![cruise track for 4 comparison cruises cruise track for 4 comparison cruises](figs/cruisetracks.png)
University of Hawaii ADCP data acquisition and processing software "UHDAS" has been installed on the Revelle since November 2004, collecting, processing, and distributing ocean velocities derived from the R.D.Instruments 150kHz Narrowband ADCP. UHDAS replaces the DOS-based system provided by the manufacturer and allows for storage and manipulation of single-ping data. A comparison between the NB150 and HDSS sonars (50kHz and 140kHz) was attempted during a CLIVAR cruise in January/February 2005, but the HDSS sonars were not functioning at full capacity.
Recent replacement of two transducers on the HDSS 50kHz system and repair of one beam (wiring?) in the HDSS 140kHz instrument as well as partial upgrades to the electronics of both HDSS systems now allow for a better comparison. The HDSS sonars were designed to maximize vertical resolution for measuring small-scale shear. With the exception of the anomalous shear in the 50kHz instrument caused by the presence of scattering layers, however, shear will not be discussed in this report. Instead, the emphasis is on measurement of water velocity relative to the earth.
Comparisons were made using 5-minute averages from all three instruments. Processing steps included single-ping editing prior to averaging, heading correction (from gyro to Ashtech heading), and scale factor and misalignment calibrations. Averages were then edited for egregious outliers. Comparisons revealed biases of some concern.
At high speeds the 50kHz instrument has very large biases in the forward direction in the presence of strong scattering layers. These biases are greater in the 50kHz (e.g. 25cm/s) instrument than in the NB150 (e.g. 5cm/s) where the profile ranges overlap. Below the NB150 range, strong scattering layers can make even larger biases in the 50kHz (40cm/s). The effect at lower speeds or with weaker scattering layers is not quantified. This source of bias in the 140kHz HDSS also has not been quantified because strong scattering layers were not found within the range of that instrument.
A troublesome characteristic shared with RDI instruments in many installations is the tendency for profiles in rough weather to contain significant along-track bias due to noise and/or bubbles. All 3 instruments on the Revelle exhibit this bias. The HDSS 140kHz instrument also seems to acquire a small bias in shear (few cm/s over its range) when underway and conditions deteriorate.
At low speeds, both HDSS instruments show small biases (1-2cm/s) relative to the NB150 and relative to each other. The cause is unknown.
Any scientist wishing to use the HDSS sonars must decide how much error they are willing to tolerate in ocean velocities, and how much work they are willing to do to obtain the final data quality for their purpose. In addition to the problems noted above, the HDSS data acquisition system and files have some bugs and annoyances.
changes in HDSS installation
HDSS 50kHz: new transducers in January 2006
HDSS new electronics of some kind
some changes in processing software
observations:
high-speed biases under adverse conditions
— NB150, HDSS 140, and 50: obvious fwd bias before editing
— HDSS 140, and 50: subtle fwd bias after editing compared to NB150
— HDSS 140: bias in fwd shear after editing while underway
effect of scattering layers:
— NB150 has inverse `S'' pattern in forward ocean velocity
— HDSS 50 has a lobe in forward ocean velocity
— effect on NB150 velocity is less than HDSS 50
low-speed biases
— HDSS 140: 2cm/s bias at low speed
— HDSS 50: 1cm/s bias at low speed
— cause and speed-dependence are unknown
Single-ping data were collected from late January to late June, 2006 (see table below). HDSS 140kHz data from the first legs (AMAT02RR and AMAT03RR) were disregarded because the instrument was improved during AMAT04RR.
Cruise | Dates | NB150 | HDSS 140 | HDSS 50 | days |
---|---|---|---|---|---|
AMAT02RR | Jan 25 - Apr 05 | yes | NA | yes | 32 |
AMAT03RR | Apr 7 - Apr 13 | yes | NA | yes | 6 |
AMAT04RR | Apr 19 - May 23 | yes | yes | yes | 35 |
AMAT06RR | June 3 - June 18 | yes | yes | yes | 15 |
Data from all three instruments were averaged in 5-minute groups in earth coordinates, and processed with University of Hawaii's open source CODAS ADCP processing code. This processing consists of the following:
select group of raw (single-ping) ADCP beam-velocity profiles
use beam orientations to convert velocities to ship coordinates
use gyro heading to transform velocities to earth coordinates
edit out bad velocities (from bubbles, underway bias, etc)
average in time (yields a profile of average measured velocity)
write the following information to disk, for subsequent loading into CODAS database:
average measured horizontal velocity (E and N)
vertical and error velocities
average signal strength
percent good (from single-ping editing of profile)
temperature, position, time, heading (end of profile)
load the information into the database
edit all averaged profiles, screening for underway bias, anomalous offsets
correct gyro heading to Ashtech heading (more accurate)
determine remaining contant angle and scale factor remaining ("watertrack calibration")
rotate the measured velocities in the database to account for the constant angle found
multiply all measured velocities by the required scale factor
use a slightly smoothed reference layer velocity estimate based on GPS positions to generated absolute ocean velocity profiles
If the ship's acceleration exceeds a specified threshold, the data near that time of acceleration is used for a calibration calculation. This calculation is referred to in CODAS processing as "watertrack" calibration, to distinguish it from "bottom track" or "reciprocal track" calibrations.
Two common sources of error in ADCP ocean velocity calculations are the total angle used to transform beam coordinates to earth coodinates, and the extent to which the measured velocities are consistently too small or too large. These errors are nearly independent, with the angle error primarily affecting cross-track velocity component and the scale factor affecting along-track component.
Both angle and scale factor errors are proportional to ship speed; the table below assumes a ship's speed of 6 m/s. Even small errors in angle or scale factor make substantial errors in the velocity estimate, compared to the typical open ocean velocities (15-40 cm/s). The errors are particularly bad for transport estimates because they are independent of depth and persistent in time—they don't average out.
error type | error direction | ocean velocity error |
---|---|---|
angle, 1 deg | cross-track | 10 cm/s |
scale, 2% | along-track | 12 cm/s |
The following figure illustrates the way in which angle and scale factor errors result in cross-track and along-track errors, respectively.
Watertrack calibration assumes that the ocean velocity is the same no matter what the ship is doing. The approach is to determine what rotation (near zero) and scale factor (near one) are necessary to minimize the difference in the ocean velocity determined before and after the acceleration. Individual estimates are noisy, but if the heading measurements and ADCP velocities are good then the noise is random, and it averages out when many estimates can be made.
Watertrack calibration opportunities usually come during bathymetric cruises, when the ship turns often, or cruises which involve station work (stopping for a station and starting again yields two opportunities for calibration).
Watertrack calibrations from AMAT04RR were obtained for all three instruments with the same reference layer, 75-200m. Results for the scale factor are:
Instrument | mean | median | stddev | num pts |
---|---|---|---|---|
HDSS 140kHz | 1.004 | 9 1.0030 | 0.0102 | 79 |
NB 150kHz | 0.999 | 0 0.9993 | 0.0069 | 87 |
HDSS 50kHz | 0.999 | 5 0.9992 | 0.0111 | 72 |
Scale factor should be very close to 1.0000 if the speed of sound at the transducer is accounted for, and if beam geometries are known. Watertrack calibrations indicate that none of the three instruments requires an anomalous scale factor.
The beams of Doppler sonars are designed to be narrow so as minimize returns from directions other than along the beam axis. Returns from very strong scatterers off the beam axis can still contaminate the signal, however. For example, the return from the ocean bottom, or from a strong scattering layer, at 300 m depth, which on average will have a Doppler shift of zero, will contaminate the along-axis return from 260-m depth, which will have a Doppler shift proportional to the ship's speed. Hence the downward-pointing sidelobe of the beams biases the velocity estimates towards zero.
The Revelle went over a seamount that was in range of the NB150 and HDSS 50kHz instruments. The signal returns (amplitude on a log scale) from both are plotted below. We are not sure what units the HDSS software provides; the values are scaled to have similar magnitudes to the NB150 for convenience.
The circled "bumps" below show the return from the downward-pointing sidelobes, and the overall shape of the returns from the bottom indicate the width of the beam patterns. Although the signal strength units differ, the shapes indicate that the HDSS 50kHz has surprisingly strong sidelobes.
This is confirmed by the velocity profiles shown below; the seamount contaminates the velocity estimates of both instruments starting at 85% of the water depth, but the artifact is much larger for the HDSS 50 than for the NB150.
A graphic example of this effect in the water is provided by the diurnal migration of various critters. Diurnal migrators swim down during the day to avoid predation, and swim back up at night to feed. These migrations are visible in the scattering return from both instruments. A 24-hour period of data from the HDSS 50kHz and NB150 instruments (where they overlap) is plotted below. The velocity artifact is obviously larger for the HDSS 50kHz than for the NB150. The gray lines in the same plots are profiles from nighttime, just before the migration. In the absence of strong scattering layers, the two instruments agree well underway in the overlap region.
The NB150 has much shorter range than the HDSS 50kHz, although they both have the same resolution. The following figure shows the same time period and the same variable (ocean velocity in the forward direction) for the HDSS 50kHz for its full range. The bump in velocity discussed above is centered around 250m (circled in blue). Below the range of the NB150, at 550m and 620m, are additional scattering layers with corresponding artifacts in HDSS 50kHz ocean velocity (in the forward direction). These are circled in red and black, below.
The scale factor correction based on watertrack calibration removes the velocity error proportional to ship's speed, but the watertrack calibration is insensitive to low-speed bias. To check for the latter, ocean velocity estimates in the along-track (forward) direction from the HDSS instruments and the NB150 were compared to each other as a function of ship's speed. To avoid biases caused by scattering layers, the comparison was made using a reference layer that is generally not affected by persistent scattering layers (100-150m).
Since ocean velocity is calculated as
u_ocean = u_ship + u_measured,
if the measured velocity (which is negative in the along-track direction)
is biased towards zero, this results in a
an ocean-velocity bias that is positive in the along-track direction.
These velocities come from calibrated datasets, and the calibrations are designed to account for speed-dependent angle and scale factor errors. Therefore, at high ship speeds, the differences are small. However, at low ship speeds, the NB150 ocean speeds exceed the HSSS 140kHz ocean speeds in the forward direction, indicating that the NB150 measured velocities are biased towards zero relative to the HDSS 140kHz. The opposite is true for the HDSS 50kHz instrument: its measured velocities are biased towards zero compared to the NB150.
The third panel compares the two HDSS instruments, in which there is a 2-3cm/s bias towards zero in the 50kHz instrument compared the the 140kHz instrument.
There are similar differences in the cross-track direction, so the bias does not seem to be simply a change in scale factor at low speeds. It is also unclear whether whatever is causing the low-speed bias operates only at low speeds; its effect at cruising speed, if any, would be indistinguishable from other factors in the watertrack calibration procedure.
The bias in the 50 kHz instrument has been present since 2003 and was seen in a comparison between the NB150 and the HDSS 50khz instrument during HOT cruise #149. The fact that the bias is largest in the comparison between the two HDSS instruments means that there is a real bias in at least one of those instruments—this is not just an NB150 problem.
All three instruments can have biases in the along-track direction when conditions get sufficiently rough. Bubbles that block the outgoing or incoming signal result in a shorter profiles, often biased towards zero beam velocity. This creates a bias towards zero in measured velocity in ship's coordinates in the along-track direction. Improved single-ping editing can help weed out the bad pings before averaging.
However, after single-ping editing and visual editing of averaged profiles, there were still along-track differences between HDSS and NB150 instruments. The following four figures show the differences between NB150 and HDSS instruments in a time series of 70-200m reference layer averages of ocean velocity in the ship's forward direction. The first figure is repeated 3 more times, with different errors annotated.
reference layer time series
This figure and the following annotated versions, have five lines:
- NB150 - HDSS 50kHz difference, cm/s (blue)
- NB150 - HDSS 140kHz difference, cm/s (red)
- ship speed, m/s (gray)
- Percent good (divided by 10) for HDSS 140kHz (green dots)
- Percent good (divided by 10) for HDSS 50kHz (black dots)
This first figure is repeated 3 more times, with different errors annotated.
low speed bias
Three annotated regions show the velocity differences hovering around 1-2cm/s for NB150-HDSS 140, and around -1-2cm/s for NB150 - HDSS 50. Conditions during acquisition were good as indicated by the fact that Percent Good (black and green dots) are around 100 (i.e. shown as values around 10, because Percent Good was scaled by 10).
good agreement underway, in good weather
Three annotated regions show the velocity differences closer to zero for underway periods compared to neighboring on-station periods. Conditions during acquisition were good as indicated by the fact that Percent Good (black and green dots) are around 90-100 (i.e. shown as values around 9-10, because PG was scaled by 10). It is common for Percent Good to decrease slightly when underway.
underway bias in bad wather
Three annotated regions show the velocity differences in which the HDSS ocean velocities were biased in the direction of travel, resulting in NB-HDSS being negative by 4-10cm/s. The weather during these periods was worse, as seen by the clouds of decending black and green dots.
The last figure illustrates that the HDSS 140 can have a larger shear in the forward direction compared to the HDSS 50, when any of the following conditions are met: underway, at the bottom of the profile, and as 140kHz profile quality (average Percent Good of the 70-200m reference layer) decreases. This same pattern is repeated when compared to the NB150 (not shown). There was no such difference between the NB150 and the HDSS 50kHz in their overlap region. However, because the HDSS 50kHz should have generally high quality data in the upper 200m but profiles much deeper than the NB150, we cannot look for shear bias in the 50kHz at the bottom of the profile similar to that found in the 140kHz.
Data are currently gathered on two Macintosh OS 9 machines, one per instrument. One newer OSX machine provides the interface between the OS 9 data-logging machines and the rest of the world. The OSX computer contains the Matlab code provided to read the single-ping beam-velocity data files that are saved to disk, the data files, some documentation, and a few other products (30-second averages stored as matlab files, and some figures).
One must specifically request single-ping data. Standard practice evidently results in smaller data files consisting of data averaged in beam coordinates. No attempt has been made to assess those files, but in general velocity estimates from any moving platform like a ship should be vector-averaged in geographical coordinates, not simply averaged in beam or platform coordinates.
Some improvements have been made to variable names and file conventions, but until the complete switch to Mac OS X for data acquisition, the present code will result in the following annoyances:
one must check for a vertical offset between instruments
ascii sorting of filenames does not match order of acquisition
filenames contain square brackets (bad for command-line users)
west longitudes are positive (are east longitudes negative??)
longitudes are parsed incorrectly (minutes recorded to only 3 decimal places vs/ 4 for latitude)
the reference year for acquisition dates is 2003
ashtech pitch and roll are recorded incorrectly (always positive)
The HDSS 50kHz instrument seems to have surprisingly fat beams, resulting in large localized biases in the presence of strong scattering layers.
The HDSS 140kHz instrument did not penetrate deep enough to run into the bottom or any strong scattering layers during the two comparison cruises. Therefore its beam characteristics and corresponding scattering layer artifacts could not be evaluated.
Both HDSS instruments and the NB150 are affected by biases in the forward direction under adverse conditions
The HDSS 140 shows a bias in forward shear underway under variably adverse conditions compared to both the NB150 and HDSS50. The HDSS 140 bias in the forward direction seems to be a continuum, not occuring only under extreme conditions. Improved single-ping editing may improve this aspect in averaged profiles.
There are unexplained biases of a few cm/s in both HDSS instruments at low speeds.