1. Introduction

The Tropical Atmosphere Ocean (TAO) array consists of nearly 70 wind and thermistor chain moorings (Fig. 1) that span the tropical Pacific from 8°N to 8°S and from 95°W to 137°E (Hayes et al. 1991; McPhaden 1993). The purpose of the TAO array is to make high-quality surface wind and temperature data available in real time to support short-term climate studies, most notably those focusing on the El Niño–Southern Oscillation (ENSO) phenomenon. In addition, near-surface currents are measured at five or six locations within the array, either by subsurface acoustic Doppler current profilers (ADCPs) or by PROTEUS (profile telemetry of upper ocean currents) moorings (McPhaden et al. 1990). PROTEUS moorings have been deployed on the equator in the eastern equatorial Pacific since 1990 as part of the Tropical Ocean Global Atmosphere (TOGA) Program and NOAA’s Equatorial Pacific Ocean Climate Studies (EPOCS) Program. These moorings (Fig. 2), equipped with telemetering RD Instruments (RDI) ADCPs, were developed as part of the TAO array to measure and transmit real-time ocean currents within 250 m of the surface.

The RDI ADCP is a four-beam system that transmits a high-frequency acoustic pulse, measures the Doppler shift in the backscattered acoustic energy as a function of time, and computes the beam-direction velocity component as a function of range. Using compass measurements, the range-gated beam velocities are then converted into vertical profiles of zonal and meridional velocity. The frequency shift measured by the ADCP is caused by the relative motion of scatterers, whose movement is assumed to be due, on the average, to oceanic advection. In most cases, the assumption of scatterers advected passively by water motion is valid. However, the presence of fish or other sound scatterers, whose mean movement is not due to oceanic advection, will bias the velocity measurement (Freitag et al. 1992; Wilson and Firing 1992). Pelagic fish are at times attracted to the vicinity of the PROTEUS moorings, resulting in large errors in the ADCP velocity measurement. Fortunately, concurrent mechanical current meter (MCM) data are available to allow evaluation and correction of the ADCP velocity errors.

2. Instrumentation

The first PROTEUS mooring was deployed in April 1990 in a stand-alone ADCP mode near 0°, 140°W. A mooring equipped with seven MCMs was deployed 17 km to the east. After these moorings were recovered in October 1990, the standard PROTEUS deployment configuration contained both the ADCP and the MCMs on the same mooring (Fig. 2). An array of PROTEUS moorings was created with the addition of moorings on the equator at 110°W, 165°E, and 156°E. The moorings were recovered and redeployed on a 5–7-month schedule.

The downward-looking 153.6-kHz ADCPs were set to collect data with 8-m bin and pulse lengths, at a 1-s sample rate for 6 min once per hour. The ADCP velocities were collected assuming a surface sound speed of 1536 m s⁻¹. Hourly surface sound speeds computed from in situ temperatures measured at the transducer head and historically averaged surface salinities were used to correct the ADCP velocities (Gordon 1996). Nominal ADCP depths, which assume a constant sound speed with depth of 1475.1 cm s⁻¹, were adjusted using historical profiles of sound speed. Early deployments were set to use a narrow low-pass filter bandwidth (300 Hz) for the entire depth profile, resulting in some skew error (Pullen et al. 1992). Beginning in fall 1991, a broad bandwidth filter (600 Hz) was used in the shallower portion of the depth profile, switching to a narrow bandwidth filter at a depth below the core of the undercurrent. The ADCPs were equipped with a KVH compass calibrated to an accuracy of ±2.5°.

The MCMs were placed in the mooring lines at six or seven depths between 3 and 300 m (Fig. 2). The MCMs were either EG&G vector-averaging current meters (VACMs) or vector-measuring current meters (VMCMs). In highly variable flows, VACMs have been found to overestimate velocity (Karweit 1974; Beardsley 1987) and VMCMs have been found to underestimate velocity by a few percent (Weller and Davis 1980). Halpern (1987) found good agreement in comparisons of VACM–VMCM pairs separated by 1 m on taut-line equatorial moorings. The largest rms differences were at 13–14 m, equal to 7.4 cm s⁻¹ or about 10% of the mean flow, with smaller differences found at deeper depths. The VMCMs use a fluxgate compass, similar to the one used in the ADCP, calibrated to an accuracy of ±2.5°. During predeployment checkout, the VACM mechanical compass linearity (compass error relative to a chosen fixed direction) was confirmed to be ±5.6° or less. In some deployments, the MCM at a specific depth failed before recovery. Most of the MCM data gaps were filled using the techniques described in Plimpton et al. (1995), which most frequently involved linear regression fits from neighboring depths.

3. Velocity comparison

Contour plots of zonal velocity from the MCMs and from the ADCP appear quite similar, as shown for the first three deployments at 140°W in Fig. 3. The westward-flowing South Equatorial Current and the eastward-flowing Equatorial Undercurrent are clearly resolved in both datasets. The measurement methods are different, however, with the MCMs measuring velocity at six or seven specific depths versus the ADCP measurement that represents a weighted average over 16 m, measured at 8-m-depth intervals. The advantages of the finer vertical resolution ADCP data are evident in the temporal variability of the currents. The maxima and minima are stronger in the ADCP data and the annual variability in core depth of the Equatorial Undercurrent is more clearly defined in the higher-resolution ADCP data.

More significant velocity differences are evident, however, in the daily averaged time series of the ADCP and MCM data. The solid lines in Figs. 4 and 5 show ADCP–MCM daily speed differences at 110° and 140°W, respectively. Deployment and recovery times are shown by the black dots at the abscissas. Data are shown at the depths of the MCMs using interpolated ADCP speeds for the comparison. The ADCP speed data at 10-m depth have been estimated by linear extrapolation based on vertical gradients between bin 1 at 14 m and bin 2 at 22 m. For deployments PR04 and PR08 at 110°W, speed differences as large as 80 cm s⁻¹ are evident at 10 m, with decreasing differences at 25, 45, and 80 m. Speed differences at 80 m and above are somewhat smaller for deployments PR06 and PR12. Speed differences appear negligible at 120 and 200 m for all the 110°W deployments. For most deployments at 140°W, speed differences appear large at the first four MCM depths, decrease slightly at 120 m, and become negligible at 200 m. These speed differences are due to ADCP bias errors caused by reflection of acoustic energy from fish schooling near the moorings.

Figure 5a also indicates that for the first deployment at 140°W (PR01), in which the ADCP and MCMs were on separate moorings, velocity differences were consistently small. To test whether fish attraction was significantly reduced in stand-alone ADCP moorings, two PROTEUS moorings were deployed within 15 km of each other in April 1993, one (PR11) in the standard ADCP–MCM configuration and the second (PR11a) as a stand-alone ADCP mooring with no instrumentation on the mooring line. As shown in Fig. 5b, both deployments exhibited significant fish bias error. Therefore, the minimal fish bias in deployment PR01 was either serendipitous or related in some unknown way to environmental factors such as temperature and biological productivity levels.

In most of the deployment periods at 110° and 140°W, significant velocity errors did not appear until 1–5 months after the mooring was deployed, suggesting that a certain time may be required in which to develop a biosystem attractive to fish. At this time the authors do not know what factors determine the arrival time and abundance of fish in the vicinity of the moorings. However, investigation of speed difference time series farther to the west at 165° and 156°E (Fig. 1) indicated that fish-bias errors were negligible in the western Pacific for most deployments (Freitag et al. 1993). This suggests a correlation between fish bias and overall levels of biological productivity which, in the equatorial Pacific, are related to the mean depth of the thermocline and nutracline.

The negative sign of the speed differences in Figs. 4 and 5 indicates that the ADCP speed measurement is low compared with the MCM speed. The bias toward low ADCP velocity measurements in the presence of fish can be caused in more than one way. For example, fish swimming randomly with zero net velocity around a mooring in the presence of a nonzero current would bias horizontal ADCP speeds low. In this case, the fish generate a signal in one or more beams that is similar in strength to that of the surrounding water, and the velocities of the fish are averaged with the currents. Alternately, bias toward zero speed would occur if fish reflections were sensed in the main lobe of one beam and were strong enough to be sensed in the side lobes of the other three beams. If the signal were much stronger than that from the surrounding water, the frequencies in opposing beams would be nearly identical and would tend to cancel in the horizontal velocity calculation.

The presence of fish in the acoustic signal will tend to increase the ADCP echo intensity, which is a function of the intensity of the backscattered signal. In the self-contained RDI system, a range-gated echo intensity level is recorded for each of the four ADCP beams. Figures 4 and 5 show that times of large horizontal speed differences were highly coincident with times of large differences in beam-to-beam echo intensity. These differences, or echo intensity range, were computed by differencing the maximum and minimum of the four beam intensities.

Diurnal signals of large speed differences are also highly correlated with times of large echo intensity range. Figure 6 shows hourly data at 0°, 110°W from 27 September to 4 October 1991 at 45-m depth. Local nighttime hours from 1800 to 0600 h are shaded and exhibit minimum fish bias signals in both the speed differences and echo intensity range. During local daytime, large differences in beam-to-beam echo intensity are coincident with large ADCP–MCM horizontal speed differences. The nighttime fish bias minimum is consistent with earlier results (Holland et al. 1990), indicating that tuna tend to leave the vicinity of fish-aggregating devices (FADs) at night.

4. Fish rejection algorithm

The hourly ensembles recorded by the moored ADCPs were averages of 360 pings taken over a 6-min interval. Thus, individual pings that were biased by acoustic reflections from fish could not be identified in post processing. To eliminate the bias problem, fish-affected pings would have to be identified and rejected during data acquisition before ensemble averaging. To investigate the fish bias signals on an individual ping basis, a moored ADCP at 0°, 110°W was set to record single ping data for 6 h on 27 October 1991 at the end of deployment PR04. The single-ping recording began at 0945 local time in an effort to begin at a probable time of maximum fish bias (Fig. 6). Unfortunately, as shown in Fig. 4, the large speed differences that existed in the later part of the PR04 deployment tended to lessen at the end of October 1991. Thus, at the time of the single-ping experiment, the fish bias was relatively small, with the average 6-h speed difference equal to 11 cm s⁻¹ at 25 m and 13 cm s⁻¹ at 45 m. For the same time of day in mid-October, the 6-h speed difference was greater than 50 cm s⁻¹ for both depths. Even so, the results of the experiment were useful in evaluating the fish bias problem.

Of particular interest in the single-ping data was the distribution of echo intensity range (EIR) to determine whether these values could be used to flag fish-affected pings. Percentages of EIR values are shown in Fig. 7 for the five depths nearest the MCM depths (note that 134 m was the deepest single-ping depth recorded). The depth distribution of the EIR values was similar to that seen in the averaged data, with larger magnitude EIR values at shallower depths. For the greatest depths, 118 and 134 m, 87% of EIR values were typically 15 dB or less, and rarely exceeded 20 dB. Slight evidence of fish occurred at 78 m, with 82% of EIR at 15 dB or less, and a few values greater than 40 dB. Conversely, at 14, 22, and 46 m, the percent of EIR that was 15 dB or less was much smaller: 58%, 43%, and 51%, respectively. In addition, EIR values at all three shallow depths at times exceeded 50 dB.

EIR, computed as the difference between the maximum and minimum of the beam echo intensities, was further evaluated by computing the percentage of echo intensity values for each of the four beams. The distribution of intensities at 46 m reveals that most of the large EIR in the single-ping data is the result of large echo intensity in beam 1 or beam 4 (Fig. 8). ADCP compass values indicate that the buoy orientation was fairly stable during the 6-h experiment (Fig. 9). The mean current was to the northeast, such that beams 1 and 4 were generally on the upstream side of the buoy. Thus, most of the fish-biased measurements occurred in the upstream beams, which is consistent with findings by Holland et al. (1990) that tuna tend to stay upstream of FADs.

An algorithm was created to test how well the EIR values could be used to identify and reject fish-affected pings before ensemble averaging. If the EIR value at a given bin exceeded a set value, then the velocity data were not used in the average. The velocity data were also rejected if the EIR at the next shallower bin was large, since echo intensity is sampled near the end of a bin. The remaining single-ping values were then averaged into 15-min velocity averages at each bin for comparison to 15-min MCM averages. If fewer than 25% of the values in a 15-min ensemble were good, then the average at that bin was flagged as bad. Speed differences between ensemble-averaged ADCP and MCM time series at 25, 45, and 120 m were computed every 15 min and then averaged over the 6 h of data. Comparisons were not made at 10 m, due to the uncertainty of extrapolation, nor at 80 m, where the MCM had failed. The percentage of good ensemble averages (referred to as the percent good in the following) was also computed.

Two factors were investigated when evaluating the algorithm: the change in the ADCP–MCM speed differences and the amount of data rejected. The maximum allowable EIR (MIR) was tested at 15, 20, and 25 dB. The speed differences were significantly reduced at the shallow depths for all three values tested (Fig. 10). Speed differences were most improved at 46 m, dropping from 13 cm s⁻¹ to near zero when the maximum range was set to 15 dB. Although the 15-dB value produced the greatest improvement in speed differences, the percent good in the upper 80 m did not exceed 60% and at some depths was as low as 30%. In addition, at deeper depths, where there is little evidence of returns from fish, the percent good remained less than 80%. This result, along with the distribution of EIR (Fig. 7), indicates that 15 dB may be too low a level for the maximum range value. When set at 20 dB, percent good exceeds 95% below 90 m, and data rejection percentages are more acceptable at the shallow depths. For a maximum range of 20 dB, speed differences are 5 cm s⁻¹ or less, which is within the amount expected due to MCM rotor speed errors and other instrument differences.

These results prompted RD Instruments, the manufacturer of the ADCP, to include an algorithm in their instrument firmware that checked the EIR for every ping of an ensemble. The algorithm is included in all firmware versions 17.09 and later with the maximum range value set by the CF command. The algorithm uses successive steps to ensure that it will not flag all data bad in the event that one beam fails, thus exhibiting a low echo intensity for all bins. First, if the EIR of a bin is greater than the set maximum range value, the lowest of the four-beam echo intensities is rejected and the EIR recomputed. Then, if the EIR of the three remaining beams is greater than the maximum allowable range, the ping is flagged as bad for that bin. Since echo intensity is sampled near the end of a bin, the next bin deeper is also flagged bad. This algorithm would work properly when elevated echo intensity is present in one or two beams. Unfortunately, in the event that echo intensity is large in three or four beams, the data may be interpreted as good.

PROTEUS moorings deployed after March 1992 had this algorithm included in the ADCP software with the CF command (maximum EIR) set for 20 dB. A decrease in the percentage of good pings per ensemble average was evident in regions of high EIR, indicating that some rejection of fish biased data was occurring. However, as seen in the ADCP–MCM speed curves in Figs. 4 and 5, ADCP data after March 1992 still had large horizontal velocity errors coincident with times of high EIR. The failure of the algorithm to adequately remove the biased data is due, in part, to the difficulty in determining the optimal EIR threshold for all circumstances encountered in a given deployment. More significantly, the assumption that, in general, high echo intensity would occur in only one or two beams did not appear valid. In many later deployments, elevated echo intensity often occurred in three or four beams.

5. ADCP data correction

Because the ADCP time series at 140° and 110°W contained large fish bias errors in horizontal velocity, both before and after the implementation of the fish bias algorithm, a method has been developed to remove the ADCP fish bias utilizing MCM data by computing empirical orthogonal functions (EOFs) of demeaned ADCP–MCM speed differences at the depths of the MCMs. As shown for PR02, mean ADCP and MCM speeds (Fig. 11a) and mean ADCP–MCM speed differences (Fig. 11b) were computed for each deployment. These mean speed differences were subtracted from the ADCP–MCM time series before computing the EOFs. The first three eigenmodes for each deployment were then computed as shown for PR02 in Fig. 12. For the mean speed differences and each eigenmode, a spline fit was performed on the differences at the MCM depths to create values at all intervening ADCP depths. For all deployments, the fish bias appeared negligible below 200 m, based on the weak variability observed in ADCP–MCM differences and EIR at 200–250 m. For this reason, and because the VACMs at 200 and 250 m may tend to overspeed by a few centimeters per second (Halpern 1987), the spline fits, for both the mean differences and individual eigenvectors, were forced to zero at 200 m.

Only the first three eigenmodes were used in this correction scheme because the higher modes tended to be noisy and the differences between the ADCP and the MCM speeds were not all due to the presence of fish. For example, the higher modes were most likely capturing more subtle forms of instrument error in the MCMs and/or the ADCPs. Table 1 lists the percent variance corrected for each of the first three vertical modes for each deployment. For most deployments, 90%–97% of the variance was explained by these three modes. For deployments only minimally affected by fish, as indicated in both the EIR values and speed differences in Figs. 4 and 5, percent variance explained by the first three EOFs was lower, for example, 83.1% for PR01 and 84.5% for PR12.

For each deployment, the mean and first three vertical modes of the ADCP–MCM speed differences were combined to produce a time series of correction profiles for the daily ADCP speeds. When the corrected ADCP speeds were compared with the MCM speeds, there was considerable reduction in the standard deviation of the ADCP–MCM speed differences for each deployment, as illustrated in Fig. 11 for PR02.

Corrected ADCP speeds were also compared for PR11 and PR11a, which were simultaneously deployed 15 km apart (Plimpton et al. 1995). Differences between the corrected ADCP speeds were generally small, with extreme values for any depth bin ranging from −13 to 16 cm s⁻¹. The deployment length mean differences in corrected ADCP records for any depth bin were less than ±1 cm s⁻¹. For the upper 80 m, where the fish bias errors were largest, the average corrected speed difference standard deviation was equal to 2.8 cm s⁻¹. Comparison of these two ADCP records provides a measure of the consistency of our correction scheme. However, absolute accuracy of the corrected ADCP data will be limited by the accuracy of the MCM data used in the analysis. Hence, we would suggest an overall accuracy of corrected ADCP speeds in regions of significant fish bias of about 5 cm s⁻¹.

The ADCP and the MCM directions were compared and a few obvious compass errors were corrected (Plimpton et al. 1995). However, the ADCP directions did not appear affected by the presence of fish. Thus, corrected ADCP speeds and ADCP directions were used to compute daily profiles of zonal and meridional velocity at both mooring locations. The time series of these profiles provides a significantly improved dataset, with finer vertical resolution and fewer data gaps than the MCM data, and greater accuracy than the uncorrected ADCP data. Contour plots of the full corrected daily averaged ADCP data for 1991–93 at 110°W and 1990–93 at 140°W are shown in Figs. 13 and 14.

6. Conclusions

Large speed errors were evident in the surface moored RDI ADCP data from equatorial moorings at 110° and 140°W due to the reflection of acoustic energy from pelagic fish attracted to the vicinity of the moorings. A test was run to determine if fish-biased data could be detected and rejected on an individual ping basis during data acquisition. An algorithm was then added to the ADCP instrument software but was unsuccessful in eliminating all the fish bias errors in later deployments. Apparently the algorithm’s assumption that fish would generally be detected in only one or two beams, as was the case in the single-ping test, was invalid for most of the PROTEUS deployment periods. A method was developed for removing the bias in post processing based on an EOF analysis of ADCP–MCM speed differences. The corrected speeds were combined with the ADCP directions to produce daily profiles of zonal and meridional velocity.

Efforts to collect ADCP data free of fish bias errors on surface-moored buoys in the eastern equatorial Pacific at 0°, 110°W and 0°, 140°W have been unsuccessful. A more recent version of the RDI ADCP, referred to as a broadband ADCP, which transmits a series of coherent pulses, would experience similar fish bias errors (RD Instruments 1995, personal communication). Large reflections from fish would produce errors in the pulse-to-pulse correlation computed along each beam.

The authors have evaluated subsurface equatorial ADCP data at 140°W from the TIWE experiment (Weisberg et al. 1991) and found no evidence of fish bias. Thus, beginning in fall 1995, the equatorial ADCPs at 110° and 140°W have been deployed on subsurface moorings. Separate surface moorings with a few MCMs continue to be deployed for backup and for continuity of the 10–15-yr-long time series at these locations in the event of ADCP failure for any reason.

The data indicated that attraction of fish to the vicinity of surface moorings in the western Pacific was far less than found in the eastern Pacific. For most deployments at 156° and 165°E, fish bias in the ADCP data was negligible. Thus, fish bias appears to be high in regions of a shallower thermocline and high biological activity. Although the fish bias problems that exist at 110° and 140°W are not universal for surface moorings, care should be exercised in interpreting ADCP data in regions of high biological productivity.