© 2003 by ICES/CIEM International Council for the Exploration of the Sea/Conseil International pour l'Exploration de la Mer
Effect of track spacing and data interpolation on the interpretation of benthic community distributions derived from RoxAnnTM acoustic surveys
a School of Conservation Sciences, Bournemouth University, Fern Barrow Poole, Dorset BH12 5BB, UK
b FRS Marine Laboratory PO Box 101, Victoria Road, Aberdeen AB11 9DB, UK
*Correspondence to E. H. Pinn; tel: +44 1202 595178; fax: +44 1202 595255. e-mail: epinn{at}bournemouth.ac.uk.
A 150 mile2 (388 km2) area in the South Minch on the Scottish west coast was surveyed acoustically using the seabed discrimination system RoxAnnTM. This site was chosen from BGS seabed sediment maps because of the wide variety of substratum types present within a relatively small area. The work presented here investigates different combinations of survey track spacing in relation to interpolation of acoustic data for mapping benthic biodiversity. Three different survey track spacings (4, 2 and 1 km) and three pixel sizes (1000, 500 and 250 m) were utilised. The results indicated considerable variations in the fine scale variations of the substratum maps produced and their accuracy in relation to ground truth data. Depending on the track spacing and level of interpolation utilised, the survey site could be considered relatively important under the UK Biodiversity Action Plan in terms of priority habitat types present or completely insignificant. These variations have serious implications for the use of this technology in site identification, conservation and management.
Keywords: data interpolation, geographical information system, habitat type, image analysis, macrofauna, RoxAnnTM
Received 6 August 2002; accepted 28 June 2003.
 |
Introduction |
|---|
 Top Introduction Materials and methodsResultsDiscussionReferences |
|---|
With the introduction of the European Community Directives on natural habitats, wild fauna and flora (92/43/EEC) and birds (79/409/EEC) (European Community, 1992), the emergence of the UK Biodiversity Action Plan and the more recent development of European NATURA 2000 network, a greater emphasis has been placed on identifying biological resources and the conservation of these resources. However, the lack of consistent and recent information on the type, location, size and quality of natural habitats has been identified as a major constraint in the implementation of biodiversity strategies (Weiers et al., 2003). Increasingly, biodiversity conservation has focused upon the protection of representative ecosystems (Alder, 1996; Banks and Skilleter, 2002). If areas are to be conserved, they require identification, monitoring and appropriate management. Such work requires accurate mapping of the constituent biological resources in terms of both habitats and species present. Habitat variability is a key factor in determining species richness within a defined area (Huston, 1994; Rosenzweig, 1995), consequently accurate determination of this variability is an important aspect in any assessment of a region's value with respect to habitat and species conservation.
The development of seabed discrimination systems such as QTC-ViewTM and RoxAnnTM has provided relatively cheap and rapid methods of mapping sublittoral sediments and, consequently, habitats and their variability (Davies et al., 1997). To date, these systems have been put to a wide variety of uses, including marine reserve monitoring (Service, 1998), investigating the physical impacts of benthic trawls (Kaiser and Spencer, 1996; Kaiser et al., 1998; Tuck et al., 1998), identification and characterisation of fish spawning and nursery habitats (Greenstreet et al., 1997; Cholwek et al., 2000; Maravelias et al., 2000; Reid and Maravelias, 2001), and the mapping of contaminated sediments (MacDougall and Black, 1999; Rukavina, 2001). The predominant use, however, has been benthic biotope mapping (e.g. Magorrain et al., 1995; Southeran et al., 1997; Pinn et al., 1998; Hamilton et al., 1999; Ellington et al., 2002; Freitas et al., 2003; Pinn and Robertson, 2003).
Acoustic remote sensing can be conducted at a variety of scales, although most uses to date are quite broad. Downie et al. (1999) reported that broad scale surveys provided good data for defining site boundaries, identifying key features of conservation importance at a coarse level and for developing management plans. Foster-Smith et al. (1999), however, stated that they are more limited with respect to fine scale features and monitoring small scale changes over time where track distances are greater than the scale of change.
RoxAnnTM acoustic mapping has been conducted at a variety of scales. Magorrain et al. (1995) and Southeran et al. (1997) used the system to map seabed sediments and biotopes in areas of approximately 20 km2, while Greenstreet et al. (1997) mapped an area of 4250 km2. However, Pinn et al. (1998) and Pinn and Robertson (2003) have conducted the largest published surveys to date, covering areas up to 17 000 km2. As surveys cover larger and larger areas, it is not feasible to have 100% acoustic coverage of the seabed. Such surveys would not only be very time consuming, but also very expensive in terms of research vessel hire. Consequently, many acoustic surveys are conducted at larger track spacing and the data interpolated to produce a seabed sediment map, e.g. Greenstreet et al. (1997), Pinn et al. (1998) and Pinn and Robertson (2003).
Acoustic surveys perform continuous sampling along a cruise track. As a result, the data are spatially autocorrelated, i.e. the data points close together are usually more correlated than those further apart. Additionally, acoustic surveys are usually carried out using a regular grid or track pattern. This type of data are rarely adaptable to classical estimation methods and random sampling theory and so the application of geostatistical methods for the analysis of acoustic data in relation to biotope survey work has been proposed by a number of authors (Greenstreet et al., 1997; Southeran et al., 1997; Pinn et al., 1998; Foster-Smith et al., 1999). The main purpose of this paper is to examine the effect of track spacing and interpolated pixel size on the interpretation of benthic community distributions. If there are wide variations in the maps produced using different survey track spacings and levels of data interpolation, this could have significant consequences for the use of this system as a tool for mapping benthic communities of an important conservation nature.
|
Materials and methods |
|---|
|
TopIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
An area of 150 miles2 (388 km2) in the South Minch, south west of the Isle of Rhum on the west coast of Scotland (Figure 1), was investigated acoustically using the seabed discrimination system RoxAnnTM. This site was selected because of the wide variety of sediments encountered in a relatively small area (see British Geological Survey seabed sediment map Sheet 56N08W).
|
The survey was conducted over a grid with a spacing of 1 km (east–west and north–south transects). The RoxAnnTM system was connected to a Simrad EK500 Scientific Echo-sounder operating a 38 kHz split beam transducer. The transducer was mounted on a towed body, and run in a position approximately 3 m from the hull, forward of the propeller and at a depth corresponding to the level with the vessel's keel at a speed of up to 10 knots. This towing arrangement was developed to decouple the transducer from ship noise, such as that generated by the vessel's propulsion system, and to lessen the effects of rough weather associated with hull mounted transducers such as cavitation and bubble generation. Data were gathered at 15 s intervals and displayed real time on an AppleMac computer running MacSea GIS software.
On return to the laboratory the acoustic data were examined on a spreadsheet and erroneous values, such as those lacking a depth record, removed. The data were then adjusted to give three sets for the area, each with a different survey track spacing; i.e. 1, 2 and 4. This enabled exact comparisons to be made between different track spacings by removing any differences that may be attributed to fractionally different cruise track positions had each set of data set been collected separately.
The data were gridded using kringing (Burrough and McDonnell, 1998) at three different pixel sizes (1000, 500 and 250 m) in Surfer (Golden Software). It was analysed using unsupervised cluster analysis in the GIS package Idrisi (Clark Labs) with the three variables E1, E2 and depth using a peak histogram technique. The track spacing and pixel sizes used were selected from previous published investigations (e.g. Greenstreet et al., 1997; Pinn et al., 1998). For the 1 km track data an additional pixel size of 100 m was also utilised, this relates to the level of interpolation employed by Southeran et al. (1997). Foster-Smith et al. (1999) advocated the use of supervised cluster analysis for the analysis of acoustic data rather than unsupervised cluster analysis. However, in the context of the current study, this was not considered appropriate. The majority of the previously published work being used for comparison in this study have used either user defined box sets (e.g. Magorrain et al., 1995; Reid and Maravelias, 2001; Rukavina, 2001) or unsupervised cluster analysis (e.g. Greenstreet et al., 1997; Tuck et al., 1998; Pinn and Robertson, 2003). In addition, Foster-Smith et al. (1999) indicate that mixed substratum/biotope types, as regularly observed in the present study, are unsuitable for signature development, a necessary component of supervised cluster analysis.
Variograms were used to check the distances over which data interpolation between acoustic tracks was appropriate (Burrough and McDonnell, 1998). Variograms illustrate the relationship between variance and the separation distance of points by plotting variance against lag distance. The maximum variance (or sill) of the variogram gives an indication of the maximum distance over which interpolation should occur (Burrough and McDonnell, 1998). Foster-Smith et al. (1999) proposed that a variance of half the local maximum was appropriate to indicate the maximum distance over which interpolation was appropriate, and that a variance of 0.2 times the local maximum gave very reliable results. The intertrack distances would consequently be twice the lag distance associated with a chosen level of acceptable variance. Variograms were constructed using Surfer (Golden Software). This programme uses a polar grid in lieu of paired comparisons to reduce the size of files (Golden Software Inc., 1999). Consequently, the maximum lag distance plotted is usually a third of the actual maximum (Golden Software Inc., 1999). For the purposes of this study, the data were plotted using two thirds the maximum distance as the sill of the variogram had been reached by this point. Interpolation distances were estimated according to the methods of Foster-Smith et al. (1999), i.e. lag distances associated with 0.5 and 0.2 times the maximum local variance.
Ground truthing of the acoustic data and investigations into the resident benthic communities were undertaken at 30 sites from throughout the survey area, using closed circuit underwater television (CCTV), grab sampling and trawling. These methods were selected to ensure that as many components of the benthic fauna as possible were sampled, both quantitatively and qualitatively, from the different habitat types encountered. A statistical comparison between the habitat type suggested by the acoustic data and that actually observed by ground truth sampling was carried out using the Chi-squared statistic. The ground truth data were considered to represent the observed data, and the acoustic data the expected data.
Over hard ground such as gravel, boulders, cobbles and bedrock, a dropframe with a vertically mounted CCTV was lowered to within 1 m of the seabed and the ship allowed to drift for 30 min. Over softer sediments, such as muds and sands, the CCTV was mounted on an epibenthic sledge which was towed at low speeds (up to two knots) for 30 min. Details of methods and equipment employed are given in Pinn et al. (1998).
On return to the laboratory, the video data were analysed using the image analysis software AnalySIS® and its associated frame grabber GrabBit PCI (Soft Imaging System GmbH). For each site, 60 individual frames were collected from the TV tow which were then analysed for species abundance and subtratum variability. In addition to CCTV at each site, on softer substrata, sediment samples for ground truthing were collected using a 0.1 m2 Day Grab. These samples were stored in 70% alcohol and analysed using a Malvern Multisizer/E laser particle size analyser. A further five grab samples were also collected for macrofaunal analysis. These samples were sieved through a 0.5 mm mesh and the residue retained on the sieve preserved in 5% buffered formalin. Identification, to species level where possible, was carried out later in the laboratory.
In addition to grab sampling, Agassiz trawling was also undertaken over mud and sand. A 2 m trawl was towed at approximately 1 knot for 30 min. The samples of fauna collected were identified on board where possible and returned to the seabed. Any animals that could not be positively named at sea were preserved in 5% formalin and returned to the laboratory for identification.
The results of the benthic community work has been presented elsewhere (Pinn et al., 1998; Pinn and Robertson, 2003) and will therefore not be considered further here.
|
Results |
|---|
|
TopIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
Figure 2 presents a depth profile image of the survey area with the ground truthing and benthic community sampling positions marked, while Table 1 gives a summary of the ground truthing data and the sampling methods used at each site. A range of sediments were observed from throughout the survey area and these varied from fine silty muds to bedrock. The fine sediments tended to be associated with the deeper water whilst the harder sediments were found at shallower depths (Table 1).
|
|
The variograms for each acoustic survey indicate relatively similar results for each of the track spacings (Table 2). Lag distances of approximately 2.8 km for the 4 km track spacing, 3.5 km for the 2 km track spacing and 3.0 km for the 1 km track spacing (Figure 3, Table 2) were obtained at 0.5 times the maximum variance. Lag distances of 900, 800 and 780 m, respectively (Figure 3, Table 2) were obtained at 0.2 times the maximum variance. These results indicate that a very good degree of spatial correlation and a suitable interpolation distance is within 850 m of each track, but that the maximum distance over which appropriate interpolation could be carried out was found to be approximately 3000 m. These interpolation distances equate to track separation distances of approximately 2–6 km. Consequently, the distances over which data interpolation was conducted in the current study is statistically appropriate for all three track spacings.
|
|
The overall pattern of sediment distribution for the three track spacings at 1000 m pixel resolution were broadly comparable, particularly in the distribution of the mud and bedrock/boulders/stones substrata (Figure 4A–C). However, the fine detail varied considerably between the different track spacings, particularly in the distribution of the intermediate substratum types such as sand and sand/stones/boulders (Figure 4A–C). A comparison of the ground truth data with the substratum type indicated by the acoustic data found 90, 83 and 77% accuracy for the 4, 2 and 1 km data, respectively. The Chi-squared statistic showed that there was no significant differences observed between the ground truth substratum type and that suggested by the acoustic data for the 2 and 4 km track spacing data (Table 3). However, significant differences (p<0.01) were observed for the 1 km acoustic data (Table 3).
|
|
Generally more acoustic clusters were identified using 500 m pixels than for the 1000 m pixel data (Figure 4). The distributions of areas identified as being mud were relatively consistent regardless of the track spacing used at 500 m pixels (Figure 4D–F). This was also true for the areas identified as being bedrock/boulders/stones at the 1 and 2 km track spacing level (Figure 4D and E). However, the total area assigned to this substratum was much smaller when analysed at the 4 km track spacing level (Figure 4F). The 4 km track data suggested a much greater coverage of sand/stones/boulders throughout the area of interest. The areas identified as sand were also larger using the 2 km track data than for either the 1 or 4 km track data (Figure 4D–F). The accuracy of each map in relation to the ground truth data was found to be 73, 93 and 83% for the 4, 2 and 1 km track spacings, respectively. Chi-squared analyses of all the data demonstrated that there was no significant difference between the ground truth information and the habitat type indicated by the acoustic survey for the 2 and 4 km track spacing levels (Table 3). A significant difference (p<0.005) was observed for the 1 km data (Table 3).
Figure 5A–C presents an analysis of the acoustic data using 250 m pixels for the three different track spacings. There was no ground truthing data for six and four of the clusters identified at the 2 and 4 km track intervals, respectively. The range of E1 and E2 values, however, suggested that they were of a sand type habitat. The overall pattern was broadly similar between each of the survey track intervals, particularly between the distributions of mud and bedrock/boulders/stones substrata (Figure 5A–C). The 1 and 4 km track data exhibited a similar distribution of sand/stones/boulders, but this substrata had a much lower coverage area with the 2 km data (Figure 5A–C). Conversely, the 2 km track data identified the presence of sand, which was not seen using the 1 and 4 km data sets (Figure 5A–C). A comparison of the ground truth and acoustic data sets revealed 77, 87 and 80% accuracy for the 4, 2 and 1 km track spacings, respectively. Chi-squared analyses (Table 3) indicated that there were significant differences between the acoustic data and the ground truth data at transect spacings of 1 km (p<0.001) and 2 km (p<0.025). No significant difference was observed in the data for the 4 km track spacing.
|
For the interpolation of the 1 km track data set using 100 m pixels, only two clusters were recognised (Figure 5D). These corresponded to substrata identified as mud and bedrock/boulders/stones. The accuracy of the map was found to be 73%, with chi-squared analysis indicating a significant difference (p<0.001) between the substratum types identified by ground truth sampling and those suggested by the acoustic data analyses (Table 3).
|
Discussion |
|---|
|
TopIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
There is very rarely a hard demarcation line between one substratum type and the next in the marine environment. Normally a gradual change is observed as one habitat gives way to another, resulting in the existence of many mixed sediment/habitat types. These gradual changes in substrata have lead to considerable problems in mapping and interpreting benthic habitats and their associated fauna. With user defined box sets, Rukavina (2001) reported RoxAnnTM to be accurate 80–85% of the time when acoustic results were compared to ground truth results. In the current study, accuracy was found to range from 73 to 90%. However, in the current study the acoustic data were analysed in more detail (unsupervised cluster analysis) before comparison, whilst Rukavina (2001) utilised repeated surveys over 6 years to define the box sets. In addition, in the current study, this comparison was made for exact matches and takes no account of near misses, e.g. sand/stones indicated by the acoustic data and sand by the ground truth data. When near misses are taken into account, the accuracy levels increase to 83–100%.
There is often a trade off between accuracy of a map and the detail shown (Foster-Smith et al., 1999). For example, small areas with few habitat types will invariably have a higher level of accuracy than larger areas of a more heterogeneous nature. Foster-Smith et al. (1999) reported that the number of ground truth sites was extremely important in determining the accuracy of substratum maps derived from acoustic data. They proposed a minimum of three ground truth positions for each substratum or biotope type was necessary to obtain at least 80% accuracy. In the present research, most habitat types were ground truthed at least three times and accuracy levels were generally greater than 80%.
Variations in survey track spacing and the interpolation of the data collected produced quite different results in terms of the indicated area of coverage of particular substratum types. This is particularly noticeable in Figure 4 where, using a 500 m pixel size, the three track spacings produced differing results in terms of the distribution of the sand and sand/stones/boulders habitats types. When investigating the effect of track spacing and consistency of interpretation, Foster-Smith et al. (1999) found that a reduction in track intensity had a dramatic effect on the map produced. As the data were interpreted differently in the two studies, and the current study also utilised a variety of pixel sizes, it is not possible to make a direct comparison between the two pieces of work. However, Foster-Smith et al. (1999) proposed that these differences were attributed to cumulative errors associated with differences in the E1, E2 and depth data. Similar conclusions were reached here.
Foster-Smith et al. (1999) found that the grid spacing (interpolation level or mapping pixel size) did not affect the accuracy of maps produced and proposed that they should be as small as possible within the limitations of the survey. However, within the current research, this was not found to be the case, and the smallest pixel size did not always produce the best results. The comparison of survey track intervals and levels of interpolation between the tracks revealed that overall gross substratum distributions were relatively similar at all levels of interpolation. It was the finer resolution of intermediate sediment types, primarily at the change over points between the mud and bedrock/boulders/stones substratum types, that the greatest variation was observed. Comparisons between the unsupervised cluster analysis of the acoustic data and the ground truth data showed that some sites were incorrectly classified by the cluster analysis. The most noticeable of these was site 29 where the site was ground truthed as being sand/gravel/stones/boulders but with gravel/bedrock dominating. In many of the maps produced using unsupervised cluster analysis this site was allocated to the "mud" cluster. The notable exceptions to this were the 1 km transect data at 500 and 250 m pixels where the site was correctly included in the sand/stones/boulders cluster.
At lower levels of interpolation (e.g. 1 km track spacing with 1000 m pixels), resolution was lost in terms of the number of habitats identified using unsupervised cluster analysis of the data in comparison to the ground truth data. This effect was clearly demonstrated in the inability of the technique to separate sand and sand/stones/boulders. At higher levels of interpolation, there was also a loss of resolution in terms of the number of substratum types identified using unsupervised cluster analysis. For example, using 1 km track intervals with a pixel size of 100 m pixels, the presence of only two habitat types was revealed as compared to the four distinct habitat types with graduations identified from the ground truth data.
The differences observed in the current study and that by Foster-Smith et al. (1999) may be attributed to differences in data analysis. This study utilised unsupervised cluster analysis to identify acoustically different areas of seabed substrata, which were then related to the ground truth results. This produces clusters that can sometimes overlap (Greenstreet et al., 1997; Kaiser et al., 1998), thereby blurring the distinction between different acoustic signatures. Foster-Smith et al. (1999) use a supervised cluster analysis technique that requires signature development and thereby produces clusters which are quite distinct from one another. This form of analysis, however, requires clearly defined substratum types for the training process (Foster-Smith et al., 1999) and not mixed substratum types, as observed in a number of the ground truth sites in the current study. Unsupervised cluster analysis, although possibly less precise, provides a suitable alternative in such situations.
The statistical analysis of the track spacing and pixel size indicated that the best fit (i.e. no significant difference between the two sets of data) between the ground truth data and the acoustic data was observed for the 2 and 4 km transects at 1000 and 500 m pixel interpolation. This indicates that interpolation of data using a pixel size 25% of the track spacing distance (e.g. 4 km transect with 1000 m pixels or 2 km transect with 500 m pixels) produced the most accurate benthic community maps with the level of ground truth sampling undertaken in the present study. At lower and higher levels of interpolation, resolution was lost in terms of the number of habitats and substratum types identified. This loss of data resolution has implications in the mapping of habitat variability and the perceived richness of a specified area. Thus, the constituent biological resources of a region may be under-valued or, equally, over-valued in any assessment of the region's value with respect to habitat and species conservation.
Why a pixel size 25% of the track spacing produced the most accurate results is unclear. It may be due to the natural variability of the site in relation to the actual variability detectable by the acoustic footprint. RoxAnnTM cannot detect substratum variability at a scale smaller than the footprint (Rukavina, 2001), the size of which is related to the depth of water (Foster-Smith et al., 1999). Consequently, at a particular level of natural variability and water depth, the ideal ratio between interpolation and track spacing will be reached. At the site investigated in the current study, this was found to be a pixel size 25% of the track spacing. Further investigations would be needed at other sites to see if this can be verified as a general relationship or whether it is site specific.
In the current study, areas of subtidal mud in deep water as well as subtidal sands and gravels were identified. Their extent of coverage, however, varied quite significantly according to the acoustic survey depending on the track spacing and level of interpolation used. For example, a considerably larger area of mud was identified using 1 km track spacing and 100 m pixel, whilst no subtidal sand and gravels were noted using this combination (Figure 5D). Both of these habitat types are listed as priority habitat types in the UK Biodiversity Action Plan. Priority habitats are those which are either rare or threatened, habitats for which the UK has international obligations or for which more than 40% of the north east Atlantic's occurrence is located in the UK (English Nature, 1998). Alternatively they may be formed from keystone species, important for rare species or are functionally critical for organisms inhabiting the wider environment (e.g. a spawning ground) (English Nature, 1998). If this area of the Minch was being surveyed as a prospective site for conservation or protection, then the acoustic results could have suggested either extreme, i.e. not important or worth protecting due to the occurrence of priority habitat types. Using 1 km track spacing and 100 m pixel size the area does not appear to contain any sand. However, using 2 km track spacing and 500 m pixel size the area appears to contain quite a large extent of sands around the bedrock outcrops. These variations have important implications in species and habitat conservation as this methodology is recognised as a standard technique by MNCR.
Foster-Smith et al. (1999) suggest the use of supervised cluster analysis of the acoustic data and verification against an external data set (e.g. a second set of ground truth data). This would overcome some of the variation by training the system to recognise certain substratum or habitat types. This, however, is considerably more time consuming, expensive in terms of data collection and requires considerably more expertise. The approach taken in the current paper and others (e.g. Pinn et al., 1998; Pinn and Robertson, 2003) is suitable for a broad scale approach where the aim is to gain an idea of the substratum types and, therefore, biotopes, present. Where a more detailed picture is required over a smaller area, e.g. when monitoring change, the methods of Sothern et al. (1997) and Foster-Smith et al. (1999) are better. A pragmatic approach is required when deciding which method to adopt, with compromise being reached between the best analytical techniques, the size of the survey area, the actual time and cost of undertaking the work and level of detail required.
|
Acknowledgements |
|---|
The authors would like to thank the crew of "FRV Clupea", as well as C. Shand and E. Armstrong for camera and sonar expertise, respectively. In addition, the authors would also like to thank the University Marine Biological Station Millport for providing access to their frame grabbing system and Dr Ian Wade (Physe Ltd) for technical advice and assistance.
|
References |
|---|
|
TopIntroductionMaterials and methodsResultsDiscussionReferences |
|---|
-
Alder J. (1996) Have tropical marine protected areas worked? An initial analysis of their success. Coastal Management 24:97–114.[Web of Science]
Banks S.A. and Skilleter G.A. (2002) Mapping intertidal habitats and an evaluation of their conservation status in Queensland, Australia. Ocean and Coastal Management 45:485–509.[CrossRef]
Burrough P.A. and McDonnell R.A. (1998) Principles of Geographical Information Systems(Oxford University Press, Oxford) (333 pp.).
Cholwek G., Bonde J., Li X., Richards C., Yin K. (2000) Processing RoxAnn sonar data to improve its categorization of lake bed surficial subtrates. Marine Geophysical Researches 21:409–421.[CrossRef][Web of Science]
Davies J., Foster-Smith R., Sotheran I.S. (1997) Marin biological mapping for environment management using acoustic ground discrimination systems and geographical information systems. Underwater Technology 22:4167–172.[Web of Science]
Downie D.W, Donnan D.W., Davidson A.J. (1999) A review of Scottish Natural Heritage's work in subtidal marine biotope mapping using remote sensing. International Journal of Remote Sensing 20:585–592.[CrossRef][Web of Science]
Ellington K.E., Gray J.S., Bjørnbom E. (2002) Acoustic classification of seabed habitats using the QTC VIEWTM system. ICES Journal of Marine Science 59:825–835.
English Nature. (1998) UK Biodiversity Action plans. Tranche 2 Action Plans, Maritime species and habitats. UK Biodiversity Group. 244 pp.
European Community. (1992) EC Council Directive 92/43/EEC on the conservation of natural habitats and of wild fauna and flora. Official Journal of the European Community L206:7.
Foster-Smith R. L., Davies J., Sotheran I. (1999) Broad scale remote survey and mapping of sublittoral habitats and biota. Technical Report of the EU supported BMAP project. 155 pp.
Freitas R., Rodrigues A.M., Quintino V. (2003) Benthic biotopes remote sensing using acoustics. Journal of Experimental Marine Biology and Ecology 285–286:339–353.
Golden Software Inc. (1999) Surfer 7 User's Guide. (Golden Software Inc, Colorado).
Greenstreet S.P.R., Tuck I.D., Grewar G.N., Armstrong F., Reid D.G., Wright P.J. (1997) An assessment of the acoustic survey technique, RoxAnn, as a means of mapping seabed habitat. ICES Journal of Marine Science 54:939–959.
Hamilton L.J., Mulhearn P.J., Poeckert R. (1999) Comparison of RoxAnn and QTC-View acoustic bottom classification system performance for the Cairns area, Great Barrier Reef, Australia. Continental Shelf Research 19:1577–1597.[CrossRef][Web of Science]
Huston A.H. (1994) Biological Biodiversity: the Coexistence of Species on a Changing Landscape(Cambridge University Press, Cambridge).
Kaiser M.J. and Spencer B.E. (1996) The effects of beam trawl disturbance on infaunal communities in different habitats after trawling disturbance. ICES Journal of Marine Science 65:346–358.
Kaiser M.J., Armstrong P.J., Dare P.J., Flatt R.P. (1998) Benthic communities associated with a heavily fished scallop ground in the English Channel. Journal of the Marine Biological Association of the United Kingdom 78:1045–1059.[Web of Science]
MacDougall N. and Black K.D. (1999) Determining sediment properties around a marine cage farm using acoustic ground discrimination: RoxAnn. Aquaculture Research 30:451–458.[CrossRef][Web of Science]
Magorrain B.H., Service M., Clarke W. (1995) An acoustic bottom classification survey of Strangford Lough, Northern Ireland. Journal of the Marine Biological Association of the United Kingdom 75:987–992.[Web of Science]
Maravelias C.D., Reid D.G., Swartzman G. (2000) Seabed substrate, water depth and zooplankton as determinants of the prespawning spatial aggregation of North Atlantic herring. Marine Ecology Progress Series 195:249–259.[Web of Science]
Pinn E.H. and Robertson M.R. (2003) Macro-infaunal biodiversity and associated feeding guilds in the Greater Minch Area, Scottish west coast. Journal of Marine Biological Association of the United Kingdom 83:433–443.[CrossRef]
Pinn E.H., Robertson M.R., Shand C.W., Armstrong F. (1998) Broadscale benthic community analysis in the Greater Minch Area (Scottish west coast) using remote and non-destructive techniques. International Journal of Remote Sensing 19:3039–3054.[CrossRef][Web of Science]
Reid D.G. and Maravelias C.D. (2001) Relationships between herring school distribution and seabed substrate derived from RoxAnn. ICES Journal of Marine Science 58:1161–1173.
Rosenzweig M.L. (1995) Species Diversity in Time and Space(Cambridge University Press, Cambridge).
Rukavina N. (2001) Mapping and monitoring of contaminated sediment geometry and stability. Science of the Total Environment 266:33–39.[CrossRef][Medline]
Service M. (1998) Monitoring benthic habitats in a marine nature reserve. Journal of Shellfish Research 17:1487–1489.[Web of Science]
Southeran I.S., Forster-Smith R.L., Davies J. (1997) Mapping of marine benthic habitats using image processing techniques within a raster-based geographic information system. Estuarine, Coastal and Shelf Science 14:Supplement A, 25–31.[Medline]
Tuck I.D., Hall S.J., Robertson M.R., Armstrong E., Basford D.J. (1998) Effects of physical trawling disturbance in a previously unfished sheltered Scottish sea loch. Marine Ecology Progress Series 162:227–242.[Web of Science]
Weiers S., Bock M., Wissen M., Rosner G. (2003) Mapping and indicator approaches for the assessment of habitats at different scales using remote sensing and GIS methods. Landscape and Urban Planning 1007:1–23.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  C. J. Brown, A. Mitchell, D. S. Limpenny, M. R. Robertson, M. Service, and N. Golding Mapping seabed habitats in the Firth of Lorn off the west coast of Scotland: evaluation and comparison of habitat maps produced using the acoustic ground-discrimination system, RoxAnn, and sidescan sonar ICES J. Mar. Sci., January 1, 2005; 62(4): 790 - 802. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||