Acoustic Doppler Current Profiler measurements near a weir with fish pass: assessing solutions to compass errors, spatial data referencing and spatial flow heterogeneity

Kriechbaumer, Thomas; Blackburn, Kim; Everard, Nick; Rivas Casado, Monica

doi:10.2166/nh.2015.095

There has been an increasing interest in the use of Acoustic Doppler Current Profilers (ADCPs) to characterise the hydraulic conditions near river engineering structures such as dams, fish passes and groins, as part of ecological and hydromorphological assessments. However, such ADCP applications can be limited by compass errors, obstructed view to navigation satellites, frequent loss of bottom tracking and spatially heterogeneous flow leading to erroneous water velocity measurements. This study addresses these limitations by (i) developing a heading sensor integration algorithm that corrects compass errors from magnetic interference, (ii) testing a Total Station based technique for spatial ADCP data referencing and (iii) evaluating a recently proposed data processing technique that reduces bias from spatial flow heterogeneity. The integration of these techniques on a radio control ADCP platform is illustrated downstream of a weir with fish pass on the River Severn, UK. The results show that each of the techniques can have a statistically significant effect on the estimated total water velocities and can strongly affect measures of vorticity. The obtained three-dimensional flow maps are suitable to describe the magnitude and orientation of the fish pass attraction flow in relation to competing flows and to highlight areas of increased vorticity.

attraction flow, compass, flow measurement, inertial measurement unit, radio control boat, Total Station

INTRODUCTION

Listen

Acoustic Doppler Current Profilers (ADCPs) have evolved as a useful tool to characterise the flow distribution of river reaches (e.g. Dinehart & Burau 2005; Rennie & Church 2010). A number of studies (Gaeuman & Jacobson 2005; Johnson et al. 2009; Jamieson et al. 2011, 2013) have illustrated the potential of ADCPs to quantify the flow field near river engineering structures as part of ecological and hydromorphological assessments. These studies have highlighted a range of ADCP data quality issues, including: (i) errors in the ADCP-internal compass data caused by changes in the local magnetic field (e.g. from steel reinforcements), (ii) limited line of sight to navigation satellites when using ADCPs in conjunction with Global Navigation Satellite Systems, (iii) discontinuous water velocity measurements caused by the loss of the ADCP bottom tracking (BT) signal, and (iv) lack of accurate three-dimensional (3D) water velocity measurements in spatially heterogeneous flows. These limitations reduce the applicability of ADCPs to characterise the hydrodynamics near engineered flow obstacles. For example, Jamieson et al. (2013) found spatial ADCP data referencing based on the global positioning system (GPS) to be insufficiently reliable when monitoring the hydraulics induced by stream barbs on a river in a heavily wooded and deep valley. Jamieson et al. (2011) experienced BT loss near a wing dike and attributed this problem to high water turbidity and turbulence and Johnson et al. (2009) found the ADCP data collected near surface flow outlets at dams to be biased because of large spatial flow heterogeneity.

This study introduces novel techniques of ADCP data collection and assesses a recently developed method of data post-processing to address these data quality issues. The proposed methods are integrated on a radio control ADCP platform and illustrated by quantifying the 3D distribution of water velocities immediately downstream of a weir with fish pass. The installation of fish passes at engineering structures designed to regulate discharge has been a wide-spread approach to restore the longitudinal connectivity of freshwater ecosystems (Katopodis & Williams 2012). Policy efforts towards restoring the ecological integrity of rivers (EC 2000, 2007) and the increasing evidence on the low efficiencies of existing fish passes (Bunt et al. 2012; Noonan et al. 2012) have led to a strong need for more post-construction assessment to gain a better understanding of the various factors determining the biological effectiveness of fish passes. The hydrodynamic conditions near fish pass entrances have been recognised as a key factor influencing the ability of fish to locate and enter these facilities (Piper et al. 2012; Williams et al. 2012; Lindberg et al. 2013). Yet, there is a lack of methods for the spatially continuous in-field quantification of near-pass hydrodynamics. To the authors’ knowledge, this paper presents the first in-field solution to rapidly quantify the spatially continuous distribution of water velocities near fish pass entrances using an ADCP.

ADCPs are mono-static sensors that measure water velocities and depths by transmitting and receiving acoustic pulses with 3–4 transducers along beams spread at an angle of usually 20–30° relative to the vertical direction. The arrangement allows for the use of a single acoustic signal to obtain measurements in multiple depths along the vertical water column (termed ‘ensemble’; Mueller & Wagner 2009). The water velocities measured in the directions parallel to each acoustic beam are processed to resolve a 3D vector describing the flow in the x, y and z directions of a coordinate system aligned with the instrument (Mueller & Wagner 2009). ADCPs have an internal fluxgate compass to determine the transformation angle required to reference these velocities to the local ambient magnetic field (magnetic north) and, after correcting for the site-specific magnetic declination, to true north. When the boat velocity is determined from ADCP-external sensors (e.g. because of BT loss), the effect of moderate errors in on the velocity components referenced to north can be large as it depends on the magnitude and direction of the actual water velocity and the ADCP boat velocity ⁠. For a ratio of 1, an error in of 10° can lead to a 17% error in the measured water velocity magnitude and an error of up to 20° in the water velocity direction (computed based on Gaeuman & Jacobson (2005)). A potential practical and low-cost solution to this limitation is the correction of ADCP compass errors with an inertial measurement unit (IMU) consisting of micro-electromechanical gyroscopes and accelerometers. Some IMUs fuse the inertial sensor data to provide orientation measurements relative to the direction of gravity, which are constrained neither in motion nor to any specific environment or location (Madgwick et al. 2011).

ADCP-measured water velocities have to be corrected for boat velocities, which are typically determined from BT (Gordon 1996). Common ADCP software flags ensembles without a valid BT signal as bad, indicating that the obtained measurements are unusable. These measurements can be recovered through the integration of external positioning systems such as GPS, based on which the boat velocity is estimated. However, fish passes and other engineered river structures are frequently installed close to river banks and these areas are particularly affected by degradation in GPS accuracy (Rennie & Rainville 2006). The problem may increase in small rivers, where the sky view can be obstructed over a large proportion of the water surface. This limitation can be addressed through the integration of ADCPs with alternative, local positioning systems such as tracking Total Stations (TS), which achieve 3D positioning precision of sub-cm level without relying on navigation satellites (Kirschner & Stempfhuber 2008).

Repeated ADCP measurements are necessary to capture the temporally averaged flow field in rivers (Muste et al. 2004). The conventional method of repeated ADCP measurements involves the averaging of multiple 3D water velocity vectors, each of which is resolved independently from the 3–4 along-beam velocities measured at the same time. This method assumes that the water velocities in the areas insonified by the beams are spatially homogeneous. The diameter of a circle enclosing the four beam footprints increases at a ratio of 0.76 m per 1 m increase in depth (calculated based on Rennie et al. (2002), for a 1,200 kHz WorkHorse RioGrande ADCP). Nystrom et al. (2002) argued that the distance between the beam footprints is comparable to the size of large-scale turbulence, so that the assumption of homogeneous flow can easily be violated in spatially complex hydraulic conditions. The data post-processing method suggested by Vermeulen et al. (2014) can avoid this bias by reducing the velocity sampling volume assumed to be homogeneous. The method uses a least squares procedure to estimate the 3D velocity vector that fits best to a set of along-beam velocities measured in similar locations during repeated cross-sectional measurements. However, the approach has not been tested in ADCP applications near flow obstacles.

The aim of this study was to integrate ADCPs with external sensors and novel data processing techniques for the accurate, in-field and rapid quantification of the spatially continuous 3D water velocity distribution near fish pass entrances. This was achieved through three core objectives:

develop an IMU-based heading sensor integration algorithm that corrects ADCP compass data biased by magnetic interference;
test a TS-based technique that provides spatially referenced ADCP data in areas of limited sky view and determines boat velocities in areas of BT loss; and
evaluate the derivation of 3D water velocities as suggested in Vermeulen et al. (2014) to address the ADCP data bias caused by spatial flow heterogeneity.

METHODS

Listen

Case study site

Listen

The study site was a 55 m reach immediately downstream of Shrewsbury Weir on the River Severn (Figure 1). The River Severn is the longest river in the UK and one of its main salmon rivers (NASCO 2009). It flows from Plynlimon, Ceredigion, in the Welsh mountains to Gloucestershire, where it discharges into the British Channel. A total of 41 obstructions, with nine of them being considered significant barriers to upstream fish migration, can be identified along the course of the river. Shrewsbury Weir is the last major migration barrier to Atlantic Salmon (Salmo salar) before spawning grounds in the upper catchments. This study focused on the fish pass installed on the right river bank, constructed in 1976 as a pool and weir pass and then refurbished in 2006 as a deep vertical slot pass. Throughout the study, the streamwise direction was defined to be orthogonal to the weir crest (Figure 1).

Figure 1

View large Download slide

Study location; the white dash-point line depicts the extent of the study area and the white dashed lines show cross sections referred to throughout the main text; the arrow pointing to the location of the fish pass entrance is orientated perpendicular to the front wall of the fish pass; and stand for the streamwise and cross-stream directions; the images on the bottom right show the study site on the day of the data collection looking in the upstream direction (20 August 2014).

Data collection

Listen

Velocity and depth data were collected using a 1,200 kHz WorkHorse RioGrande ADCP (Teledyne RDI 2007) deployed from an ARC-Boat radio control platform (HR Wallingford 2014). The data were collected along 13 cross-sectional and eight longitudinal profiles spaced approximately four meters apart (Figure 2). Each profile was repeatedly sampled to capture the mean 3D velocity patterns. Hereafter, each cross-sectional repetition is referred to as transect. Cross sections within 28 m from the weir foot as well as longitudinal profiles were sampled six times to account for larger turbulence. Cross sections further than 28 m were sampled four times. The ADCP recorded velocity and depth data at an average frequency of 1.5 Hz and with a mean boat speed of 0.42 ms^–1. The vertical measurement resolution was set to 0.12 m. The discharge was assumed to be constant and equal to 7.1 m³ s^–1 based on records from the nearest gauging station (Figure 1).

Figure 2

View large Download slide

Sampling strategy and technical survey setup (not to scale).

A Leica Nova MS50 (Leica Geosystems 2015) with TS capability was used to automatically track a 360 ° prism installed directly above the centre of the ADCP (Figure 2). To support the accurate implementation of the sampling strategy a software application was developed in Matlab to display the real-time boat positions against the planned cross-sectional path. This ensured that the spatial variation of the individual transects of a measurement section and the resulting loss in spatially dependent flow features (Jamieson et al. 2011) were minimised. On average, 81.0% of all ensembles were at distances below 1 m to a straight line fitted through the ensemble locations of the respective measurement section.

All data were recorded on a laptop mounted on the ARC-Boat and controlled on shore from another laptop via Windows Remote Desktop Connection (Figure 2). The TS data were transmitted wirelessly to the on-board laptop using a MOXA NPort W2150 wireless device server. Bespoke software was developed in C ++ to record the data from the MS50 and an x-IMU inertial measurement unit (x-io Technologies 2012). The ADCP data were recorded using the ADCP software WinRiver II by Teledyne RD Instruments Inc. To enable temporal synchronisation of the sensors, their data were time stamped with the Windows PC time of the logging computer (for TS and IMU) and the ADCP-internal real-time clock (for the ADCP). To keep the accumulated drift of the real-time clock below 0.05 s, the absolute time of the clock was set by the Windows PC time of the logging computer at least every 30 minutes in WinRiver II. The error of the time synchronisation depends on the recording frequencies of the sensors, which were 1.5, 5.4 and 64 Hz on average for the ADCP, TS and IMU, respectively. In total, 0.56% of all ensembles had a temporal offset to the nearest TS sample above 0.15 s. These were excluded from the analysis to limit the error in spatial data referencing.

Compass correction

Listen

Temporary compass errors were corrected by integrating the absolute heading data from the ADCP-internal fluxgate compass with relative heading data from the ADCP-external IMU. The ADCP-IMU integration algorithm detects biased ADCP compass data through cross-correlation analysis of the time synchronised compass and IMU data within a shifting window of length w and for lags of –1, 0 and 1 (Figure 3). The data within a window were considered biased if none of the three cross-correlation coefficients was positive and significant ⁠. The window was shifted by until the end of the data series was reached. Data detected as biased were then replaced by corrected heading values ⁠, which were computed as shown in Equation ():

1

where are the ADCP compass heading data, are the IMU heading data, i is the ADCP ensemble index and d is the distance in the data series from i to the centre position of the closest previous window with unbiased ADCP compass data. If the beginning of the compass data series was biased, was computed as follows:

2

with d becoming the distance from i to the centre position of the closest subsequent window with unbiased compass data. Figure 3 illustrates the implementation of the algorithm using data collected at a river cross section with a steel hulled narrowboat moored on one of the river banks and affecting the local magnetic field.

Figure 3

View large Download slide

ADCP-IMU integration illustrated using data collected on a cross section of the River Thames at Eynsham, Oxfordshire (UK), with a moored steel hulled vessel acting as source of magnetic interference. (a) Time synchronised data of ADCP compass and x-IMU. (b) Detection of biased ADCP compass data; the two inlay plots show the results of the cross-correlation analysis for unbiased (left plot) and biased (right plot) compass data, where stands for cross-correlation function. (c) Results of the compass error detection. (d) Results of the compass error correction.

The error in the compass heading

was defined as shown in Equation ():

3

The effects of the circular nature of degrees were accounted for (e.g. if

and

⁠,

⁠).

Spatial data referencing

Listen

The ADCP data were spatially referenced using the tracking TS. The ADCP positions were transformed to global positions in the UTM coordinate system based on reference measurements with a differentially corrected GPS (Trimble GeoExplorer 6000 GeoXH). The positioning error caused by the temporal offset between ADCP and TS data was estimated as follows:

4

where is the total BT-based boat velocity for ensembles with a valid BT signal. The TS positions were used to estimate the boat velocity for ensembles (i) affected by BT signal loss or (ii) with unrealistically high BT-based total boat velocity magnitudes (>1.4 ms^–1). The error in the total TS-based boat velocity was estimated as shown in Equation ():

5

To assess whether was directionally biased by a non-stationary channel bed, moving bed tests were performed in three locations of the study area (Figure 2) for durations of at least 400 s each.

3D water velocity estimation

Listen

3D water velocities were estimated from the along-beam velocities using the Matlab ADCP tools implementing the method by Vermeulen et al. (2014). The method combines the along-beam velocity samples located within the same 3D cell of a fitted mesh. For the study at hand, the longitudinal

⁠, lateral

and vertical

mesh cell dimensions were chosen to be 2.00 m, 0.40 m, and 0.15 m, respectively (Figure 4(a)). The cell size selection determines the volume for which spatially homogeneous flow is assumed, which is in contrast to conventional repeated transect ADCP data processing, where the minimum size of this volume is fixed and determined by the ADCP beam spread and measurement depth (Figure 4(b)). The sensitivity of the 3D velocity estimates to the mesh cell size was quantified by comparing the average number of along-beam velocity samples per cell

and the average of the total water velocity magnitudes of the mesh cells of a cross section

for 36 different mesh cell sizes. The effect of the number of transects taken along a cross section on

was quantified by calculating the mean change in

caused by including another transect.

Figure 4

View large Download slide

(a) Plane view of the 3D velocity projection mesh on the ADCP track (shown for section b in Figure 1); and stand for the longitudinal and the lateral mesh cell dimensions. (b) Volume for which spatially homogeneous flow is assumed in the processing method by Vermeulen et al. (2014) with the cell dimensions used in this study and the minimum W in conventional processing of data from a 1,200 kHz WorkHorse RioGrande ADCP with a vertical measurement resolution of 0.12 m; an instrument draft of 0.11 m was assumed for both methods.

Depths and water velocities in unmeasured locations of the study area were estimated through ordinary kriging as suggested by Jamieson et al. (2011, 2013) and Rennie & Church (2010), using a 0.25 × 0.25 m² grid for depths and a 0.50 × 0.50 × 0.15 m³ grid for velocities. The error

in the water velocities introduced by the spatial interpolation was quantified through cross-validation (Webster & Oliver 2007) and defined as follows:

6

where

and

are the water velocities obtained through the method by Vermeulen et al. (2014) and predicted through kriging in the same location, and

⁠, which are the velocity directions in the stream coordinate system. The cross-validation was carried out for a sample of 1,000 points randomly selected out of the 10,371 measurements.

Effect of data correction techniques

Listen

To assess the effects of the suggested techniques, namely (i) IMU-based compass correction, (ii) TS-based recovery of ensembles with BT loss and (iii) the water velocity estimation by Vermeulen et al. (2014), the 3D distribution of water velocities in the case study reach was quantified with and without the application of each of these techniques. As counterpart to the 3D water velocity estimation by Vermeulen et al. (2014), the 3D velocities were resolved using the conventional repeated transect processing method as implemented in the Matlab application VMT (Parsons et al. 2012). For each processing variant, the total water velocity magnitudes and the absolute area-weighted vorticity measure (Crowder & Diplas 2002) were computed from the 3D flow distribution obtained after kriging interpolation and compared through descriptive statistics. Wilcoxon signed-rank tests were used to test whether the differences in caused by each of the techniques were statistically significant. The vorticity measure was computed as suggested by Shields & Rigby (2005):

7

8

where ⁠, and are the changes in the streamwise, cross-stream and vertical water velocities in the streamwise, cross-stream and vertical directions ⁠, and ⁠.

These hydrodynamic measures were chosen because they reflect the absolute water velocity magnitudes and the strength and abundance of spatial velocity gradients ⁠, both of which are known to affect fish swimming behaviour near fish passes (e.g. Larinier 2002; Enders et al. 2009). To explore spatial variations in the effects of the techniques, the analysis was carried out for the cross sections b, d and f shown in Figure 1 and for the horizontal planes at depths of 0.35 and 1.10 m.