Data Quality of Nortek Vector ADV in the North East Channel of the Gulf of Maine May 2007. 29 June 2007 Analysis and Report Prepared by David S. Ciochetto Department of Oceanography Dalhousie University Halifax, N.S. Canada Abstract A Nortek Vector ADV was included in a benthic deployment on the continental shelf (173 m water depth) as an examination of the instrument’s performance in this environment. A previous deployment of this class of instrument in this environment was unknown to date to the author. Based on previous experience with the instrument, the performance was not as good as expected by the author. Data quality results are presented which characterize the deployment. At the time of this report, comparisons to the acoustically rich environment of Bedford Basin and particle characteristics are not available. The experiment was not exhaustive and the deployment settings were not optimal. Further investigation of the ADV operation in this environment is encouraged. Based on the data taken here, an ADV has a high probability (not quantified) of obtaining mean velocity measurements in this environment. The main source of particles was hypothesized to be sediment re-suspended from the bottom due to tides. The signal to noise ratio appeared to slightly lag the spring-neap tidal variation indicating a history effect in sediment suspension of a few tidal cycles. 1. Introduction In May 2007 a Nortek Vector Acoustic Doppler Velocimeter (ADV) was deployed on the RALPH frame among a suite of other instruments at a depth of 170 m. The deployment site was at (42.05313o, -65.96678o) in the North East channel of the Gulf of Maine (Fig. 1). The ADV in the experiment had probe serial number VEC 4527 and hardware serial number VEC 1072. Data was provided to the author for comment on the quality of the data. This report details that work. 2. Results Data were obtained from 13 May 2007 at 00:00 GMT to 29 May 2007 at 14:10 GMT. The ADV was set to sample in burst mode at 2 Hz for the first 10 minutes of every 2 hour period. During the time that the instrument was submerged, there were no error flags reported by the instrument. Data bursts during the experiment were acquired either at depth or on deck. Data reported in the sensor file (*.sen) were broken into the time on deck and the time of the experiment via the error code, status code, compass and tilt sensor data. Prior to 28 May 2007 the probe pointed down as indicated by the status code. The data files (*.dat) were analyzed as a continuous time series rather than as bursts. In this format, the pressure signal was used to 1 determine the time period that the instrument was on the bottom. Determination of the actual experiment times was simpler and more reliable when determined from the pressure signal. Matlab code was generated to read the important header info into a structure in the Matlab workspace and to manipulate Nortek format ADV burst data. Scripts were written to generate the time variable for the bursts based on the information provided in the header data. The data was analyzed as a continuous time series rather than through the selection of specific bursts. Matlab scripts to read the burst format ASCII data were started but to date have not been completed. All times reported in figures in the remainder of this report will be in elapsed year day relative to 1 May 2007 at midnight GMT (e.g. 12.25 indicates that 12 and ¼ days have gone by so the calendar date would be 13 May 2007 at 06:00:00.) The instrument settings included a nominal velocity range of 1 m s-1, a sampling volume (length) of 14.9 mm, high power level, ENU coordinates for the velocity data, sound speed calculated based on the measured temperature (T) from the ADV sensor and a fixed salinity (S) of 35.0 ppt. The implication of some of these settings will be discussed later. The sensor data (NEC70701.sen) are taken at 1 Hz and for a static deployment serve primarily as a quality check on the deployment. The Lithium-Ion battery pack held steady at 10.7 +/- 0.1 V throughout the experiment with an initial and final on deck value of 11.3 and 11.0 V respectively. The temperature and sound speed (Fig. 2) dropped over the period of the experiment. The sound speed data is only used to post process the velocity data to correct for a more accurate sound speed. Only ADV data existed for this analysis and the focus was not on the velocity but rather the data quality, thus Fig. 2 is presented for reference and comparison to other data from RALPH. The compass and tilt sensor data indicated significant variation prior to 13 May at 00:00:00 and two hours prior to the end time indicated by the error codes. This data established the time of the start of the experiment with respect to the sensor data. As mentioned above, it was simpler to determine the period of good data solely from the pressure sensor for this analysis. The compass and tilt sensor were stable on the bottom with a heading of 87.3o +/0.13o, pitch of 1.4o +/- 0.08o and a roll of -0.9o +/- 0.1o where the best estimates are the means of 112998 samples and the reported uncertainties are simply the standard deviations. 2.1 Background Noise The ADV measures the ambient noise during the first cycle of every burst. The instrument is powered up and activates its receivers for the first second to measure the background noise level before each burst. The noise amplitude [counts] and correlation [%] are summarized for the 188 noise samples in Fig. 3. The noise correlation distribution has a median for all three receiver beams near 4%. The noise amplitude is steady near 47.5 counts. The conversion from the instrument-referenced amplitude unit of counts to dB is x [counts] * 0.43 = x [dB]. The noise amplitude measured prior to each burst is the reference used to determine the SNR of each sample based on its amplitude SNRi = amplitudei − amplitude noise j . (1) No significant temporal trend was discerned in the noise data throughout the extent of this experiment. 2 2.2 Pressure signal and tides As the instrument was mounted on a stationary frame, the pressure signal will reflect the surface elevation of the ocean. The Matlab script sw_dpth.m (Morgan 1994) was used to convert the pressure signal in db to a depth in m. The data are shown in Fig. 4. To compare this to the surface tidal expression, Web Tide (Hannah and Chaffey 2007) was used to estimate the surface elevation and currents due to the tides at the location shown in Fig. 1, (42.05313o, -65.96678o). The pressure data, when compared to the prediction, illustrate that the major surface expression was due to the tide. Only qualitative comparison was accomplished at this stage due to the requirement to match the data prediction times to the pressure data or to average the pressure data to the relevant tidal prediction times. Neither approach was followed in the analysis for this report. The experiment captured the spring and neap variability as well as the semidiurnal variability in this location (Fig. 4a.). A detail view of two days of data is illustrated in Fig. 4b. Qualitatively, the Web Tide prediction and the pressure sensor data have a good match. Variability from the prediction will arise from atmospheric forcing such as swell, chop and local storm influences. Meteorological forcing data were not a focus of this analysis so that data were not sought or examined. The pressure sensor produced non-zero pressure readings on the deployment and recovery vessels. Prior to the experiment, the median of the pressure signal was 0.192 db. The post-deployment pressure median was 0.113 db. Combining the data from these times produces a bimodal histogram with peaks around 0.1 and 0.21 db with a median of 0.157 db. Since we are near the surface, this is an offset of about 16 cm. It will not be considered for the remainder of the analysis. The Nortek Vector ADV has a feature on deployment to zero the pressure. It is recommended to zero the pressure sensor at the start of the deployment to remove this offset. 2.3 Correlation Correlation is the primary measure of signal data quality in an ADV. Generally accepted levels are 70% for most applications. No theoretical predictions are available in the literature either for general or specific acoustic environments. The ADV measures the scattered sound from consecutive pulse pairs. After signal processing, the phase difference determines the velocity (Voulgaris and Trowbridge 1998) and the correlation is a measure of the similarity of the two pulses. The echoes are generated by acoustic differences between particles in the sample volume and the water in which they are suspended. Size and shape of the particle with respect to the operating frequency of the sound are also a factor in the strength and character of the sound energy returned to the instrument. Low correlation can indicate high levels of turbulence on the scale of the sample volume, high advection of particles through the sample volume, particles that are poor reflectors or a lack of particles. The correlation for this experiment is shown in Fig. 5. Note that Fig. 5 is a plan view of a two dimensional histogram where the intensity of the black color indicates the number of samples in that correlation-time bin. To enhance the contrast of the figure, the black upper limit was reduced to 70 samples per bin. There are a total of 676,800 samples represented in Fig. 5. By limiting the color scale to 70, only 11 bins are set to black compromising their real magnitude relative to the other bins. This represents 770 samples or 0.1 % of the data represented. As the point of the figure is to indicate the temporal trend in correlation with the fact that the large values are still represented, there is no compromise in the scientific integrity of the plot. The data at the start of the experiment have a mean correlation near 75%, which drops steadily toward the 15th of May. There is a significant rise in the correlation values from a mean of 3 approximately 68% near 15 May 00:00 GMT to an approximate mean of 85% near 17 May 12:00 GMT. The data that follow the 17th of May are mostly above the 70% correlation limit with a subjective mean of about 85%, which is relatively steady with time. Several spikes in correlation occur; the most significant one reaches a strong mean correlation near 93% on approximately 26 May 00:00 GMT. The scattered sound propagates along the path of the receiving transducers. When the data are converted into an orthogonal coordinate system, each beam contributes to each resulting velocity component. Thus, if the data is deemed poor in one beam, then that entire velocity sample must be deemed poor and discarded. Data with correlation less than 70% in any beam were removed. Figure 6 illustrates the percentage of data in a given burst that had correlation > 70% in all three beams and hence “good” data for the first pass of data quality control. Data is seen to be poor in the early part of the experiment prior to 15 May with “good” data per burst ranging from 10 – 30%. A dramatic rise in data quality begins near 14 May 12:00 GMT and concludes near the 16th of May near 12:00 GMT. On average, over 80% of the data taken after this rise pass the correlation filter. Spikes in the percentage of good data occur as seen in Fig. 6. 2.4 SNR The signal to noise ratio (SNR) is the relationship (Eq. 1) of the signal amplitude of each sample to the ambient acoustic noise measured at the start of each burst (Fig. 3). The SNR serves as a secondary data quality check. Loose recommendations by the manufacturers of ADVs state that once SNR > 15 the data are unquestionably good (given good correlation) and that SNR > 10 will give acceptable data; one manufacturer states that data with SNR as low as 5 may be used to estimate mean velocity values. As with the correlation, there have been no theoretical or experimental evidence presented in the literature to support these rule of thumb limits. The SNR gives an indication of the strength of the echoes relative to the local environment. High SNR indicates acoustically favorable particles in the sample volume or a sufficient number of scatterers of acoustic properties to generate a high return. Low SNR indicates either a lack of acoustically favorable particles to the instrument, instrument damage or no particles in the scattering volume at the desired sampling rate. The ADV was developed for use in a laboratory and, as such, when SNR is low, additional seeding is added to the flow until the signal strength is adequate. Field applications do not share this feature and are limited to working with available marine particles. The literature concerning the use of the ADV in acoustically unfriendly environments does not exist to date with the only submission coming close stating that the data quality was low but it did not halt the publication of the data (Zhang, Streitlien, Bellingham, and Baggeroer 2001). The ADV has been used successfully in the field for coastal benthic applications (e.g. Anderson and Lohrmann 1995: Barbhuiya and Dey 2004: Voulgaris and Meyers 2004). The temporal results for SNR are presented in Fig. 7. The limits of SNR of 10 and 5 discussed above are indicated. The data presented in Fig. 7 only include data with correlation > 70%. The general average SNR for the experiment is around 7.5. Near the beginning of the experiment, prior to 15 May, the SNR is generally low. The light colors indicate that much of the data did not pass the 70% correlation threshold as seen in Fig. 6. Spikes of higher SNR and more data (darker) visually correlate with the spikes in correlation seen in Figs. 5 and 6. Beginning near the 15th of May, as seen previously, the SNR increases to a level with an approximate mean of 10 until 23 May where it declines to an approximate mean of 7.5. The remainder of the experiment maintains this SNR with the exception of a pulse of strong data quality near 25 May 4 at 18:00. The mean and variability about the mean were calculated. They follow the darkest parts of Fig. 7. To maintain clarity of the figure, this information is not included. No limitations were placed on the color range in Fig. 7 as were applied to Fig. 5. 2.5 Velocity The velocity is the desired data product from an ADV. It is derived from the speed of sound in the water, the probe geometry and the phase difference between pulse pairs. After filtering the data for low correlation or SNR, the data must be checked for phase errors. With the Nortek Vector ADV, the setup software indicates the actual velocity range that it can detect based on the selected nominal velocity range. As the phase difference between pulses in pulsepairs determines the velocity of the particles in the sample volume, if a particle travels too fast, a velocity ambiguity may occur due to the inability to determine the proper phase. Any data that exceeds the limits is therefore known to be bad. It and the others in that velocity vector sample must be discarded during post-processing. A preliminary analysis of the ADV velocity data was performed. Barotropic tidal velocities were predicted by Web Tide. Figure 8 illustrates the comparison between the prediction and the measurements. The ADV velocities are significantly less than the Web Tide prediction, even in magnitude. The ADV did not measure a superposition of local velocity on the barotropic tidal velocity. The reduction in velocity is suspected to result from the presence of the bottom boundary layer. Only a point velocity record at the ADV sample volume exists for this analysis so this assumption can not be verified. Figure 8 indicates that the magnitude of the Northward velocity exceeds that of the Eastward velocity where Web Tide predicts similar magnitudes for these velocity components. The difference is assumed to be due to topographic and stratification effects not accounted for in the model. The amplitude of u in Fig. 8c indicates difficulty in determining phase differences between the ADV and the prediction. Figure 8d however indicates that the ADV data actually lead the prediction by approximately 0.05 days or 1.2 hours. This difference is assumed to be due to topographical or stratification effects. In Figs. 8b and 8d the velocity tends toward a Southward mean rather than zero mean. This is indication of a persistent flow out of the North East Channel or a geostrophic flow about George’s Bank. 3. Discussion 3.1 Comparison to expected instrument response Data taken by the author previously at station 2 of the Halifax Section of the Atlantic Zonal Monitoring Program in 155 m of water indicated better data quality than were observed in the current experiment (Ciochetto 2007). These previous data were taken on 3 Sep 2003 from a CTD cage with cast depths to 140 m. The site was located 30 km from the coast of Nova Scotia. Figure 2 of Ciochetto (2007) illustrates that SNR rose to 10 at a depth of 128 m and linearly increases from 12 to 16 (average of all three beams). Figure 2 (Ciochetto 2007) indicated that correlation in this depth range started near 87, quickly reached a constant 94 near 128 m and remained constant. The trial indicated in Ciochetto (2007) clearly predicted better data quality than was observed in the present experiment. There are several reasons why the data quality may have differed in the data in this report. The possibilities include a smaller sampling volume, different particle characteristics, different particle load in suspension, different ADV probe and different turbulence levels. 5 The setup of the instrument indicated that the sampling volume was set to 14.7 mm height where in the previous experiment, the sampling volume was set to 18.0 mm. The sampling volume is determined by the diameter of the transmit pulses and the geometry of the return paths (Voulgaris and Trowbridge 1998). The beams may be considered cylindrical in the acoustic nearfield (Borg 2002). Thus the sampling volume was 0.82 of the previous. No research in the literature indicates how an increased sampling volume of 22% impacts the data quality. Acoustic instruments rely on sound scattering from particles. There are many characteristics of the seawater and the particles that are important (Hay 1991). It is assumed that there are sufficient numbers of particles in the sampling volume such that their individual orientations are insignificant. It is also assumed that the particles are mainly re-suspended from the bottom. Figure 2 of Ciochetto (2007) supports this assumption in the fact that the data quality increases greatly as the bottom boundary layer is approached. This assumption will be investigated further in section 3.2. Future data massaging should include comparison of the tidal forcing as predicted by Web Tide for the data of Ciochetto (2007). A look at the OBS data should indicate the particle load for this experiment and can be compared to Ciochetto (2007). Another possible problem might be that too many particles are in the sampling volume and the echoes are too strong. This can be ruled out based on Fig. 7 and Fig. 9 (a summary of the data quality parameters which will be discussed in more detail in part 3 of this section). A lower correlation and a high SNR would support this hypothesis but the data clearly show that the correlation and SNR drop together. A 6 MHz acoustical instrument is most sensitive to particles with a diameter of 78 μm. Particle size spectra were determined for the data presented in Ciochetto (2007). The maximum particle diameter of a single particle in the sample volume at 140 m depth was measured to be 30 μm which increased from 20 μm at 100 m depth. The particles decreased in diameter with a power law. The particle load was estimated to be 230 +/- 22 ml-1. If a dominant particle generation mechanism exists, a Gaussian particle size spectrum will result. Without an estimate of the particle size spectra it is difficult to predict the particles near the bottom in the Gulf of Maine. The assumption will be made that the particle size spectra is a power law similar to the previous experiment. The most difficult acoustical qualities to access are the density and speed of sound contrast between the particles and the water in which they are suspended. This would require a sample of the suspected sound scattering material for every deployment and throughout time. This is impractical and technically difficult. For that reason, ADVs can not be calibrated to characterize the particles in suspension and rely on the rule of thumb levels set out above. The assumption is made here that the characteristics of the particles observed by the instrument are similar to those of Ciochetto (2007). High levels of turbulence in laboratory settings have been shown to cause de-correlation in ADV measurements (Lohrmann, Cabrera, and Kraus 1994). In these situations, the lateral motion of the particles and motion into and out of the sample volume in the time between pulses cause significant differences in the echoes received reducing the pulse-pair correlation. In this case, the correlation would be independent of the SNR. Comparing Figs. 5 and 7 and in the summary shown in Fig. 9, it is obvious that this is not the case. For this data, correlation is a strong function of SNR or signal strength. Thus high levels of turbulence are ruled out as a cause of poor data quality. Other data quality issues may be considered but are not included in this report as they are deemed secondary causes of loss of signal quality. 6 Without a measure of particles in the water column, the particle hypotheses presented here can not be investigated further. There is no evidence at this time to support differences in particle composition or number. In fact, as the measurements were taken closer to the bottom boundary, one would expect a greater particle load with particles of a greater density and compressibility contrast to water, i.e. more acoustically friendly. If the assumption is made that the particles and suspension due to forcing are similar to Ciochetto (2007), the remaining cause for difference in data quality is the smaller sampling volume. As the instrument employs various internal signal processing techniques when the data are taken at sampling rates < 64 Hz, only a lab experiment with similar particles can determine the difference. The final difference between the experiments is the fact that different probes were used. The probe used in Ciochetto (2007) was hand picked for applications in low particle environments by the owner of Nortek and co-developer of the ADV class of instrument. If we assume that manufacturing quality differences are small, then the only conclusion that can be made with regard to the differences between these two experiments was due to the difference in the sampling volume of 22%. 3.2 Temporal variability The temporal data represented in Figs. 5 – 7 present a mystery. Apart from the variability in each burst of data seen throughout the record or the strong spikes of high quality data, the most striking feature is the significant increase in data quality near 15 May 2007 in each figure. The RALPH frame was seated on the seabed and did not move perceptibly during the experiment. There are a few reasons why this pattern may have occurred. The observed change does not result from any instrument failure or settings change. The only remaining change can come from the suspended particles. The particles may have several sources from either the lateral or vertical directions. Lateral sources could include turbidity currents or advection of water masses. Vertical sources may include re-suspension due to bottom shear and vertical particle flux due to detritus, atmospheric dust or vertical migration of plankton. Lateral movement of particles would be accompanied by a change in the water mass properties. CTD data were not available for this analysis and characterization of the water mass was not attempted. The vertical sources include re-suspension from the bottom, particle ‘rain’ from detritus and Aeolian input or biological migration. The spring bloom was observed in the Northwest Arm of Halifax Harbor on 8 March and terrestrial sources caused an echo bloom on 3 April. The spring bloom on the Scotian Shelf typically consists of diatoms which have a silicious shell. Diatoms are on the order of 10 – 50 μm in diameter. They may provide adequate particles to an ADV. The characteristics of the spring bloom for this year over the NE Channel in the Gulf of Maine are unknown to the author. If the timing were similar to Halifax Harbor then by 15 May one may not expect to see particles at 178 m. The end of the spring bloom is mitigated by zooplankton grazing as well as loss of nutrients from consumption. The quantity of diatoms required to raise the SNR from 5 to 10 is unknown. If they are large diatoms, it is possible that one test per pulse-pair in the 2.8 x 10-6 m3 sample volume could produce good signal. Fallout from the spring bloom should show the opposite trend than that observed in Figs. 5 – 7, the bloom snow would be expected to arrive earlier at 178 m and be falling off rather than ramping up. Vertical migration by plankton of a size favorable to the ADV is rare. Zooplankton have effective spherical diameters > 100 μm and smaller marine particles don’t migrate that deep. Migrations are typically on a diel cycle contrary to the change observed in Figs. 5 – 7. The other 7 vertical source of particles is due to shear from tidal currents. Depending on the strength of the flow, various sizes of sediment are re-suspended. Comparison of Figs. 4a and 7 indicate that the spring-neap cycle of the tides at this location may be the main forcing of the SNR signal. The SNR starts low as the tidal elevations are increasing in magnitude with the spring cycle and then slowly decrease as the neap tide is approached. The SNR data appear to lag the tidal elevation spring-neap cycle by 2.5 days. The more energetic tides would produce shear that can re-suspend larger particles at the height of the ADV. Note that the correlation remains high throughout the later part of the data. It is hypothesized that a few smaller than optimal particles of sufficient acoustic properties will generate a good correlation with a low SNR. A competing hypothesis was the result of previous work by the author where high correlation and low amplitude data results from a cloud of small particles with acoustically undesirable properties scattering sound not unlike light scattering from fog. The present data are not able to test these hypotheses. Based on the comparison of Figs. 4a and 7, it is strongly suggested by this data that the tides are the strongest source of particles during the experiment through sediment re-suspension. The data record nearly encompasses one full spring-neap cycle. A stronger case for this conclusion could be made if the data record were at least 1.5 – 2 springneap cycles. Analysis of the velocity lends support to the above conclusions. A comparison of the ADV data that pass the > 70% correlation filter weakly support the tides as the main source for scatterers as illustrated in the joint histograms in Fig.s 10 and 11. SNR as a function of velocity was interrogated. It was discovered that no relationship existed between SNR and |uADV|, uADV (Eastward velocity), |uWebTide|, uWebTide or vWebTide (Northward velocity). However a discernable rise in SNR is indicated when compared to vADV (Fig. 10). SNR is stronger with more “good” data when the Southward velocity is higher. The joint-histogram of SNR and vADV (Fig. 10) approximately follows a parabolic shape where the mean of the lowest values is near a SNR = 8 for vADV = 0 m s-1 and gets greater as the measured velocities approach + 0.2 m s-1. Both extreme velocities illustrate a rise but the rise in SNR for southward velocity was more prominent. Considering the bathymetry in Fig. 1, flow along the topography, which is mostly North-South at the measurement location would be expected. Flow mediated by the tides and including a Southward discharge from the continental shelf to the continental rise would account for the enhanced performance with the Southward flow. The Southward discharge may be transporting particles offshore with it. This evidence supports the hypothesis of scatterer supply due to bottom re-suspension by tidal and current friction. Figure 11 illustrates SNR as a function of the ADV pressure signal. The pressure data is presented as depth – mean depth for the deployment in meters. The color bar is limited to a count of 200 bin-1 to enhance the contrast. Figure 11 does not indicate an increase in SNR with any depth (phase of the tide). However it does indicate that for depths ranging from -3 m to 3 m that the SNR is of higher quality. The dark patches in the joint-histogram indicate regions where more data are retained as good either due to passing the correlation filter or due to a peaky SNR d − d i histogram at that ith location. Considering Fig.s 4a and 7, the dark spots could indicate that the SNR was simply greater during the neap cycle of the tide beginning near 21 May. The maximum tidal excursion during this time was on the order of + 0.5 m (Fig. 4a) and SNR was lower near a mean of 8 (Fig. 7). Both of these effects are not reflected in Fig. 11. Figure 7 indicates shorter time scale fluctuations in SNR which appear to be on a time scale shorter than diurnal variability. It is hypothesized that this variability is the significant contributor to the pattern seen in Fig. 11. If the SNR is enhanced for a particular phase of the ( ) 8 tide, then an increase in SNR would be expected at a particular elevation corresponding to that phase. The semi-diurnal nature of the tides with the spring-neap variability would produce variability in that elevation. Repeated cycles would enhance the signal at semi-discrete elevations. These arguments with respect to Fig. 11 lend support to the hypothesis that the main source of scattering material is tidal re-suspension of sediment leading to enhanced signal quality. The prediction from Web Tide was averaged to the ADV burst periods and the mean SNR in a burst was compared to these elevations. The results are similar to Fig. 11. They indicate a weak trend that the SNR is greater for d − d < 0 when sea level is lower than the mean. The trend may be due to the location of the station with respect to a tidal amphidrome. The magnitude of the difference ranges from a SNR of 7 for the high elevation to 9 for the low elevation. ( ) 3.3 Overall performance The data are generally filtered based on correlation and signal strength or SNR. Figure 9 summarizes these data for the entire experiment. The rule of thumb limits are indicated with a line for both SNR of 5 and 10 since there does not appear to be general consensus on this limit. In this figure, good data fall in the upper right hand quadrant, data on the lower half are discarded in primary correlation filtering and data in the upper left quadrant have suspect signal strength. The data from this experiment are seen to primarily lie in the good correlation but questionable signal strength region with SNR range from 8 to 9 and correlation from 80% to 82%. Correlation is seen to be a function of signal strength with the correlation decreasing as SNR decreases. The function, for these particles and state of turbulence, pass through the 70% correlation limit at a SNR just above 5. If correlation is trusted as the only indicator of poor data quality, this experiment suggests that the mean functional dependence supports SNR as low as 5.5. This is simply suggested by the correlation. To verify this result, tests with similar particles would be required in a controlled facility where the velocity can be verified against another source. Figure 9 indicates that most of the data in this experiment fall into the category of good data quality with weak signal strength. The desired data product from an ADV is velocity. The question that is asked in this analysis is do we trust the velocity measurements. Without calibration with the exact particles in a controlled manner, this is a very difficult question. The Web Tide prediction can be used as a comparison, however it is not a calibration. Included in the ADV velocity signal are motions on time scales that differ from that of the tides. This question can not be answered with significance from the data available from this report. Thus users of ADVs in the field must learn how to trust the data quality parameters in the results, the correlation and SNR. This analysis and report is a step toward that goal. Overall, based on the data quality analysis presented here, the ADV data from this experiment appear reliable enough to estimate the mean velocity. A sampling rate of 2 Hz is not deemed sufficient to resolve the scales of turbulence that may be expected. If the data are trimmed with respect to poor correlation, they should represent the mean velocity. Further averaging may be required as analysis dictates. Any repeat of this experiment in similar conditions should double check that the sampling volume is as large as possible, currently 18.0 mm in length. The sampling rate of 2 Hz may be increased but that will depend on data storage capacity and deployment length. It would be interesting to deploy the instrument for more than one spring-neap cycle to test the hypothesis 9 that the SNR follows this cycle indicating that the primary forcing of suspended sediment in the bottom boundary layer at this location is the envelope of the tide signal rather than the diurnal tidal forcing. 4. Conclusions This report documents Nortek Vector ADV data taken on the continental shelf and discusses the data with respect to its quality. The following conclusions were arrived at in this analysis. · Data obtained in this deployment should be adequate to estimate the mean velocity. · Correlation was found to be a function of SNR. · Data with SNR > 8 produce acceptable correlation in this instrument with the particles encountered. · The difference in data quality from Ciochetto (2007) is due to the smaller sampling volume and that with a larger sampling volume, the data quality would be significantly increased. · The main source of particles was re-suspension of benthic particles due to the tide. · The temporal behavior of the SNR followed the tidal Spring-Neap variation. Acknowledgments This work was accomplished in collaboration with Angus Robertson of the Bedford Institute of Oceanography (BIO) and Eric Siegel of Nortek USA. Funding was not provided for the work by the author but contributions in kind toward the instrument were made by both collaborators and Nortek AS. This data and work will contribute to a J. Atm. And Ocean. Tech. paper on the use of the ADV instrument in acoustically unfriendly environments. Even though the manufacturer of the instrument have collaborated and contributed, the author attests that the results presented here are unbiased in any manner and reflect on the true performance of the instrument. The author would like to thank Angus Robertson, Eric Siegel, Dr. Barry Ruddick (Dalhousie University, DAL) for permission to include the instrument in the experiment and for resources to accomplish this work, Dr. Michael Li (BIO), Dr. Dan Kelley (DAL) for data interpretation consultation, Audrey Barnett (DAL) for financial support and to the DFO crews of the vessels that launched and recovered the RALPH frame. This report will be included with comments and corrections as an appendix in the author’s dissertation. All feedback is appreciated and can be sent to david.ciochetto@dal.ca 10 Figures Fig. 1. Location of the deployment (square) generated by Web Tide. 11 Fig. 2. Sound speed (calculated) and temperature (recorded). 150 Histograms of Noise Amplitude [counts] 150 150 100 100 100 50 50 50 0 0 47 48 0 47 48 Histograms of Noise Correlation [%] 40 40 40 30 30 30 20 20 20 10 10 10 0 0 46 47 48 5 10 Beam 1 15 0 0 5 10 Beam 2 15 0 0 5 10 Beam 3 Fig. 3. Noise characteristics for the ADV shelf deployment in May 2007. 12 15 Fig. 4a. Depth from ADV pressure signal (dots) and Web Tide prediction for tidal elevation (line) at (42.05313o, -65.96678o) as indicated in Fig. 1. The tidal prediction is shifted to the mean depth from the ADV pressure sensor of 172.5 m. Fig. 4b. Detail of tidal elevation comparison with ADV pressure data. 13 Fig. 5. Correlation as a function of time represented by the plan view of a two dimensional histogram. The time is binned into 1-minute bins. The data are taken at 2 Hz. For each segment of time, the data from all three beams are combined and then binned at the correlation bit resolution. The color bar indicates the number of samples that fall into each bin. Fig. 6. Summary of the data that pass the correlation filter as a function of burst. 14 Fig. 7. SNR v. time represented by a two-dimensional histogram with 0.43 dB bin resolution in SNR and 1 minute bin resolution in time. Only data with correlation > 70% is shown. The color bar indicates the number of samples that fall in a bin. Fig. 8. ADV velocity data and Web Tide barotropic tidal velocity prediction. The data presented have passed the correlation > 70% filter. 15 Fig. 8 (cont.). Detail view of velocity comparison. 90 3500 80 3000 Correlation [%] 70 2500 60 2000 50 40 1500 30 1000 20 500 10 5 10 15 SNR 20 25 Fig. 9. Correlation as a function of SNR represented by a two-dimensional histogram. The color bar represents the number of samples in a respective bin. 16 Fig. 10. SNR as a function of Northward Velocity. Gaps in the axes appear where data exist. Fig. 11. SNR as a function of surface elevation. 17 References Anderson, S. and A. Lohrmann, 1995: Open water test of the SonTek acoustic Doppler velocimeter. Proceedings of the IEEE, Fifth Working Conference on Current Measurement, St. Petersburg, Florida, U.S.A., William S. Sullwold Publishing, 188 - 92. Barbhuiya, A. K. and S. Dey, 2004: Turbulent flow measurement by the ADV in the vicinity of a rectangular cross-section cylinder placed at a channel sidewall. Flow Measurement and Instrumentation, 15, 221-237. Borg, K., 2002: Coherent doppler profiler observations from a ripple migration event at the long term ecosystem observatory LEO-15, Oceanography, DALHOUSIE UNIVERSITY (CANADA), xix, 149 leaves. Ciochetto, D. S., 2007: Use of a Nortek ADV on the continental shelf. Dal - BIO Internal Report, 6 pp. Hannah, C. G. and J. Chaffey, 2007. http://www.mar.dfompo.gc.ca/science/ocean/coastal_hydrodynamics/WebTide/webtide.html. Hay, A. E., 1991: Sound Scattering from a Particle-Laden, Turbulent Jet. Journal of the Acoustical Society of America, 90, 2055-2074. Lohrmann, A., R. Cabrera, and N. C. Kraus, 1994: Acoustic-Doppler velocimeter (ADV) for laboratory use. Conf. on Fundamentals and Advancements in Hydraulic Measurements and Experimentation, Buffalo, NY, American Society of Civil Engineers, 351–365. Morgan, P. P., 1994: Seawater A library of Matlab computational routines for the properties of sea water, version 1.2222, 29 pp pp. Voulgaris, G. and J. H. Trowbridge, 1998: Evaluation of the acoustic Doppler velocimeter (ADV) for turbulence measurements. Journal of Atmospheric and Oceanic Technology, 15, 272289. Voulgaris, G. and S. T. Meyers, 2004: Temporal variability of hydrodynamics, sediment concentration and sediment settling velocity in a tidal creek. Continental Shelf Research, 24, 1659-1683. Zhang, Y., K. Streitlien, J. G. Bellingham, and A. B. Baggeroer, 2001: Acoustic Doppler Velocimeter Flow Measurement from an Autonomous Underwater Vehicle with Applications to Deep Ocean Convection. Journal of Atmospheric and Oceanic Technology, 18, 2038-2051. 18