This is a repository copy of Assessment of surface currents measured with high-frequency phased-array radars in two regions of complex circulation.

Transcription

1 Ths s a repostory copy of Assessment of surface currents measured wth hgh-frequency phased-array radars n two regons of comple crculaton. Whte Rose Research Onlne URL for ths paper: Verson: Accepted Verson Artcle: Wyatt, L.R. orcd.org/ , Mantovanell, A., Heron, M.L. et al. ( more authors) (07) Assessment of surface currents measured wth hgh-frequency phased-array radars n two regons of comple crculaton. IEEE Journal of Oceanc Engneerng. ISSN Reuse Unless ndcated otherwse, fulltet tems are protected by copyrght wth all rghts reserved. The copyrght ecepton n secton 9 of the Copyrght, Desgns and Patents Act 988 allows the makng of a sngle copy solely for the purpose of non-commercal research or prvate study wthn the lmts of far dealng. The publsher or other rghts-holder may allow further reproducton and re-use of ths verson - refer to the Whte Rose Research Onlne record for ths tem. Where records dentfy the publsher as the copyrght holder, users can verfy any specfc terms of use on the publsher s webste. Takedown If you consder content n Whte Rose Research Onlne to be n breach of UK law, please notfy us by emalng eprnts@whterose.ac.uk ncludng the URL of the record and the reason for the wthdrawal request. eprnts@whterose.ac.uk

2 Assessment of surface currents measured wth hgh-frequency phased-array radars n two regons of comple crculaton L. R.Wyatt, Senor Member, IEEE, A. Mantovanell, M. L. Heron, Fellow, IEEE, M. Roughan and C. R. Stenberg Abstract Surface velocty data from two WERA hgh frequency (HF) ocean radar systems, deployed as part of the Australan Integrated Marne Observng System (IMOS), are compared wth near surface currents obtaned from drfters and ADCPs (acoustc Doppler current profler). We evaluate data from two contrastng locatons n the frst detaled evaluaton of the IMOS HF radar surface veloctes. HF radar measurements are generally robust but demand qualty-control procedures to elmnate obvous errors and outlers that appear temporarly or systematcally n the data. A number of dfferent qualty control procedures and flters are appled and assessed ncludng Taylor dagrams, Hampel and Savtzky-Golay flters. In addton the need for and effect of averagng are dscussed. The radar measurements of surface current agreed better wth the near-surface drfter currents than wth the subsurface ADCP currents. Nonetheless the ADCP comparsons are consstent wth those prevously reported n other regons. The value of the Taylor Dagram for comparng dfferent surface current data sets and processng approaches s demonstrated. Nose levels n the radar current spectra are used to estmate the error n the measurements and n some cases, these errors were found to approach the precson of the radar estmates. Our results gve gudance on the most useful temporal samplng resoluton. In partcular we show that, at these stes and these operatng frequences, usng 0-mnute samplng wthout further averagng does not provde addtonal nformaton because the hgher frequences are domnated by nose. Averagng the radals over 30-mnutes may be suffcent for many applcatons. Inde Terms ADCP, coastal currents, Coffs Harbour, drfter, Eastern Australa, Great Barrer Reef, Hampel flter, HF radar, qualty-control, Savtzky-Golay flter, Taylor Dagram, WERA ACORN and ANMN are Facltes of Australa s IMOS. IMOS s a natonal collaboratve research nfrastructure, supported by Australan Government. It s led by Unversty of Tasmana n partnershp wth the Australan marne and clmate scence communty Addtonal fundng has been provded by the Australan Research Councl through a LEIF grant whch pad for the CBG radar purchase, James Cook Unversty, the Australan Insttute of Marne Scence, the Unversty of Queensland, the NSW State government. L. R. Wyatt s wth the School of Mathematcs and Statstcs, The Unversty of Sheffeld, Sheffeld, S0 TN, UK and s also an Adjunct Professor n the College of Scence, Technology and Engneerng, James Cook Unversty, Townsvlle, QLD 48, Australa (e-mal: l.wyatt@sheffeld.ac.uk). A. Mantovanell was wth the School of Mathematcs and Statstcs, The Unversty of New South Wales, NSW 05, Australa, the Australan Insttute of Marne Scence (AIMS), Townsvlle, QLD 480, Australa; AIMS@JCU and The College of Scence, Technology and Engneerng, James Cook Unversty, Townsvlle, QLD 48, Australa(e-mal: alessandra.mantovanell@gmal.com). M. L. Heron s wth The College of Scence, Technology and Engneerng, James Cook Unversty, Townsvlle, QLD 48, Australa (e-mal:mal.heron@eee.org). M. Roughan s wth the School of Mathematcs and Statstcs, The Unversty of New South Wales, NSW 05, Australa (e-mal: mroughan@unsw.edu.au). C. R. Stenberg s wth the Australan Insttute of Marne Scence (AIMS), Townsvlle, QLD 480, Australa (e-mal: c.stenberg@ams.gov.au).

3 I. INTRODUCTION Near real-tme nformaton on the spatal varablty of ocean currents s essental for coastal management (e.g., ol, pollutants and larvae trackng), navgaton and search and rescue operatons. Hgh-frequency ground-wave ocean radars (HF radars) are unque n provdng measurements of the two-dmensonal surface current feld over large coastal areas (from nearshore to km offshore) wth hgh temporal resoluton (Paduan and Washburn []). HF radars are remote sensng platforms nstalled on land that transmt rado waves (at frequences of 3-30MHz) n a radal path to the ocean surface and receve the backscattered sgnal from sea surface gravty waves of one-half of the ncdent rado wavelength approachng or recedng wth respect to the radar look drecton. Ths s the Bragg-scatterng effect frst descrbed by Crombe []. The Bragg-scatterng generates two dscrete peaks n the radar backscattered sgnal power (often referred to as Doppler) spectrum that are shfted (Doppler-shft) by the supermposed current feld. The dfference between the epected Bragg-peak (the phase speed of the Bragg-wave as determned from frst order ocean wave hydrodynamcs) and the observed Doppler-shft s used to calculate the speed of the radal component of the surface current movng toward or away from the receve antenna (Paduan & Washburn []). Radal currents produced by two or more radars, spatally separated but lookng at the same patch of water at dfferent vewng angles, can be summed to calculate the two-dmensonal surface current velocty vector (Gurgel et al [3]); the separaton between two radar statons determnes the doman of the mapped regon. Here we evaluate the performance of two phased-array WERA HF ocean radar systems (Gurgel et al [3]; avalable from Helzel Messtechnk GmbH, Germany) operatng on contrastng contnental shelves wth comple crculaton, and assess the errors n current vectors after combnng two radals wth dfferent errors. In addton, qualty-controlled current data etracted from HF radar, acoustc Doppler current proflers (ADCP) and satellte-tracked surface drftng buoys (drfters) at the same locaton and nstant of tme are compared to cross-valdate the measurements. Bottom-mounted ADCPs provde hgh temporal resoluton measurements of both the horzontal and

4 vertcal current components but they are sngle pont platforms that cannot resolve the near surface currents (top 6-5% of the depth profle) due to sde-lobe contamnaton (Gordon [4]). Drfters provde near realtme Lagrangan measurements of the flow (ntegrated over the drogue depth) gvng pont-by-pont nformaton about the horzontal currents along the path of moton; drfter data are however sparse n tme and space. Ocean HF radars provde the two-dmensonal current feld wth hgh temporal (0-60 mnutes) and spatal (0.3-0 km) resolutons (Gurgel et al [3]; Paduan et al. [5]) but ther measurements are restrcted to the surface layer thus ncorporatng wnd and wave-drven near surface currents (Graber et al. [6]; Cosol et al. [7]). In addton, each HF radar data pont s an average over a target patch of a few square klometres whle an ADCP samples over depth cells and beam separatons of a few metres. Typcally HF radar-derved surface currents are valdated aganst subsurface ADCP currents and currents estmated from drftng buoys wth both surface and subsurface drogues, despte the dfferences n ther spatal and temporal resolutons and samplng depths. Shay et al [8] was one of the early studes of ths knd and used a number of dfferent statstcs to quantfy the dfferences between these measurements, ncludng root-mean-square error (rms), correlaton coeffcents from lnear regresson for radals and vector components and comple correlatons for the vectors. Lke many other such studes, only short (-3 month) data sets were avalable allowng ad-hoc methods to be used to de-spke and nterpolate data for subsequent spectral analyss. Robnson et al [0] summarse a number of these valdatons as well as the results of ther much longer ( year) tme seres comparson. A smple automated de-spkng procedure was appled and least squares fttng used to obtan tdal coeffcents thus avodng the need to fll gaps. For the scalar comparsons n these studes, r of 0.3 to more than 0.9 and rms dfferences between m s - are typcal and reflect the epected spatal-temporal varablty of the currents and the nherent nstrumental errors and resolutons (Graber et al. [6]; Shay et al. []). Although ocean radar technology has been prevously valdated n other locatons, we provde here evaluaton of the WERA radar performance at two stes along the Eastern coast of Australa subjected to comple coastal dynamcs n order to provde confdence n these HF radar measurements. Many prevous

5 studes have looked at the accuracy of the radals measured by HF radar. Here we focus on the accuracy of the data product of most nterest to users,.e. the vector currents. Some of the methods we have appled are not wdely used n the HF radar communty and part of the am of ths paper s to demonstrate ther value for ths applcaton. We eplore the use of the Taylor dagram as a tool to smplfy the comparson of dfferent qualty control (QC) and averagng strateges by avodng long and dense tables of comparson statstcs. We test the applcaton of flters to automatcally remove outlers and smooth data sets whch often contan gaps. Smple lowpass flters appled n e.g. Shay et al [], Cosol et al [7] can only be appled to evenly sampled data wth no gaps. We use spectral methods to determne nose levels and hence lmts to the accuracy of the surface current measurements. The HF radar and ADCP data sets used here are from the southern Caprcorn Bunker Group (CBG ~4 S) area of the Great Barrer Reef (GBR) and Coffs Harbour (COF ~30 S) on the coast of eastern Australa. The CBG and COF regons both ehbt comple crculatons and large spatal and temporal varablty of the coastal currents mposed by wnds, tdes, large scale crculaton (East Australan Current, EAC) and topography (Mao and Luck [3]; Schaeffer et al. [4,5]; Mantovanell et al. [6] ; Roughan and Mddleton [7,8]). Secton II of the paper presents a descrpton of the data sets followed n Secton III by a descrpton of the error analyss from radal to vector currents and the HF radar QC procedures that have been used here. In Secton IV, we evaluate the effect of the QC procedure on data precson and accuracy and perform a detaled statstcal comparson of the current data obtaned from the dfferent platforms (HF radar, ADCP, drfters) to assess the qualty of the HF radar surface current measurements. The results are dscussed n Secton V and ths s followed by some concludng remarks. A. HF radar data II. INSTRUMENTATION AND DATA SETS A system of two hgh-frequency (HF) ground-wave WERA phased-array radars, operate both at the CBG (-4 S; Fg. a) and COF (30-3 S; Fg. b), as part of the IMOS Australan Coastal Ocean Radar

6 Network (ACORN). The radar domans of the CBG and COF radar systems are shown respectvely n Fg. a and Fg. b, ncludng only regons wth acceptable geometrc dluton of precson (GDOP, see Secton IIIA), and ther confguraton detals are summarzed n Table. The radar doman n the CBG ncludes the reef lagoon on the shallow contnental shelf (~70 km wde and less than 50 m deep), the shelf break, whch drops steeply from 50 m to 400 m depth, and a small porton of the offshore regon (Fg. a). The COF radar doman covers the narrow contnental shelf (< 30 km), the shelf-break and deeper off-shore waters (up to 50 km from the coast; Fg. b). Data analyses encompassed the perod between November 009 and 30 November 00 (CBG system) and September 0 and 30 September 03 (COF system); tmes are gven n Coordnated Unversal Tme (UTC). The radar azmuthal resoluton s determned assumng a / spacng (λ s the rado wavelength) between the N receve antenna elements gvng an appromate half-power beamwdth of 0/N (Skolnk [9]). Ths beamwdth was modfed by the wndow used n the beam-formng and a Dolph-Chebyshev wndow (Harrs [0]) wth alpha 3 and used for the calculaton of the azmuthal resoluton n Table. Ths ncreased the non-wndowed beamwdth by a factor of.68 and reduced the sde lobes. The mamum azmuthal range of the radars s 90 at CBG and 0 at COF (Table ). The radar data are processed usng standard WERA software onto a rectangular grd wth horzontal resoluton of 4 km (CBG) and.5 km (COF) usng a combnaton of dgtal beam-formng to grd locatons and nterpolaton from the orgnal range-resolved data. Ths has the effect of over-samplng at long ranges and under-samplng at short ranges relatve to the ntrnsc (azmuth and range cell) radar resoluton. Radal components from two radar statons are combned to calculate the east-west (u) north-south (v) current components at each grd locaton (dscussed below n secton IIIB). Radal (every 0 mnutes) and vector (obtaned from hour averages of the radals) data are freely avalable from the IMOS data portal ( n both real tme (FV00) and as Level (FV0) qualty-controlled (QC) NetCDF fles. The analyses here use the radal FV0 fles. The results are compared wth the monthly-aggregated hour averaged current vector FV0 fles wth and wthout takng account of the addtonal qualty flags

7 contaned n these fles. Errors n the determnaton of the radal currents may occur due to rado-wave nterference, movng shp reflectons, mproper determnaton of the Bragg peaks and/or of the angle of arrval from the scatterng patch through nadequate calbraton of the beam pattern, azmuthal and range resoluton and addtonal uncertantes arsng durng the vector mappng (Găcć et al. []). The HF radar data presented here were qualty-controlled as descrbed n Secton III. TABLE I DETAILS OF THE HF RADAR SYSTEM CAPABILITIES AND CONFIGURATION FOR THE CBG AND COF SYSTEMS CBG COF Operatng frequency (MHz) Azmuthal resoluton ( ) Range cell resoluton (km) Measurement depth (m) ( λ radar /8π) Samplng perod each staton (mnutes) 5 5 Integrated samplng perod (mnutes) 0 0 Dstance between stes (km) Mamum confgured offshore dstance (km) Number of Transmttng antennas 4 4 Number of Recevng antennas (phased array) 6 Transmtter power (W) Data acquston near real tme near real tme Note that both systems have undergone a frequency change snce these measurements were made, wth the frequences beng 9.33(CBG) and 3.5(COF) at tme of publshng. Ths s set to be larger than the mamum epected range for sea echoes at the gven frequency. B. ADCP data ADCP data sets were obtaned from IMOS ( for the GBR and Coffs Harbour (CH) moorngs (see locatons n Fg. ). Three ADCPs are moored n shallow waters (45-60 m) n the southern GBR, namely Heron Island North (HIN), Heron Island South (HIS) and One Tree Island (OTE), wth bn szes (-4 m) and samplng rates (0-30 mnutes) varyng among the deployments (for detals see Append, Table A). The two Coffs Harbour ADCPs, located on the 70 m (CH070) and 00 m (CH00) sobaths, have bn szes of 4 m and samplng rate of 5 mnutes. All moorng statons were equpped wth upwardlookng four-beam broadband ADCPs (RDI Workhorse Sentnel Instruments operatng at 307. and 64.4 khz; Table A), wth beam angles slanted at 0. The set-up enabled bn-mappng correcton for tltng, three beam soluton and Earth coordnates (East/North/Up), wth correspondng velocty components

8 denoted by u, v, w. Fg.. Locaton of the radar statons ( ), and the HF radar doman for surface current measurements for (a) the southern Great Barrer Reef (CBG) and (b) Coffs Harbour (COF) HF radar systems, respectvely. The ADCP moorngs are ndcated by black squares and moorng names are labelled; the drfter clusters release postons are ndcated by crcles. Spatal and temporal data coverage (%) for (a) the GBR regon between November 009 and 30 November 00 and (b) Coffs Harbour regon between September and 3 September 03 for areas wth acceptable GDOP are ndcated wth the colour-shadng (see colorbar). ). Bathymetry s ndcated by the black lnes. The ADCP data were corrected for magnetc declnaton and qualty-controlled to remove data of poor qualty.e. () contamnated by sde-lobes near the surface (frst 6% of the water column), () wth low sgnal to nose (.e. correlaton magntude for two or more beams < 64 counts), () subjected to nstrument tltng larger than 0 and (v) dentfed as outler when the measured value was larger than the tmeaveraged value of velocty magntude for each depth stratum plus 5 tmes ts standard devaton. ADCP data were posterorly re-mapped nto unform depth strata (as wde as the bn), takng nto account the temporal changes of the bn depth n relaton to the surface due to tdal varatons measured by the bultn pressure sensor (Mantovanell et al. []). When the bn sze of a partcular moorng changed durng the analysed year, data were averaged over a depth stratum correspondng to the sze of the largest bn for consstency. Data from the two shallowest depth strata wth more than 50% of vald observatons durng each

9 deployment after QC were used to compare wth HF radar data. There were small dfferences n the comparsons for these two depth strata. The cases wth better agreement are presented here, wth centre of the depth strata at 5.5 m (HIN), 5 m (HIS), 0 m (OTE), 0 m (CH070) and 4 m (CH00). ADCP u-and v- velocty components were averaged over 30 mnutes and nterpolated nto the same tmeframe as the HF radar data for unformty and the standard devatons of the averaged values are used as a proy of the ADCP errors. The 30 mnute averaged ADCP data are used for the comparsons wth the 0 mnute and hour averaged radar data as well as wth the 30 mnute averaged radar data. C. Drfter data Drfters, used for comparson wth the CBG radar, were bult, released and montored by the Australan Insttute of Marne Scence (AIMS). Four drfters separated by - km were released n a square array (less than 50 mnutes apart) () on Aprl 00 n the slope regon ( m water depth) and () on 5 Aprl 00 n shallow shelf waters (depths < 50 m) south of Heron Island (Fg. a). All drfters stayed nsde the HF radar doman for 6-0 days. Drfters conssted of a Davs drogue (0.45 m wde and 0.9 m long) attached to a 60 cm long PVC cylnder (that houses the electroncs) lnked to a surface float (dameter ~0 cm) wth only 0 cm emergng n ar (drag area rato 40). The entre drfter assembly occuped the frst m of the water column n a good appromaton to the HF radar depth of measurement of.4 m (Table ). Drfters were satellte-tracked by a GPS Satellte Messenger gvng a poston (UTM n m, Unversal Transverse Mercator projecton) every 0 mnutes, wth nomnal accuracy of 6.4 m. The actual accuracy, based on postons transmtted for 4 days by a drfter kept n a fed locaton, was wthn ±0 m of the mean poston 80% of the tme, whch corresponds to an error n the speed of about ±0.03 m s -. Postons were nterpolated nto dscrete tme ntervals of 0 mnutes to match the HF radar tmeframe. Drfter current velocty components (u, v) were calculated by centred dfference of the Eastng (Northng) UTM coordnates dvded by tme, and posterorly averaged over one hour. Coarse outlers were manually removed.

10 III. HF RADAR QUALITY-CONTROL AND ERROR ANALYSIS A. Radal accuracy The Level QC radal data (FV0 fles) uses a swarmng method (Prytz [3]) to fnd and categorse the Bragg peaks and hence flag the radal speed and drecton data. For the analyss presented here we retan data wth an ntal QC flag of (good data) or (probably good data). The standard devatons of the radals reported n the FV0 fle are determned from the wdth of the prmary Bragg peaks. The radal velocty resoluton s 8. cm s - at COF and 3.5 cm s - at CBG based on the frequency resoluton n the radar Doppler spectrum ( Hz); the radal velocty accuracy s mproved to 3.8 cm s - (COF) and 6.3 cm s - (CBG) by the standard WERA processng, whch averages 5 overlapped Doppler spectra and uses a centrod method to determne the locaton of the Bragg peak (Barrck [4]). In ths paper we consder the mpact of addtonal averagng by comparng () the 0-mnute radal data set wth () the 30 mnute runnng-averaged radal dataradals, whch further mproves the accuraces to. cm s - (COF) and 3.6 cm s - (CBG) and () the hour-averaged radal data (sngle value per hour) obtaned from IMOS portal,wth correspondng resolutons of.6 cm s - (COF) and.6 cm s - (CBG). The 30 mnute runnng-averages on the radals are over 3 measurements stored n the mddle nstant of tme, thus retanng the measurement dscrete tme nterval of 0 mnutes. B. Calculaton of the current velocty components and errors The two radals are combned n the usual way (Paduan and Washburn []) to gve the u (east-west) and v (north-south) current components.the errors n the current vectors are calculated assumng that the errors n the two radals (from the two separate radars) are uncorrelated and potentally dfferent. The varances u and v n the u and v current components respectvely can be estmated as (Wyatt [5]): cos θ + a a u () sn ( θ θ ) sn θ + cos θ a a v, () sn ( θ θ ) sn θ

11 where θ, θ are the bearng angles and a, a are the varances of the radal speeds (.e. squared values of the radal standard devatons reported at each dscrete tme n the FV0 fles) from statons and, respectvely. For the cases where addtonal averagng s carred out, the averaged radal varances are the averages of the ndvdual varances,.e. t s assumed that there s no correlaton between the data sets contrbutng to the average. The varance, V, of the velocty magntude, V, s gven by: + +, sn cos + sn cos, where s the covarance of u and v (Wyatt [4]). The square roots of the calculated varances gve the standard devatons ( u, v, V ) of u, v and V that are used to assess the measurement errors n the HF radar current vector. The same procedure was used to calculate the standard devatons n the current vectors reported n the hourly averaged IMOS fles. C. The geometrc dluton of precson (GDOP) GDOP reflects the effect of the radar geometry on the accuracy of the current vectors derved from the radals measured at the two radar stes. It s related to the angle between the radars at each locaton (as shown n the denomnator n Equatons (3) and (4)). GDOP values can be calculated separately for the u (GDOP east) and v (GDOP north) components of the current vector (followng Chapman et al. [6] termnology). Here we use ther modulus gven by: GDOP, (3) θ where θ s the ntersecton angle between the beams from the two radars to the cell locaton gven by θ θ θ. Note that ths result assumes that the errors n the two radals are the same (.e. a a n Equatons () and ()), so ths procedure s not eact but provdes a useful ndcator of the lkely error. The GDOP spatal dstrbuton for CBG (Fg. a) and COF (Fg. b) radar stes show areas of favourable geometry/hgher precson (small GDOP, blue regon) and areas of largest sgnal degradaton along the antenna par baselne and n the far feld (hgh GDOP, red and yellow regons) for smlar lookng angles. The thck grey lnes n Fg. show locatons where the ntersecton angle between the radal beams of the sn

12 two dfferent radars s wth the range: 30 < θ < 50, correspondng to GDOP 3. Ths has been shown to be a reasonable bound for good data qualty data (Shay et al. []; Cook and Shay [7]; Robnson et al. [0]) and s the value used n ths paper. Fg.. Maps of geometrc dluton of precson (GDOP) for (a) the CBG and (b) COF radar systems. The thck grey lnes show locatons where the ntersecton angle between the radal beams of the two dfferent radars s wthn the range: 30 < θ < 50 D. Average current and standard devaton maps Fg. 3 (CBG) and Fg. 4 (COF) show the standard devatons of the annual averages of the HF radar 30- mnutes averaged varances of the current components ( u, v ) and current magntude ( V ). These show an overall decrease n the accuracy away from the radar statons (where temporal coverages are also lower; see Fg. a,b) and n areas wth hgher GDOP (see Fg. a,b). The standard devatons for areas wth over 50% of temporal coverage (Fg. a,b) have the same order of magntude as the theoretcally predcted current resolutons gven n Secton IIIA. Annual averages of velocty magntude show weak currents (< 0. m s - ) and domnant onshore flow for the CBG regon (Fg. 3d) and a predomnantly southward flow n the COF regon wth stronger currents east of the slope under hgher nfluence of the EAC (Fg. 4d).

13 E. The HF radar data qualty-control and processng In secton IV, we compare a number of dfferent QC approaches appled to the HF radar data. The QC steps adopted for the radar current data analyses startng from IMOS FV0 radals are outlned n Table, and those appled n the data analyses startng from the IMOS current velocty vectors are descrbed n Table 3. The cases n Table all have a tme step of 0 mnutes, those n Table 3 have a tme resoluton of hour. The QC steps n Table nvolve: () applyng GDOP (Equaton (3) referred to as RAW0 n Table ); () usng a mamum threshold for both radal and vector speed magntude of 3 m s - to remove coarse outlers and settng a mnmum acceptable Bragg sgnal to nose rato of 0 db (.e. rato 0 log SNR, where SNR s the value gven n the NetCDF fle) for each grd pont (QCT0 n Table ) to ensure a good separaton between the Bragg peaks and the background nose (Fernandez et al. [8]); () reducng spkes by applyng a Matlab hampel flter, whch calculates the medan and the standard devaton of 7 data ponts replacng the md-pont (unless t s NaN) wth the medan f the md-pont vares from the medan by more than 5 tmes the standard devaton (QCH0 n Table ); (v) startng from the 30 mnute runnng-averaged radals (after GDOP and SNR tests, and dscardng radal speeds above 3 m s -, as n QCT0 and QCH0 n Table ), plus addtonal removal of outlers n the current components (u, v) by acceptng mamum speeds less than the tme-averaged value for the analysed perod plus fve tmes ts standard devaton calculated separately for each velocty component(qc30, Table ). TABLE STEPS FOR THE QC PROCEDURE STARTING WITH FV0 RADIAL DATA. THE CASES BEING COMPARED ARE LABELLED WITH THE LETTERS (B-E, H-I) FOR CONSISTENCY WITH THE RESULTS SHOWN IN FIG. 7 AND FIG. 9 Radal Steps Vector Steps Case Apply GDOP Calculate u,v (usng Equatons () and ()) RAW0 (B) Apply speed and SNR threshold to B Calculate u,v QCT0 (C) Apply hampel flter to C Calculate u,v QCH0 (D) 30 mnute runnng-averaged radal speeds and Calculate u,v and apply QC30 (E) varance usng C current speed threshold Apply -hour Savtzky-Golay lnear flter to D Calculate u,v SGL60 (H) Apply -hour Savtzky-Golay quadratc flter to D Calculate u,v SGQ60 (I)

14 Fg. 3. Mean standard devaton calculated as the square root of the mean varance between 0//009 and 3//00 for the (a) u-component ( u, m s - ), (b) v-component ( v, m s - ) and (c) current absolute magntude ( V, m s - ) of GBR radar system; (d) averaged current magntude (V, m s - ; see colour bar) and drecton (arrows) between 0 November 009 and 3 November 00 for areas wth temporal coverage above 50% over the analysed year.

15 Fg. 4. Mean standard devaton calculated as the square root of the mean varance between 0/09/0 and 30/09/03 for the (a) u-component ( u, -- ), (b) v-component ( v, m.s - ) and (c) current absolute magntude ( V, m.s - ) for the Coffs Harbour radar system; (d) averaged current magntude (V, m.s - ; see colour bar) and drecton (arrows) between 0 September 0 and 30 September 03; for areas wth temporal coverage above 50% over the analysed year. We assessed here the hour averaged current (u, v) data (IMOS FV0) (RAW60 n Table 3) wth flags or (QC60 n Table 3) n the NetCDF fles. These flags mean that the orgnal radal data had sgnal to nose s > 8 db, radal speeds less than 3 m.s - at COF or m.s - at CBG and at least 3 such radal measurements durng the hour; the vector speeds satsfy the same thresholds and the GDOP crteron s also satsfed.

16 TABLE 3 STEPS FOR THE QC PROCEDURE STARTING WITH IMOS CURRENT VECTORS. THE CASES BEING COMPARED ARE LABELLED WITH LETTERS (F, G) FOR CONSISTENCY WITH THE RESULTS SHOWN IN FIG. 7 AND FIG. 9. Vector Steps Case Read FV0 hour averaged current vector data fles RAW60 (F) Apply QC flags and. These account for GDOP and sgnal to nose thresholds. QC60 (G) HF radar data often contan temporal and spatal gaps generated by nstrument falure, weak sgnal-tonose, rado frequency nterferences (such as onosphere varatons) and ocean wave condtons (mamum range s reduced n hgh seas) and power avalablty (Wyatt et al. [9]). Therefore, the spatal coverage of HF radar current velocty data over tme serves as a performance ndcator. A hgher data return and spectral qualty s epected near the radar statons, wth the sgnal avalablty beng reduced at longer ranges because of sgnal propagaton losses (Wyatt et al. [9]). Percentages of vald HF radar current data (over one year) for case E at CBG (Fg. a) and COF (Fg. b) regons show mamum coverages between the two radar statons and a gradual reducton of the coverage wth ncreasng range, as epected. The presence of temporal gaps prevents the use of many numercal flters to further smooth the data. An ecepton to ths s the Savtzky-Golay flter (Orfands [30]) whch can be appled to data wth gaps n the tme seres. It also has an mplementaton n Matlab, although modfcatons were requred n order to calculate the varances of the smoothed radal current estmates (see Append for the detals). We have assessed the mpact of applyng ths flter to smooth the 0 mnute (QC0) data over a hour wndow, consderng both ts lnear (SGL60); Table ) and quadratc (SGQ60; Table ) forms. Cases QC0, QC30, QC60, SGL60 and SGQ60 n Tables and 3 are the new approaches that are beng assessed n ths paper. We have also compared, 3, 7, 3, and 5 hour smoothng of the () 30-mnute averages (QC30) and () hour IMOS current data (QC60) aganst the ADCP data wth smlar smoothng, wth a vew to the potental use of ths flter for some oceanographc applcatons where more smoothng may be warranted (see Secton IV-B-5).

17 IV. CROSS-VALIDATION AMONG THE DIFFERENT PLATFORMS A. Methods for data comparson The centred root mean square dfference (E; Equaton (4)) and the centred correlaton coeffcent (R, Equaton (5)) were appled n the comparsons among the dfferent data sets, snce these parameters, together wth the standard devaton of the radar tme seres, can be combned n one plot, namely the Taylor dagram (Taylor [3]). The Taylor dagram has not been wdely used n HF radar evaluatons but t has the advantage of makng the comparsons between the dfferent data sets and qualty control procedures relatvely straght forward. In Equatons (4) and (5), f s used genercally for u- and v-components of the flow and current speed (V) and the subscrpts r, m are respectvely used for the HF radar (f r ), and ether the ADCP moorng or the drfter data (dependng on the comparson) (f m ) for n,,.. N observatons n tme. Smlarly, the standard devatons for the u, v, V speeds of the HF radar and ADCP or drfter tme seres are genercally represented by s r and s m, respectvely. The parameters E and R are calculated as follows: N E [( f ( n) f ) ( f ( n) f )] (8) N n r r m m R N N n ( f ( n) f ) f ( n) r r r s s m ( f ) m m. (9) where f s the parameter beng compared (e.g. u), s s ts standard devaton; subscrpts r denotes the radar measurements and m the comparator. N s the number of observatons n the comparson. In addton, the bas, standard devaton, root mean square dfference (rms), slope and ntercept of lnear regressons wth 95% confdence lmts and the coeffcent of determnaton (and equvalently the correlaton coeffcent) were also determned. Three methods were used for the comparson of drectonal and vector dfferences. The frst method calculates the mean dfference between two current measurements wth ts 95% confdence nterval and concentraton (Bowers et al. [3]). The concentraton parameter s large f the spread n the drectonal

18 dfference dstrbuton s small (.e, agreement s good) and vce-versa. The second method gves the crcular correlaton coeffcent between the two drectons (Fsher and Lee [33], Fsher [34]). Fnally, the vector correlaton was determned as descrbed by Kundu [9]. These analyses were appled to the dfferent moorng locatons n the CBG (HIN, HIS, OTE) and COF (CH070, CH00) regons and to the drfter comparson. In addton, comparsons between ADCP and radar spectra are used (Secton IV-B-3) to estmate the ntrnsc uncertanty n the u and v components derved from the spectral nose levels. B. Total current comparson ) HF radar versus ADCP comparson for Coffs Harbour (COF) Comparsons between HF radar and ADCP u- and v-components usng the hour averaged IMOS current data (QC60) at the CH070 and CH00 moorng locatons are shown n Fg. 5. For ths fgure each par of ADCP and radar u- or v-components was placed nto bns (across the speed range shown) and the numbers n each bn were counted. Bns wth less than % of data were ecluded to remove the effect of outlers. Therefore, Fg. 5 shows 99% of the data. The colour codng of the remanng bns ndcates the numbers of observatons n each bn normalsed by the number of observatons n the bn wth the mamum number of data ponts. Lghter colours ndcate bns contanng larger proportons of the data. Hgher correlaton coeffcents were found for the v-component (roughly algned wth the EAC) n both locatons (CH070 and CH00 moorngs). Smlar standard devatons were observed n the v- and u- component comparsons, ecept at the CH70 locaton. The CH70 moorng s closer to the coast and therefore the radar data may be nfluenced by antenna sde lobe problems or small scale current structure n ths locaton. Lower correlaton coeffcents for the u-components are related to the smlar scatter but smaller range than the v-components. The Taylor dagram (Fg. 6) compares the dfferent QC procedures:() case B (red),.e. RAW0, the vectors obtaned from the 0 mnute data wth only GDOP appled; () case C (green),.e. QCT0, the 0 mnute vectors after applyng a 3 m.s - speed and 0 db sgnal-to-nose thresholds to RAW0; () case D

19 (blue) QCH0 vectors after applyng the hampel flter to case C; (v) case E (magenta),.e. QC30, the 30 mnute averagng after applyng GDOP, SNR, radal and current components thresholds; (v) RAW60, vectors from the standard IMOS hour averaged current data before (F, cyan), and QC60 after applyng QC flags and (G black); (v) SGL60 current vectors after Savtzky-Golay lnear (H, red) and SGQ60 quadratc (I, green) flterng of case D. Note that -hour Savtzky-Golay flterng has also been appled to the 30 mnute averages (QC30) and ths s referred to n secton IV-B-4. The centred rms dfferences and radar tme seres standard devatons are both normalsed wth respect to the standard devatons of the ADCP tme seres for the u, v, and speed measurements (as specfed n Fg. 6). Very lttle dfference was found between RAW60 and QC60 at both locatons (CH070 and CH00) because both GDOP and sgnal to nose rato are wthn good ranges. The addtonal steps n QC60 wll be more mportant around the edges of the radar coverage. The fgure confrms the qualtatve remark above that the u-component comparson at CH00 s better than that at CH070. The u-component data beneft most from the averagng (see the E-G cluster compared wth B-D) because ampltudes are smaller than those for the v-component and hence nearer the nose level. The IMOS -hour current averagng (QC60) provdes a small beneft over QC30 (.e., 30 mnutes average), as evdenced by the ncrease n the correlaton coeffcent, the decrease n centred rms and normalsed radar standard devaton closer to. Therefore, G (QC60) s closer to A (pont of perfect agreement) than E (QC30), but both cases (G and E) are clearly better than the 0-mnute non-averaged data. (QC0, D) The Savtzky-Golay flter ams to smooth data sets whlst retanng hgh-frequency content at the epense of not removng as much nose as the average of the data ponts over hour ntervals (see Fg. A. n the Append). Therefore the hour Savtzky-Golay smoothng (cases H and I) s not epected to perform as well as the hour IMOS average (case G) and ths can be seen n the u-component comparsons at both CH70 and CH00 locatons (Fg. 6 left). The fgure shows that the flter s not even as effectve as the 30 mnute average although t has some postve mpact relatve to the non-averaged data. Dfferences n varablty wth frequences of about 48 cycles per day whch are not suppressed as much by the Savtzky-

20 Golay flter may eplan the dfference n performance. The averagng/flterng and addtonal QC had lttle mpact on the v-component, whch s the larger of the two components at Coffs Harbour locatons (CH070 and CH00; Fg. 6 mddle). The drectonal and vector comparsons were better at CH00 than CH070, showng hgher correlaton and concentraton (Table 4). It s clear that the 0 mnute radar data (QCT0, no average) has the worst comparson n each case (.e., the lower correlaton coeffcents and concentraton; Table 4) although the applcaton of the hampel flter (QCH0) provdes a small mprovement. The hour averaged (AC60) statstcs were only margnally better than the 30 mnute average (QC30) statstcs. The Savtzky-Golay statstcs (not ncluded n the Table) were generally between those of QCH0 and QC30 as was the case for the Taylor dagram analyss. TABLE 4 COF DIRECTION AND VECTOR COMPARISON OF RADAR AND ADCP CURRENTS. CH070 CH00 Case QCT0 QCH0 QC30 QC60 QCT0 QCH0 QC30 QC60 N Drecton dfference ( ) ± 95% CI 0.6 ± ± ± ± ± ± ± ± 0.7 Concentraton Comple Correlaton Coeffcent Phase Dfference (º) Crcular correlaton ) HF radar versus ADCP comparson for the Great Barrer Reef (CBG) Fg. 7 shows the CBG u- and v-components of HF radar currents aganst the ADCP currents at the three CBG moorngs, usng the same processng parameters as descrbed n Fg. 5 for COF. In all cases there was a hgher correlaton for the u-component whch has the larger range of the two components n ths ste. The correlatons for the v-component were sgnfcantly lower, partcularly at HIN and HIS stes, closest to the reefs. Note that the u- and v-components at OTE were closer n magntude and ther statstcs le between those of the other stes. The Taylor dagrams for CBG are presented n Fg. 8, keepng the same notaton as used for COF n Fg. 6. The averagng and addtonal QC had lttle mpact on the u-component at the CBG stes, the results of

21 whch are rather smlar to those for the v-component at COF and vce versa. The drectonal and vector comparsons for CBG are shown n Table 5. A larger bas between the radar and ADCP current drectons was observed at the HIS ste as evdenced by the larger drectonal dfference and the phase dfference from the comple correlaton analyss. Whether ths s evdence of a surface shear or a bas n ths partcular ADCP measurement s not yet establshed. The north-south currents (vcomponent) are very small at HIS, and the man currents are algned n the east-west drecton. Therefore, small errors n the v-components wll lead to large errors n the derved angle for both measurement systems and ths could be the reason for the bas. The comple correlaton coeffcent s hgher at HIS than at the other stes ndcatng that the vector currents are n better agreement wth the ADCP notwthstandng the apparent bas n drecton. It s clear that the 0 mnute radar data has the worst comparson n each case.

22 Fg. 5. Bnned scatter plots of HF radar versus ADCP at Coffs Harbour (COF). The fgure ncludes 99% of all the data. The frst and second rows show respectvely the u- and v-components at CH070 (left) and CH00 (rght) moorng locatons. The black lne shows the lnear regresson. The colour codng of the remanng bns ndcates the numbers n each bn relatve to the mamum number of data ponts n a bn. Some of the calculated statstcs are shown nsde the fgures: bas, std (standard devaton), CC (correlaton coeffcent), N (number of observatons), the lnear regresson equaton and rsq (coeffcent of determnaton).

23 Fg. 6. Taylor dagrams showng the mpact of the dfferent QC steps for u, v, speed magntude (V) for CH070 (top three fgures) and CH00 (bottom three fgures). A s the pont of perfect agreement; B (red) s the 0-mnute data wth no addtonal QC, C (green) s the case where a 3 m.s - speed and 0 db sgnal-to-nose threshold have been appled; D (blue) ncludes the hampel flter; E (magenta) wth 30 mnute averagng QC; F (cyan) wth standard IMOS hour averagng, G (black) wth addtonal QC; H (red) lnear and I (green) quadratc Savtzy-Golay flters. Red dashed lnes show normalsed centred rms dfference decreasng towards A, blue dashed lnes the correlaton coeffcent decreasng from north to east and black dashed lnes the normalsed radar standard devaton ncreasng from the orgn. The normalsatons used for u, v, and speed are 0.7, 0.36, 0.8 and 0.8, 0.39, 0.3 m.s - at CH070 and CH00 respectvely. Note that the top left fgure has a dfferent vertcal scale. TABLE 5 CBG DIRECTION AND VECTOR COMPARISON OF RADAR AND ADCP CURRENTS. ; HIN HIS OTE Statstc QCH0 QC30 QC60 QCH0 QC30 QC60 QCH0 QC30 QC60 N Drecton Dfference ( ).3 ±.36 ±.33 ± -.34 ± -.68 ± -.57 ± ± 95% CI ±0.56 ±0.5 ±.50 Concentraton Comple Correlaton Coeffcent Phase Dfference (º) Crcular correlaton

24 Fg. 7. Bnned scatter plots of radar versus ADCP at CBG. The u- and v-components are shown respectvely n the frst and second rows for HIN (left), HIS (mddle) and OTE (rght). Colour-codng and statstcs notatons are as defned n Fg. 5

25 Fg. 8. Taylor dagrams showng the mpact of the dfferent QC steps for CBG u- (above) and v-(below) components. Notaton s the same as Fg. 6. The normalsatons used for u and v are respectvely are 0.36 and 0.4 m.s - at HIN; 0.8 and 0.08 m.s - at HIS and 0. and 0.9 m.s - at OTE. 3) Spectral comparsons Wyatt et al. [35] calculated spectra from HF radar current data usng a tme seres created by fllng gaps n the orgnal data set wth a tdal predcton, obtaned usng a least squares analyss, plus the mean of the resduals. A dfferent approach has been adopted n ths study because there s a strong wnd nfluence n the radar data at the tdal frequences and the radar and ADCP tme seres have long gaps. The full data set was splt nto segments wth gaps no longer than 3 hours and wth at least 4096 samples per segment for the 0 mnutes samplng nterval, and at least 768 samples, for -hour samplng nterval. Ths guarantees that the analysed tme perods are roughly the same ( 8-3 days), gvng smlar spectral frequency resolutons and averagng. The short gaps (< 3 hours) were lnearly nterpolated. Data that dd not conform to these requrements were not ncluded n the analyss. Each segment was then processed usng the Matlab pwelch

26 functon (50% overlap, Hannng wndow) usng FFT lengths of ether 048 for the 0 mnutes samplng) or 56 (for the -hour samplng) and the resultng spectra were then averaged over all segments. The Matlab pwelch functon s also used to calculate the confdence ntervals for the spectral estmates. Note that 0 (60) mnute samplng mples a Nyqust frequency of 7 () cycles per day. We also used a smoothed Lomb-Scargle spectrum (Lomb [36]; Scargle [37]), whch s avalable as plomb n Matlab. Smoothng of the Lomb-Scargle spectra can be acheved by applyng a Bartlett wndow n a runnng average usng flter n Matlab. Ths method can provde a spectrum for data wth ether mssng samples or sampled unevenly, and therefore, avods the need to check for contnuty n the data set. The Lomb-Scargle spectrum gves smlar results (not shown) to the pwelch method used here but the confdence ntervals for the resultng spectrum are not as easy to determne. Fg. 9 shows a v-component pwelch spectrum from COF CH00 and the u-component spectrum from CBG HIS, these beng the larger ampltude component n each case. The other spectra (not shown) are qualtatvely smlar. Fg. 0 shows the same data sets but lmted to the frequency range, cycles per day, of the IMOS data (case G). All spectra clearly show sem-durnal and durnal sgnals; the sem-durnal sgnal beng partcularly strong for the CBG data. At these locatons these sgnals are both tdal and related to the wnd forcng whch also has sgnfcant energy at these frequences. Note that at COF the durnal and nertal frequences are very smlar. Fg. 0 also shows the IMOS spectra rescaled (multplyng by the 4 cycles per day sample rate dvded by the 44 cycles per day sample rate of the other cases) n order to drectly compare nose levels (see analyss below). We refer to Schmd [38] for a more detaled dscusson about the nterpretaton of the Matlab pwelch spectrum n terms of the scalng requred to get sgnal, nose and power levels. The spectra at hgh frequences (the nose floor) can provde an estmaton of the rms fluctuatons n the u- and v-components under the assumpton that t s whte nose. The square root of the nose power (varance) s obtaned from the mean of the spectral densty at the upper eghth of the frequency range multpled by the mamum frequency (spectral bandwdth). Ths range was selected to avod the nulls

27 assocated wth the smoothng appled (Append ). These nose levels are plotted as dashed lnes over the range 9 to 7 cycles per day n Fg. 9 and 0. A smlar calculaton has also been done lmtng the frequency range to 0.5 cycles per day,.e. the upper eght of the IMOS spectrum frequency range (as n Fg. 0). The results of both calculatons are shown n Table 6. The hour averaged nose levels are shown wth a red dashed lne n Fg. 9 and 0. At some locatons (e.g. CBG) these two estmates are smlar for the 0-mnute sampled data (QCH0) and Fg 9 shows that the 0 mnute spectrum s farly flat from about a frequency of 4 cycles per day for all cases (ncludng those not shown n the fgure). Ths result suggests that there s lttle to be ganed n usng the data at ths temporal resoluton at these stes wthout addtonal averagng because the magntude of any flow components at hgher frequences are below the nose level. The ADCP and 30-mnute radar data (QC30) estmates over the two frequency ranges are dfferent as epected for runnng averages whch are lowpass flters as s the Savtzky-Golay flter. In these cases, the two estmates are dfferent as epected (Table 6) wth the Savtzky-Golay flter gvng smlar results to those of QC30. The standard IMOS result (QC60) has lower rms fluctuatons than the most of the other cases, a few of the hgh frequency QC30 cases are smlar. For some of the cases the estmates are smlar to the precson n the radal measurements dscussed n Secton IIIA. The estmates for the COF and CBG HIS ADCP data are smlar. Those at CBG HIN and OTE are larger and the lmted frequency range estmates are sometmes hgher than the radar values.

28 TABLE 6 ESTIMATED RMS NOISE LEVEL (M/S) IN THE U- AND V-COMPONENTS AS DETERMINED FROM THEIR SPECTRA. UPPER ROW USE THE UPPER EIGHTH OF THE IMOS BANDWIDTH ( CYCLES PER DAY) TO ESTIMATE THE RMS NOISE LEVEL, LOWER ROW USES THE UPPER EIGHTH OF THE OF THE HIGHEST FREQUENCY RANGE (7 CYCLES PER DAY). Ste varable ADCP QCH0 QC30 QC60 SGL60 SGQ60 COF CH070 u v COF CH00 u v CBG HIS u v CBG HIN u v CBG OTE u v Fg. 9. Velocty component pwelch spectra for (a) the v-component at COF CH00 and (b) the u-component at CBG HIS. Blue ADCP, magenta 0 mnute data hampel flter (QCH0); black 30 mnute QC (QC30); red hour QC IMOS data (QC60); green - Savtsky-Golay (SGQ60). 95% confdence ntervals are shown usng the same colour codng. Nose levels are shown wth dashed lnes usng the same colours.

29 Fg. 0. Velocty component pwelch spectra for the v-component at COF CH00(a) not scaled and (b) scaled and for the u- component at CBG HIS (c) not scaled and (d) scaled. The IMOS spectra n (b) and (d) have been rescaled to enable drect comparson of the hgher nose levels (dashed lnes). Same colour codng as Fg. 9 4) Smoothng wth the Savtzky-Golay Flter In ths secton we dscuss the applcaton of the Savtzky-Golay flter to the IMOS QC data (QC60) and to the ADCP data over tme wndows of, 3, 7, 3, and 5 hours (note that the flter, as mplemented n Matlab, requres an odd number of data ponts), as llustrated n Fg. that compares tme seres of the u- components of the radar and ADCP currents. Increasng the flter order (.e. from st order lnear to nd order quadratc), retans more of the hgh frequency content n the data. Analyses usng the 30-mnute averages (QC30) as the startng pont produced smlar results (not shown). The same comparsons between

30 radar and ADCP have been made, namely: () the Taylor dagram statstcs, () the drectonal statstcs and () the spectral analyss. The Taylor dagrams for the CH00 moorng at COF and the HIN moorng at CBG are shown n Fg.. They ehbt rather dfferent behavour. The COF results show a reducton n the normalsed rms and ncrease n correlaton coeffcent as the smoothng perod ncreases although the standard devaton of the radar measurements mantan a very smlar rato wth that of the ADCP. Dfferent normalsatons are appled to each case (as descrbed n Fg. ) snce the ADCP standard devatons also decrease wth smoothng (albet very slowly n the COF case). In the CBG case the longer perod smoothng cases start to devate from the ADCP measurements possbly due to the domnance of the sem-durnal sgnal at ths locaton whch s beng fltered out n these cases. Fg. 3 compares the 5 hour fltered tme seres at the HIN and OTE moorngs and shows that the HIN radar u-component s stll rather nosy n spte of the flterng. Snce the ampltudes of the resdual currents are small n both cases, the nosy radar data are domnatng the statstcs of the comparson at HIN. Ths nose n surface currents s lkely to be due to subcell scale current varablty at HIN whch s n shallower water and closer to the reefs than OTE (Fg. a). We appled three dfferent estmates of the uncertanty n the radar measurements to all the analyss presented n ths paper: () the measure obtaned from the spectrum (the rms nose level assumng whte nose), () the root-mean-square dfference between radar and ADCP and (3) the standard devaton n the radar u, v components derved from the nformaton suppled n the data fles usng the methods descrbed n Secton IIIB for the averages and Append.3 for the flter. Estmates and 3 (eclusvely based on the radar) are shown n Fg. 4a for these fve longer perod fltered cases. At COF the values relate to the v- component and at CBG to the u-component snce these are the larger n both cases. The errors decrease as the length of smoothng ncreases, so that the fve dfferent cases are easy to dentfy n Fg. 4. For a gven perod of smoothng the lnear flter (order n the legend) seems to reduce both estmates more than the quadratc flter (order ) probably because of the dfferences n hgh frequency content referred to above. The standard devatons estmated from the data (tem 3 above and shown n the Fg 4a) are generally

31 lower at COF and hgher at CBG than those obtaned from the spectra. The results suggest that the COF estmates provded by the processng from radar data to radal may be a lttle optmstc although the estmates are clearly correlated at both stes. The tme seres n Fg. 3 confrms the low spectral estmates at OTE relatve to those at HIN. Estmates and are compared n Fg. 4b. The pattern at HIS and HIN s very dfferent from elsewhere wth an ncrease n the root-mean-square dfference for longer smoothng perods (as observed n the Taylor dagrams and attrbuted to hgher nose levels). The root-mean-square dfferences between radar and ADCP are hgher than the spectral estmates ndcatng that there are real dfferences between the measurements. Ths s dscussed further n Secton V. 5) Drfter comparson Taylor dagrams for the comparsons between the h averaged drfter and radar currents etracted along the tracks of the eght drftng buoys are shown n Fg. 5, both for u- and v-components. The comparson statstcs use data normalsed by the standard devaton of each drfter (as specfed n Fg. 5). Drfters B- E (released on the shelf; Fg. 5 and 6) showed the best agreement for both u- and v-components, the comparson beng partcularly good for the east-west currents (u-component) wth the larger range. The results were smlar for three of the shelf drfters (Fg. 5, drfters C, D and E). Drfter B remaned close to the launch ste and dfferences wth the radar u-component are a lttle hgher. The slope drfters, released n deeper water, showed a hgher varablty for the v-component measurements than s present n the radar data (normalsed radar standard devatons < ). The slope drfters travelled along the CBG reef matr and developed several loops around the slands, beng more nfluenced by small scale crculatons observed near the reefs (Fg. 6). In addton, they travelled more dstant from the radar stes and some of ths varablty maybe eplaned by the smaller radar spatal resoluton at long ranges (as shown wth the black boes n Fg. 6). The dfferences n the radar spatal resolutons between the shelf (smaller samplng cells) and slope (larger samplng cells) measurements may have contrbuted to the dfferences observed n the comparsons. Scatter plots of the comparson of drfter

32 D and for all the cases together are shown n Fg. 7. The comple correlaton coeffcents found for these cases were 0.97 to 0.98 for the shelf drfters and 0.93 to 0.94 for the slope drfters (not shown). Fg.. Tme seres of the u-component of current at the CH00 moorng at COF for the IMOS -hour averaged data (upper panel); after applcaton of the 5 hour Savtzky-Golay flter wth flter order (mddle) and (bottom). Radar data are n blue and ADCP (at 4 m depth) n red.

33 Fg.. Taylor dagrams showng the comparson between radar and ADCP data for dfferent smoothng perods and flter orders denoted by the letters where B ( h), D (3 h), F (7 h), H (3 h), J (5 h) are order and C ( h), E (3 h), G (7 h), I (3 h), K (5 h) are order. Upper row: CH00 at COF, lower row: HIN at CBG. Notaton s otherwse the same as Fg. 6. The normalsatons n m/s used at CH00 for u are 0.8(B), 0.8(C), 0.7(D), 0.8(E), 0.7(F), 0.8(G), 0.6(H), 0.7(I), 0.5(I), 0.6(J) and for v are 0.39(B), 0.39(C), 0.39(D), 0.39(E), 0.38(F), 0.39(G), 0.37(H), 0.39(I), 0.37(I), 0.38(J) at HIN for u are 0.35(B), 0.36(C), 0.33(D), 0.36(E), 0.(F), 0.34(G), 0.09(H), 0.6(I), 0.07(I), 0.(J) and for v are 0.4(B), 0.4(C), 0.3(D), 0.4(E), 0.0(F), 0.3(G), 0.07(H), 0.(I), 0.07(I), 0.08(J)

34 Fg. 3. Tme seres of the u- components of current at the HIN (above) and OTE (below) moorngs at CBG after applcaton of the 5-hour Savtzky-Golay order flter. Radar data are n blue and ADCP n red Fg. 4. Comparsons of dfferent estmates of uncertanty n the radar measurements for the dfferent smoothng perods (, 3, 7, 3, 5 hours). The numbers followng the moorng locaton n the legend refer to the order of the Savtzy-Golay flter. Spectral estmates are compared wth (a) the radar standard devaton and (b) the rms dfference between radar and ADCP. The smaller black symbols o and denote the samplng varablty estmate of uncertanty for -hour averaged radal data, dscussed n secton IIIA, at CBG and COF respectvely. In some cases the 3 hour smoothng overlad the h smoothng pont and thus there appear to be only 4 symbols for those cases on the fgure. The : lne (dashed) s shown n (a) for reference.

35 Fg. 5. Taylor dagrams showng the comparson between drfter and radar data. Notaton s manly the same as Fg. 7 ecept the letters now refer to dfferent drfters B-E released on the shelf and F-I on the slope. The normalsatons used are u: 3.4, 3.0, 3.0, 3.0, 3.6, 3.3, 4.3, 3.7 v: 8.0, 6.9, 6.8, 7.5, 4.9, 4.9, 4.5,.8 cm s - respectvely. Fg. 6. Map showng the drfter tracks, 4 from each release pont(); radar stes (). Black boes mark nomnal radar resoluton from each radar at grd cells centred n the mddle of the overlap.

36 Fg. 7. Scatter plots of the comparson between drfter and radar data for case D (on the Taylor dagram above) and (as bnned plots, notaton as n Fg. 6) for all the data combned (below). V. DISCUSSION The accuracy of HF radar current measurements at two coastal regons (Eastern Australa) wth comple crculatons was evaluated for the frst tme n ths study usng the Taylor dagram, the drectonal statstcs and spectral analyss. Tdal and wnd forcng domnate n the southern regon of the GBR lagoon, whle the