1. Introduction
For many years, most United States open ocean research ships have been equipped with acoustic Doppler current profilers (ADCPs) from R. D. Instruments (RDI); specifically, the 150-kHz model VM-150 that was first produced around 1984. This instrument uses single unmodulated pings to measure the Doppler shift of sound scattered by plankton and nekton along each of four narrow acoustic beams. Because the pings are uncoded, they have the narrowest bandwidth that is consistent with a given vertical resolution (typically 8 m); the VM-150 is referred to as a narrowband (NB or NB-150) instrument. Typical VM-150 range varies from 250 to 450 m, depending on ambient noise and the availability of scatterers in the water column.
As a possible successor to their NB instruments, RDI developed broadband (BB) ADCPs. In each ping, the phase of the carrier is modulated with a repeated pseudo- random code, resulting in a relatively broad bandwidth for a given pulse length. The advantage is higher short-term accuracy and/or vertical resolution (Pinkel and Smith 1992); disadvantages include reduced depth range and an inherent ambiguity in the velocity estimates (appendix A). Although RDI's original BB instruments can perform well in applications such as moored and lowered profilers, and can also be satisfactory in shipboard applications when the weather is good and maximum range is not required, their reduced depth range and robustness makes them inferior to the NB for blue-water ships.
RDI now offers an alternative, called the Ocean Surveyor (OS). The NB and the BB instruments use a separate transducer for each beam, but the Ocean Surveyor generates all four beams at once from a single, flat, phased array of transducer elements. Because the phased array is roughly half the diameter of a cluster of four discrete transducers of the same frequency, much lower frequencies of operation become feasible; OS systems are offered at nominal operating frequencies of 150, 75, and 38 kHz. Range and velocity measurement noise vary inversely with frequency, so cutting the frequency in half, from 150 to 75 kHz, roughly doubles the expected range but requires that more pings be averaged for a given accuracy. The OS can be operated in either broadband or narrowband mode; one might choose broadband mode for improved vertical resolution in shallow water in good weather, and narrowband mode as the default. The OS-75 transducer is similar in size to the NB-150, so a replacement normally will not require enlarging the transducer as well.
The purpose of this paper is to compare the old standard (a 150-kHz RDI NB ADCP) with a possible successor: a 75-kHz Ocean Surveyor. Simultaneous measurements from both instruments were made on a transit of R/V Endeavor in February 2001. We compare the OS measurements in broadband and narrowband mode to each other and to the NB, show how the data quality of all three can be improved by editing, and discuss the strengths and weaknesses of the OS.
2. Instruments and setup parameters
The two sonars, NB-150 and OS-75, were installed on R/V Endeavor, a 56-m ship operated by the University of Rhode Island, Graduate School of Oceanography, and scheduled through the University–National Oceanographic Laboratory System (UNOLS). The OS-75 transducer was mounted in an existing well, about 2 m forward of the NB-150 transducer well. Both transducers were near the keel, about 20 m from the stern. Beam 3 of the NB-150 pointed about 45° clockwise of forward; this orientation minimized the extreme values of Doppler shift when the ship was underway. The OS-75 forward beam was directed about 8° from its intended fore–aft alignment.
Position fixes were provided at 1 Hz as National Marine Electronics Association (NMEA) “$GPGGA” messages from a P-code global positioning system (GPS) receiver. Primary heading measurements came from the ship's gyrocompass. Secondary attitude measurements were provided at 2 Hz via NMEA “$PASHR,ATT” messages from an Ashtech ADU-2 4-antenna GPS attitude sensor. ADCP velocity profiles were initially rotated into earth coordinates using the gyrocompass heading. In later processing, this rotation was corrected based on the difference between Ashtech and gyrocompass heading measurements. [See King et al. (2001) for references and discussion of shipboard ADCP processing and accuracy.]
The OS-75 was controlled by RDI's relatively new Vessel-Mounted Data Acquisition System (VMDAS, version 01.2.0.6), running under Microsoft Windows. VMDAS records raw single-ping data and American Standard Code for Information Exchange (ASCII) data streams from up to two serial ports, in addition to several stages of processed data.
As on most UNOLS ships, the NB-150 was controlled by RDI's MSDOS-based software, DAS 2.48, with data acquisition supplemented by the memory-resident programs agcave.exe and ue4.exe. The latter was configured to record raw single-ping data and coincident gyrocompass and ADU-2 attitude measurements to a networked disk.
To minimize acoustic interference between the two instruments, they were configured to ping simultaneously. Because the OS-75 is capable of greater range, it has a longer minimum interval between pings than the NB-150; hence, the OS-75 was configured to serve as the master and the NB-150 as the slave. This required modification of the latter by RDI. Special firmware (version 17.12, replacing 17.10) was developed by RDI, and a modified Peripherals-3 board was installed to receive the trigger pulse.
It is convenient to treat the OS as two virtual instruments: the “OSN” when in narrow bandwidth mode, and the “OSB” when in broad bandwidth mode. The OS can alternate pings between broadband and narrowband modes, so that the two virtual instruments, OSN and OSB, can operate simultaneously. In this study, either kind of OS ping triggered a ping in the NB-150, resulting in two more virtual instruments: “NBN” or “NBB” for pings triggered by OSN or OSB, respectively (Table 1). Concurrent NBN and OSN pings (or NBB and OSB pings) therefore sampled the same ship's velocity and nearly the same ocean velocity. The latter differed primarily because of the differences in azimuthal alignment of the NB and OS transducers.
Both the OS and the NB can measure the velocity of the ocean bottom relative to the ship. When this option is turned on, each water track ping group is preceded by a bottom track ping. Normally there is a single water track ping per group, but when the OS is operating with interleaved broadband and narrowband pings, the ping sequence is bottom track, broadband, and narrowband. The length of a bottom track ping is proportional to the water depth, and is usually much longer than a water track ping.
The length of each water track ping (pulse length) and the length of each time slice for processing the returned signal (bin length, or depth cell size) are also important parameters governing ADCP performance. Shorter bin and pulse lengths improve vertical resolution but reduce accuracy and range. Although bin and pulse lengths are fundamentally time intervals, they are specified as the depth intervals to which they correspond. The NB-150 is usually operated with 8-m bins and 8-m or 16-m pulses; occasionally 4-m bins and/or pulses are used (Chereskin and Harding 1993). Reasonable choices for the OSN-75 are 8- and 16-m bin and pulse size; because of its higher short-term accuracy, the OSB-75 can be operated with 4-m bins and pulses in addition to 8 and 16 m. Although it is not necessary, it is common to keep the pulse length equal to the bin length, and this was done for both instruments and all tests.
For both instruments, the specified depth bins are nominal; the actual sampling depths depend on the sound speed profile. The NB instruments use a nominal sound speed of 1470 m s−1 for all depth calculations, but the OS instruments use either a user-specified sound speed or a sound speed calculated from measured temperature and specified salinity. Pulse lengths in both the NB and the OS must be integer multiples of fixed time intervals, so the depth range ensonified by a given pulse increases with sound speed. The NB uses fixed time intervals for depth cells also, but as estimated sound speed increases, the OS reduces the sampling time so as to maintain a constant depth interval, taking into account the variation of the beam angle with sound speed at the transducer. For the purposes of our instrument comparison, we corrected the OS depths to match the NB depths; we did not correct for the variation of sound speed with depth.
After a ping, processing of the received sound must be delayed long enough to allow the reverberation to die down. This “blanking interval” varies inversely with sonar frequency; for the NB-150 it was set to the most common value, 4 m, and for the OS-75 it was set to 8 m. These are the defaults. Larger values may be needed for transducer installations subject to excessive reverberation, as can occur when the transducer is behind an acoustic window, and/or under conditions of very low acoustic backscatter. Because the blanking interval is longer for the OS-75 than the NB-150, the shallowest bin is also deeper unless the bin and/or pulse are shortened to compensate. The center of the first bin on Endeavor for the typical NB configuration, with 8-m bin and pulse length, is 17 m; this can be achieved with the OS by using 4-m bin and pulse lengths in broadband mode. More typical OS-75 bin and pulse lengths would be 8 m, for a 21-m top bin center. To maximize range and accuracy one might use 16-m bin and pulse, yielding a shallowest velocity estimate centered at 29 m.
All three virtual instruments correctly measure the velocity component along each beam only if it falls within a range that depends on the instrument type and setup. There is an inherent ambiguity in velocity estimation with coded pulses (appendix A), so the OSB reports velocities within a maximum “ambiguity interval” of υ0 ± 3.9 m s−1, where the offset υ0 is a user-settable parameter to compensate for the ship's speed. If the actual velocity falls outside this interval, it wraps around; for example, with υ0 = 0, a velocity of 4 m s−1 would be reported as −3.8 m s−1. Although the NB does not have such an inherent ambiguity, there is still a wraparound. The usual NB-150 interval is ±5.12 m s−1, because velocities in units of 0.25 cm s−1 are stored as signed 12-bit integers; hence, the NB ambiguity interval results from design decisions about data precision and data storage requirements. The OSN has a velocity interval of υ0 ± 4.5 m s−1. If the velocity in the upper part of the profiling depth range lies outside this interval, the ping is rejected; otherwise, a bin-to-bin tracking loop follows the signal as a function of depth, permitting deeper velocity estimates to be outside the initial acceptance interval. Under no circumstances do the velocity estimates wrap around to the other end of the allowable velocity interval. In calm seas, these velocity intervals rarely would be exceeded by a research ship, although the problem has been identified in NB-150 data from the NOAA ship Oceanographer working in the eastern equatorial Pacific, owing to the combination of a fast ship and strong current shear (P. Plimpton 1987, personal communication). The speed threshold can be maximized through suitable choice of υ0 for the OS, and on either instrument by mounting the transducer with the beams oriented 45° from the keel. In rough seas, however, transducer heave may trigger ambiguity errors at much lower ship speeds.
Although the normal ambiguity interval for shipboard NB systems is 5.12 m s−1, it can be reduced to 2.56 m s−1 (with velocity recorded in units of 0.125 cm s−1) by a switch setting on the Peripherals-2 circuit board. About halfway through the cruise, it was discovered that some time at least two years earlier the switch on the Endeavor's ADCP had been set to the lower value. The error was corrected on 20 February 2001. While the low setting was in effect, all averaged NB data during which the ship's speed relative to the water exceeded 6–13 kt, depending on sea state, were corrupted.
For each depth bin of each ping, both ADCPs use two criteria to accept a velocity estimate. First, a measure of signal strength or quality must exceed a threshold. For the NB and the OSN this is an estimate of signal-to-noise (S/N) ratio, and for the OSB it is correlation. The NB and OSB return the quality number on an arbitrary scale from 0 to 255; OSN values range from 0 to 190. The instrument default for the threshold is 120 counts for both OSB and OSN. Second, the error velocity (scaled difference between the two independent estimates of vertical velocity) must not exceed a threshold, which was set to 1.5 m s−1 on the NB (although 1 m s−1 is used more commonly) and was left at its default value of 1 m s−1 on the OSB and OSN. A third velocity screening algorithm, designed to reject returns from strong discrete scatterers such as fish, was left disabled by default. (The fish rejection algorithm is more useful for moored current profilers than for vessel-mounted instruments.)
3. Data and processing
Both single-ping and averaged data from all three virtual instruments (NB, OSB, and OSN) were recorded. Single-ping data included axial velocity components for each beam. All ensemble averaging began with single-ping velocity profiles transformed to earth coordinates using a transformation matrix based on the array geometry together with an estimate of heading, but ignoring pitch and roll. Initially, 5-min averages from each instrument were processed independently using standard University of Hawaii software (available online at http://currents.soest.hawaii.edu).
Preliminary analysis of the averaged data showed puzzling differences among the velocity estimates from the OSN, OSB, and NB, so additional analysis was done using the single-ping data. OSN and OSB pings were matched to the NB pings they triggered (labeled NBN and NBB, respectively). The matched sets of pings were then divided into 5-min ensembles. The key point is that when comparing the NBN to the OSN or the NBB to the OSB, the temporal sampling is identical, but when comparing the OSN to the OSB, it is not; some of the velocity difference comes from differences in the times at which the ship's wave-induced motion is sampled. During one period of rough weather, for example, this aliasing caused a 13 cm s−1 standard deviation of the difference between 5-min averaged OSN and OSB velocities.
Ensemble averaging of pings into 5-min profiles was done using an iterative reference layer averaging scheme. Each ensemble is approximated as a single function of depth, with a zero-average over a reference layer (chosen here as 50–150 m), plus a reference layer velocity for each ping (appendix B). Adding the average of the single-ping reference layer velocities to the function of depth yields the ensemble-average velocity profile.
The NB and OS transducer orientations were calculated using standard bottom tracking methods (Joyce 1989; Pollard and Read 1989), after correcting the bottom track velocity for the gyro compass error as determined by the ADU-2 attitude sensor.
Because our purpose is to compare two instruments rather than to measure ocean currents, we did not use standard methods to find scale calibration factors independently for the OS and the NB. Instead, an approximate scale factor for the NB, 0.9976, was calculated from bottom track measurements at the end of the cruise. All NB velocities were multiplied by this factor. Then the OS scale factor, 1.003, was chosen for best agreement between the OS and the NB during a test when the weather was good and the ship speed was varied in steps. The same scale factor was used for both broadband and narrowband modes of the OS. The NB velocities were corrected for variations in sound speed at the transducer. No such correction was made for the OS; because the phased array maintains a constant horizontal wavenumber component, the scale factor for the horizontal velocity component does not depend on sound speed.
4. Tests
The ADCP comparison took place during a transit from the shipyard in Tampa, Florida, to Endeavor's home port, Narragansett, Rhode Island (Fig. 1), 17–23 February 2001. The ship track included both shallow coastal regions and deep water. Much of the track was along the Gulf Stream; two sections across the Gulf Stream were made near the end of the cruise.
Three types of test were performed. First, the OS and NB were operated separately and simultaneously, with and without triggering, to check for interference. Second, the ship's speed was varied to compare the OSN, OSB, and NB in terms of data return, depth range, and accuracy. Third, during most of the cruise, when the ship was underway at full speed, the test consisted of simply gathering data with several reasonable configurations, in order to get as broad a comparison as possible among the three instruments. Most of our analysis effort has concentrated on a period when adverse weather reduced data quality (Fig. 2; decimal day 52.11–52.65).
5. Results
a. Bottom tracking
Most of the potential bottom tracking data for the NB were corrupted by the incorrect Peripherals-2 board scale factor setting; only a small data sample (26 5-min ensembles) from the end of the cruise was available. Therefore, NB bottom tracking was useful for calculating the NB scale factor, but not for direct comparison to OS bottom tracking.
The OS tracked the bottom during more than half the cruise, but the velocities were sometimes erratic. After the cruise, the problem was identified as a bug in the OS firmware, subsequently corrected by RDI, which caused invalid velocity estimates to be set to zero instead of the value used to identify bad data (−36728). The problem was evident mainly during bad weather (decimal day 49–49.8). Although we have not done a careful analysis of the OS bottom track data, a cursory look at the raw data with the bad values removed indicates that it is still subject to small, slowly varying bias. In our experience with many NB-150 systems on other ships, bottom tracking is subject to bias varying with weather, bottom depth, and bottom topography, so we are not surprised to see similar behavior in the OS.
b. Beam sidelobes
Although the OS-75 and the NB-150 have similar nominal beamwidths of about 4° (D. Symonds 2000, personal communication), the NB-150 appears to have weaker sidelobes. Evidence is provided by the amplitude of the return from the ocean bottom. The earliest reflection comes from the downward-pointing sidelobe, followed by the peak amplitude along the main lobe, and then more sidelobe energy. In a plot of amplitude versus depth, the return from the ocean bottom shows up as a strong local maximum, roughly 40 dB above the return from scatterers in the water (Fig. 3). Comparing the amplitude profiles from NB and OS, we see in the latter a slower decline to the noise floor below the maximum amplitude, and returns from the nearly vertical sidelobe that are 12–15 dB higher.
We suspect that the sidelobes of the OS-75 transducer on the Endeavor are worse than on some other OS systems. Bottom returns from an OS-75 mounted temporarily on the R/V Oceanus and from an OS-38 mounted permanently on the Japan Marine Science and Technology Center (JAMSTEC) R/V Kaiyo, although broader in depth range than those from the Endeavor's NB, lack the sharp increase at 85% of the water depth characteristic of the Endeavor's OS.
c. Acoustic interference
When the NB and the OS were pinging asynchronously, acoustic interference was evident as isolated high amplitudes in the single-ping data from both instruments. More spikes showed up in NB data than in either OSN or OSB, perhaps because the first harmonic of the OS is the operating frequency of the NB. Some amplitude spikes yield random velocity estimates, others are rejected. In broadband mode, the OS rejects most of the high-amplitude samples, presumably because of low correlation. In all virtual instruments, some spikes were eliminated by the error velocity screening criterion used during acquisition, but we do not know how many. Editing in postprocessing, as discussed later, also removes some, but not all, of the velocities corresponding to amplitude spikes. Although the affected velocity samples that pass the editing criteria are biased toward zero velocity relative to the ship, there are so few of them (less than 1%) that their effect on a 15-min average profile was too small to see in our tests. The bias after editing appears to be 1 cm s−1 or less. (Asynchronous pinging with bottom tracking was not tested.)
Bottom tracking with synchronous pinging causes very serious interference. The synchronization mechanism does not distinguish between bottom track and water track pings, and it is the start, not the end, of the pings that are simultaneous. Especially in deep water, bottom track pings are longer than water track pings. Therefore, an OS bottom track ping, for example, coinciding with an NB water track ping, can corrupt the top 200 m of the latter.
Synchronous water track pings can also interfere with each other. If the pulse length from one instrument exceeds the sum of the pulse and blanking length of the second, and if the transducers are close together, then the top depth bin of the second instrument will be corrupted by the ping from the first. This occurred frequently in our tests with a 16-m pulse from the OS together with an 8-m pulse and the default 4-m blanking interval on the NB.
d. Range and resolution
As expected, in narrowband mode the OS-75 was able to profile about twice as deep as the NB-150 under a variety of conditions (Fig. 2). The NB range varied from 200 m (8-m bins, bad weather) to over 400 m (8-m bins, good weather), the OSB varied from 400 m (8-m bins, bad weather) to 650 m (8-m bins) or 700 m (16-m bins), and the OSN varied from 600 m (8-m bins, bad weather) to 750 m (8-m bins) or 800 m (16-m bins). The OSB range was generally about 85% of the OSN range when configured with the same bin and pulse, dropping to 80% when configured for the same short-term standard deviation (e.g., 8-m bin and pulse for the OSB, 16-m for the OSN). The ranges of all instruments and modes were reduced in rough weather, particularly as the ship's pitching increased. The percentage of good velocity estimates (after editing in each instrument and data acquisition system, but with no additional editing in postprocessing) was most dramatically reduced in the OSB over the entire potential profiling range. Both OSN and OSB maintained their highest percent good in their top bins, but NB percent good was reduced in the top few bins and in the deeper bins. (In the context of ensembles of pings, the phrase “percent good” refers to the percentage of all pings for which the velocity estimate in a given depth bin is accepted as valid, based on a given set of editing criteria; hence, it is a function of depth. This will be distinguished later from a “layer percent good,” which is calculated for a single ping at a time and is not a function of depth.)
e. Short-term accuracy
The theoretical maximum accuracy of any Doppler sonar is a function of the pulse type and length, and the signal processing (Pinkel and Smith 1992; Chereskin and Harding 1993). In practice, the standard deviation of velocity estimates is always larger than the theoretical minimum, typically by about a factor of 2. To measure the standard deviation directly, one would need an ensemble of measurements made under perfect conditions, with uniform motion of scatterers. Lacking such perfect conditions, we can get an upper bound estimate from the redundancy inherent in four-beam sonars. RDI scales the error velocity so that “in horizontally homogeneous flows, the variance of the error velocity will indicate the portion of the variance of each of the nominally-horizontal components (X and Y) attributable to instrument noise (short-term error)” (R. D. Instruments 1998, p. 11).
The error velocity standard deviations for the NB and OSB instruments with 8-m pulse and bin, and for the OSN with 16-m pulse and bin, are similar: about 0.15–0.18 m s−1 over most of the profiling range (Fig. 4). The standard deviations imply that to obtain 1 cm s−1 accuracy one must average about 290 OSB pings or 360 OSN pings. The OSB standard deviation increases rapidly near the bottom of the profiling range, as the signal-to-noise ratio and the correlation drop. No such increase at depth is seen in the NB or OSN profiles.
f. Velocity comparison
The error velocity provides an estimate of short-term variance, but it is insensitive to significant long-term bias. Therefore, we compare averaged profiles among the three virtual instruments. Instrument-related bias usually increases with measured speed, so we will look for it primarily in the forward velocity component of the water relative to the ship. Defining forward as positive, the relative velocity is typically −4.5 to −6 m s−1 when the ship is underway.
With good weather allowing nearly 100% good data in the top 250 m for all instruments, agreement among the OSN, OSB, and NB was excellent (Fig. 5). No significant biases could be detected. The good agreement is partly a result of our selection of the OS scale calibration factor for best agreement in a fair-weather data sample, but that intercalibration did not guarantee that the OSN would agree with the OSB, or that the agreement among all three would be good throughout their depth ranges.
In rough weather, with reduced percentages of good data for all instruments, the velocity profile estimates differed. Relative to the NB, both the OSN and OSB were biased toward zero by about 5 cm s−1 below 100 m, and up to 10 cm s−1 at shallower depths; most of the differences in vertical shear were above 100 m. Recall that this comparison is made with matched pings, so it reflects differences in the way the instruments estimate the same velocity field, not differences in sampling times.
To investigate the differences between NB and OS during rough weather, we calculated the reference layer velocity difference between matched pings, that is, between simultaneous OSN and NB (NBN) pings, and between simultaneous OSB and NB (NBB) pings. We then grouped the pings according to the percentage of good depth bins within the 50–150-m reference layer, as a rough indicator of a priori profile reliability. (We will refer to this as the layer percent good, calculated for each ping, to distinguish it from the usual percent good, which is the percentage of pings with good estimates in a given depth cell.) The histogram of NB minus OS difference in the forward velocity component is skewed to negative values; most of the tails, or outliers, are instances when the NB is measuring a larger (more negative) flow of water past the ship (Fig. 6). Although this could be a bias of the NB toward large values, it seems more likely that it is a bias of the OS toward zero. The skewed distribution is clear in the unedited data when the NB pings have high percent good (top panels in Fig. 6) and the OSB pings have medium or low percent good; but it is found even with high percent good in the OSN pings. This is at least partly a consequence of the normally greater range of the OSN; the reference layer is a small part of the normal range, so a bad ping (i.e., a ping with velocity estimates in a much smaller fraction of the potential depth range than would occur under ideal conditions) may still have a high layer percent good. Nevertheless, we are left with the impression that the OSB algorithm includes more stringent acceptance criteria than the OSN algorithm, and that the latter leaves more outliers (bad estimates) that need to be removed by editing before averaging the pings into ensembles.
There is also evidence of bias toward zero in the weak NB pings, the bottom panels in Fig. 6. The red curves, showing OS pings with high layer percent good, are skewed to the right; that is, when good OS pings coincide with bad NB pings (those with low layer percent good), the velocity difference suggests a bias toward zero in the NB. Looking at all panels in Fig. 6, however, we see many more negative outliers than positive ones; without editing, the bias toward zero occurs more frequently in the OS than in the NB.
g. Editing
Although it has long been known that ADCP velocity measurements could be improved by editing the single-ping profiles prior to averaging (Zedel and Church 1987; Trump 1989), the editing capabilities of common commercial data acquisition systems (DASs) such as DAS 2.48 are minimal. Therefore, we have revisited the question of single-ping editing, both for the NB and for the OS.
To develop editing criteria that can be applied to single pings, and to single bins within pings, we used the subset of NB and OSB depth bins that most closely matched corresponding OSN depth bins (usually differing by less than 2 m). The depth-matched difference in forward velocity component between NBB and OSB and between NBN and OSN was then analyzed as a function of possible editing criteria. Three criteria were found to be useful: error velocity, correlation (OSB) or S/N (OSN), and the percentage of valid velocity estimates within a specified shallow layer (the layer percent good). The first two criteria are independently applied to each bin in a ping; the latter is a measure of the “strength” or “quality” of the ping as a whole.
Binwise editing criteria are shown in Fig. 7 for the same “bad weather” data shown in Fig. 5. Reducing the error velocity threshold to 0.5 m s−1 for OSB and OSN was highly effective at removing outliers from the velocity differences (Fig. 7, second panels). The distribution of points removed is highly skewed; very few velocity differences near zero are falsely rejected by this criterion. (With different instrument parameters and ocean conditions, however, a different threshold might be required.)
Correlation or S/N in the OS was very effective at removing bins with bad velocity estimates. A single threshold of 150 was applied to all bins (Fig. 7, bottom panels). This removed many bad OS velocity estimates, most of which were biased toward zero relative to the corresponding NB estimates. A more stringent criterion (180) was used for the top few bins of the OSB. Because the top three bins of the OSN had very high S/N regardless of the quality of the velocity estimate, one threshold (186) was used for the top three OSN bins, and another (170) was used for bins 4–6.
After applying the more stringent error velocity and correlation or S/N criteria, velocity difference outliers remain, particularly in the top few depth bins. These outliers are often found in profiles with very few good bins, most of which are shallow (Fig. 8). Most of these can be eliminated by rejecting entire profiles with too few good bins in a specified shallow layer. Using a layer extending from 50 to 150 m, suitable percentages of good bins required within the layer are 50% for the OSB and the NB, and 80% for the OSN.
These three criteria, applied independently to each profile, were not completely effective, so we added a fourth criterion based on an ensemble of profiles: rejection of single-bin velocity outliers. The ensemble-averaged profile of velocity relative to the reference layer and the reference layer velocity for each ping are calculated as described in appendix B. The sum of the relative velocity profile and the reference layer velocity is the model profile, and the difference between this model and the measured velocity is the residual. The standard deviation of the residuals is calculated for all bins in an averaging interval; residuals in excess of two standard deviations or 1 m s−1, whichever is smaller, trigger rejection of their corresponding velocity estimates.
The four editing criteria are applied in sequence as follows: first, the error velocity and correlation or S/N thresholds weed out individual velocity estimates; second, the layer percent good criterion is used to delete weak profiles; third, ensemble outliers are removed; and fourth, the layer percent good criterion is applied a second time. Except for the S/N criterion, the editing was applied to both the NB and the OS datasets.
The effect of this editing procedure is negligible in calm weather but becomes large as the weather deteriorates (Fig. 5). In our bad-weather example, the editing removes the 0.05 m s−1 offsets between NBN and OSN and between NBB and OSB, and almost eliminates the differences in the top 100 m (Fig. 5, lower middle panel). Below 100 m, editing shifts the NB curves to the left, toward more negative velocities, by about 0.05 m s−1, and shifts the OS curves to the left by about 0.1 m s−1. By removing 10%–20% of the velocity estimates (Fig. 5, rightmost panels), the editing also reduces the effective range of the OSN by about 100 m in this example. A weak scattering layer near 400 m boosts the percent good, helping to maintain the OSB range after editing.
An important question at this point is, to what extent might the negative shift of the velocity profiles by editing reflect biased sampling in addition to instrument bias? Do “weak profiles” tend to occur when the ship has temporarily been slowed by plowing into a wave, for example? With the present dataset we cannot answer this conclusively, but the available evidence suggests that the selective sampling bias is small compared to instrument bias. To estimate sampling bias, we look at the edited NB reference layer forward velocity for each ping minus the 5-min ensemble average. The NBB pings are divided into two groups: those with corresponding OSB pings that pass all the editing criteria, and the remainder. The latter group consists of “good” NBB pings at times of “bad” OSB pings. The histogram of their velocity deviations is only slightly biased to the right (Fig. 9, left panel, thick black curve), with a mean of 13 cm s−1. The corresponding histogram for good NBN velocity deviations coinciding with rejected OSN pings is biased even less, with a mean of 7 cm s−1. The corresponding sampling bias in the mean of the accepted pings is 1.6 cm s−1 for the NBB and 0.4 cm s−1 for the NBN. It is likely that some of this is not actually sampling bias; residual bias toward zero in the good NB pings would shift the histograms in the same direction.
6. Discussion
The test cruise has demonstrated that the OS-75 can be used to measure ocean currents at ranges up to 800 m under good conditions, with a nominal vertical resolution of 16 m or better. A similar instrument operating at half the frequency, the OS-38, can profile even deeper; in the western equatorial Pacific, where scattering levels are low, the instrument on the JAMSTEC ship Kaiyo routinely profiled to 1000 m or more (Firing and Kashino 2002). This improved range relative to the NB-150 will be of great value. For example, use of shipboard ADCP measurements as a reference for geostrophic calculations is often limited in accuracy by the near-inertial velocity signal, which is noise for the purpose of this calculation. The noise is reduced by vertical averaging, and/or by using a deeper reference layer, where the near-inertial energy level is lower.
The most significant concern about the OS-75—an a priori concern that was confirmed by the test—relates to its acoustic beam pattern. Beamwidth and sidelobe response are functions of the effective aperture of a transducer and of its spatial uniformity. Single-ceramic transducers, such as those used in the NB-150, tend to have nearly ideal characteristics for a given aperature (or transducer diameter measured in wavelengths at the operating frequency). Phased arrays, as in the OS-75, are inherently inferior because they are made up of many elements; any differences in properties from one element to the next, or in their respective electronic circuits, degrade the beam pattern. The best indicator of beam pattern from the test cruise is the amplitude of the return as a function of time (plotted as nominal depth) when the ocean bottom is in range. Comparing the NB-150 and the OS-75, we found that the latter has higher sidelobes; its return from the sidelobe pointed straight down is about 12–15 dB higher than that from the NB. One expected consequence of this poorer beam pattern is increased severity of artifacts in the velocity profile in the presence of strong acoustic scattering layers. Unfortunately, although such layers are ubiquitous in many ocean regions, they were not encountered during the test cruise, so this prediction could not be confirmed. Beam patterns are expected to be more variable from one transducer to another, and over time as the ceramics age, for phased arrays than for single ceramics. It is not yet clear to us, however, whether the particular transducer on the Endeavor is typical of the OS series; cursory inspection of data from the OS-75 on the Oceanus and from the OS-38 on the Kaiyo suggests that their sidelobes are intermediate between the Endeavor's OS and NB, but a more thorough study is needed.
To address this concern about beam sidelobes, we would like to see published specifications for the beam pattern, together with one or more acceptance tests. These might include preinstallation tank or lake testing, and they should include a reasonably easy postinstallation test that can be repeated periodically to check for degradation due to aging. Analysis of the amplitude of the mainlobe and sidelobe return from the ocean bottom may suffice, once test areas with known bottom properties are established.
A second concern about the OS profilers is that even in NB mode they have a smaller axial velocity range (υ0 ± 4.5 m s−1) than that of the NB-150 (±5.12 m s−1). We suspect that this has been a little-known but significant problem for some NB-150 installations and that it may continue to cause trouble with the OS series. RDI has attempted to reduce the problem by shifting the zero point of the velocity range. It is not clear how much this will help. The problem, if it occurs, will be manifest in rough weather, and in installations subject to large heave, most likely with the transducer mounted near the bow. We have not found evidence of velocity wraps in the data from this cruise, however.
A third concern is velocity bias and reduced depth range in moderately rough weather, such as during the test cruise when pitch maxima exceeded 3°. The OS-75 performance degrades more quickly than the NB-150. The bias is toward zero relative to the ship; when earth-relative velocity profiles are calculated, the bias is therefore in the direction of ship's motion. Because we do not understand the cause of the bias yet, we cannot predict how it will behave under different weather and scattering conditions, and on different ships.
Although we have developed an editing procedure that seems to have been quite effective in minimizing bias in both the NB and the OS, it has important limitations. It was tuned for this particular dataset; we do not know how well it will work on other ships and under other conditions. Furthermore, it is not easy to judge; on this cruise we were able to compare two instruments with identical temporal sampling (matched simultaneous pings), but we did not have the large number of ship accelerations that could provide an independent check.
Editing out the bad velocity estimates is only half of the battle; the other half is calculating an accurate time-averaged velocity from the estimates that remain. This problem is especially severe in bad weather, when many pings in a row can be blocked by bubbles under the transducer, and the ship's speed can vary substantially over periods of several seconds to a minute or more. One solution would be to supplement the ADCP with short-term inertial navigation. A package of tilt sensors and accelerometers near the transducer would allow accurate estimation of the velocity fluctuations between valid pings and, hence, of the time-averaged velocity over a typical 5-min averaging interval. The accelerometers would also permit detection and resolution of velocity ambiguity errors.
The Ocean Surveyor systems work well enough, and offer enough advantages, that we expect they will gradually replace many elderly NB systems. A question that arises is, which particular system, or combination of systems, should be installed? The advantage of phased arrays over discrete transducers is their smaller size for a given operating frequency, making low-frequency systems feasible. The trade-off for improved range is reduced accuracy, vertical resolution, and ability to profile close to the bottom of the ship. The OS-75 is a good compromise, but a better system for many ships may be the combination of an OS-38 and a sonar operating at 150 kHz or higher. (Another alternative to consider is the combined 50- and 140-kHz system designed and built by Dr. R. Pinkel and collaborators at Scripps Institution of Oceanography and installed on the R/V Revelle.) For best performance, the high-frequency sonar should use discrete transducers for each beam, unless future tests demonstrate equivalent sidelobes from a high-frequency phased array. The high-frequency sonar would provide high vertical resolution and accuracy in the top 200 m or so, while the low-frequency sonar would provide range exceeding 1000 m. Tests of acoustic interference between the OS-75 and the NB-150 on the Endeavor indicate that synchronization of the two sonars probably would not be necessary. Interference might be minimized by mounting the two transducers as far apart as possible, given other constraints. If synchronization is used, either the pings should be interleaved or the start of the high-frequency ping should be delayed relative to the low-frequency ping so that both pings end at the same time. If the pings start at the same time, the longer low-frequency ping will interfere with the top of the high-frequency profile.
The Endeavor tests, together with experience on the Kaiyo, show that the OS-75 and OS-38 can track the bottom at depths exceeding their water tracking range, but we suggest that this capability is best left unused. Tracking the bottom in deep water cuts the water track ping rate in half (assuming that the system is operating in NB mode or BB mode, not interleaved), thereby reducing velocity profile accuracy, while providing no advantage over GPS navigation for referencing the profile to the earth. Bottom tracking in shallow water is useful for calculating the transducer orientation relative to the ship's fore-aft reference, such as the GPS attitude sensor's antenna array. This can be done when leaving and approaching port but is not necessary at other times during a cruise when GPS attitude measurements are available. Another motivation for using bottom tracking on NB-150 systems is to take advantage of the instrument's ability to clip each profile at 85% of the water's depth, as determined by bottom tracking, thereby eliminating sidelobe interference from the bottom. This editing step can be done with our present software on single-ping data, with the bottom depth determined from the water-track amplitude profile, so bottom tracking is not needed.
7. Conclusions
The most basic conclusion is that the OS-75 works; that is, it can be used to measure ocean currents at ranges up to 800 m under good conditions, with nominal vertical resolution of 16 m or better. Broadband and narrowband modes both work roughly as expected, with the BB pings yielding higher short-term accuracy or better vertical resolution at the cost of about 15% in range under good conditions. In comparison to the fleet-standard NB-150, the OS-75 in NB mode achieves about twice the range (e.g., Fig. 10).
Editing of single-ping data prior to averaging is essential for achieving maximum accuracy under less than ideal conditions. In rough weather, we found a forward velocity bias of about 0.05 m s−1 in the NB-150 and about twice that in both BB and NB modes of the OS-75. The causes of this bias are not understood. The bias can be reduced with a combination of editing criteria involving error velocity, correlation or signal strength, the number of valid depth bins in a ping, and removal of outliers from an ensemble of pings.
In addition to the problem of bias in bad weather, which was evident on our test cruise, there are two problems that we did not see in our particular dataset but that may appear in the future. First, transducer heave in rough weather may cause the axial component of beam velocity to exceed the ambiguity interval, leading to data loss and/or corruption. Second, the downward-pointing sidelobes of the OS beams on Endeavor are 12–15 dB higher than the NB sidelobes. In regions with strong scattering layers, such as the eastern equatorial Pacific, the beam sidelobes cause systematic horizontal velocity artifacts of order 0.1 m s−1 with NB systems; we expect the problem will be worse with the OS.
Our findings lead to two recommendations.
Because the beam sidelobes of a phased array may vary with time and from instrument to instrument, and because instrument accuracy in the presence of scattering layers may depend critically on sidelobe levels, beam pattern specifications and tests are needed.
Three-axis tilt and acceleration sensors should be mounted in or near the transducer. This additional instrument package would permit resolution of velocity ambiguities and would improve the accuracy of velocity measurement.
The tests reported here, together with early experience with a 38-kHz Ocean Surveyor mounted on the JAMSTEC ship Kaiyo, demonstrate the potential scientific value of these low-frequency Doppler sonars. We look forward to taking advantage of their improved depth range on future cruises.
Acknowledgments
We thank Darryl Symonds of R. D. Instruments for making the Ocean Surveyor available for this test, for his invaluable help before and during the cruise in setting up the instrument, and for patiently answering questions before, during, and after the cruise. We thank Bill Fanning for his technical assistance during and after the cruise. Thanks go to John Freitag for working with RDI and with Sandy Shor at NSF to arrange and fund the comparison. This work was funded in part by the National Science Foundation under Grant OCE-0072128.
REFERENCES
Chereskin, T. K., and Harding A. J. , 1993: Modeling the performance of an acoustic Doppler current profiler. J. Atmos. Oceanic Technol., 10 , 41–63.
Firing, E., and Kashino Y. , 2002: Energetic deep currents observed east of Mindanao. Eos, Trans. Amer. Geophys. Union, 83 (Suppl.) OS30.
Joyce, T. M., 1989: On in situ “calibration” of shipboard ADCPs. J. Atmos. Oceanic Technol., 6 , 169–172.
King, B. A., Firing E. , and Joyce T. M. , 2001: Shipboard observations during WOCE. Ocean Circulation and Climate: Observing and Modelling the Global Ocean, G. Siedler et al., Eds., Academic Press, 99–122.
Pinkel, R., and Smith J. A. , 1992: Repeat-sequence coding for improved precision of Doppler sonar and sodar. J. Atmos. Oceanic Technol., 9 , 149–163.
Pollard, R. T., and Read J. F. , 1989: A method for calibrating shipmounted acoustic Doppler profilers and the limitations of gyro compasses. J. Atmos. Oceanic Technol., 6 , 859–865.
R. D. Instruments, 1998: ADCP coordinate transformation, formulas and calculations. 27 pp. [Available from R. D. Instruments, 9855 Businesspark Ave., San Diego, CA 92131.].
Trump, C. L., 1989: Three practical hints on using vessel mounted ADCPs. Mar. Technol. Soc. J., 23 (3) 23–35.
Zedel, L. J., and Church J. A. , 1987: Real-time screening techniques for Doppler current profiler data. J. Atmos. Oceanic Technol., 4 , 572–581.
APPENDIX A
Ambiguity Velocity
APPENDIX B
Profile Averaging
A shipboard ADCP measures the velocity of the water relative to the ship; its variability during an averaging period, say 5 min, is usually dominated by variations in ship motion, including the short-term wave-induced motions and any changes in speed and heading. Because profiling range is also variable, simple ensemble averaging of all available velocity estimates as a function of depth leads to artifacts in the bottom of the profile, particularly when profiling range is correlated with velocity. To minimize this problem, it is common to calculate the average in two parts, using a reference layer. Ideally, the reference layer is a depth range with 100% good data. The average of all valid velocity estimates within the reference layer is subtracted from each profile; the residual velocity profiles are temporally averaged, the reference layer velocities are temporally averaged, and the two parts are added back together to yield the time-averaged profile. This procedure, or a slight variation, has been used in most ADCP data acquisition software (with the notable exception of RDI's Transect program) during the last 20 years (e.g., Trump 1987). Although it works very well when a reference layer with 100% good data exists, it is suboptimal when no such layer can be found. This is usually the case on at least part of any cruise. Therefore, we have implemented an improved averaging procedure that is less sensitive to missing data.
R/V Endeavor cruise track and tests, Tampa, FL, to Narragansett, RI. Velocity vectors are averages over the top 100 m. Bottom depth is contoured at 1000-m intervals.
Citation: Journal of Atmospheric and Oceanic Technology 20, 6; 10.1175/1520-0426(2003)020<0872:ADCOTR>2.0.CO;2
Ship motion and depth range of the three virtual instruments. The abcissa is time in days from the start of 2001. (top) The 5-min maximum absolute values of roll (red circles) and pitch (black plus signs), together with ship's speed (green); (lower three panels) the percentage of good velocity estimates as a function of time and depth for the NB, OSB, and OSN virtual instruments. Data are as collected by the shipboard data acquisition systems and averaged in 5-min ensembles, with no additional editing
Citation: Journal of Atmospheric and Oceanic Technology 20, 6; 10.1175/1520-0426(2003)020<0872:ADCOTR>2.0.CO;2
Automatic gain control counts as a function of depth showing returns from the bottom. One count is roughly 0.45 dB. These data are taken from the acoustic interference test, with unsynchronized pinging but with most of the resulting amplitude glitches filtered out. Prior to averaging, the OS profiles were shifted slightly in depth to make the maxima coincide and were also shifted laterally to have the same maximum AGC count as the average NB profile. Horizontal bars show plus/minus one standard deviation. Sidelobe returns from the bottom are expected to begin at 85% of the water depth (thick gray line). Just below this line, the OS amplitude increases sharply to a plateau, then increases again to the maximum AGC. The first sidelobe increase is weaker in the NB. Below the AGC maximum, the NB amplitude falls to the noise floor more steeply than the OS amplitude does
Citation: Journal of Atmospheric and Oceanic Technology 20, 6; 10.1175/1520-0426(2003)020<0872:ADCOTR>2.0.CO;2
Percent good and std dev of error velocity from 2 h of data collected during good weather with typical configurations of the three virtual instruments. NB8, OSB8, and NBN16 refer to the narrowband instrument with 8-m bin and pulse, OS in BB mode with 8-m bin and pulse, and OS in NB mode with 16-m bin and pulse, respectively
Citation: Journal of Atmospheric and Oceanic Technology 20, 6; 10.1175/1520-0426(2003)020<0872:ADCOTR>2.0.CO;2
Comparison of (left) velocity profiles and (right) percent good pings from 2-h intervals of (top) good weather (decimal days 51.64–51.74) and (bottom) rough weather (decimal days 52.24–52.34); (middle) the velocity differences, with the unedited and edited versions plotted on separate axes within the same panel. The rightmost three panels represent (from left to right) the percentage of good pings for each bin in the unedited and edited datasets, and percentage of pings removed by editing. The velocity profiles were averaged from matched sets of single pings divided into 5-min ensembles, using a reference layer averaging algorithm as described in the text and appendix B. Only the forward component of the velocity relative to the ship is shown. NBB and NBN denote NB-150 pings synchronized with the OSB and OSN pings, respectively
Citation: Journal of Atmospheric and Oceanic Technology 20, 6; 10.1175/1520-0426(2003)020<0872:ADCOTR>2.0.CO;2
Histograms of ping-by-ping reference layer velocity differences, NB minus OS, stratified by the percentage of good bins within the 50–150-m reference layer. The two columns show OSB panels on the left, OSN on the right; within each column, comparisons between unedited and edited data are on the left and right, respectively. (top) The cases with high NB percent good; red curves show cases with high OS percent good. Only the forward velocity component of the water relative to the ship is shown, so negative velocity differences represent a bias of the OS toward zero or of the NB toward large negative values. Note that the histograms use a log scale on their y axes to emphasize the outliers. NBB and NBN denote NB-150 pings synchronized with the OSB and OSN pings, respectively
Citation: Journal of Atmospheric and Oceanic Technology 20, 6; 10.1175/1520-0426(2003)020<0872:ADCOTR>2.0.CO;2
Editing parameters for virtual instrument pairs [(left) OSB and NBB, (right) OSN and NBN] are plotted as a function of forward velocity difference. Data shown are the ping-matched, depth-matched subset of the same bad-weather segment used in Fig. 5. Error velocity (top four panels) is plotted for each instrument with a threshold of 0.5 m s−1 shown as a gray line. After removing bins that fail the error velocity threshold, (left) shallow OSB correlation and (right) OSN S/N are shown in the third panels (again plotted as a function of forward bin-matched velocity difference); deep OSB correlation and OSN S/N are shown in the bottom two panels. Gray lines indicate the thresholds used for correlation and S/N editing
Citation: Journal of Atmospheric and Oceanic Technology 20, 6; 10.1175/1520-0426(2003)020<0872:ADCOTR>2.0.CO;2
Number of good bins remaining in the reference layer for each profile after applying the error velocity and correlation (or S/N) thresholds, plotted against forward reference layer velocity difference for (left) NBB and OSB and (right) NBN and OSN. Data are the ping-matched bad weather data from Fig. 5. For each profile, the same velocity difference is plotted twice, first (top) with NB layer percent good on the y axis, second (bottom) with OS layer percent good on the y axis. Circled areas show the velocity bias toward zero associated with bad profiles, which have only a few bins in the reference layer. Because the forward velocity of the water relative to the ship is negative, the bias toward zero shows up in the NB-OS difference (top) as a positive velocity when the NB pings are bad, and (bottom) as a negative velocity when the OS pings are bad
Citation: Journal of Atmospheric and Oceanic Technology 20, 6; 10.1175/1520-0426(2003)020<0872:ADCOTR>2.0.CO;2
Histograms of reference layer velocity deviations from their 5-min ensemble means. In a small fraction of pings, the NB profile was deemed good but the corresponding OS profile was rejected. The histograms of NB reference layer velocity deviations in these cases (heavy lines) are centered close to zero: the means are 0.13 m s−1 for good NBB pings coinciding with bad OSB pings, and 0.08 m s−1 for good NBN pings coinciding with bad OSN pings. Hence, there is at most a weak correlation between ping rejection and short-term velocity fluctuations of the ship
Citation: Journal of Atmospheric and Oceanic Technology 20, 6; 10.1175/1520-0426(2003)020<0872:ADCOTR>2.0.CO;2
The northbound section crossing the Gulf Stream illustrates the (left) NB-150 and the range advantage of the OS-75 [(center) broadband mode and (right) narrowband mode]. The zonal velocity component is plotted with a percent good threshold of 20%. Weather was moderately rough during the crossing and improved beyond the north wall of the Gulf Stream
Citation: Journal of Atmospheric and Oceanic Technology 20, 6; 10.1175/1520-0426(2003)020<0872:ADCOTR>2.0.CO;2
Data identifiers
School of Ocean and Earth Science and Technology Contribution Number 6119.
