PAPER Performance Evaluation of SeaSonde High-Frequency ... · Performance Evaluation of SeaSonde High-Frequency Radar for Vessel Detection AUTHORS Hugh J. Roarty Erick Rivera Lemus

P A P E R

Performance Evaluation of SeaSondeHigh-Frequency Radar for Vessel DetectionA U T H O R SHugh J. RoartyErick Rivera LemusEthan HandelScott M. GlennCoastal Ocean ObservationLaboratory, Rutgers University

Donald E. BarrickJames IsaacsonCODAR Ocean Sensors, Ltd.,Mountain View, CA

A B S T R A C THigh-frequency (HF) surface wave radar has been identi!ed to be a gap-!lling

technology for Maritime Domain Awareness. Present SeaSonde HF radars havebeen designed to map surface currents but are able to track surface vessels in adual-use mode. Rutgers and CODAR Ocean Sensors, Ltd., have collaborated onthe development of vessel detection and tracking capabilities from compact HF ra-dars, demonstrating that ships can be detected and tracked by multistatic HF radarin a multiship environment while simultaneously mapping ocean currents. Further-more, the same vessel is seen simultaneously by the radar based on different pro-cessing parameters, mitigating the need to preselect a !xed set and therebyimproving detection performance.Keywords: radar, detections, maritime domain awareness, dual-use

Introduction

The U.S. Integrated Ocean Ob-serving System (IOOS®) led by theNational Oceanic and AtmosphericAdministration (NOAA) has designed(Interagency Working Group onOcean Observation, 2009), is con-structing, and has recently begun oper-ating the more advanced portions of anational high-frequency (HF) radarnetwork focused on the real-timemap-ping of surface currents. The primaryoperational users of the resulting sur-face current maps are the U.S. CoastGuard for Search and Rescue and theNOAA HazMat team for ocean spillresponse. The IOOSMid-Atlantic Re-gion’s CODAR SeaSonde HF RadarNetwork, led by Rutgers University,is the !rst region in the United Statesto achieve operational status by con-structing and operating the end-to-end system that produces andlinks validated real-time surface cur-rent maps to the Coast Guard’s Searchand Rescue Optimal Planning System(Roarty et al., 2010b).

The Department of HomelandSecurity has called for the developmentof tools to provide wide-area surveil-lance from the coast to extend beyondthe horizon (Department of HomelandSecurity Science and Technology,2009). Rutgers and CODAR OceanSensors, an academic-industry partner-ship established in 1997, have workedtogether for over a decade to expandthe capabilities of compact CODARHF radars to include the dual-use appli-cation of detecting and tracking shipswithout compromising the network’sability to map surface currents. Initialdevelopment focused on the demon-stration and evaluation of a non-real-time end-to-end system for dual-usevessel tracking in the New York Bightmultifrequency HF radar testbed(Roarty et al., 2010a). Technologydemonstrations determined (a) thatvessels could be detected, (b) thatmultilook detections could be associ-ated with a known ship, and (c) thatthe associated detections could then beinput to a range of tracking algorithms

whose output produced tracks and pre-dicted trajectories on a computer screen,providing useful information to oper-ators. Radar hardware developmentfocused on developing network "exi-bility beyond monostatic backscatteroperations, demonstrating (a) that bi-static and multistatic operations werepossible with a shore-based networkand (b) that buoy-based bistatic trans-mitters can be operated at all three ofthe commonly used HF radar frequen-cies (5-6, 12-13, and 24-25 MHz).The greatest challenge in developinga robust ship surveillance capabilityfor any HF radar is the development ofthe initial vessel detection algorithm.This conclusion focused the initial re-search on the mathematical problemof identifying and extracting the radarreturn of a surface vessel hidden withina highly variable and noisy back-ground, requiring additional detectionalgorithm development, testing, andsensitivity analysis in a variety of envi-ronments with different noise charac-teristics. It is the aim of this paper to

14 Marine Technology Society Journal

analyze the parameters and settings ofthe vessel detection algorithm that areoptimal for !nding those ship echoesamong the other signals that are sentback towards the radar. In Methodol-ogy, we describe the HF radar net-work, the SeaSonde HF radar used inthis test, the Automatic Identi!cationSystem (AIS) network used to groundtruth the radar detections, and the shipdetection software used to process theradar data for ships. In Results, we dis-cuss the results of the ship detection testthat was conducted. Lastly, the perfor-mances of the various ship detectionprocessing methods against the avail-able targets are discussed in Discussion.

MethodologyHF Radar Network

These experiments were conductedwithin the New Jersey Shelf ObservingSystem (Glenn & Scho!eld, 2002). Amajor component of this observingsystem is an HF radar network. Thenetwork was created in 1998 with theplacement of two 25-MHz systems onthe southern coastline of New Jersey(Kohut & Glenn, 2003). The networkwas then expanded with the placementof four 5-MHz systems spanning theNew Jersey coastline (Gong et al.,2009). The 25-MHz network wasmoved north in 2003 in support of theLagrangian Transport and Transforma-tion Experiment (Chant et al., 2008).The work discussed here utilizes the lat-est addition to the network: the place-ment of a 13-MHz system outside theentrance to New York Harbor in SeaBright, New Jersey. This network alsocontributes to the Mid-Atlantic Re-gional Coastal OceanObserving System(MARCOOS), which has a total of30 radars from Cape Hatteras to CapeCod that are operated by eight univer-sities (Roarty et al., 2010b).

13-MHz SeaSonde SystemThe radar was deployed in Sea

Bright, New Jersey, 40 km south ofthe Battery in New York City. Theradar was a direction-!nding typeradar, SeaSonde Remote Unit SSRS-100, manufactured by CODAR OceanSensors and was installed in October2008. The radar’s primary functionwas the measurement of surface cur-rents, which are provided in real timeto the NOAA National HF RadarNetwork (Temll et al., 2006). Theradar also has the dual-use capabilityto detect the location of ships at sea.

The radar consists of a compact re-ceive antenna with three elements: twodirectional crossed loops and an omni-directional monopole, a monopoletransmit antenna, and a hardwarehoused within a climate-controlled en-closure (Figure 1). The radar transmitsa radio wave with a center frequencyof 13.46 MHz and a bandwidth of50 kHz. The bandwidth of the radarsets the spatial range resolution of thesystem, which was about 3 km for thisparticular bandwidth. The details ofthe waveform are given in Table 1.Separate transmit and receive antennaswere used for this study spaced at leastone wavelength apart, which is approx-imately 23 m at the 13-MHz radioband. A ship with a vertical structureof a quarter wavelength (6 m) is theminimum-sized optimal re"ector(Ruck et al., 1970).

Figure 2 shows the spatial and tem-poral radial vector coverage for oceancurrents of the radar over a 1-weekperiod, which coincided with the shipdetection exercise. The radar collectedrange data, which are a time series ofthe complex echo signal voltagesbefore Doppler processing, from00:00GMT to 01:00 GMT on Febru-ary 26, 2009. These range !les are theresult of the !rst fast Fourier transform

FIGURE 1

Picture of the (A) transmit antenna, (B) receiveantenna, and (C) equipment enclosure for theSeaSonde 13-MHz radar.

TABLE 1

Characteristics of the radar waveform used inthe study.

Waveform Characteristic Value

Center frequency 13.46 MHz

Bandwidth 50 kHz

Blank 668.8 !s

Blank delay 8.55 !s

Sweep rate 2 Hz

Pulse shaping On

May/June 2011 Volume 45 Number 3 15

(FFT) of the frequency-modulatedcontinuous wave received signal. Therange data were collected usingan FFT length of 512 points. With a2-Hz sweep of the radar, each range!le encompasses 256 s of coherent in-tegration time. There were a total of15 range !les over the hour-long pe-riod. The time on the computer andall the subsequent !les it generates aresynchronized to atomic time via GPSby the Macintosh operating system.

AIS ReceiversRutgers also operates an AIS receiver

network, which was utilized in thisstudy; this allows transponders on

vessels to broadcast the ship’s positionand identi!cation. When earlier workwas performed (Roarty et al., 2010a),the authors were limited to verifyingdetections of vessels where a self-recording GPS could be placed on avessel by the researchers or when theGPS information could be providedby other researchers (Rossby &Gottlieb, 1998). The ability of the re-searchers to obtain AIS position dataon the vast majority of ships at seahas greatly accelerated the research.The Rutgers AIS network has receiv-ers, which are manufactured by ShineMicro, Inc., located at its !eld sta-tion in Tuckerton, Sandy Hook, andLoveladies, New Jersey, as shown in

Figure 3. The AIS transmissions areused as ground truth for the HF radarship detections. The range of theshore-received AIS signal is typically30 nautical miles, but under certainatmospheric conditions, range can beupwards of hundreds of nauticalmiles (International Maritime Or-ganization, 2006). Data from the in-dividual AIS receivers were sent backto the Rutgers Coastal OceanObserva-tion Laboratory (Glenn & Scho!eld,2009), where it was archived usingthe Coast Guard software ‘AISSource.’ The data were then timestamped using the clock on the com-puter. The computer kept time witha software tool to synchronize withthe National Institute of Standardsand Technology Internet Time Ser-vice. Figure 4 shows the tracks of ves-sels sent via AIS over a 3-week period,indicating that this is a target-rich en-vironment and that vessels do notalways stay in the shipping lanes.

Detection SoftwareThe ship detection algorithm is ex-

plained in Roarty et al. (2010a). Theship detection code is written in theMATLAB programming languageand is designed to run of"ine in abatch-processing mode. The rangedata that were collected by the radarwere read by the software to processfor the hard targets. The ship detectioncode utilizes a constant false alarm rate(CFAR) to !nd targets. A signal that isabove the background by some thresh-old on the monopole and at least oneof the two loops is considered a pos-sible detection. Figure 5 shows thespectra from the monopole of the SeaBright radar site at 00:15:21 GMTon February 26, 2009. The x axis de-notes Doppler shift, the y axis showssignal strength, and the z axis denotes

FIGURE 2

Radial coverage map of the SeaSonde at Sea Bright, New Jersey, over a 1-week period. The colormap illustrates the temporal coverage along the radial grid (black = 75%, red = 50%, pink = 25%).(Color versions of !gures available online at: http://www.ingentaconnect.com/content/mts/mtsj/2011/00000045/00000003.)


the radar range cell, which was 3 kmfor these spectra. The Bragg peaksfrom which surface currents arederived (Barrick, 1972) are shown±0.4 Hz. The large signals at zeroDoppler are signals returned to theradar from stationary objects. Shipsignals can be seen between theBragg peaks and the zero Dopplersignals, with positive Doppler shiftdenoting a ship moving towards theradar and the corollary with a signalon the negative Doppler measuring aship moving away from the radar.The ship detection code is designedto identify these signals. The code uti-lizes two schemes as the basis forthresholding in its CFAR peak-picking, where one averages in time,in!nite impulse response (IIR), andthe other averages in Doppler andrange space using a median to createthe signal background. In its currentform, the code is able to process thedata using three combinations ofthreshold and integration time foreach background simultaneously.This results in a set of six detectionpackages after each software run. Thedetection code performs a slidingFFT on the range data so a new detec-tion !le is output every 32 s. This set-ting is adjustable so that the user caninput the desired update rate for thedetections. The output of the detec-tion code is a series of !les that containrange, range rate, and bearing of possi-ble detections from the radar. The !lesalso include the uncertainties in theabove quantities, the signal-to-noiseratio for each antenna, and an estimateof the radar cross section of the possibletarget.

ResultsThe range data that were collected

at the radar site were transported back

FIGURE 4

Map of the study area showing tracks of vessels (red lines) sent via AIS over a 3-week period.The Nantucket, Hudson Canyon and Barnegat (clockwise from right) shipping lanes are shownin the bottom right.

FIGURE 3

Location of the three AIS receivers operated by Rutgers University shown as green circles.(Color versions of !gures available online at: http://www.ingentaconnect.com/content/mts/mtsj/2011/00000045/00000003.)


to the laboratory for of"ine processing.The user sets three combinations ofintegration time and threshold foreach of the two backgrounds. Hence,the output of one run of the rangedata through the ship detection coderesults in six concurrent data streamsof possible ship detections. Becausethe ship detection code only outputssix packages (three for the IIR methodand three for the Median method) foreach software run, the software is runseveral times to !ll out our desiredprocessing matrix. The threshold(dB)/FFT points that were initiallytried using the IIR background were6/16, 7/32, 8/64, 9/128, 10/256,and 11/512. The threshold (dB)/FFTpoints that were tried using theMedian background were 8/32, 9/64,10/128, 11/256, 12/512, and 13/1024.The threshold was increased becausethe average of signal to noise increasedby !1 dB for each doubling of the FFTlength (Roarty et al., 2010a). A plot ofrange (km), range rate (m/s), and bear-ing (degrees clockwise from north) ofpossible detections using the IIR back-ground, 256-point FFT, and 10-dBthreshold is shown in Figure 6. Thetrails of vessels can be seen in therange and range rate subplots. Thereare also false positives in the data

stream as noted by the single detec-tions with no adjacent detections inspace or time.

The AIS data were then used to seehow the radar was performing when itcame to detecting the speed and loca-tion of vessels at sea. The AIS data were!rst !ltered by time to coincide withthe measurements, then geographicproximity to the radar site (a 60-kmthreshold was used) and then binnedby ship identi!cation. Eighteen shipspassed this !rst stage of !ltering.Then any ship that was located onthe bay side of the radar or had zeroradial velocity was removed. This leftfour ships for possible detection, onetug boat (the Dolphin), two cargo con-tainers (theMaas Trader and theMOLEf!ciency), and a tanker (the Joelmare).The latitude, longitude, and time

FIGURE 5

Picture of power spectra for Antenna 3 of the SeaSonde at 00:15 GMT on February 26, 2009. Thex axis is Doppler shift (Hz), the y axis is signal strength (dB), and the z axis denotes the range bin fromthe radar (scalar). Vessel echoes are observed between the two sea-echo Bragg peaks at approxi-mately ±0.4 Hz.

FIGURE 6

Plot of target detections from 00:10 to 00:55 GMT on February 26, 2009. Panels from top to bot-tom are range, range rate, and bearing. The yellow horizontal lines in the middle are the expectedpositions of the very strong Bragg sea clutter echoes that would mask vessel detection.


reports of these four ships were used tocalculate the ship range, radial velocity,and bearing relative to the radar at SeaBright, New Jersey. A text !le of shipposition and time was processed usingthe program ‘GPSTracker’, which ispart of the SeaSonde detection softwarepackage. This software is normallyused to perform the same task whenmeasuring the antenna pattern of thereceive antenna with a transponderon a vessel. This software generates a!le for each vessel that contained therange, radial velocity, and bearingof that particular vessel as shown inFigure 7. These data were then usedfor comparison with the range, radialvelocity, and bearing calculationsfrom the radar.

The next step was to compare thedata from the radar with data obtainedvia AIS. If the calculated range iswithin half the width of a range bin(1.5 km in this case) and within twoDoppler bins (varied between 0.02and 1.4 m/s, depending on the lengthof the FFT window) of the actual ves-sel, then the detection is considered ahit. A detection rate is then calculatedas the number of times the radar de-tected the target divided by the totalnumber of time sample possibilities,which was every 32 s for this experi-ment. The detection rates for Joelmareusing the IIR and Median backgroundare shown in Tables 2 and 3, respec-tively. The detection rates for MaasTrader using the IIR and Medianbackground are shown in Tables 4and 5, respectively. The detectionrates for the Dolphin using the IIRand Median background are given inTables 6 and 7, respectively. In thecase of theMaas Trader, once the high-est detection rate was found for theinitial runs, then the thresholds werevaried along the FFT length with thehighest detection rate. The detection

FIGURE 7

Range, range rate, and bearing plot of four vessels in the vicinity of the Sea Bright HF radarstation on February 26, 2009 as measured by AIS.

TABLE 2

Detection rate in percent for the Joelmare with different combinations of threshold (columns) andFFT points (rows) using the IIR background.

6 dB 7 dB 8 dB 9 dB 10 dB 11 dB

16 NSD

32 NSD

64 0.6

128 15.6

256 13.2

512 10.8

NSD stands for “no ship detected.”

TABLE 3

Detection rate for the Joelmare with different combinations of threshold (columns) and FFTpoints (rows) using the Median background.

8 dB 9 dB 10 dB 11 dB 12 dB 13 dB

32 0.2

64 3.3

128 34.0

256 33.3

512 12.5

1024 7.6


rate for theMOLEf!ciencywas not cal-culated due to a short AIS record, butthere are results that will be discussedin the next section.

An example of the “association” ofthe detections with the ground truthdata is shown in Figure 8. This !gureshows the detectionsmade by the radarassociated with the GPS position ofthe cargo ship the Maas Trader fromFigure 6. The detections are shownas blue diamonds with error boxesaround the detection, with half theheight signifying the uncertainty ofthe measurement and the width ofthe box denoting the length of theFFT window. The uncertainty ! isgiven by the equation

! ! "!!!!!!!!!!!!SNR3

p ;

where " is the range, range rate, orbearing bin size and SNR3 is thesignal-to-noise ratio on Antenna 3(the monopole). This !gure is char-acteristic of the uncertainty providedby the SeaSonde HF radar that deter-mines bearing using direction !nding,i.e., low uncertainty in range and rangerate measurements and higher uncer-tainty for the bearing estimate.

DiscussionWe will now discuss detection re-

sults of the radar with four vesselsthat passed in front of the radar duringthe test period.

Cargo Container MOL Ef!ciencyThe MOL Ef!ciency, International

Maritime Organization (IMO) ShipNo. 9251365, is a cargo containerwith a length of 294 m and a beamof 32 m. The vessel was exiting NewYork Harbor on a southeast course

TABLE 4

Detection rate in percent for theMaas Trader with different combinations of threshold (columns) andFFT points (rows) using the IIR background.

6 dB 7 dB 8 dB 9 dB 10 dB 11 dB

16 NSD

32 NSD

64 NSD

128 58.7

256 70.3 70.3 68.9 66.2 64.9 64.9

512 38.5


TABLE 5

Detection rate for the Maas Trader with different combinations of threshold (columns) and FFTpoints (rows) using the Median background.

8 dB 9 dB 10 dB 11 dB 12 dB 13 dB

32 13.8

64 23.6

128 55.0

256 65.0 65.0 62.5 61.3 58.8 58.8

512 46.5

1024 30.8

TABLE 6

Detection rate for the Tugboat Dolphin with different combinations of threshold (columns) andFFT points (rows) using the IIR background.

6 dB 7 dB 8 dB 9 dB 10 dB 11 dB

16 NSD

32 0.5

64 5.4

128 92.4

256 73.1

512 60.0



and was approximately 15 km fromthe radar. The AIS record of theMOL Ef!ciency only spans from

00:00 to 00:23 GMT on February26, 2009. It is unclear to the authorsas to why the AIS record terminated.

Did the operators of the vessel stoptransmitting the signal, or were the re-ceivers unable to record the signal? Theradar was able to make detections onthe vessel coincidental with the AISdata. If we assume that the vessel main-tained its course and the radial velocitymaintained its rate of change, then theradar did indeed make the additionaldetections of the vessel as clearlyshown in Figure 9.

Tanker JoelmareThe Joelmare, IMO Ship No.

9288019, is a tanker with a length of228m and a beam of 32m. The tankerJoelmarewas exiting New York Harborbut then turned around and headedback into the harbor. The radar wasable to detect the vessel 34% of thetime with the Median backgroundand 16% of the time using the IIRbackground. We propose two ex-planations as to why the radar didnot detect this vessel very well. First,from Figure 7: The radial velocity ofthe vessel was noisiest of all the vessels.This lack of a constant radial velocitywould spread the energy of the re-turned signal over several Dopplerbins. This would cause the signal tonot be detected, because its amplitudefalls below the threshold set in the soft-ware. The second reason as to why theradar did not detect this vessel wasthat the vessel was north-northeast ofthe radar site, and the signal had topropagate over large sections of landthat attenuated the signal in thosedirections.

Cargo Container Maas TraderThe Maas Trader, IMO Ship No.

9308625, has a length of 139 m, abeam of 23 m, and a gross tonnage of9981.More particulars on this vessel aswell as theDolphin are given in Table 8.

TABLE 7

Detection rate for the Tugboat Dolphin with different combinations of threshold (columns) andFFT points (rows) using the Median background.

8 dB 9 dB 10 dB 11 dB 12 dB 13 dB

32 10.8

64 31.0

128 93.2

256 78.6

512 56.2

1024 32.8

FIGURE 8

Plot of target detections (blue dots) and corresponding uncertainty values (blue squares) asso-ciated with GPS track of the Maas Trader (solid aqua line). The panels from top to bottom arerange (km), radial velocity (m/s), and bearing (degrees clockwise north). The uncertainty valuesfor each measurement are shown as the height of each blue box; the length of the FFT is thewidth of the box. (Color versions of !gures available online at: http://www.ingentaconnect.com/content/mts/mtsj/2011/00000045/00000003.)


The authors were unable to !nd theheight above the water for any of thevessels detected. The cargo containerMaas Trader was heading south inthe Barnegat shipping lane for thistest. The radar at Sea Bright had thebest detection rate of 70% using theIIR background and 65% with the Me-dian background. Both of these casesoccurred with the lowest thresholdand yielded the highest number offalse positives. It would be up to a po-tential user of the detection !les to de-termine where the threshold should beset. If the detection !les were to bepassed on to a tracker, the performanceof the data in the tracker could helpdetermine what the threshold levelshould be. Another option is to runall combinations of FFT length andthreshold and let a tracker determinewhich detections are authentic andwhich ones are false. The radar wasnot able to detect the vessel at 00:40.This was due to the fact that the vesselwas crossing through the zero Dopplerarea, where there are large echoes fromstationary objects. A plot of the FFTlength and threshold combinationsversus detection rate, which is a sum-mary of Tables 4 and 5 for the MaasTrader test case, is given in Figure 10.A peak in the detection rate is foundbetween the 128- and 256-pointFFT.With a 2-Hz sweep, this convertsto a 1- to 2-min averaging period asoptimal for the detection of these ves-sels with the HF radar.

Tugboat DolphinThe Dolphin , IMO Ship No.

7319010, has a length of 41 m, abeam of 10 m, and a gross tonnage of198. The tugboatDolphinwas headingnorth into New York Harbor inside ofthe Barnegat shipping lane. Figure 4shows that this is a heavily transited

FIGURE 9

Plot of target detections for range, range rate, and bearing. The solid aqua line from 00:10 to00:23 GMT shows the path of theMOLEf!ciency from the AIS signal. There are additional detectionspast 00:23 on the !gure, but the AIS data were not available to compare with the radar detections.(Color versions of !gures available online at: http://www.ingentaconnect.com/content/mts/mtsj/2011/00000045/00000003.)

TABLE 8

Characteristics of vessels detected in this study.

Vessel Characteristic Maas Trader Dolphin MOL Ef!ciency Joelmare

MMSI No. 237956000 366920980 351166000 477738400

IMO No. 9308625 7319010 9251365 9288019

Type Cargo Tug Cargo Tanker

Length (m) 139 41 294 228

Beam (m) 23 6 32 32

Hull type Single Single NA NA

Gross tonnage 9981 198 NA NA

Depth (m) 11.8 5 NA NA

Draught (m) 8 4.25 12.1 6.6

Freeboard (m) 3 0.75 NA NA

Hull material Steel Steel NA NA

NA stands for “not available.”


route as well. The Sea Bright radar hadthe highest detection rate on this vesselwith a 92% using the IIR backgroundand 93% using the Median back-ground. The vesse l had a large

superstructure, which made it anideal target for the radar. The bestcase detections from the Dolphin andMaas Trader were placed on a map inFigure 11.

SummaryA case study has been performed

using a SeaSonde HF radar to detectvessels at sea in a dual-use mode. Theselected 13-MHz HF radar that op-erates within the MARCOOS andprovides radial current data to theNOAA National HF Radar Networksimultaneously detected the speedand location of several ships at sea.The detections made with the HFradar were checked against the GPSposition of the target sent via the AISsystem. An optimal integration timefor this type of radar with this classof vessel is between 1 and 2 min. Thedetection rates for some vessels wereabove 90%. Lower thresholds resultedin higher detection rates but also led tohigher false alarm rates. Overall, themedian background performed betterthan the IIR background, but therewere instances where the IIR back-ground was the best.

One bene!t of HF systems forvessel detection/tracking is that theyprovide over-the-horizon detectioncapability (Khan et al., 1994). The sys-tems being developed and evaluatedwithin the 5-MHz and 13-MHzbands regularly see vessels between50 and 110 km. However, the focusof this paper was on the optimal pro-cessing parameters that would enablethe best detections. Because of this,the authors focused on vessels thatwere close to shore and would providethe most signal to test these parameterswithout introducing other parametersfrom the radar equation. The authorswill, in future work, use these optimalparameters to test the range limits ofthe vessel detection system as a sepa-rate study.

A signi!cant !nding was that thesame targets were seen by the detectionalgorithm simultaneously, at different

FIGURE 10

Response of detection rate to the variance of FFT length and threshold level.

FIGURE 11

Detections of the Maas Trader (IIR background, 256-point FFT, and 10-dB threshold) andDolphin (median background, 256-point FFT, and 10-dB threshold) overlaid on the path ofthe vessel from GPS/AIS.


FFT/coherent-integration times, withdifferent thresholding and back-grounds. Thus, one is not forced topreselect a !xed processing param-eter suite. A properly optimized as-sociation algorithm (which is underdevelopment) would search for de-tections of the same vessel amongall of the output combinations andthereby yield a much improved de-tection. This would increase thedetection rate seen by an individuallook of 90%, for example, to perhaps98%, while reducing the overall falsealarm rate.

This offers the opportunity to con-vert the National HF Radar Networkinto a dual-use system that wouldprovide surface currents to the U.S.Coast Guard as well as provide widearea surveillance in the maritime do-main to the Department of HomelandSecurity.

AcknowledgmentThis material is based on work

supported by the U.S. Departmentof Homeland Security under AwardNo. 2008-ST-061-ML0001.

The views and conclusions con-tained in this document are those ofthe authors and should not be inter-preted as necessarily representing theof!cial policies, either expressed orimplied, of the U.S. Department ofHomeland Security.

Lead Author:Hugh J. RoartyCoastal OceanObservation LaboratoryInstitute of Marine andCoastal SciencesRutgers UniversityNew Brunswick, NJ 08901Email: [email protected]

ReferencesBarrick, D. 1972. First-order theory andanalysis of MF/HF/VHF scatter from the sea.IEEE Trans Antennas Propag. 20(1):2-10.doi: 10.1109/TAP.1972.1140123.

Chant, R.J., Glenn, S.M., Hunter, E.,Kohut, J., Chen, R.F., Houghton, R.W., …Scho!eld, O. 2008. Bulge formation of abuoyant river out"ow. J Geophys Res.113(C1):C01017.

Department of Homeland Security Scienceand Technology. 2009. High-Priority Tech-nology Needs. Washington, D.C. 24 pp.Available at: http://www.dhs.gov/xlibrary/assets/st_high_priority_technology_needs.pdf.

Glenn, S.M., & Scho!eld, O. 2009. Growinga distributed ocean observatory: Our viewfrom the COOL room. Oceanography.22(2):128-45.

Glenn, S.M., & Scho!eld, O.M.E. 2002.The New Jersey shelf observing system.OCEANS ’02 MTS/IEEE. 3:1680-7.

Gong, D., Kohut, J.T., & Glenn, S.M.2009. Seasonal climatology of wind-drivencirculation on the New Jersey shelf. J GeophysRes. 109(C7):C07S07.

Interagency Working Group on OceanObservation. 2009. A Plan to Meet theNation’s Needs for Surface Current Mapping.Washington, D.C. 64 pp.

International Maritime Organization. 2006.Automatic Identi!cation Systems. London:IMO Publishing. 102 pp. Available at: http://www.ioos.gov/library/surfacecurrentplan9_3lowres.pdf.

Khan, R., Gamberg, B., Power, D., Walsh, J.,Dawe, B., Pearson, W., & Millan, D. 1994.Target detection and tracking with a highfrequency ground wave radar. Journal ofOceanic Engineering. 19(4):540-8.

Kohut, J.T., & Glenn, S.M. 2003. ImprovingHF radar surface current measurementswith measured antenna beam patterns.J Atmos Oceanic Technol. 20(9):1303-16.doi: 10.1175/1520-0426(2003)020<1303:IHRSCM>2.0.CO;2.

Roarty, H.J., Barrick, D.E., Kohut, J.T., &Glenn, S.M. 2010a. Dual-use compact HFradars for the detection of mid- and large-size vessels. Turk J Electr Eng Comput Sci.18(3):373-88.

Roarty, H.J., Glenn, S.M., Kohut, J.T.,Gong, D., Handel, E., Rivera Lemus, E., …Seim, H. 2010b. Operation and applicationof a regional high-frequency radar network inthe Mid-Atlantic Bight. Mar Technol Soc J.44(6):133-45.

Rossby, T., & Gottlieb, E. 1998. TheOleander Project: Monitoring the variabilityof the gulf stream and adjacent waters be-tween New Jersey and Bermuda. Bull AmMeteorol Soc. 79:5-18. doi: 10.1175/1520-0477(1998)079<0005:TOPMTV>2.0.CO;2.

Ruck, G.T., Barrick, D.E., Stuart, W.D.,& Krichbaum, C.K. 1970. Radar CrossSection Handbook. New York: Plenum Press.949 pp.

Temll, E., Otero, M., Hazard, L., Conlee, D.,Harlan, J., Kohut, J., … Lindquist, K.2006. Data management and real-timedistribution in the HF-radar national network.In: OCEANS 2006 MTS/IEEE Proceedings,pp. 1-6. Boston, MA. doi: 10.1109/OCEANS.2006.306883.


Documents

PAPER Performance Evaluation of SeaSonde High-Frequency ... · Performance Evaluation of SeaSonde High-Frequency Radar for Vessel Detection AUTHORS Hugh J. Roarty Erick Rivera Lemus