing [23]. can be found in [23].

Figure 1.1. Coverage of the HFR systems in the RottnestRottnest Shelf deployment, WA, withwith thethe dominantdominant bathymetric features. Red (black) markers are used for the WERA (SeaSonde) FRE and GUI stations (GHED and LANC); red (black) areas show the typical coverage of each HFR station. Mooring locations are marked as black diamonds inside the WERA HFR footprint. Remote Sens. 2018, 10, 1476 4 of 15

The QC metrics used here involve the calculation of correlation coefficient (R) and the root-mean-square difference (RMSD) between subsurface currents and HFR radial velocity time series at the current meter bin closest to surface and free from surface interference, and the radar radial velocity at the grid cell closest to the mooring locations. The following equations are used for R, RMSD, and the standard deviation σ: ∑n (x − x´)(y − y´) R = i=1 i i (1) (n − 1)σxσy s 2 ∑n (x − y ) RMSD = i=1 i i (2) n s 2 ∑n (x − x) σ = i=1 i . (3) x n − 1 Subsurface current data are projected onto the direction between the mooring location and the radar receiver before any comparisons. A synthetic SNR field that represent the typically observed effects of the 50 Hz contamination is used to test the capabilities of the iterative identification algorithm in a controlled environment and to test the effects of different background noise levels on the detection rate.

3. Results

3.1. Velocity Threshold Optimization Data QC on radial velocity is primarily based on the application of radial velocity thresholds. The conventional approach involves setting threshold velocity values a priori, using either long-term statistics of HFR data at each radar grid cell or quantitative analyses and comparisons with near-surface current data. Data can be derived from moorings, drifters, or even glider observations, provided that they are available within the radar footprint and that differences and limitations of each individual sampling strategy are properly accounted for. In the specific deployments across Australia, both approaches have proven to be reliable and complementary: under the assumption of normally distributed radial velocity data, 99% confidence levels (3 times the deviation from the mean, σ) can be set from the velocity distributions and anomalous radial velocity data can be identified and either flagged or removed. Examples of radial current distributions from the radar stations and the current meter data for the FRE HFR system are reported in Figure2. Regardless of the relative distances between near-surface radar data and mooring observations, which are 13–56 m below the surface, there is a surprisingly good agreement in terms of mean radial velocity and 99% confidence levels across all moorings. Before applying any velocity threshold, R values are in the range R = 0.15 to R = 0.86, with RMSD in the range 10.3 cm/s to 40.8 cm/s (Table2). After applying a 1.5 m/s (absolute value) on radial velocity, comparison metrics improved in general for both FRE and GUI HFR stations, with R increasing up to 25% or more and RMSD decreasing by up to 50% (Table2)

Table 2. Effects of velocity thresholds on radar data: a, no threshold applied; b, 1.5 m/s (absolute value) threshold applied.

Mooring Distance from Surface (m) FRE R/RMSD a FRE R/RMSD b GUI R/RMSD a GUI R/RMSD b NRSROT 15 0.44/11.6 0.43/11.3 0.52/26.2 0.71/21.6 WACA20 27 0.59/14.5 0.60/14.1 0.43/30.9 0.49/30.3 WATR04 14 0.67/17.1 0.86/9.7 0.72/10.3 0.72/10.3 WATR50 56 0.86/16.5 0.89/14.7 0.80/14.3 0.79/14.2 WATR10 43 0.30/32.0 0.54/21.4 0.25/12.5 0.24/12.4 WATR20 32 0.79/18.6 0.88/14.4 0.78/11.8 0.78/11.8 WACA20 32 0.63/10.3 0.62/10.0 0.36/30.8 0.54/20.8 WATR04 13 0.46/27.7 0.72/14.7 0.65/12.7 0.66/11.9 WATR10 43 0.15/40.8 0.36/21.5 0.31/10.7 0.30/9.7 Remote Sens. 2018, 10, x FOR PEER REVIEW 5 of 15

WATR10 43 0.30/32.0 0.54/21.4 0.25/12.5 0.24/12.4 WATR20 32 0.79/18.6 0.88/14.4 0.78/11.8 0.78/11.8 WACA20 32 0.63/10.3 0.62/10.0 0.36/30.8 0.54/20.8 RemoteWATR04 Sens. 2018 , 10, 1476 13 0.46/27.7 0.72/14.7 0.65/12.7 0.66/11.9 5 of 15 WATR10 43 0.15/40.8 0.36/21.5 0.31/10.7 0.30/9.7

Comparison metrics metrics are are in in good good agreement agreement with with previously previously reported reported findings findings in different in different ocean regionsocean regions [24–26 [24–26].]. Significant Significant differences differences and poor and poor statistical statistical comparison comparison results results can becan observed be observed for thefor radar-mooringthe radar-mooring pairs pairs FRE—WATR04 FRE—WATR04 (distance (distance from from surface surface 13 m), 13 FRE—WATR10 m), FRE—WATR10 (distance (distance from surfacefrom surface 43 m), 43 GUI—NRSROT m), GUI—NRSROT (distance (distance from surfacefrom surface 15 m), 15 and m), GUI—WACA20 and GUI—WACA20 (distance (distance from surfacefrom surface 32 m). 32 Differences m). Differences arise primarily arise primarily from radar from radial radar velocity radial often velocity exceeding often exceeding 1.5 m/s, which 1.5 m/s, are inconsistentwhich are inconsistent with both with HFR both data HFR at nearby data at locations,nearby locations, or in comparison or in comparison to mooring to mooring velocity velocity data. Atdata. these At locations,these locations, comparison comparison metrics metrics improve improve significantly significantly when 99%when thresholds 99% thresholds are applied are applied to the HFRto the observations HFR observations with increased with increased (decreased) (decreased) R (RMSD) R (RMSD) values values (Table2 (Table). 2).

(a)

(b)

Figure 2. 2.Examples Examples of radial of radial velocity velocity distributions distributions for FRE for HFR FRE station, HFR and station, corresponding and corresponding subsurface radialsubsurface velocity radial components, velocity components, at two deep-water at two mooring deep-water locations mooring within locations the radar within domain: the (a ) radar 56 m fromdomain: the surface;(a) 56 m (fromb) 32 the m fromsurface; the ( surface.b) 32 m Unitsfrom the for radialsurface. velocity Units for are radial m/s. velocity are m/s. 3.2. SNR Threshold Optimization 3.2. SNR Threshold Optimization The SNR of the Doppler lines in a Doppler spectra is in general considered to be a good proxy The SNR of the Doppler lines in a Doppler spectra is in general considered to be a good proxy for HFR data quality and, in combination with velocity threshold, are more than adequate to identify for HFR data quality and, in combination with velocity threshold, are more than adequate to identify and flag the majority of suspect anomalous data. For direction-finding systems such as the SeaSonde systems, it was shown in particular that low SNR values are useful for identifying anomalous data and that data quality generally improves as SNR increases [27]. Analyses of the five-month data set Remote Sens. 2018, 10, x FOR PEER REVIEW 6 of 15

and flag the majority of suspect anomalous data. For direction-finding systems such as the SeaSonde systems, it was shown in particular that low SNR values are useful for identifying anomalous data Remote Sens. 2018, 10, 1476 6 of 15 and that data quality generally improves as SNR increases [27]. Analyses of the five-month data set used here suggest that similar assumptions hold for phased-array radars such as the WERA systems. usedDistribution here suggest of radial that velocities similar assumptions for different hold classes for phased-array of SNR (not radarsshown suchhere) as suggest the WERA thatsystems. suspect Distributionradial velocities of radial exceeding velocities 1.5 m for/s different (absolute classes value) of magnitude SNR (not shown tend to here) be clustered suggest that between suspect 5 and radial 10 velocitiesdB, while exceedingradial velocities 1.5 m/s between (absolute [−1 value).5, −1] magnitudem/s and [1, tend 1.5] toor be+/− clustered 1 m/s span between similar 5 andranges 10 dB,for whileSNR, radialincluding velocities values between below the [− 1.5,10 dB−1] thresho m/s andld. As [1, 1.5]for radial or +/− velocity,1 m/s span SNRsimilar thresholds ranges are for usually SNR, includingdefined a valuespriori belowunder thethe 10rul dBe of threshold. thumb that As for data radial quality velocity, improves SNR thresholds with increasing are usually SNR definedvalues. aThresholds priori under are the also rule generally of thumb thatassumed data qualityto be time improves- or space with-invariant increasing; SNRi.e., it values. is assumed Thresholds they areare alsoconsta generallynt with assumedtime or space, to be and time- regional, or space-invariant; spatial, or temporal i.e., it is assumed variations they due are to constant such factors with as time noise or space,levels andor the regional, transmit spatial, patterns or temporal are not variations taken into due account. to such factorsRestrictive as noise thresholds levels or can the have transmit the patternsundesirable are noteffect taken of removing into account. valid Restrictive observations; thresholds on the can othe haver hand, the undesirable the opposite effect would of removing happen validwith poorly observations; constrained on the thresholds. other hand, the opposite would happen with poorly constrained thresholds. AnAn optimizationoptimization ofof thethe SNRSNR thresholdthreshold isis possiblepossible andand desirabledesirable usingusing independentindependent datasets: differentdifferent thresholdthreshold values values are are used, used, poorly poorly SNR-constrained SNR-constrained radial radial velocities velocities are removed,are removed, and Rand and R RMSDand RMSD metrics metrics are computedare computed with with the the aim aim of of finding finding that that particular particular value value for for which which aa significantsignificant changechange inin (R, (R, RMSD) RMSD) is obtained is obtained with anwith acceptable an acceptable data loss. data Results loss. for Results the Rottnest for the HFR Rottnest deployment HFR aredeployment summarized are summarized in Figure3. Figure in Figure3 in particular3. Figure 3 shows in particular that the shows (R, RMSD) that the for SNR(R, RMSD) thresholds for SNR in thethresholds 0–40 dB in range, the 0 while–40 dB Table range3 shows, while resultsTable 3 only shows for theresults default only threshold for the default used in threshold the proprietary used in softwarethe proprietary (6 dB) and software the 10 dB(6 dB) threshold and the for 10 which dB threshold no more statistically for which significantno more statistically changes are significant observed. changes are observed.

(a) Figure 3. Cont.

Remote Sens. 2018, 10, 1476 7 of 15 Remote Sens. 2018, 10, x FOR PEER REVIEW 7 of 15

(b)

FigureFigure 3.3. CorrelationCorrelation (R; (R; black black solid solid lines), lines), RMSD RMSD (dashed (dashed grey grey lines), lines) and, percentand percent data availabledata available (solid grey(solid line) grey as line) a function as a function of different of different SNR threshold SNR threshold values values for (a) GUIfor (a and) GUI (b )and FRE (b HFR) FRE stations, HFR stations, for all thefor mooringall the mooring deployments deployments in Figure in 1Figu. Unitsre 1. for Units RMSD for areRMSD cm/s. are cm/s.

InIn general, (R, (R, RMSD) RMSD) values values are are consistent consistent with with other other radar radar validation validation analyses analyses performed performed in indifferent different environments environments [24 [–2426].–26 Little]. Little or no or noeffect effect is introduced is introduced for forSNR SNR values values below below 4 d 4B, dB, in inagreement agreement with with the the minimum minimum threshol thresholdd applied applied during during the the inversion inversion of ofthe the Doppler Doppler spectra spectra to toradial radial currents. currents. For For both both HFR HFR stations stations used used here here (FRE (FRE and and GUI), GUI), the most statisticallystatistically significantsignificant changeschanges occuroccur between between 5 5 and and 10 10 dB. dB. In thisIn this range, range, RMSD RMSD values values drop drop from 40from to 10–1240 to cm/s,10–12R cm values/s, R increasevalues increase from R from < 0.2 R to< 0.2 R >to 0.7, R > while0.7, while data data loss loss is within is within 10–15%. 10–15%. The The analysis analysis also also showsshows thatthat somesome locationslocations areare lessless sensitivesensitive thanthan othersothers toto variationsvariations inin thethe SNRSNR thresholds,thresholds, andand nono statisticallystatistically significantsignificant changeschanges cancan bebe detecteddetected regardlessregardless ofof thethe appliedapplied SNRSNR thresholds.thresholds. 3.3. Advanced Artifact Removal 3.3. Advanced Artifact Removal When correctly chosen or fine-tuned, thresholds on radial velocity or SNR (Sections 3.1 and 3.2) When correctly chosen or fine-tuned, thresholds on radial velocity or SNR (Sections 3.1 and 3.2) are in general capable of handling the majority of anomalous velocities in a radial map. There may be are in general capable of handling the majority of anomalous velocities in a radial map. There may cases, however, when artifacts appear in the radial velocity field, such as unrealistic velocities at fixed be cases, however, when artifacts appear in the radial velocity field, such as unrealistic velocities at and known range cells, i.e., distances, from the HFR receiver. Most of the time, similar artifacts are fixed and known range cells, i.e., distances, from the HFR receiver. Most of the time, similar artifacts introduced by either external radio frequency interference (RFI) or by modulations of the 50–60 Hz are introduced by either external radio frequency interference (RFI) or by modulations of the 50–60 220 V power line. As a consequence, strong signals appear in the range-Doppler spectra at multiples of Hz 220 V power line. As a consequence, strong signals appear in the range-Doppler spectra at the 50 Hz frequency, for instance at approximately 20, 40, 60, 80, 100, and 120 km offshore (Figure4a); multiples of the 50 Hz frequency, for instance at approximately 20, 40, 60, 80, 100, and 120 km at higher harmonics, this signal also tends to spread over frequency and potentially interfere with the offshore (Figure 4a); at higher harmonics, this signal also tends to spread over frequency and detection of the Doppler peaks, thus introducing spurious radial currents. potentially interfere with the detection of the Doppler peaks, thus introducing spurious radial There are several possible ways of dealing with this contamination. A first approach, presently currents. implemented and operational across the Australian Ocean Radar HFR network, makes use of an There are several possible ways of dealing with this contamination. A first approach, presently iterative method that fits a 1-D or 2-D reference signal to the radial SNR map, and identifies and implemented and operational across the Australian Ocean Radar HFR network, makes use of an flags anomalous data that clearly stand out from the background SNR values. Other approaches, iterative method that fits a 1-D or 2-D reference signal to the radial SNR map, and identifies and currently under investigation, act on the I/Q time series of the range-gated signals at each antenna, flags anomalous data that clearly stand out from the background SNR values. Other approaches, and subsequently identify and filter any anomalous signal in the frequency domain in a manner similar currently under investigation, act on the I/Q time series of the range-gated signals at each antenna, to [28–31]. The conventional beam-forming and the Doppler spectra inversion steps are then applied and subsequently identify and filter any anomalous signal in the frequency domain in a manner on the filtered data. This section focuses on the first approach, which is operationally implemented similar to [28–31]. The conventional beam-forming and the Doppler spectra inversion steps are then applied on the filtered data. This section focuses on the first approach, which is operationally implemented across the network for both NRT and DM operations. Other methods still need to be

Remote Sens. 2018, 10, 1476 8 of 15 Remote Sens. 2018, 10, x FOR PEER REVIEW 8 of 15 acrossimplemented the network across for the both network NRT for and both DM NRT operations. and DM Other operations. methods Other still methods need to be still optimized need to be in orderoptimized to account in order for to the account spreading for the over spreading Doppler over for the Doppler higher-order for the higher-order harmonics (Figure harmonics4a). (Figure 4a).

(a) (b)

Figure 4. Range-DopplerRange-Doppler spectra spectra for for Receive Receive Antenna Antenna #09 #09 at atthe the Red Red Rock Rock (RRK) (RRK) HFR HFR station station (a) (showinga) showing both both the the contamination contamination from from the the 50 50 Hz Hz across across range range (distance) (distance) from from the the receive receive array array in the ﬁrstfirst order order Bragg Bragg peaks peaks and and the thespectral spectral spreading spreading over Doppler; over Doppler; (b) the (radialb) the velocity radial map velocity extracted map fromextracted this from data set,this documentingdata set, documenting the spatial the inconsistencies spatial inconsistencies in radial in velocity radial mapsvelocity at ﬁxedmaps annularat fixed rangesannular from ranges the from antennas. the antennas. Peak SNR Peak values SNR in values these annularin these regionsannular typicallyregions typically exceed 40 exceed dB. 40 dB.

An example of heavily contaminated range-Doppler spectra and corresponding radial map is given inin FigureFigures4a,b, 4a,b, respectively. respectively. Data Data shown shown here here refer refer to the to HFR the HFR station station at Red at Rock Red (RRK),Rock (RRK), NSW, whereNSW, where the 50 the Hz 50 contamination Hz contamination is particularly is particularly significant, significant, although although it is commonly it is commonly observed observed at all theat all installations the installations with the exception with the of exception Cape Wiles of (CWI), Cape SA. Wiles The range-Doppler(CWI), SA. The spectrum-derived range-Doppler followingspectrum-derived [32,33] (Figure following4a) shows [32,33] the (Figure expected 4a) Bragg shows peaks the from expected the backscattering Bragg peaks ocean from waves, the butbackscattering also strong ocean signals waves, at 20, 40,but 60, also 80, strong 100, 120, signals and 140at 20, km 40, offshore, 60, 80, with100, SNR120, and values 140 comparable km offshore, in magnitudewith SNR values to the comparable dominant Bragg in magnitude peaks. While to the at dominant closer ranges Bragg this peaks. feature While is wellat closer separated ranges from this thefeature true is Doppler well separated peaks, the from spreading the true overDoppler frequency peaks, thatthe spreading is observed over with frequency range biases that theis observed Doppler peakwith range detection biases at 60,the 80,Doppler 100, 120, peak and detection 140 km at offshore. 60, 80, 100, Effects 120, are and evident 140 km in offshore. both the radialEffects andare SNRevident maps, in both where the well-defined radial and rings SNR are maps, clearly where detected well-defined and inconsistent rings are radial clearly current detected patterns and areinconsistent found. Since radial the current SNR values patterns associated are found. with Since radial the currents SNR values at these associated range cells with typically radial currents exceed 30–40at these dB, range any spikecells typically identification exceed based 30–40 on dB, the any conventional spike identification SNR threshold based wouldon the failconventional detection, removeSNR threshold potentially would valid observations, fail detection, and remove propagate potentially anomalous valid observations observations, through and to propagate the vector mappinganomalous step. observations Similarly, through a conventional to the vector threshold mapping on radialstep. Similarly, velocity woulda conventional possibly threshold remove theon obviouslyradial velocity erroneous would data possibly points; remove however, the theobviously method erroneous would fail data to identify points; thehowever, problematic the method range cells.would For fail this to identify specific example,the problematic poorly SNRrange constrained cells. For this radial specific currents example, are still poorly within SNR realistic constrained ranges ([radial−1.5, currents1.5] m/s). are Radial still velocities within realistic exceeding ranges a maximum ([−1.5, value 1.5] m/s). of 1.5 Radial m/s are velocities flagged; some exceeding of them a aremaximum associated value with of poor1.5 m/s SNR are values, flagged; whilst some others of them exceed are the associated 10 dB threshold. with poor Peaks SNR at 80values, and 120whilst km cannotothers exceed be identified the 10 using dB threshold. standard Peaksconstraints at 80 on and SNR 120 or km radial cannot velocity. be identified In contrast, using most standard of the observationsconstraints on associated SNR or radial with thevelocity. peak atIn the contrast, 140 km most range of may the beobservations flagged and associated removed with SNRthe peak and speedat the 140 combined. km range may be flagged and removed with SNR and speed combined. The artifact removal procedure developed at the Ocean Radar facility fitsfits either a 1-dimensional polynomial to to the the SNR SNR distribution distribution along along each each radial radial direction, direction, or ora 2-D a 2-D surface surface on the on SNR the SNR map mapand, and,on the on basis the basisof a “distance” of a “distance” between between the model the model and the and data, the data,identifies identifies and flags and flagssuspect suspect data. data.The Theprocedure procedure is iterative is iterative and and continuously continuously updates updates the the distance distance between between the the fit fit model model and and the observations until no suspect data points are found, or a maximum number of iterations is reached. Detailed steps are provided below:

Remote Sens. 2018, 10, 1476 9 of 15 observations until no suspect data points are found, or a maximum number of iterations is reached. DetailedRemote Sens. steps 2018, are10, x provided FOR PEER below:REVIEW 9 of 15

1.1. fitfit a polynomial model to the SNR distribution along each radar beam (1-D case) or fitfit a 2-D2-D surfacesurface to to the the SNR SNR map; map; 2. estimateestimate a a “distance” “distance” between between the the data data and and the the polynomial fit fit and set confidence confidence levels as n * σ(data-fit),σ(data-fit), where n = 2, 2, 3; 3; 3. removeremove suspect suspect SNR SNR data data outside of the confidence confidence levels; 4. movemove to to the the next next radial radial beam (1-D case); 5. repeatrepeat Steps Steps 1–4 1–4 until ( i) no more anomalous data are detected or (iiii) aa maximummaximum numbernumber of iterationsiterations is is reached. reached.

For the the 1-D 1-D case, case, the actualthe actual implementation implementation of the algorithm of the algorithm allows for allows the choice for of the a second-order, choice of a third-order,second-order, or third-order, exponential or function, exponential while function, for the while 2-D casefor the two 2-D options case two that options fit either that afit third- either or a fifth-orderthird- or fifth-order surface to surface the SNR to mapsthe SNR are maps available. are available. An example An example of the results of the of results the 1-D of iterativethe 1-D identificationiterative identification and removal and removal along one along specific one specific radial beam radial (92 beam◦ NCW) (92° NCW) is provided is provided in Figure in Figure5 for the5 for cubic the cubic fit (Figure fit (Figure5a), the 5a), quadratic the quadratic fit (Figure fit (Figure5b), and 5b), exponential and exponential decay (Figure decay 5(Figurec). No other5c). No thresholdsother thresholds are applied are applied to this to example. this example.

(a)

(b)

Figure 5. Cont.

Remote Sens. 2018, 10, 1476 10 of 15 Remote Sens. 2018, 10, x FOR PEER REVIEW 10 of 15

(c)

Figure 5. Example of the effects of the proposed 1-D QC methodologymethodology on radial velocity map from RRK HFR station along one radial beam: ( a) cubic 1-D fit; fit; (b) quadratic 1-D fit;fit; (c) exponential decay 1-D fit.fit. Units are km forfor offshoreoffshore range,range, dBdB forfor SNR,SNR, andand m/sm/s for radial velocity (color scale).

In order toto testtest thethe effectivenesseffectiveness of thethe 1-D1-D andand 2-D2-D formulationformulation andand optimizeoptimize the choicechoice of thethe fitfit function, a series of sensitivitysensitivity tests have been performed on a synthetic SNR map with realistic patterns, forfor instanceinstance matching matching the the one one that that generated generated the the radial radial current current shown shown in Figure in Figure4b. The 4b. SNR The modelSNR model used used here ishere a simple is a simple linear linear decay decay over over range range for all for bearings, all bearings, which which starts starts from from 60 dB 60 atdB the at gridthe grid cell closestcell closest to the to receiverthe receiver site andsite decreasesand decreases to 4 dBto at4 dB 180 at km 180 offshore. km offshore. A Gaussian-shaped A Gaussian-shaped noise functionnoise function is added is added to range to cells range 90–95 cells and 90–95 120–128 and with120–128 a peak with amplitude a peak amplitude of 50 dB, andof 50 to dB, range and cells to 150–159range cells with 150–159 a peak with amplitude a peak of amplitude 15 dB, so asof to15 reproduce dB, so as ato typical reproduce pattern a typical in the data.pattern An in additional the data. noiseAn additional term is included noise term in theis included sensitivity in the tests, sensitivity with variance tests, with levels variance of 0 (no levels noise), of 1,0 2,(no 5, noise), and 10 1, dB. 2, No5, and assumptions 10 dB. No areassumptions made in regard are made to directional in regard to patterns. directional Sensitivity patterns. analysis Sensitivity aimed analysis to determine aimed theto determine speed, the the number speed, of the iterations number required of iterations to flag required and remove to flag the and known remove artifacts, the known and the artifacts, number ofand “false-flag” the number i.e., of incorrect“false-flag” detections, i.e., incorrect as a function detections, of the as noisea function level. of the noise level. Results for the 1-D case for both the noise-free and randomly distributed noise cases can be summarized as follows. In In the the absence absence of of noise, the quadratic, cubic, and exponential fitfit equally succeed in detecting and flaggingflagging more than 70% of thethe anomalies,anomalies, whilewhile minimizingminimizing the false-flagfalse-flag detection (0% (0% in in this this specific specific realization). realization). On the On other the hand, other the hand, exponential the exponential fit requires fit a significantly requires a largersignificantly number larger of iterationsnumber of to iterations achieve resultsto achieve comparable results comparable to the quadratic to the quadratic and cubic and function; cubic however,function; however, there is no there major is improvementno major improvement in the detection in the ratedetection as the rate number as the of number iterations of increases.iterations Detectionincreases. ratesDetection decrease rates when decrease noise when is added noise to is the added SNR maps,to the SNR but the maps, lower but success the lower rate success is related rate to theis related missed to identification the missed identification of the weaker of SNR the peaksweaker at SNR the outer peaks ranges at the rather outer than ranges an increasingrather than rate an ofincreasing false-detection. rate of false-detection. When applied to fieldfield data, results suggest that a third-order polynomial (Figure 66c)c) isis moremore effective in in identifying identifying and and removing removing anomalous anomalous observations observations from the from radial the maps radial than maps the quadratic than the (Figurequadratic6b) (Figure or exponential 6b) or exponential fit (Figure6 fitd). (Figure For this 6d). specific For this example, specific the example, artifact removalthe artifact is applied removal to is radialapplied currents to radial from currents one station from (RRK), one station and additional (RRK), and thresholds additional have thresholds been applied have to been radial applied SNR and to radial velocitySNR and before radial the velocity polynomial before fit.the polynomial fit. Results for the 2-D fitfit improve with respect to the 1-D case when noise is added to the synthetic SNR test test case, case, but but no no improvement improvement is obtained is obtained when when a noise a term noise is term excluded. is excluded. In general, In the general, detection the ratedetection decreases rate with decreases the increasing with the noise increasing level similarly noise level to the similarly 1-D case, to but the the 1-D false case, detection but the rate false is significantlydetection rate lower is significantly in the 2-D caselower for in the the same 2-D noisecase for level. the Processing same noise speed level. also Processing increases speed slightly also in respectincreases to slightly the 1-D in case respect when to noise the 1-D is added case when to the noise SNR is map. added When to the applied SNR map. to a realWhen radial applied velocity to a map,real radial no major velocity differences map, no can major be found differences between can the be third- found or between fifth-order the surfaces, third- or yetfifth-order processing surfaces, speed greatlyyet processing improves—up speed greatly to a factor improves—up of 10 compared to a factor to the of 1-D 10 case.compared to the 1-D case.

Remote Sens. 2018, 10, 1476 11 of 15 Remote Sens. 2018, 10, x FOR PEER REVIEW 11 of 15

(a) (b)

Figure 6.6. Effects of the proposed QC methodology on radial velocity map from the RRK HFR station: ((aa)) nono QCQC applied;applied; ((bb)) quadraticquadratic 1-D1-D ﬁt;fit; ((cc)) cubiccubic 1-D;1-D; ((dd)) exponentialexponential decaydecay 1-D1-D ﬁt.fit. (b–d)(b–d) alsoalso include thresholdthreshold onon radialradial velocityvelocity andand SNR.SNR. UnitsUnits forfor radialradial velocityvelocity areare m/s.m/s.

4. Discussion Routinely, QC QC procedures procedures for for HFR HFR radial radial velocity velocity maps maps rely onrely the on definition the definition of threshold of threshold values thatvalues are that applied are applied to SNR to or SNR radial or velocityradial velocity speed. speed. Examples Examples of operational of operational procedures procedures applied applied in RT modein RT formode HFR for in HFR general in general and WERA and systemsWERA insystems particular in particular can be found can be in [found28,29], in while [28,29], advanced while methodologiesadvanced methodologies capable of handling capable ofcomplex handling cases complex of external cases radio of frequency external interferencesradio frequency are describedinterferences in [ 13are]. Indescribed most cases, in [13]. these In values most are cases, defined these “a values priori” are without defined any “a assumption priori” without on the realany rangeassumption of variability on the or real the physicalrange of processes variability in the or theregion. physical This work processes suggests in that the aregion. fine-tuning This of work the thresholdssuggests that is beneficiala fine-tuning to improving of the thresholds the overall is beneficial quality ofto theimproving HFR data the set. overall Velocity quality thresholds of the HFR can bedata set set. through Velocity comparisons thresholds with can availablebe set through data fromcomparisons independent with sampling available devicesdata from or, alternatively,independent usingsampling long-term devices statistics or, alternatively, of radar observations. using long-term A proper statistics definition of ofradar the radial observations. velocity thresholdA proper significantlydefinition of improvesthe radial thevelocity comparison threshold metrics significantly at several improves locations the across comparison the HFR metrics domain at for several both FRElocations and GUIacross radar the stationsHFR domain (Table for2). both FRE and GUI radar stations (Table 2). InIn agreementagreement with previous studies, SNR can be considered a a valid valid proxy proxy for for HFR HFR data data quality. quality. However, similarly similarly to toradial radial velocity, velocity, the SNR the threshold SNR threshold needs to needs be tuned to for be each tuned specific for each deployment, specific deployment, in order to avoid removing potentially valid observations. Tests performed across several mooring locations also suggest that a single threshold value may not be adequate for the

Remote Sens. 2018, 10, 1476 12 of 15 in order to avoid removing potentially valid observations. Tests performed across several mooring locations also suggest that a single threshold value may not be adequate for the entire domain and that this threshold may vary over time as a result of varying environmental factors, external noise, interference sources, or other problems. The analysis performed in Section 3.2 suggests that small variations in the SNR threshold can greatly improve data quality without signiﬁcantly compromising the data availability. For example, setting a minimum SNR of 7–8 dB instead of the default threshold results in signiﬁcant improvements (Table3). Operationally, this SNR threshold was increased to 10 dB for the Australian HFR systems, and the combined effects of SNR and radial velocity thresholds are documented in Table4.

Table 3. Effects of SNR thresholds on radar data: a no threshold applied; b threshold applied.

Mooring Distance from Surface (m) FRE R/RMSD a FRE R/RMSD b GUI R/RMSD a GUI R/RMSD b NRSROT 15 0.44/11.6 0.44/11.5 0.52/26.2 0.75/20.9 WACA20 27 0.59/14.5 0.58/14.3 0.43/30.9 0.50/30.5 WATR04 14 0.67/17.1 0.89/8.4 0.72/10.3 0.72/10.2 WATR50 56 0.86/16.5 0.88/14.5 0.80/14.3 0.80/14.2 WATR10 43 0.30/32.0 0.59/20.4 0.25/12.5 0.24/12.5 WATR20 32 0.79/18.6 0.88/14.2 0.78/11.8 0.78/11.8 WACA20 32 0.63/10.3 0.63/10.1 0.36/30.8 0.66/17.1 WATR04 13 0.46/27.7 0.86/9.0 0.65/12.7 0.71/11.5 WATR10 43 0.15/40.8 0.66/10.8 0.31/10.7 0.31/10.6

Table 4. Effects of SNR and velocity thresholds on radar data: a no threshold applied; b threshold applied.

Mooring Distance from Surface (m) FRE R/RMSD a FRE R/RMSD b GUI R/RMSD a GUI R/RMSD b NRSROT 15 0.44/11.6 0.45/11.4 0.52/26.2 0.77/20.6 WACA20 27 0.59/14.5 0.82/13.8 0.43/30.9 0.50/30.5 WATR04 14 0.67/17.1 0.89/8.4 0.72/10.3 0.72/10.1 WATR50 56 0.86/16.5 0.89/14.3 0.80/14.3 0.81/13.9 WATR10 43 0.30/32.0 0.59/20.4 0.25/12.5 0.25/12.4 WATR20 32 0.79/18.6 0.89/13.9 0.78/11.8 0.78/11.8 WACA20 32 0.63/10.3 0.84/9.9 0.36/30.8 0.66/17.1 WATR04 13 0.46/27.7 0.88/8.7 0.65/12.7 0.72/11.3 WATR10 43 0.15/40.8 0.69/10.1 0.31/10.7 0.32/10.2

Thresholds on radial velocity or SNR may both be inadequate when more complex cases arise, such as the contamination of the 50 Hz signal, that can significantly bias direction and speed of the resulting current vector when propagated to the step of vector mapping. For this purpose, a fast and effective iterative approach has been developed that iteratively identifies and removes anomalous radial observations from a radial map based on the spatial distribution of the SNR maps and the basic assumption that its spatial distribution can be modeled as a cubic polynomial (1-D case) or a third-degree surface in a Cartesian (X, Y) plane. The first method can be considered an improvement to the method proposed in [10] in its application to ship detection and tracking. In combination with recent advances in the waveform design and implementation [13], the proposed method has the potential to improve radial data quality for WERA HFR systems. The actual implementation of the procedure uses a Python code for the NRT applications where computational speed needs to be optimized to run every 5 min across the network. An equivalent version in MATLAB language, which also includes the 2-D fit, can be run both in NRT and in DM for debugging, testing, or development purposes.

5. Conclusions NRT and DM QC procedures developed for ocean HFR systems can greatly improve the quality of HFR station data and resulting vector maps, thus facilitating data uptake and usage. This study describes new methods that optimize data quality using a combination of radial velocity and SNR thresholds. Thresholds can be set through quantitative analyses and comparisons with independent data sets, when available, or a self-validation approach that uses distributions and statistics from HFR Remote Sens. 2018, 10, 1476 13 of 15 data. Methodologies can be applied both in NRT and DM operations, and, although designed for radial velocity data collected from commercial-type WERA HF systems, they can easily be extended to any other type or HFR for oceanographic use.

Author Contributions: Conceptualization: S.C.; Methodology: S.C.; Software: S.C. and B.G.; Validation: S.C. and B.G.; Formal Analysis: S.C.; Resources: S.C. and B.G.; Data Curation: S.C., S.d.V., and Y.H.; Writing-Original Draft Preparation: S.C.; Writing-Review & Editing: S.C., Y.H.; Project Administration: S.C.; Funding Acquisition: S.C. Funding: This research was funded by Integrated Marine Observing System IMOS—Ocean Radar—2017–2019 grant number 53000300. Acknowledgments: Data was sourced from the Integrated Marine Observing System (IMOS)—IMOS is supported by the Australian Government through the National Collaborative Research Infrastructure Strategy and the Super Science Initiative. Subsurface current data were collected by the IMOS Australian National Mooring Network (ANMN) Facility. The WERA data used here were collected by the Ocean Radar Facility at the University of Western Australia. Conﬂicts of Interest: The authors declare no conﬂict of interest.

References

1. Archer, M.R.; Keating, S.R.; Roughan, M.; Johns, W.E.; Lumpkin, R.; Beron-Vera, F.; Shay, L.K. The kinematic similarity of two western boundary currents revealed by sustained high-resolution observations. Geophys. Res. Lett. 2018, 45, 6176–6185. [CrossRef] 2. Archer, M.R.; Roughan, M.; Keating, S.; Schaeffer, A. On the Variability of the East Australian Current: Jet Structure, Meandering, and Inﬂuence on Shelf Circulation. J. Geophys. Res. Oceans 2017, 122, 8464–8481. [CrossRef] 3. Kerry, C.; Powell, B.; Roughan, M.; Oke, P. Development and evaluation of a high-resolution reanalysis of the East Australian Current region using the Regional Ocean Modelling System (ROMS 3.4) and Incremental Strong-Constraint 4-Dimensional Variational (IS4D-Var) data assimilation. Geosci. Model Dev. 2016, 9, 3779–3801. [CrossRef] 4. Mantovanelli, A.; Keating, S.; Wyatt, L.R.; Roughan, M.; Schaeffer, A. Lagrangian and Eulerian characterization of two counterrotating submesoscale eddies in a western boundary current. J. Geophys. Res. Oceans 2018, 122, 4902–4921. [CrossRef] 5. Mihanovic, H.; Pattiaratchi, C.; Verspecht, F. Diurnal Sea Breezes Force Near-Inertial Waves along Rottnest Continental Shelf, Southwestern Australia. J. Phys. Oceanogr. 2016, 46, 3487–3508. [CrossRef] 6. Schaeffer, A.; Gramoulle, A.; Roughan, M.; Mantovanelli, A. Characterizing frontal eddies along the East Australian Current from HF radar observations. J. Geophys. Res. Oceans 2017, 122, 3964–3980. [CrossRef] 7. Wandres, M.; Wijeratne, E.M.S.; Cosoli, S.; Pattiaratchi, C. The effect of the Leeuwin Current on offshore surface gravity waves in southwestwestern Australia. J. Geophys. Res. Oceans 2017, 122, 9047–9067. [CrossRef] 8. Manual for Real-Time Quality Control of High Frequency Radar Surface Current Data. Available online: https: //cdn.ioos.noaa.gov/media/2017/12/HFR_QARTOD_Manual_05_26_16.pdf (accessed on 20 August 2018). 9. Gurgel, K.W.; Barbin, Y.; Schlick, T. Radio frequency interference suppression techniques in FMCW modulated HF radars. In Proceedings of the IEEE/Oceans’07 Europe 2007, Aberdeen, UK, 18–21 June 2007. 10. Dzvonkovskaya, A.; Gurgel, K.W.; Rohling, H.; Schlick, T. HF radar WERA application for ship detection and tracking. Eur. J. Navig. 2009, 7, 18–25. 11. Middleditch, A.; Cosoli, S. Operational data management procedures for the Australian Coastal Ocean Radar Network. In Proceedings of the OCEANS 2016 MTS/IEEE Monterey, Monterey, CA, USA, 19–23 September 2016. [CrossRef] 12. Gomez, R.; Helzel, T.; Merz, C.R.; Liu, Y.; Weisberg, R.H.; Thomas, N. Improvements in ocean surface radar applications through real-time data quality-control. In Proceedings of the 2015 IEEE/OES Eleventh Current, Waves and Turbulence Measurement (CWTM), St. Petersburg, FL, USA, 2–6 March 2015. [CrossRef] 13. Dzvonkovskaya, A.; Petersen, L.; Helzel, T. HF Ocean Radar with a triangle waveform implementation. In Proceedings of the 19th International Radar Symposium IRS 2018, Bonn, Germany, 20–22 June 2018; ISBN 978-3-7369-9545-1. Remote Sens. 2018, 10, 1476 14 of 15

14. Integrated Marine Observing System (IMOS), 2018, WACA20 September 2014. Available online: http://data. aodn.org.au/IMOS/ANMN/WA/WACA20/Velocity/IMOS_ANMN-WA_AETVZ_20140919T080000Z_ WACA20_FV01_WACA20-1409-Continental-194_END-20150225T040128Z_C-20150408T024315Z.nc (accessed on 7 April 2018). 15. Integrated Marine Observing System (IMOS), 2018, WATR04 September 2014. Available online: http://data.aodn.org.au/IMOS/ANMN/WA/WATR04/Velocity/IMOS_ANMN-WA_AETVZ_ 20140925T080000Z_WATR04-ADCP_FV01_WATR04-ADCP-1409-Workhorse-ADCP-41_END- 20150420T030000Z_C-20150422T054233Z.nc (accessed on 7 April 2018). 16. Integrated Marine Observing System (IMOS), 2018, WATR50 September 2014. Available online: http://data. aodn.org.au/IMOS/ANMN/WA/WATR50/Velocity/IMOS_ANMN-WA_AETVZ_20140925T080000Z_ WATR50_FV01_WATR50-1409-Workhorse-ADCP-496_END-20150517T184000Z_C-20150522T092647Z.nc (accessed on 7 April 2018). 17. Integrated Marine Observing System (IMOS), 2018, WATR10 November 2014. Available online: http://data. aodn.org.au/IMOS/ANMN/WA/WATR10/Velocity/IMOS_ANMN-WA_AETVZ_20141117T080000Z_ WATR10_FV01_WATR10-1411-Aquadopp-Profiler-94_END-20150420T064500Z_C-20150422T065435Z.nc (accessed on 7 April 2018). 18. Integrated Marine Observing System (IMOS), 2018, NRSROT February 2015. Available online: http://data.aodn.org.au/IMOS/ANMN/NRS/NRSROT/Velocity/IMOS_ANMN-NRS_AETVZ_ 20150203T080000Z_NRSROT-ADCP_FV01_NRSROT-ADCP-1502-Workhorse-ADCP-42_END- 20150517T033000Z_C-20150702T032647Z.nc (accessed on 7 April 2018). 19. Integrated Marine Observing System (IMOS), 2018, WATR20 February 2015. Available online: http://data. aodn.org.au/IMOS/ANMN/WA/WATR20/Velocity/IMOS_ANMN-WA_AETVZ_20150213T080000Z_ WATR20_FV01_WATR20-1502-Continental-194_END-20150424T134002Z_C-20150925T092052Z.nc (accessed on 7 April 2018). 20. Integrated Marine Observing System (IMOS), 2018, WACA20 March 2015. Available online: http://data. aodn.org.au/IMOS/ANMN/WA/WACA20/Velocity/IMOS_ANMN-WA_AETVZ_20150319T080000Z_ WACA20_FV01_WACA20-1503-Continental-194_END-20150925T023000Z_C-20150925T081259Z.nc (accessed on 7 April 2018). 21. Integrated Marine Observing System (IMOS), 2018, WATR04 April 2015. Available online: http://data.aodn.org.au/IMOS/ANMN/WA/WATR04/Velocity/IMOS_ANMN-WA_AETVZ_ 20150417T080000Z_WATR04-ADCP_FV01_WATR04-ADCP-1504-Workhorse-ADCP-40_END- 20150915T034000Z_C-20150918T025944Z.nc (accessed on 7 April 2018). 22. Integrated Marine Observing System (IMOS), 2018, WATR10 May 2015. Available online: http://data. aodn.org.au/IMOS/ANMN/WA/WATR10/Velocity/IMOS_ANMN-WA_AETVZ_20150521T080000Z_ WATR10_FV01_WATR10-1505-Aquadopp-Profiler-94_END-20151125T042547Z_C-20151126T051809Z.nc (accessed on 7 April 2018). 23. Western Australia Moorings. Available online: http://imos.org.au/facilities/nationalmooringnetwork/ wamoorings/ (accessed on 7 April 2018). 24. Graber, H.C.; Haus, B.K.; Chapman, R.D.; Shay, L.K. HF radar comparisons with moored estimates of current speed and direction: Expected differences and implications. J. Geophys. Res. 1997, 102, 18749–18766. [CrossRef] 25. Liu, Y.; Weisberg, H.R.; Merz, C.R. Assessment of CODAR SeaSonde and WERA HF Radars in Mapping Surface Currents on the West Florida Shelf. J. Atmos. Ocean. Technol. 2014, 31, 1363–1382. [CrossRef] 26. Wyatt, L.R.; Mantovanelli, A.; Heron, M.; Roughan, M.; Steinberg, C. Assessment of Surface Currents Measured with High-Frequency Phased-Array Radars in Two Regions of Complex Circulation. IEEE J. Ocean. Eng. 2017, 43, 484–505. [CrossRef] 27. Cosoli, S.; Bolzon, G.; Mazzoldi, A. A Real-Time and Offline Quality Control Methodology for SeaSonde High-Frequency Radar Currents. J. Atmos. Ocean. Technol. 2012, 29, 1313–1328. [CrossRef] 28. Levkov, C.; Mihov, G.; Ivanov, R.; Daskalov, I.; Christov, I.; Dotsinsky, I. Removal of power-line interference from the ECG: A review of the subtraction procedure. BioMed. Eng. Online 2005, 4, 50. [CrossRef][PubMed] 29. Suchetha, M.; Kumaravel, N.; Jagannatha, M.; Jaganathan, S.K. A comparative analysis of EMD based filtering methods for 50 Hz noise cancellation in ECG signal. Inform. Med. Unlocked 2017, 8, 54–59. [CrossRef] Remote Sens. 2018, 10, 1476 15 of 15

30. Verma, A.R.; Singh, Y. Adaptive Tunable Notch Filter for ECG Signal Enhancement. Procedia Comput. Sci. 2015, 57, 332–337. [CrossRef] 31. Nguyen, P.; Kim, J.-M. Adaptive ECG denoising using genetic algorithm-based thresholding and ensemble empirical mode decomposition. Inf. Sci. 2016, 373, 499–511. [CrossRef] 32. Thomson, D.J. Spectrum estimation and harmonic analysis. Proc. IEEE 1982, 70, 1055–1096. [CrossRef] 33. Percival, D.B.; Walden, A.T. Spectral Analysis for Physical Applications: Multitaper and Conventional Univariate Techniques; Cambridge University Press: Cambridge, UK, 1993.

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).