Assessing the Sulfide Footprint of Mussel Farms with Sediment Profile Imagery: A New Zealand Trial

Growing numbers and increased stocking of marine mussel farms make reliable techniques for environmental effect assessment a priority. Previously, we showed how the color intensity of soft sediment could be used to estimate its acid volatile sulfide (AVS) content, a product of the anaerobic microbial degradation of organic matter deposits. We then proposed to include assessments of the AVS farm footprint in marine farm monitoring, in particular, to investigate temporal changes in the extent of the seafloor area of elevated sediment AVS content. Such assessment requires accurate detection of the AVS footprint boundary. Here, we demonstrate how to detect this boundary with analyses of sediment color intensity. We analyzed 182 sediment profile images taken along three transects leading from approximately 50 m inside to 200 m outside a long-line mussel farm in New Zealand and found that the mean sediment color intensity inside the farm boundary was almost one third lower than that of the sediment distant from the farm. Segmented regression analysis of the combined color intensity data revealed a breakpoint in the trend of increasing grey values with increasing distance from the farm at 56 ± 13 m (± 95% confidence interval of the breakpoint) outside the mussel farm. Statistical analyses indicated that the extent of the color intensity footprint was a function of water column depth, as was shown visually using mapping methods; organic particles disperse further in a deeper seawater column. We conclude that for soft coastal sediments, our sampling and data analysis techniques may provide a rapid and reliable supplement to existing benthic surveys that assess environmental effects of mussel farms.


Introduction
Global food production from marine farms has increased on average 6.2% per year since 2000, rising to 90.4 million tons in 2012 [1,2]. Of all marine-cultured species, bivalves contributed over 15 million tons. The introduction of new farms to the coast, and increased stocking of existing farms necessitates the development of rapid and reliable techniques for the assessment of the effects that such farms have on coastal ecosystems.
Suspended bivalve farms can alter their ecosystems to various degrees depending on the farm's size, age, and stocking density, the seawater column depth and flow regimes, season, and climatic conditions [3][4][5][6]. Ecosystem effects may arise from mussel feeding habits, farm structures, and activities associated with mussel cultivation and harvest. Documented effects include changes in local hydrodynamics [7], phytoplankton depletion [8,9], the spread of invasive organisms [10], and the deposition of farm-derived organic matter (mussel feces and pseudofeces, [11]). The latter can increase the sulfide and ammonium content of the sediment below mussel farms altering the structure and composition of benthic species assemblages [4,12,13]. Mattsson and Lindén [3], for example, reported that the dominant heart urchin and brittle star had been replaced by opportunistic polychaetes 15 months after the introduction of a suspended mussel farm in~15 m deep water of a sheltered bay on the Swedish west coast. Other effects may result from the provision of additional hard substrate due to dropping shells: aggregation of sessile suspension feeder including ascidians, bryozoans, sponges, bivalves, and calcareous polychaete species. Such alterations increase the surface roughness and heterogeneity of the seafloor and create a reef-like habitat for a variety of mobile species including fish, crustaceans, and various echinoderms [13,14].
These and other alterations of the benthic environment are horizontally confined to an area beneath and possibly around the farm, which hereafter we refer to as the "footprint". In New Zealand, for the purpose of marine farm monitoring, environmental managers ask if and how this footprint changes over time. Once a new farm is fully operational its footprint may not change over time if the interaction of this farm with its surrounding environment reached a steady state. Alternatively, the extent and/or the intensity of the footprint may increase over time. To identify, and if existing, quantify such increase, marine farm monitoring should assess two variables: (1) the size of the affected seafloor area, and (2) the degree to which the affected seafloor differs from the unaffected seafloor. Various approaches have been used worldwide to describe these variables, for example: detecting mussel debris with side scan sonar [15] or sediment grab samples [14], identifying genetic differences in sediment microbial communities [16], modeling biodeposit dispersion [17], and measuring the total free sulfide content of the sediment, the sediment redox potential, and water and organic matter contents [18].
In recent years, environmental scientists have attempted to assess the footprint of mussel [19] and fish farms [20][21][22] by measuring the depth of the apparent redox potential discontinuity (aRPD) with sediment profile imagery (SPI). SPI analyses can be supplemented with that of sediment surface images [23] and other visual indicators, such as the presence or absence of fauna and their burrows, in addition to the depth of the aRPD. A combination of these parameters can be included in analyses to calculate benthic indices, for example, the organism-sediment index (OSI, [24,25]), the benthic habitat quality index (BHQ, [26,27]), and the Galway Bay index of habitat quality (GBHQ, [28]). The latter index is site-specific allowing greater differentiation of intermediate environmental states along a local enrichment gradient (see also [29]).
Bull and Williamson [30] used sediment profile image analysis to quantify sedimentary minerals based on a relationship between a specific sediment color property and the mineral concentration. The authors reported two linear correlations for the subtidal sediment of a New Zealand estuary: one between sediment color intensity and acid volatile sulfide (AVS) concentration (R 2 = 0.67), and the other between sediment color saturation and iron oxyhydroxide (FeOOH) concentration (R 2 = 0.62). Sediment color intensity and AVS concentration are inversely related, such that a decrease in sediment color intensity (darker sediment) accompanies an increase in AVS concentration. The authors established each correlation by slicing sediment cores vertically and photographing the exposed surface. They then analyzed subsamples of the sediment for AVS and FeOOH content, which were correlated with the corresponding color property.
The correlation between sediment color intensity and AVS concentration is of particular interest for assessing the environmental effect of marine farms because the sediment AVS concentration is a function of the organic matter deposition rate [12,[31][32][33][34][35][36][37]. To the best of our knowledge, however, AVS measurements are not used in routine monitoring of marine farms presumably because of the laborious analytical process. Estimating the AVS concentration from sediment profile images is rapid [38], however, and could make AVS surveys a valuable supplement to existing benthic monitoring techniques.
Wilson and Vopel [38] further developed the approach of Bull and Williamson [30] and used their improved technique to establish a site-specific correlation between sediment color intensity and AVS concentration. They studied soft subtidal sediment affected by a long-line mussel farm that had been operating since 1980 in Awakiriapa Bay, Waiheke Island, New Zealand. The authors reported a strong quadratic relationship between sediment color intensity and AVS concentration (R 2 = 0.93) and suggested to use this correlation in the monitoring of long-line mussel farms, that is, to investigate temporal changes in (a) the color intensity of sediment underneath such farms and (b) the extent of the footprint, that is, the seafloor area of decreased sediment color intensity (elevated sediment AVS content). The latter requires an approach to accurately detect the location of the footprint's boundary.
Here, we demonstrate such approach with a series of sediment profile images obtained along transects leading from inside to outside of a long-line mussel farm in Awakiriapa Bay, Waiheke Island, New Zealand (Fig 1).

In situ SPI survey
We acquired 182 profile images of soft subtidal sediment during three days in April 2013 and one day in June 2013 along three transects at the Awakiriapa Bay long-line mussel farm, Waiheke Island, New Zealand (S36°48.085', E175°10.022' ; Fig 1). Each transect started approximately 50 m inside the farm boundary and extended north,~200 m beyond the farm boundary. The farm boundary is defined as the position of the end buoy of the mussel line. We didn't require any specific permission to obtain images from this site as it was in public waters and there were no endangered or protected species.
We acquired sediment profile images using SPI-Scan, a rotational imaging device (Benthic Science Ltd., see Wilson and Vopel [38]). The sediment penetration depth of the instrument was adjusted using two approaches, (1) attaching 3-8 × 1 kg weights to the top of the device and (2) controlling the speed at which the instrument penetrated the sediment by either releasing the instrument approximately 2-3 m from the seafloor, or slowly lowering the instrument to the seafloor. We made such adjustments so that the sediment filled two-thirds of the image. Sediment profile scanning was started immediately after the device was in place.
The software (SPI-Scan) digitally embedded the GPS coordinates, date, and time of the acquired image in the metadata of each image. This information is also presented in the upper right-hand corner of the image (Fig 2).
A color calibration strip was included on the right hand side of each profile-image. We used the calibration strip during image analysis to adjust the color of the image so that the color reproduction of each image was identical. Each profile image covered an area of 117 × 216 mm at a resolution of 300 dpi (0.08 mm pixel -1 ). Each scan took approximately 30 seconds to complete; this excludes the time required to lower the instrument from the boat or move the instrument, while underwater, to a new location. An example of the color-range of imaged sediments is shown in Fig 3.

SPI analyses
To measure the color intensity (grey value) of each sediment profile image, we first imported all images into the software analySIS FIVE (Olympus Soft Imaging Solutions, LS Research version 3.3). We then converted the color image from the red, green, blue (RGB) color space to  the hue, saturation, intensity (HSI) color space and measured the average grey value of the intensity channel of a defined area in the sediment profile image, as described below. We observed during image analysis that the sediment surface in the profile images was rarely horizontal and therefore trialed two different methods for selecting an area for image analysis, (1) a rectangle starting approximately 1 cm below the sediment surface that extended 4 cm down, and (2) a polygon that followed the contour of the sediment surface, starting 1 cm below the sediment surface and extending 4 cm down (Fig 4).
Because the mean grey values of the rectangular image area did not differ from those of the polygonal area (paired-samples t-test, t(174) = 0.49, p = 0.62), both methods would have sufficed. In this study, however, we used the larger polygon (~400,000 versus~260,000 pixels 2 ) because the selection of this area was less ambiguous than that of a rectangular area.
Finally, we converted measured sediment color intensities to sediment AVS concentrations using the equation published by Wilson and Vopel [38] for the Awakiriapa Bay long-line mussel farm: [AVS] = 0.0024 × GREY 2 − 0.5249 × GREY + 28.392. Please note that not all measured grey values could be accurately converted to AVS concentrations because some were outside the range of grey values used by Wilson and Vopel [38] to establish this equation.

Statistical analyses
We investigated whether there was any statistically significant difference between transects by analyzing grey values inside the farm (distance from farm boundary <0 m), outside of the farm (distance from farm boundary >0 m), and overall separately with ANCOVA. Where there was a significant difference, we additionally explored the effect of seawater depth, transect, and distance from the farm boundary on the grey value with a full factorial ANCOVA.
To identify the boundary of the color intensity footprint, we performed a segmented regression analysis with the 'segmented' package in R [39,40]. We performed this analysis for each of the three transects and on a dataset created by combining the data from all three transects. The analysis used an iterative procedure to fit two linear regressions and find a breakpoint in the data trend by minimizing the sum of squares of the differences between observed and calculated variables. We constrained the slope of the second linear regression to zero. We did so because we were trying to identify a color change from the unaffected sediment surrounding the farm (background). We assumed that although the color intensity of this sediment will naturally vary, there would be no significant long-distance trend of increasing or decreasing color intensity. We ran the segmented model using distance from the farm boundary as the predictor and a starting (psi) value of 100. We varied psi from 50 to 150 to ensure this value was not biasing the result.

Color intensity mapping
We used marine farm location data (Land Information New Zealand) in ArcMap (ESRI Arc-GIS, version 10.2) to determine the distance of each sampling point from the northern boundary of the mussel farm. To do so, we imported the GPS coordinates embedded in each sediment profile image and then calculated the distance from the image location to the northern edge of the mussel farm with the Generate Near Table tool in ArcMap. The depth information for each data point was extracted from a 20 m resolution gridded bathymetric dataset [41]. We mapped the sediment color intensity over an area extending from~50 m inside the northern farm boundary to~200 m north of the farm boundary with the polynomial interpolation model in ArcMap (see S1 Table for model parameters). The model extrapolated the measured sediment color intensities to cover this area.

Results
Sediment color intensities (grey values) were high and varied little (93 ± 1 grey values; mean ± 95% CI, n = 83) at distances >50 m north of the northern farm boundary. This defines the "background" sediment color intensity in Awakiriapa Bay. The mean sediment color intensity inside the farm, that is, south of the northern farm boundary, was almost one third lower (61 grey values) than the background color intensity. This lower sediment color intensity defines the footprint of the farm.
Conversion of the measured sediment color intensities to AVS concentrations with the AVS/color intensity correlation published by Wilson and Vopel [38] revealed AVS concentrations far above the range of those in the published correlation for nearly half of all profile images inside the farm, and negative concentrations for some profile images of sediment outside the farm (Fig 5). The latter concentrations were all within the margin of error of the AVS/ color calibrations if the actual AVS concentrations were at or near 0 μmol g −1 ; the mean 95% individual CI of the AVS/color calibration was 0.5 μmol g -1 [38].
In the following we locate the position of the AVS footprint boundary using color intensity data instead of AVS data to avoid uncertainty from extrapolating the calibration in Wilson and Vopel [38]. Segmented regression analysis of the color intensity data from all transects located the northern boundary of the footprint at 56 ± 13 m (± 95% confidence interval of the breakpoint) outside and north of the mussel farm ( Fig 6A). Furthermore, segmented regression analysis on individual transects revealed that the footprint extended 35 m less on the shore side than it did in the middle of the farm (Fig 6B-6D, Table 1).
The mean sediment color intensity inside the farm was lower (darker sediment) along the Shore and Mid transects than it was along the Channel transect (Table 2).
To elucidate the intensity increase of the footprint towards the shore side of the farm, we investigated how the variables Depth, Distance from the farm boundary, and Transect influenced sediment color intensities inside the mussel farm and found that all three variables, and some interactions between these variables, had a significant effect ( Table 3). The Shore transect was closest to the shore and in seawater two meters shallower than that of the other two transects ( Table 4).
The color intensity footprint map in Fig 7 supports the results of our segmented regression analysis: sediment color intensities were >80 grey values at distances greater than 50 m from the farm boundary and <77.5 grey values at distances less than 50 m from the farm boundary. This map also indicated that the color intensity footprint didn't extend as far on the western side of the farm as it did on the eastern side.

Discussion
We demonstrated the use of in situ SPI to assess the sediment color intensity footprint of a long-line mussel farm. Two variables were of interest, the intensity of the footprint, that is, the magnitude of the difference between the color intensity of the sediment underneath the farm and that of the surrounding unaffected sediment, and the size or spatial extent of the footprint. For our trial mussel farm, we determined the latter with segmented regression analysis of our color intensity data (Fig 6, Table 1); the footprint extended approximately 50 m beyond the northern boundary of the mussel farm. This distance lies within the range of numerical model predictions of the dispersion and erosion of feces and pseudofeces at a nearby mussel farm, 25 km southeast of Awakiriapa Bay. Here, Giles et al. [17] found that mussel feces released by this farm into a 10 m deep seawater column were deposited up to 60 m away from the mussel farm boundary. Incorporating resuspension by erosion into their model increased this distance to 130 m. Clearly, the deposition of small amounts of feces and pseudofeces away from the farm does not necessarily result in a detectible decrease of sediment color intensity, but such deposition modeling can inform the design of future AVS and color intensity footprint surveys. In particular, it will help to ensure that survey transects extend far enough from the mussel farm boundary to include the unaffected seafloor.
Two advantages of this SPI analysis technique are that it is rapid and it does not require indepth training and experience of the operator to measure and adequately interpret the ecological significance of image parameters. The operator simply uses software routines to measure one property of the sediment, its color intensity. Selecting an image area to determine this property is the only step in our routine that requires the operator to make a decision. We believe that its simplicity and automation make this method suitable to complement other types of assessments of the effects of organic enrichment of sediments underneath mussel or fish farms on the soft-sediment ecosystem because it will ensure that data acquisition remains consistent over the lifetime of a monitoring program. The placement of color intensity survey transects at the onset of such program must consider local differences in seawater depth and current speed-two variables that affect the dilution and dispersion of feces and pseudofeces after release [42,43].  For our trial mussel farm, differences in the sediment color intensities between the three transects and the sediment color intensity map in Fig 7 indicate that the farm AVS footprint did not extend as far beyond the farm boundary close to the shore as it did close to the channel, logically because the seafloor underneath the shore side farm blocks is less deep and thus biodeposits would not disperse as far. Hartstein and Stevens [44] support this reasoning; their sediment trap measurements and dispersal model revealed that because particles released from a farm in shallower seawater spend less time in the water column they deposit over a smaller area of seafloor than particles released in deeper seawater. Consequently, deposition in shallow seawater will elevate the sediment organic matter content more than deposition in deeper seawater. Differences in water depth cannot explain the differences in the sediment mean color intensity between the Mid and Channel transects within the farm as their depths did not differ significantly. We suggest that this difference resulted from a higher flow speed along the Channel transect, which has a similar effect as a greater water depth, increasing the dispersion of suspended particles from the mussel farm.
Over half of the sediment color intensities measured inside the mussel farm were outside the range of those color intensities included in the AVS/color intensity correlation presented by Wilson and Vopel [38]. Clearly, a new calibration covering a wider range of sediment color intensities is required to accurately predict the entire range of AVS concentrations and any conclusions made from the extrapolated AVS concentrations should be conservative. Many of the AVS concentrations derived for the sediment in the area between the farm boundary and the footprint boundary, however, are within the calibrated range and so suitable for comparison with future measurements. Such measurements outside the farm boundary may reveal that the sediment color intensities have decreased over time and this will indicate that the intensity and extent of the AVS footprint of this farm has increased. Environmental managers may use such trend to request more detailed investigations of the effect of organic matter deposition on benthic ecosystem functioning. We mapped sediment color intensities to visually represent the two-dimensional color intensity footprint. Such representation may simplify any assessment of the shape of the footprint but it cannot provide any more detailed information than the three transect measurements from which it was derived. The availability of resources for mussel farm monitoring often limit the number of sites, measured parameters, and the frequency at which measurements are conducted. If so, we recommend focusing data acquisition to produce sufficient sediment profile images along one transect rather than aiming for greater spatial coverage. The minimum number of scans will depend on local conditions and this should be investigated with a trial before commencing monitoring. In this study, segmented regression identified the footprint boundary along the Channel transect with as few as 28 data points. Segmented regression analyses of, for example, annual data sets will then allow investigators to assess if the position of the footprint boundary in the direction of this transect or the intensity of the footprint, that is, the color intensity of the sediment underneath the farm change over time. If the shape of the footprint and local differences in its extent are of interest, however, the sampling design should ensure sufficient spatial coverage to produce color intensity maps as shown in Fig 7. Deposition of farm-derived organic matter alters both the redox chemistry of the receiving soft-sediment seafloor and the composition of its benthic species assemblages [45][46][47]. Changes in these two variables are linked; soft sediment with elevated organic matter often supports assemblages dominated by few opportunistic species. The seafloor underneath mussel farms, however, accumulates hard substrate in form of dropped mussel shells and this accumulation can lead to contrasting results. In a previous study of our trial mussel farm in Awakiriapa Bay, for example, Wong and O'Shea [14] demonstrated that species richness and diversity of the macrofaunal assemblage were higher beneath than outside the mussel farm. The authors suggested that the boundary of the farm footprint (based on attributes of the benthic macrofaunal assemblage) was marked by the outer limit of benthic clumps of mussel shells, which extended 30 m past the boundary of the mussel farm. Here, instead of using clumps of dropped mussel shells as an indicator, we used a property of the soft-sediment ecosystem that is linked Table 3. Full factorial ANCOVA used to investigate the effect of Distance from the farm boundary, Transect, and Depth on the color intensity of the sediment inside the farm (distance <0 m from the farm boundary). with the deposition of farm-derived organic matter and demonstrated that the footprint of this farm extended beyond 30 m.
Supporting Information S1 Table. Parameters for the local polynomial interpolation model in ArcMap (ESRI Arc-GIS, version 10.2). Measured sediment color intensities along three transects running across the boundary of the mussel farm in Awakiriapa Bay, Waiheke Island, New Zealand were