Broad scale mapping of sublittoral habitats in The Sound of Barra, Scotland

Transcription

1 COMMISSIONED REPORT Commissioned Report No. 005 Broad scale mapping of sublittoral habitats in The Sound of Barra, Scotland (ROAME No. F01AA401B) For further information on this report please contact: Evanthia Karpouzli Maritime Group Scottish Natural Heritage 2 Anderson Place EDINBURGH EH6 5NP Telephone: evanthia.karpouzli@snh.gov.uk This report should be quoted as: Bates, C. R., Moore, C. G., Malthus, T., Harries, D. B., Austin, W., Mair, J. M. and Karpouzli, E. (2004). Broad scale mapping of sublittoral habitats in The Sound of Barra, Scotland. Scottish Natural Heritage Commissioned Report No. 005 (ROAME No. F01AA401B). This report, or any part of it, should not be reproduced without the permission of Scottish Natural Heritage. This permission will not be withheld unreasonably. The views expressed by the author(s) of this report should not be taken as the views and policies of Scottish Natural Heritage. Scottish Natural Heritage 2003.

2

3 Scottish Natural Heritage Commissioned Report No.005 (ROAME No. F01AA401B) This report was produced for Scottish Natural Heritage by the Sedimentary Systems Research Unit, University of St Andrews, the School of Life Sciences Heriot-Watt University and the Department of Geography, University of Edinburgh on the understanding that the final data provided can be used only by these parties and SNH. Dr Richard Bates Sedimentary Systems Research Unit School of Geography and Geosciences University of St Andrews St Andrews Scotland Dr Colin Moore School of Life Sciences Heriot-Watt University Edinburgh Scotland Dr Tim Malthus Department of Geography University of Edinburgh Edinburgh Scotland

4

5 COMMISSIONED REPORT Summary Broad scale mapping of sublittoral habitats in The Sound of Barra, Scotland Commissioned Report No. 005 (ROAME No. F01AA401B) Contractors: University of St Andrews, Heriot-Watt University and University of Edinburgh Background The Sound of Barra is situated between the islands of South Uist and Barra in the Outer Hebrides. The site has been selected as a possible Special Area of Conservation (psac) on the basis of the nationally important colony of common seals Phoca vitulina and also for the wide variety of habitats associated with shallow sandbanks which are slightly covered by seawater all the time. A number of the sandbank habitats are of considerable conservation value, most notably the extensive beds of the eelgrass Zostera marina and tide-swept maerl beds composed of the coralline red alga Phymatolithon calcareum. In order that a comprehensive management plan can be developed to ensure the sustainable use of resources within the marine psac it is essential to obtain an estimation of the geographic distribution and extent of the habitats of interest. To collect these spatial data a comprehensive biotope mapping survey of the sublittoral habitats within the Sound was undertaken in August 2001, by a collaborative research group comprising staff from the University of St Andrews, Heriot-Watt University, Edinburgh University and Scottish Natural Heritage (SNH). Mapping of the shallow (predominantly <5m) Sound of Eriskay and northwestern region of the psac was accomplished principally through the employment of IKONOS satellite imagery. Deeper areas of the site were surveyed by rapid broad scale remote acoustic mapping techniques, with ground truth data collected to enable the interpretation of the acoustically classed sea floor maps. The ground-truthing data were collected in the field using a range of sampling techniques including diver-based observations with video and still cameras (20 stations), snorkel video, glass-bottom bucket sampling and drop-down video (58 stations), video imagery collected by remotely operated vehicle (ROV) (23 stations) and grab or core sampling of sedimentary habitats for infaunal and sediment analyses (23 stations). This information supplemented the existing knowledge on the distribution of marine communities within the Sound of Barra psac and all this information was synthesised into a record of habitat information depicted in a series of biotope classification maps. Main findings The shallow northwestern region of the psac is composed of rock and sand biotopes. The more exposed western fringe includes extensive areas of kelp-dominated rock and medium-coarse sand (IGS.Sell). Passing eastwards, increasing shelter results in dominance by fine-medium sands (IMS.EcorEns) incorporating extensive eelgrass beds (IMS.Zmar). The area of eelgrass beds was estimated, by the use of multispectral image classification, to be 75ha. Patches of dense living maerl, Phymatolithon calcareum, are present in the

6 Scottish Natural Heritage Commissioned Report No.005 (ROAME No. F01AA401B) deep channel passing through the Sound of Eriskay (IGS.Phy.R). In deeper water to the east of Eriskay the sea bed is floored predominantly by coarse sediments of gravel, pebbles and maerl (CGS.Ven.Neo, IGS.PhyHEc), infralittoral kelp biotopes and circalittoral scour-tolerant faunal turfs (MCR.FluHByS). A similar range of biotopes is found in the more exposed eastern fringe of the southern region of the psac, with muddy sand biotopes (CMS.AfilEcor, IMX.LsacX) predominant in more sheltered conditions. The southern region contains the only extremely sheltered subtidal biotopes in the psac and also supports more extensive areas of steep and vertical rocky reef biotopes. Much of the psac was found to be composed of biotopes with a very high species diversity. For further information on this project contact: Evanthia Karpouzli, Maritime Group, Scottish Natural Heritage, 2 Anderson Place, Edinburgh EH6 5NP. Tel: For further information on the SNH Research & Technical Support Programme contact: The Advisory Services Co-ordination Group, Scottish Natural Heritage, 2 Anderson Place, Edinburgh EH6 5NP. Tel: or ascg@snh.gov.uk

7 Contents Summary 1 INTRODUCTION Sound of Barra psac Site background Site habitat description Geological background 4 2 BROAD SCALE HABITAT MAPPING AND MARINE SACs 5 3 METHODS Scientific staff Acoustic technologies for habitat mapping Survey technologies Echo-sounder or single beam sonar Survey technologies Bathymetric sidescan Acoustic survey equipment Survey methodology Acoustic survey calibration Data recording errors Data processing Data processing errors/combined survey errors Spatial mapping of acoustics in the GIS Combination of linetrack data and full coverage data Combining line track data with spatial coverage data to produce acoustic classed seafloor maps Ground truth observations Field procedures Biological laboratory procedures Sediment laboratory analysis Satellite and air photography Aims and objectives Field methods Aerial photography Satellite Imagery 33 4 RESULTS Acoustic Maps Infaunal analysis Sediment particle size and loss on ignition results Satellite and air photography results Field methods Analysis of image quality Approaches to water depth correction Bottom classification Detecting change comparison of aerial photography and IKONOS data 68

8 5 DATA INTEGRATION Distribution patterns of biotopes Northern area Distribution patterns of biotopes Southern area 77 6 DISCUSSION General Acoustic mapping Satellite and photographic mapping Satellite and photography remote sensing field measurements Aerial photography Satellite imagery Comparisons of the relative advantages and disadvantages of aerial photography and high spatial resolution remote sensing Bathymetric mapping Change detection 86 7 REFERENCES 87 Appendix A Biotope ground truth data 90 Table A1 Table A2 Table A3(a) Table A3(b) Details of ground truth and survey stations worked during the current survey of the Sound of Barra psac 91 Biotope station records and video images for the Sound of Barra psac 103 Species abundance (no./0.1m 2 ) for infaunal samples collected by grab in the southern region of the Sound of Barra psac 109 Species abundance (no./0.1m 2 ) for infaunal samples collected by grab in the northern region of the Sound of Barra psac 125 Appendix B Acoustic Maps 141 Figure a Slope calculated from the Sound of Barra bathymetric model (north area) at 5m bins 142 Figure b Slope calculated from the Sound of Barra bathymetric model (south area) at 5m bins 143 Figure c Sidescan amplitudes from Submetrix System 2000 (north area) 144 Figure d Sidescan amplitudes from Submetrix System 2000 (south area) 145 Figure e Echoplus AGDS survey tracks (north area) 146 Figure f Echoplus AGDS survey tracks (south area) 147 Supporting Information: Scottish Natural Heritage hold all other non-published data products arising from this mapping project including raw sediment PSA data, photographic slides, video footage, completed MNCR diver survey forms, raw acoustic data and GIS products. Please contact SNH Maritime Group for further information.

9 List of tables Table 3.1 Scientific Staff involved in the 2001 Sound of Barra mapping survey 6 Table 3.2 Bathymetric sidescan resolution 17 Table 3.3 Characteristics of the IKONOS and QuickBird satellites and sensors 25 Table 3.4 IKONOS image data acquisition characteristics 33 Table 4.1 Table 4.2 Diversity and abundance for infaunal grab and core samples collected in the northern and southern regions of the Sound of Barra psac 39 Granulometric parameters associated with infaunal grab samples collected in the northern and southern regions of the Sound of Barra psac 45 Table 4.3 Calculated broad-band Blue and PAR attenuation coefficients 46 Table 4.4 Variation in reflectance (x 10 2 ) along a transect over deepwater 53 Table 6.1 Costs for varying levels of IKONOS and QuickBird multispectral data acquisition (from Space Imaging Inc. and Eurimage, respectively) for the standard 60 day acquisition window 85

10 List of figures Figure 1.1 Site location Sound of Barra 2 Figure 1.2 Sound of Barra psac 3 Figure 3.1 Schematic of acoustic acquisition hardware 11 Figure 3.2 Echoplus AGDS and Hypack Max navigation software 11 Figure 3.3 Figure 3.4 Submetrix System 2000 Bathymetric Sidescan, bow mounted together with TSS DMS-05 Motion Reference Unit 12 Navigation with dgps, forward looking sonar and echosounder on RV Serpula 13 Figure 3.5 Submetrix System 2000 and GIS QA/QC 13 Figure 3.6 Measurements of broad-band attenuation being undertaken 27 Figure 3.7 Gimble-mounted pyranometer recording fluctuations in incident 27 irradiance Figure 3.8 Measurements of in situ spectral attenuation using the spectroradiometer fitted with fibre optic probe 28 Figure 3.9 Measurement of reflectance properties of a Zostera marina sample 28 Figure 3.10 Measurement of reflectance properties of attached seaweeds in situ 29 Figure 3.11 Location of echosounder based depth transects 30 Figure 3.12 Sequence of air-photo acquisition over the Sound of Eriskay 30 Figure 3.13 Locations of GCP s (triangles) and tie-points (squares) over the Barra photo image block 31 Figure 3.14 Orthorectified mosaicked photographic dataset 32 Figure 3.15 Mosaicked and masked aerial photographic dataset 32 Figure 3.16 Distribution of ground control points used for geocorrecting the Sound of Barra IKONOS image datasets 34 Figure 3.17 Empirical line relationships for the individual IKONOS wavebands 35 Figure 4.1 Bathymetric model of northern area using Submetrix System Figure 4.2 Bathymetric model of southern area using Submetrix System Figure 4.3 Figure 4.4 Figure 4.5 Figure 4.6 3D view of sidescan and AGDS draped on the bathymetric model for the north area 38 3D view of sidescan and AGDS draped on the bathymetric model for the south area 38 Ground truth stations in the northern region of the Sound of Barra psac 40 Ground truth stations in the southern region of the Sound of Barra psac 41

11 Figure 4.7 Figure 4.8 Figure 4.9 Figure 4.10a Figure 4.10b Cluster analysis of species abundance data from infaunal samples from the Sound of Barra psac 42 Detrended correspondence analysis of species abundance data from infaunal samples from the Sound of Barra psac 43 Locations of broad-band Blue and PAR (red) and spectral (green) attenuation measurements 47 Measured spectral variation in downwelling light at 0.5m intervals at Station A 48 Measured spectral variation in downwelling light at 0.5m intervals at Station B 48 Figure 4.11 Calculated spectral attenuation coefficients 48 Figure 4.12a Figure 4.12b Figure 4.13a Figure 4.13b Figure 4.14a Figure 4.14b Figure 4.15a Measured spectral reflectance s from exposed kelp, seagrass and Cladophora sp. 49 Measured spectral reflectance s from exposed brown intertidal algal species 49 Measured in situ spectral reflectance s over algal species, using the fibre optic 49 Measured in situ spectral reflectance s over algal species, using the fibre optic 49 Measured in situ spectral reflectance s over Zostera, using the fibre optic 49 Measured in situ spectral reflectance s over Zostera, from the water surface 49 Measured spectral reflectance s from exposed bare substrate surfaces 50 Figure 4.15b Measured in situ reflectance from sand, made using the fibre optic 50 Figure 4.16 Figure 4.17 Figure 4.18 Figure 4.19 Figure 4.20 Figure 4.21 Geocorrected and masked panchromatic IKONOS image, stretched to enhance differences in the water areas 51 Close-up of panchromatic data to show noise and striping problems in the raw data, at the scale of individual pixels. Each individual pixel represents 1m 52 Close-up of same region as above after a median filter has been applied 52 Relative spectral sensitivity of the IKONOS panchromatic band (courtesy of Space Imaging Inc.) 53 Geocorrected and masked and atmospherically corrected multispectral IKONOS image 54 Close-up of IKONOS multispectral data to show level of noise in the raw data, at the scale of individual pixels. Each individual pixel represents 4m 55

12 Figure 4.22 Figure 4.23 Figure 4.24 Locations of digitised hydrographic soundings for interpolation to produce the interpolated bathymetric map 58 Depth contours for Sound of Eriskay using the triangulation interpolation 59 Depth contours for Sound of Eriskay using the radial basis function interpolation 60 Figure 4.25 Location of the transects used for the depth comparisons (E1 to E4, echosounder transects; S1, S2, sonar data transects) 61 Figure 4.26 Figure 4.27 Figure 4.28 Comparison of field-based echosounder determinations of water depths to the interpolated (radial basis function) hydrographic data. Depths are from Chart Datum 62 Comparison of depth estimates from two transects from the channel obtained using the sonar to the interpolated (radial basis function) hydrographic data. Depths are from Chart Datum 63 Image of bottom reflectance calculated using interpolated (radial basis) depth surface 65 Figure 4.29 Image of bottom reflectance calculated using Lyzenga approach 66 Figure 4.30 Figure 4.31 Figure 4.32 Figure 4.33 Figure 5.1 Figure 5.2 Figure 5.3 Figure 5.4 Spectrally based classification of bottom habitat types performed on the bottom reflectance (depth corrected) image 67 Direct comparison of the three image datasets (left, 1999 aerial photography; centre, 2001 panchromatic IKONOS image; right, 2001 multispectral IKONOS image) 69 Detected increased reflectance change (green) and decreased reflectance change (red) after comparison of the photographic and multispectral datasets 70 Close-up comparison of causeway on the south side of the channel to show changes in bottom reflectance and bottom morphology as a direct result of causeway construction 71 Distribution of biotopes in the shallow western sector of the northern region of the Sound of Barra psac 74 Distribution of biotopes in the eastern sector of the northern region of the Sound of Barra psac 75 Classification of biotopes in the shallow western sector of the northern region of the Sound of Eriskay, performed on the basis of manual interpretation and depth corrected bottom reflectance 76 Distribution of biotopes in the southern region of the Sound of Barra psac 78

13 1 INTRODUCTION This report details the studies undertaken by the University of St Andrews, Heriot-Watt University and the University of Edinburgh to survey and map sublittoral communities within the Sound of Barra possible Special Area of Conservation (psac). The aim of this project was to generate a record of the sedimentological, faunal and floral characteristics of the site. The Sound of Barra represents a particular challenge for habitat appraisal because of the existence of very shallow conditions over a large portion of the area to the north and west of the island of Eriskay. This severely limited the use of Acoustic Ground Discrimination Systems (AGDS) over much of the area and alternative techniques including airborne and satellite imagery had to be utilised. The data for the project were collected during a cruise conducted in August, 2001 using the research vessel Serpula (Heriot-Watt University) with support in the shallow (northern) areas from small inflatable boats. Ground validation of the acoustic and aerial remote sensing data was provided through diver, drop-video, remotely operated vehicle (ROV) and grab sampling methodologies. Specific objectives of the programme were to: Obtain full coverage bathymetric charts of the eastern half of the north area and all of the southern area at high resolution (sub 10m bins). Undertake an acoustic ground discrimination survey over the eastern half of the north area and all of the southern area of the site. Acquire and analyse airborne and satellite data of the western half of the northern area utilising ground truth information for image calibration. Acquire ground truth information for subsequent biotope classification. Produce biotope distribution maps calibrated by ground truth data. In addition a number of research objectives were fulfilled: An assessment of methodologies suitable for mapping the different Zostera marina and maerl bed biotopes listed under the MNCR classification scheme. An assessment of the applicability of airborne and satellite techniques for mapping shallow marine areas in clear water areas on the west coast of Scotland. An investigation of methodologies for the integration of airborne, satellite and acoustic data with ground truth information. 1.1 Sound of Barra psac Site background The Sound of Barra is situated between the islands of South Uist and Barra in the Outer Hebrides (Figure 1.1). The coastline is characterised by rocky shores interspersed with small inlets and larger sandy bays. The region experiences some of the strongest winds and largest waves in the UK. However, many of the marine communities in the Sound of Barra are indicative of a degree of shelter from the prevailing westerly weather conditions. In contrast, much of the sea bed is exposed to moderate or strong tidal currents. 1

14 Figure 1.1 Site Location Sound of Barra The Sound of Barra psac is a composite site composed of two blocks containing the major islets, skerries and sandbanks that provide haul-out and pupping areas for the common seal (Phoca vitulina) colony (Figure 1.2). The northern block encompasses the Sound of Eriskay, the southern shores of South Uist, the islands of Calvay and Hartamul and the northern shores of Eriskay. The southern block encompasses the convoluted coastline of north-east Barra and the skerries, islets and offshore islands in the Sound of Hellisay. The diversity of marine habitats within the site is a reflection of the variation in exposure and tidal streams around the islands, the range and types of substrata and the varied topography. 1.2 Site habitat description The Sound of Barra psac is notable for a wide variety of shallow sandbank habitats, including gravelly and clean sands, muddy sands, extensive beds of the eelgrass Zostera marina and tide-swept maerl beds composed of the coralline red alga Phymatolithon calcareum. The diversity and types of communities associated with these habitats are determined by sediment type and a variety of other physical factors including geographic location, relative wave exposure of the coast and differences in depth and salinity of the surrounding water. Z. marina beds and maerl beds are distinctive communities that are characterised by a disproportionately high diversity and abundance of plant and animal species in comparison to adjacent sedimentary biotopes. 2

15 Figure 1.2 Sound of Barra psac Land psac boundary Dense beds of the eelgrass Z. marina characterise sheltered medium sands in the Sound of Eriskay. The eelgrass beds constitute an important reservoir of coastal biodiversity and create a productive and diverse habitat that provides shelter and food for a wide variety of marine species. Maerl beds have previously been found to occur on sands and shell gravels in the main channel of the Sound of Eriskay, and to the south of Flodday on the north-east coast of Barra. Maerl is a term used to denote loose-lying coralline red algae, which develops when crust-forming species become free-living due to fragmentation. The habitat is of high intrinsic value, with an extremely species-rich assemblage of flora and fauna living amongst or attached to its branches, or burrowed in the coarse gravel of dead maerl beneath the surface living layer. The Sound of Barra holds one of the largest discrete group of common seals in the Outer Hebrides and the colony is representative of the region. Adults consistently use the site to rest, pup and moult. The highest concentrations of seals in the Sound of Barra are usually found on offshore islands to the south-east of Bruernish, at Greanamul and Bogha nan Sgeirean Móra and on island shores in the Sound of Hellisay. Eel Rock and rocks to the south of Rubha na Mòine in the Sound of Eriskay are also favoured haul-out areas. The shallow sandbanks and sublittoral reefs throughout the site are of considerable importance in maintaining a food supply for the seals; common seals are known to have a varied diet comprising a wide range of fish species, octopus, squid and various shellfish. 3

16 Aerial surveys of the Sound of Barra common seal population are undertaken approximately every 4 5 years as part of a routine programme by the Sea Mammal Research Unit. The coastal and marine habitats in parts of the northern and southern regions of the psac have been recently surveyed by Cordah Limited (Cordah, 2000; Comhairle nan Eilean Siar, 1999a,b,c) as a precursor to the construction of a causeway from Rubha na Mòine on South Uist to Rubha Bàn on the Isle of Eriskay and for proposed ferry terminals at Aird Mhor on Barra and Ceann a Gharaidh on Eriskay. The Marine Nature Conservation Review also located a small number of sublittoral survey stations within what has now become the psac (Connor and Little, 1998). 1.3 Geological background The Western Isles are composed of a variety of metamorphic, sedimentary and igneous rocks that range in age from Precambrian to Tertiary. The southern end of the Outer Hebrides is dominated by biotite and hornblende gneiss s of the Lewisian Complex with minor amounts of granitic gneiss s, granite and pegmatite on Barra. The Lewisian Complex consists of rocks ranging in age from billion years. Three stages are recognised within the Complex, an early phase of basic and ultrabasic gneiss development, a phase of dyke emplacement and a final phase of granitic veining. Metamorphic activity was associated with the whole of the Complex. The Complex weathers to a distinctive subdued hummocky landscape. The most prominent structural feature of the Outer Hebrides is the Outer Hebrides Fault Zone. The zone is marked by a series of fault related rock types including mylonites and phyllonites. This zone extends from Barra through Eriskay to South Uist and continues to the eastern side of the Isle of Lewis. Between Barra and South Uist a number of W-E trending dykes of Permo-Carboniferous age have been noted. 4

17 2 BROAD SCALE HABITAT MAPPING AND MARINE SACs Survey work in the subtidal marine environment has historically been based on a point sampling approach with often widely distributed locations investigated utilising diving or grab sampling methodologies. Existing data of this nature were sufficient to advise the initial SAC selection process, allowing comparisons of sites to be made across the UK. However, in order that comprehensive and defensible management plans can be developed for important marine sites it is essential to obtain estimates of the geographic distribution and extent of the biological resources in the form of broad scale habitat/biotope maps (Downie et al., 1999). Scottish Natural Heritage requires this information on the natural heritage resource for tackling statutory casework issues and for the implementation of the EC Habitats Directive (European Community, 1992) through the identification, selection, management and monitoring of areas of importance. To date 34 marine SACs have been put forward in Scotland for a number of habitat types and certain species that are listed within the Directive. The Sound of Barra has been proposed for the Annex I habitat type Sandbanks which are slightly covered by seawater all the time and for a nationally important colony of common seals Phoca vitulina, an Annex II species. To help implement the Habitats Directive in the UK, the Marine SACs LIFE Project was established ( ). One of the key tasks of this project was the identification and development of appropriate methods for recording, monitoring and reporting on the habitats and species present within marine SACs. An important element of this was the testing of acoustic based survey techniques for habitat mapping and the monitoring of long-term habitat change within SACs. A number of sampling techniques make it possible to create images of large areas of the sea floor and provide broad scale maps of seabed habitats. Broad scale implies that large areas are mapped and show the approximate disposition of broadly defined classes of habitats or biotopes. Fine scale implies that small areas are mapped to a higher level of detail and accuracy (Foster-Smith et al., 2000). Such fine scale mapping would require an intensive and prohibitively expensive survey programme. The approach taken to broad scale mapping in the marine environment is based on remote sensing. An image of an area is obtained using remote sensing techniques (inc. satellite observations, aerial photography or acoustic surveys) and information about certain attributes of the sea bed is then collected by direct or remote sampling at selected sites (a programme of ground-truth validation of the initial remotely sensed data). It must be remembered that every habitat system is in a state of continuous change and that baseline mapping surveys represent the conditions at the time of data acquisition. Work undertaken through the LIFE project and associated studies demonstrated that the repeat mapping of a near- shore SAC is achievable and that full coverage bathymetric charts can be produced with positional accuracies of 5m or better, object identification at sub-metre scale and habitat identification within 5m grids (EN, 2000; Bates and Whitehead, 2001). In the past, the mapping of areas at such a fine spatial resolution had necessitated the use of acoustic based remote sensing techniques on close line spacings with data extrapolation over areas not covered by the survey (EN et al., 2000). The broad scale remote sensing and mapping of sublittoral habitats and biota has become common place over the last 5 years with much of the work in the UK initiated in response to the 1992 EC Habitats Directive. The Marine Monitoring Handbook (JNCC, 2001) has synthesised the results from many of the individual studies undertaken. 5

18 3 METHODS Bathymetric sidescan and single beam sonar remote acoustic mapping techniques were utilised at near 100% coverage to develop acoustically classed seafloor maps of the Sound of Barra psac. High fidelity ground validation data were collected in the field using diver, drop-video, ROV and grab sampling methodologies to enable the subsequent interpretation of the acoustic maps. Each of these methodologies are discussed in further detail in the following sections. 3.1 Scientific staff A number of research scientists from all institutes were involved with the field surveys and subsequent data analysis. These individuals together with the SNH staff involved in the survey work are listed in Table 3.1. Table 3.1 Scientific Staff involved in the 2001 Sound of Barra mapping survey Scientist Field Responsibility Data Processing Responsibility Academic Institution Dr W. Austin Sediment sampling/isotope St Andrews Dr R. Bates Project Management/Acoustic Project Management/Acoustic/GIS St Andrews Ms K. Polozek Sediment analysis St Andrews Dr J. Mair Groundtruthing Biological Heriot-Watt Dr C. Moore Project Management and groundtruthing Project Management/Biological/GIS Heriot-Watt Dr D. Harries Groundtruthing Biological/GIS Heriot-Watt Dr A. Lyndon Groundtruthing Biological Heriot-Watt Dr M. Wilkinson Algal analysis Heriot-Watt Dr T. Malthus Satellite Imagery Edinburgh Mr B. James Project management and groundtruthing SNH Dr D. Donnan Groundtruthing SNH Dr S. Downie Groundtruthing SNH Dr J. Khan Groundtruthing SNH Ms E. Karpouzli Groundtruthing Satellite Imagery SNH 3.2 Acoustic technologies for habitat mapping Before using any survey technology for either mapping or object identification an appreciation of the system capabilities is needed. A key element is the system resolution or fidelity with which the system can identify objects on the sea floor. The system resolution will dictate the size of object that is recognisable at a particular distance from the survey instrument. The majority of instruments have both theoretical resolution limits and manufacturer defined values from testing under ideal conditions. Unfortunately, these are rarely achieved in real surveys. Achieving a particular resolution will depend on the precision to which the instrument can measure electronic signatures, however, it is the precision with which each measurement of the sea floor is made with a particular instrument that is of real interest to the user. A discussion of an 6

19 instrument in terms of the accuracy of object identification on the bottom can be misleading as this is a function of not only the sonar instrument specifications but also of all the navigation errors, the location errors for the instrument, the acoustic noise that is recorded and most importantly the identification or interpretation logic. Accuracy in interpretation requires consistent ground truth information and reliable positioning of both the acoustic data and ground truth data. The theoretical resolution for each survey instrument is briefly given followed by a discussion of the use of the instrument in SAC habitat mapping. Acoustic methods for habitat surveying have traditionally relied on single beam echo-sounder type instrumentation and a number of ground discrimination systems have been developed around this method (Greenstreet et al., 1997; Foster-Smith and Sotheran, 1999). More recently, sidescan sonar has been used for not only object identification but also for mapping different areas of the sea floor and classifying them by type (Curran, 1995; EN, 2000; Foster-Smith et al., 2000). In this project the latest development in acoustic techniques, namely bathymetric sidescan, was used as it has been shown to have distinct advantages in habitat mapping through obtaining near full coverage data (Bates and Byham, 2001) Survey technologies Echo-sounder or single beam sonar The echo-sounder or single beam sonar has been used for a number of decades to measure bathymetry and also to record reflecting objects such as fish within the water column. More recently, the acoustic amplitude variations have also been processed for seafloor classification (Chivers et al., 1990; Foster-Smith et al., 2000). An echo-sounder consists of a single sonar transducer that is used to both transmit and record an acoustic energy pulse directly beneath the sonar. The energy pulse or acoustic wave travels from the sonar and is reflected or echoed from boundaries in its path. The intensity of reflection depends on the impedance ratio between water and the reflector and the angle that the reflector makes with the acoustic pulse. For example, a hard, flat sea floor will reflect more energy than a soft sea floor or one that is at an angle to the transmitted acoustic pulse. The sonar produces a number of acoustic energy pulses per second as it passes over the sea floor thus measuring a line track of data. The range to the bottom and velocity of the acoustic pulse in the water will determine the number of samples or pings off the bottom per second as there is a finite time that must be observed for the energy to travel to, and reflect off, the bottom. The fidelity of recording changes on the bottom is determined by the ping rate with respect to the speed of boat over the bottom. Thus, for most surveys, not only are large areas of the bottom covered with the echo-sounder but there is also significant averaging of data between each ping. Despite this, the echo-sounder has been shown to produce high resolution, repeatable depth data along survey line tracks. However, because the echo-sounder only produces information for targets directly beneath the sonar, it is necessary to extrapolate between survey line tracks in order to produce area coverage maps of the sea floor. Seafloor classification with single beam sonar The strength of acoustic energy reflected from the sea floor with single beam sonar has also been used to classify the bottom type. The basis of all these techniques is that different amounts of energy will be reflected or scattered from the sea floor based on the contrast in acoustic impedance between the bottom type and the water column. For example, a soft bottom such as mud will have a different reflection signature than a hard bottom such as rock. In general, a hard surface will produce stronger echoes than a soft bottom or a bottom that is covered in overgrowths of biota. A number of methods have been proposed for this and 7

20 include those by Jackson and Briggs (1992) who used the backscattered energy from the echo to infer bottom roughness. Orlowski (1984) used a method which integrated parts of the multiple echo signature from the sea floor to provide information on the seafloor characteristics. Burns et al. (1985) developed a classification system based on the first echo and the second echo or first multiple echo from energy that bounces between the sea surface and the sea floor. Two commercial systems, RoxAnn and the Echoplus have been developed from this work and it is recommended that the systems are used with transducer beam widths of The first echo has been related to the roughness of the sea floor with the rougher the sea floor the more energy that is backscattered to the transducer. The second echo is interpreted in two ways. Chivers et al. (1990) suggest that the dominant ray paths for the second echo undergo two reflections at the sea bed and a single scattering at the sea surface. The amount of energy returned is related to the acoustic impedance contrast at the sea bed and thus a harder bottom will reflect more energy. A second theory proposed by Heald and Pace (1996) envisages the transmitter-receiver configuration as a bi-static system but with the energy reflected still dependant on the hardness of the bottom. As the second echo is reflected by the sea surface it can be influenced by the sea state especially in rough weather. QTC View is an alternative system based on single beam echo sounders that uses only the first echo (Simpkin and Collins, 1997; Collins and McConnaghey, 1998). The technique records both transmitted and received waves in order to perform compensations for beam spreading before Principal Component Analysis (PCA) is used to identify key parameters of the echo shape (Collins et al., 1996). This is equivalent to analysis using the Cartesian plot in RoxAnn but with more variables. A problem that has been experienced in many recent surveys using echosounders for seafloor classification has been the repeatability of measurements during a single survey and between surveys using different sonars and different survey vessels. Numerous authors (for example Foster-Smith et al., 2000) have noted a drift in instruments as temperature and humidity changes, and a different response has been noted from different sonars when used on different vessels. Some of these issues have been addressed with careful quality control during the survey and some issues have been addressed with new instrument development such as the Echoplus. The Echoplus has digital compensation adjustments within the instrument for frequency variations, depth from the bottom (signal strength losses), pulse length differences and power level fluctuations. None-the-less, procedures are recommended where repeat surveys are made over two or more areas of known sea floor in order to calibrate the amplitude response. Acoustic maps of seabed character from AGDS The final output of single beam echo sounders is a depth chart and line plot of backscatter or reflection characteristics of the sea bed. In order to produce a map of bottom character it is first necessary to extrapolate the line data into a grid of values. This interpolation marks the point in the interpretation from where the investigator can influence the outcome by their choice of processing parameters. In order to ensure a robust interpretation it is vital at this stage that the processing is conducted under the close scrutiny of both survey and biological experts. The extrapolation can also potentially introduce large errors into the system as the assumption must either be made that conditions vary uniformly between real data points or alternative methods of finding boundaries must be invoked. The numerous methods for achieving full coverage maps and predictions from them on likely distribution of different seafloor types has been extensively studied over the last 5 years (Foster-Smith et al., 2000). While the use of AGDS has been proved highly effective in mapping habitats, this project tested the use of AGDS together with full coverage information from 8

21 bathymetric sidescan systems so that extrapolation of the AGDS was not necessary across unknown boundaries. In general, once an acceptable extrapolation of single line track AGDS data has been made, it is possible to apply image processing procedures and modelling with geographic information systems based on acoustic signature correlation to known bottom type. The final step in image classification is the prediction of likelihood of finding similar or dissimilar bottom types across the survey area Survey technologies Bathymetric sidescan A bathymetric sidescan system is one that is used to measure the depth to sea floor and amplitude of sonar return from the sea floor along a line extending outwards from the sonar transducer at right angles to the direction of motion of the sonar (Geen, 1998). As the sonar platform moves forwards, a profile of sweeps is defined as a ribbon-shaped surface of depth measurements known as a swath in a similar manner to a sidescan image of the sea floor. The acoustic signal is produced by the sonar in a similar manner to a sidescan acoustic pulse and is narrow in azimuth (that is, viewed from above), and wide in elevation (viewed from the side). The difference between a sidescan and bathymetric sidescan system is in the recording of the acoustic energy. In a bathymetric sidescan system a number of transducers or transducer staves are used to record the returned energy that is back-scattered from the sea floor. When this back-scattered sound is detected at the transducers, the angle it makes with the transducer is measured by recording the phase difference between transducers and a reference signal. Multiple staves ensure that both the angular measurement and the overall phase resolution are measured with high precision. The range for a reflector is calculated from the travel time to the reflector and back and the range and phase angle pair enable the location of the ensonified sea bed patch to be known relative to the sonar transducer thus creating a 3D bathymetry map of the sea bed. A typical far range limit is about 7.5 times the water depth giving a total swath width of approximately 15 times the water depth. In addition to a determination of the location of a reflecting target on the sea floor, the amplitude of the returned signal can be measured with the bathymetric sidescan system. These amplitude data can be used in one of two ways. Either they can be treated as a sidescan record, that is as a time series to produce a qualitative image, or geo-referenced picture of the sea floor or they can be processed using the bathymetric information for the point on the sea floor from which each individual reflection is measured. In this latter case the recorded amplitude is compared with the source signal after compensation for energy losses during the travel path such as loss of signal to the water column, spherical beam spreading and the incidence angle for scatter or reflectance from the sea floor. It is only since the development of this type of sonar with high fidelity co-location of bathymetry and amplitude that these compensations for amplitude loss have been possible. Acoustic maps of seabed character from bathymetric sidescan The final output of a bathymetric sidescan is two products a bathymetric chart and a sidescan or amplitude map (Bates and Byham, 2001). The bathymetric chart is calculated from all of the points on the sea floor where a reflection signal was obtained. In gently undulating sea floor the points form a continuous cover at sub-metre fidelity. However, in areas of large bathymetric variation, some areas will have a higher density of reflection cover than others. Furthermore, it is possible that some areas, those in the shadow of large upstanding features on the sea floor may have no reflection points on them. Great care must be taken in 9

22 surveying to ensure that this condition is kept to a minimum. For the majority of sites, the bathymetric sidescan technique provides better than 95% coverage of the sea floor. Using a bathymetric map alone, even with the high resolution obtainable with the bathymetric sidescan, for habitat mapping is not recommended as not all seafloor sediment or biological zones are depth defined. However, the broad range of many individual species can be limited by depth and thus the maps provide a very useful additional tool for habitat appraisal. In addition, because the resolution of the bathymetry from these techniques is high, it is possible to use very small scale features to aid in mapping boundaries between contrasting surfaces such as rock to sediment where small slope changes become very apparent. The second product of the bathymetric sidescan is the amplitude map. This map can be produced in two forms. In the simplest version, only amplitude data are preserved where an acceptable bathymetric value has been recorded. As many of the bathymetric data points are filtered out during processing this decimated data set may loose much of the texture information that is important to a high contrast sidescan image and thus a second data set is usually also preserved where all the amplitude data are saved and separately processed. Again two routes are available for processing these data. In the first, the sidescan recorded is treated like a typical sidescan data set with amplitude correction based on time varied gains or some form of angular corrections (eg Danforth, 1997). The single swaths of amplitude data are mosaicked to produce an amplitude map or geo-referenced image that can be overlaid with the bathymetric information thus providing a powerful basis for interpretation of seafloor type. As the exact position and attitude of the sonar are known this approach offers a new way forward for broad scale mapping projects that hitherto was not possible with the sidescan and single beam line track AGDS data of the past. A further method of amplitude data processing is currently being investigated at a research scale in St Andrews with collaborators from the University of Oxford and Trondheim Marine Laboratory. In this method, alternative information is processed from a combination of the amplitude data that has been corrected for the bathymetric model together with wavelet signature extraction Acoustic survey equipment Both AGDS and bathymetric sidescan data were simultaneously acquired during this project with all information collected on a single PC system. An additional PC was used for navigation during the project and this was connected in a mini-network with the acoustic acquisition PC. A further PC was used for regular quality control, running the GIS and for downloading the acoustic data at the end of each survey day for processing during the evenings. A block diagram of the acquisition set up is shown in Figure 3.1 and each individual component is discussed in further detail below. AGDS The AGDS system chosen for this project was the Echoplus manufactured by SEA Ltd. (Figure 3.2). The Echoplus was chosen in order to have the ability to use more than one frequency echosounder simultaneously for recording AGDS and also as the electronics within the instrument offer the latest in digital signal processing for consistency and repeatability of measurements. The Echoplus was coupled to either a Furuno FCV292 dual frequency echosounder with 28kHz and 200kHz transducers or a Lowrence 300A with 50kHz transducers. The use of alternative frequencies with different echosounders was undertaken in order to test AGDS methods for discriminating particular biotopes. 10

23 Figure 3.1 Schematic of acoustic acquisition hardware Figure 3.2 Echoplus AGDS and Hypack Max navigation software 11

24 Bathymetric sidescan The bathymetric sidescan used for this project was a Submetrix System 2000 with 117kHz transducers. Acquisition settings varied with transmit lengths of cps (77 424µsec), a ping rate of 3 5 per second and 2048 sample receiver length. Sound velocity measurements were acquired at the site but no depth stratification was noted and thus velocities of 1498ms -1 were used for ray tracing throughout the site. The transducers were bow-mounted on the survey vessel with the motion reference united permanently fixed immediately above the transducers (Figure 3.3). Figure 3.3 Submetrix System 2000 Bathymetric Sidescan, bow mounted together with TSS DMS-05 Motion Reference Unit 12

25 The motion reference unit was a TSS DMS-05 dynamic motion sensor which used solid state sensing elements to measure instantaneous linear accelerations and angular rates of motion change to 0.05º. This information is critical to correct positioning of seafloor reflection positions especially at far offsets at the end of each swath. The information from the DMS-05 is supplemented by navigational input from the differential Global Positioning System (dgps) and also a magnetic compass. The magnetic compass used was a Aximuth 1000 produced by KVH Industries, Inc. This fluxgate digital compass provides azimuth information to 0.5º accuracy after compensation. All data were recorded on a PC with RTS2000 acquisition software (SEA Inc.). Navigation Navigation was accomplished using Hypack Max Survey software supplied by Coastal Oceanographics Inc. with background charts from C-Map Norway (Figure 3.4). Real-time positioning was accomplished using differential GPS from the Racal Landstar system Mk III receiver. This provided continuous correction data from a sequence of base stations around the coast of Scotland relayed via satellite to the Landstar for positional resolution of less than 2m. The survey navigation was accomplished on a separate PC that was networked with the sonar acquisition computer and also the QA/QC computer with the GIS (Figure 3.5). Figure 3.4 Navigation with dgps, forward looking sonar and echosounder on RV Serpula Figure 3.5 Submetrix System 2000 and GIS QA/QC Survey methodology The survey area was initially divided into a number of zones based on known bathymetric variations and anticipated weather patterns during the survey period. This ensured that natural land features, and the shelter they create, could provide optimum survey conditions with the minimum swell, wave and wind action. The survey line spacing was then chosen that would give a minimum of 50% overlap on the bathymetric sidescan data. The line spacing was further reduced where either the sea floor type changed rapidly or there 13

26 were specific sea floor biotope characteristics of important interest. Surveys were conducted by the helmsman following a course indicated on the navigation computer. The acoustic data were continually monitored on the acquisition computer and a bottom coverage map produced in real-time in order to ensure full sea floor coverage. No attempt was made to obtain ground truth information during the acoustic survey rather the ground truth locations were chosen following preliminary analysis of the acoustic data. All survey data were acquired in the field on the acquisition computer hard drive and also backup disks were made on a magneto-optic drive Acoustic survey calibration If acoustic data are to be acquired over a number of days, and furthermore if the data are to be compared to previous and subsequent acoustic data, it is of paramount importance that careful calibration of the instruments is undertaken on installation. Calibration of the AGDS and bathymetric sidescan was achieved using the following procedures: AGDS The AGDS was calibrated for depth by repeat surveying of areas of known depth over different states of the tide. At the beginning and end of each day a 2 minute record was made of these data near the vessel mooring site. Amplitude data for E1 and E2 were also recorded at the beginning and end of each day over the same section of sea floor for comparison of E1/E2 values. An example of the results of these data over a two day period are shown in Box 1. Bathymetric sidescan Roll calibration An area of sea floor that was relatively flat was chosen for the roll calibration. Across this area, 5 survey lines were acquired with 100% overlap of port and starboard transducers between the lines. These data were then processed and compared thus allowing adjustment of the transducer angles to give coincident reflections to less than 0.05º. Skew calibration was accomplished by using recognisable objects on the sea floor and surveying them with both transducers at different offsets. Pitch calibration was achieved by surveying up and down a relatively uniform slope. Known objects were recognised on the slope and these used to calculate the angular pitch calibration. Once calibrated for roll, pitch and skew, amplitude variations within the Submetrix system are recorded in the data and therefore can be analysed and compensated for at the processing stage. None-the-less, it was still survey practice to record and review data at the beginning and end of each day over known seafloor conditions while the data were being acquired for AGDS calibration. Tidal corrections Tidal corrections were applied to both the bathymetric sidescan data and the AGDS from 10 minute tide curves modelled using information from the Admiralty Tide Tables and the Hydrographic Office. Where data were not available for ports or areas within the survey site, alternative locations were used and these calibrated with local information. The tidal models were entered directly into the navigation computer and to the bathymetric sidescan acquisition software during acquisition so that real-time corrections could be made. 14

27 Box 1 Daily Comparison of Echoplus (AGDS) E1/E2 values on 7/8/01 and 8/7/01. The Echoplus line track data for E1 and E2 are combined to give a total value that is colour coded into a range of values between and

28 3.2.6 Data recording errors AGDS There are a number of potential system errors that are generated with AGDS but many of them relate to the particular use of AGDS together with the navigation errors, and style of deployment (line spacing, water depth survey speed). These combined errors are discussed in further detail below. Potential errors resulting from the echosounder and the Echoplus and their impact on final survey resolution are discussed below. There are two main sources of error with the AGDS, namely the precision with which depth can be measured and the resolution of the amplitude measure on the bottom. The depth resolution is a function of the echosounder frequency, pulse width, digitising rate and water depth. For most environmental purposes this typically gives errors of depth well within the overall survey errors for water depths between 3 150m. The resolution of the amplitude measure is also a function of the echosounder but is controlled by the beam width and sample rate. The beam width is set by the manufacturer and the sample rate is dictated by the water depth or time of travel for an acoustic pulse between the echosounder and the sea floor. Typical beam widths of between 8 20º will result in very different areas of ensonification on the sea floor and thus different degrees of fidelity to which the sounder can map boundaries between different seafloor conditions. Bathymetric sidescan The range in a bathymetric sidescan is measured using travel times to typically better than 0.05m and transducer angles to better than 0.05º. As the transmit beam spreads in the water away from the sonar in a similar manner to the sidescan, the size of the footprint will also increase. Thus a footprint can be calculated with a 234kHz sonar to about 0.87m at near range and 5.2m at 300m range along track and 5cm across- track. The 117kHz transducer has an along- track footprint of 1.5m at near range and 8.9m at 300m range with a 7.5cm across- track dimension. Because phase difference is recorded, a major advantage is realised with use of the bathymetric sidescan sonar in that there is no footprint spreading along the beam ie in the across- track dimension. However, it should be noted that it may not be possible to achieve these across- track dimensions in practise at far offsets due to energy loss. Details of the bathymetric sidescan resolution are given in Table 3.2. The maximum range limit is dictated by the nature of the sea floor and the grazing-angle limit where most of the energy is reflected away from the sea floor. Bottom types such as soft mud or peat can reduce the expected range by as much as 30%. Sand, rock and shingle all give good sonar backscattering. For seafloor classification this is an important issue as it is vital that similar size areas of the bottom are surveyed across a sonar record in order to be able to make meaningful comparisons. It should also be remembered that if there are slopes on the sea floor that fall away from the direction of the sonar beam, these areas will fall into shadow zones and it is unlikely that they will be ensonified. Thus once more, obtaining true 100% coverage of the sea floor is rarely achieved. Similar to the sidescan sonar, the number of pings or hits on a target is defined by the ping rate and speed advance of the sonar over the sea floor of survey. The ping rate is determined by the furthest range limit and speed of beam in the water. High survey speeds will result in poorer target definition or poorer quality images of the target. 16

29 Table 3.2 Bathymetric sidescan resolution Across- track and Along- track resolution Across- track range (m) Along- track range (m) Frequency (khz)/beam width (º) / / Distance between pings (alternate pinging for port and starboard transducers) Range (m) Survey Speed (kts) Data processing AGDS The AGDS data were recorded in line data format using the same PC as for the bathymetric sidescan. In the field these data were extracted from the bathymetric sidescan data for plotting as unedited E1/E2 values in Arcview in near real-time. This allowed for site quality control on the data and also provided information for locating ground truth sites. Subsequently a number of line editing functions were conducted on the data using simple spreadsheet editing functions. Depth Editing this was used to highlight erratic changes in depth where large jumps in depth (greater than 5m) were evident between individual records Navigation Jumps instability in dgps can sometimes cause large navigation errors to be recorded in data. These were removed by comparison of positions along track Erratic Changes in E1/E2 large changes in E1/E2 were edited together with values at either extreme end of the range of possible values for E1/E2 Bathymetric sidescan The bathymetric sidescan data were processed using the acquisition software RTS2000 on-line during acquisition as the speed of sound profiles, calibration settings and tidal information had been calculated and input before commencing the survey. This enabled preliminary bathymetric models of the site to be produced during the field work. In the field, the data were processed using broad bathymetric filters with large depth acceptance windows (+/-5m) therefore subsequent to acquisition all the data were re-processed in order that the bathymetric filters could be refined. Processing of the data at this stage involved the following steps: Input corrected bathymetric data Filter for along- track and across- track anomalies to 1m bins 17

30 Export navigation filtered data (filters out large navigation jumps) Import processed data to mosaic programme Grid 2000 Filter and smooth data to 5m bins Export grid data at 5m bin resolution for whole survey area and 1m bin resolution for specific areas of interest Sidescan The sidescan data were processed separately to the bathymetry data post-acquisition using SonarWeb Pro (Cheasepeake Inc). SonarWeb Pro uses amplitude corrections to the amplitude time series based on the work of the USGS (Danforth, 1997). The processing method incorporates the following steps: Import raw data from the bathymetric sidescan together with the full navigation information. The lines are imported at the desired output resolution to match the bathymetric model 5m for the whole site with 1m and 0.25m for specific areas of detail Geometrical correction and amplitude adjustment for offset angles from the transducers. Nadir is removed using bottom tracking algorithms with manual adjustment in areas of rapidly changing bathymetry Line projection onto the relevant datum (OSGB36) and overlapping data is combined to give a mosaic of the whole site. Overlap data points are averaged to give mean amplitude values from all crossing tracks Final output in the form of geo-referenced TIFF and geo-referenced JPEG files Data processing errors/combined survey errors Mapping error The production of the final predictive maps is subject to further errors as a result of the individual errors from each system (the acoustics and the ground truth observations) and the inherent approximations necessary when combining the results. The final error can be thought of as the resolution of the maps. Here resolution is used to refer to the level of detail to which habitats or biotopes are discriminated. This level of detail is therefore a combination of both absolute detail and combined errors within recording systems and the level of interpretative discrimination that it is possible to put to an observation of biotope sequence. Finally, the map output has a finite resolution in both paper form and electronically within a GIS project. Error in ground truth position All of the ground truth positions are subject to error from the positional error for the dgps (typically less than 2m) and more importantly from the position of the direct measurement (grab, diver, video or ROV) with respect to the vessel and dgps. The uncertainty of the position of the ground truth sampling with respect to the vessel is related to the depth of sampling. In general, the deeper the sampling the greater the uncertainty especially in strong drift conditions arising from currents and wind forces. For typical surveys, the length of cable paid out for a grab, video or ROV is 1.25 times the water depth and a very approximate position for the sampling device might be within a circle centred on the vessel that has a radius of half the water depth. 18

31 Track spacing Prior to it being possible to obtain close to 100% seafloor coverage with acoustic techniques, it was necessary to survey areas with close line tracks in order to record small changes in bottom type across track. Methods for extrapolation between lines were then applied to the data based on the spatial correlation of the data. When lines are close together relative to the heterogeneity of the sea floor then the particular method of extrapolation between lines is of little consequence, however, when the lines are widely spaced or there are significant gaps in the lines then the final results become extremely sensitive to the extrapolation method. Numerous methods of extrapolation have been tested such as distance-weighting and kriging but all must assume some form of averaging and smoothing of change between known data points and thus the mapping of discrete boundaries is difficult. Because of these shortcomings, all the line track data from AGDS acquired in this project were integrated with the bathymetric sidescan which allowed the line data to be extrapolated using knowledge of the seafloor changes from the continuous coverage data. This represents a significant advance in technologies for remote monitoring using acoustic methods. AGDS The maximum resolution of the AGDS is a combined function of the particular echosounder used (its frequency and beam width) which defines the acoustic footprint or ensonified area, the water depth and speed of sound in water which defines the number of pings recorded per second, and the vessel speed over the sea floor which defines the spacing between the ensonified patches. Thus for a vessel working in 10m of water at 3ms -1 with a beam angle of 15º and a dgps error of less than 2m AGDS values could be recorded 3 times a second giving an ensonified area of 5.4m 2. Values would be recorded at 1m intervals across the sea floor and overlaps of 30% would be achieved between readings. Thus discrete boundaries on the sea floor could be recorded to +/-3m. Increases in speed of travel, depth or ping rate will decrease this value. Bathymetric sidescan While the minimum seafloor ensonification area is relatively small with bathymetric sidescan sonar (less than 1m 2 ), when the sonar is used in practice together with navigation error and with averaging between swaths, it is more practical (in terms of processing size and run time on typical computers) to bin data at a minimum of 2m 2 for the 117kHz transducers with large areas binned at 5m 2, 10m 2 and 20m 2 for working models. One advantage of the bathymetric sidescan, however, is the fact that this seafloor resolution can be maintained at all water depths for the bathymetric information. The sidescan information can also be presented at a range of scales depending on the size of area that is being analysed. Typically bin sizes of 1m 2 are used for the study of large areas but this is reduced to 0.25m 2 for smaller areas of particular interest. As all the bathymetric information is surveyed with the a motion reference unit with angular resolution to less than 0.05º, relative error in positioning within any part of the bathymetric sidescan model is typically less than 0.5m, however, due to dgps errors, the absolute survey bin position error is less than 2m Spatial mapping of acoustics in the GIS Combination of linetrack data and full coverage data The output from the ADGS line track data is an edited file containing position, depth, and E1/E2 values. The output from the bathymetric sidescan are an edited file containing position and depth, 19

32 and a georeferenced amplitude image. These text files are input to the standard GIS package, Arcview, together with other background information such as OSGB land DEM (digital elevation model) and admiralty charts. Following this, a number of procedures are applied to the data in order to produce final maps of acoustics and seafloor type. Creation of a Digital Bathymetric Model (DBM) The digital bathymetric model was created using a triangulated irregular network (TIN) to represent the seafloor surface. The TIN represents a surface with vector features (points, lines and faces) and thus it can precisely model discontinuities in the surface with breaklines. This is important in analysis of the sea floor as it is anticipated that many significant changes in seafloor type will occur along discrete boundaries such as breaks of slope for example between a submerged rock cliff and the sediment plane at the bottom of a loch. A disadvantage of the TIN is that it cannot represent vertical cliffs, overhangs or caves. However, because the faces in the TIN can be defined as a plane, a slope and a slope direction, it is possible to calculate secondary maps from the TIN for seafloor aspect and slope. Both slope and aspect together with depth can be important for controlling biotope type. The TIN is produced from the continuous coverage grid file of bathymetric information with a boundary set at low water mark projected from the OSGB DEM. From this map the slope aspect and slope angle are calculated. The slope angle map is defined with a scale range expanded for small slopes (between 1 15º). Sidescan image data No processing of the sidescan image data was necessary in the GIS as the images are georeferenced raster files at the spatial resolution with which they were originally created using the sidescan sonar mosaic programme. This resolution varied between 1m bins to 0.25m bins for particular areas of interest. AGDS line track data The AGDS line track data are entered as a table of values and plotted in the GIS with single data points representing each acoustic set of E1/E2 values. In the field, the E1 and E2 data were combined in order to produce an in-field rapid assessment of the data range. This was achieved by taking the square root of each value, in order to expand the low end values below unity that dominate the acoustic returns over smooth soft seafloor material, and summing these values. This method of analysis does not, however, do justice to the information that is present in the individual variations in E1 and E2. A more critical examination of the data is necessary and it is usual to first produce a scatter plot of E1/E2. The scatter plots typically show a broad trend of data from smooth and soft sea floor to rough and hard sea floor. In the GIS, both the E1 and E2 data are the plotted as separate line tracks and each set of data is classed based on natural breaks in the data that were calculated from variogram analysis of the data ranges. A further discussion of methods for classifying the AGDS data where ground truth information is assimilated into the classification is given later, however, for the preliminary survey maps it was decided that the acoustic data should be analysed and combined to give a representation and classification of the sea floor based entirely on the acoustic information. 20

33 Combining line track data with spatial coverage data to produce acoustic classed seafloor maps The production of final acoustic classed seafloor maps was achieved through combining information from the following set of maps: DBM Slope angle Sidescan image AGDS line track Changes in AGDS class type were noted and these were compared to the bathymetric model, the slope angle map and the sidescan image. These maps provided an explanation for the change in AGDS such as a textural change from the sidescan image or a change in slope from the bathymetric model and slope map. The feature was then traced to create a separate polygon or Arcview Shape File that contained all of that discrete class of AGDS data that fell within this zone defined by the full coverage data. The process was repeated for all changes in AGDS down to the smallest class change that could be identified as a discrete feature on the bathymetric model or sidescan image. The result is a set of shape files of different AGDS class from E1 and E2 projected across the full DBM for the site. This range of classes represents the acoustic final map product but can also be viewed in 3D for better representation of the relationship between seafloor type and bathymetry. 3.3 Ground truth observations Field procedures Details of the stations used for groundtruthing the acoustic and satellite surveys are provided in Table A1 (Appendix A) and their locations mapped in Figure 4.5 for the northern region and Figure 4.6 for the southern region of the psac. The distribution of ground truth stations was dictated by the requirement to relate biotopes to the acoustic classification, to give good geographical coverage over the entire psac, and, in the case of the shallow waters in the north, to provide good coverage of areas likely to support maerl and Zostera marina biotopes. Cognisance was also taken of previous MNCR and Cordah surveys in the area. Biotope groundtruthing of the Sound of Eriskay and the shallow area to the west of the causeway was carried out principally from two RIBs employing non-differential GPS for position fixing. The main techniques used here were drop down and snorkel diver video and glass-bottom bucket observations. A drop down digital video system was deployed at 34 stations for 3 5 minutes from a slowly drifting vessel. Digital video was also taken at 15 stations by snorkel diver for c.15 minutes, with snorkel diver observations only on substrate and biota abundance s at a further 3 stations. Notes on the substrates and biota were also recorded for 16 stations from observations through a glass-bottomed bucket from a slowly drifting vessel for 3 10 minutes. MNCR phase 2 diving surveys were carried out at 5 stations in the north and 15 stations in the southern region of the psac from RIBs and RV Serpula. The level of survey quality was set at adequate for both 21

34 fauna and flora. In situ recording was supplemented by the collection of biological specimens and by digital video at 6 stations. MNCR survey, site and habitat forms were completed. The SNH Highball ROV was deployed at 13 stations in the north and 10 stations in the south from RV Serpula using dgps position fixing. Video footage was recorded at each station for minutes. Samples for macrobenthic infauna were taken at 23 stations. At all these stations data on epibiota were also available from diver or ROV observations, thus permitting an holistic impression of the biotopes to be acquired, which could be used in the ascription of biotopes to stations where no infaunal data were available. At four of the MNCR phase 2 stations in the shallow northern sector, eight 10 15cm long cores of diameter 10.2cm were taken by diver. At ten stations in the north and nine stations in the south single 0.1m 2 Van Veen grab samples were taken from RV Serpula. A subsample of around ml of sediment was taken from each sediment sample for grain size analysis before sieving the remainder through a 0.5mm mesh. The screenings were preserved in borax-buffered 10% formalin. All depths were converted to depths relative to chart datum using the Admiralty Tidecalc program, with Castle Bay as the chosen port. All position fixing used the WGS84 datum Biological laboratory procedures Examination of the video material involved recording details of the substrate type and estimations of the abundance of the biota, using the SACFOR scale (Connor et al., 1997) where possible, for each distinct biotope. This information, together with allocated biotope, depth, position, date, time and sampling gear was entered into an Excel spreadsheet, which also contained similar data for the non-video ground truth methods. Biological material from the infaunal grab sampling programme was identified to species level (with the exception of a few problematical taxa) and enumerated by Environment and Resource Technology, Edinburgh, who also supplied a reference collection of species. Analysis of the species abundance data involved the determination of total abundance, Shannon-Wiener species diversity (using both log 2 and log e ), Pielou evenness and species richness (as the number of taxa per sample). To aid in the allocation of biotopes the log-transformed infaunal species abundance data were also subjected to multivariate analyses. Sample similarities were investigated by cluster analysis using the Bray- Curtis similarity coefficient and group average sorting. Trends and similarities amongst the samples were also examined by two ordination techniques, to ensure that only robust patterns were identified: non-metric multidimensional scaling, using the Bray-Curtis dissimilarity matrix, and detrended correspondence analysis. Similarities between stations revealed by multivariate analyses, together with consideration of the raw species abundance data and species records from the observational groundtruthing methods, were used, in conjunction with depth, substrate and exposure, in the allocation of biotopes. The biotope classification scheme followed was that of Connor et al. (1997). The distribution of biotopes amongst survey stations was tabulated in two ways. In Table A1 (Appendix A) the data were ordered by station and then biotope. Positional and physical data are also provided within this table. Table A2 (Appendix A) lists the stations where each biotope was recorded, together with video frame dumps (supplied by SNH) of the biotopes. 22

35 Substrate and biotope data for the survey stations were entered into an ArcView GIS to aid in the groundtruthing of the acoustic survey. The GIS was also used to develop maps of the distribution of infralittoral and circalittoral coastal fringing biotopes, to aid in the description of the distribution of biotopes throughout the psac and to provide maps of the location of survey stations Sediment laboratory analysis Particle Size Analysis (PSA) Sediments, removed from the Van Veen grab samples, were processed for particle size analysis (<2mm) from wet sediment sub-samples. These materials were refrigerated at 4 C prior to analysis. All of the analyses were made using a Coulter LS230 Laser Particle Sizer at the School of Geography and Geosciences, University of St Andrews. Three sediment sub-samples were measured according to the following processing methods (see Austin and Evans, 2000 for details): 1) Raw Sample. Approximately g of sediment was added to the Coulter Counter and the particle size distribution determined after initial ultrasonification of the sample (60 seconds). 2) Organic-free Sample. Approximately g of sediment were treated with 10ml Hydrogen Peroxide solution (20% vol.) to remove organic material. The process was accelerated by heating the sample on a sand-bath (approximately 80 C) and repeated if the reaction continued after 2 hours. The samples were subsequently centrifuged and washed with deionised water; this rinsing procedure was repeated twice. Prior to measurement, 2ml of a Sodium Hexametaphosphate solution was added. The particle size distribution was determined after initial ultrasonification of the sample (60 seconds). 3) Biogenic-free Sample. In order to remove all the biogenic components from the sediment sample, the organic component is removed as in step (2) and was subsequently treated with Hydrochloric Acid (20% vol.) to remove Calcium Carbonate. Many of the sediment samples were coarse, shell-rich gravels and sample weights of up to 10g were required. After acid treatment, the samples were centrifuged and washed with deionised water; this rinsing procedure was repeated twice. Prior to measurement, 2ml of a Sodium Hexametaphosphate solution was added. Again, the particle size distribution was determined after initial ultrasonification of the sample (60 seconds). 4) For material coarser than 2mm, dry sieving was accomplished. The samples were oven dried for 24 48hrs and then subjected to sieve analysis using the following rages of sieve/grain size: >63 125µm, µm, µm, µm, µm and >2000µm. Each of the above sub-samples were analysed three times. Between analyses, the system repeats a 60 second ultrasound agitation of the sample in order to separate sediment aggregates. Where there is a clear and progressive dissaggregation of the sample, the third and final particle size distribution has been used in any further consideration of the sample. Loss on ignition measurements Approximately 4 5g of wet sediment were accurately weighed and heated overnight at 100 C. Samples were transferred to a desiccator to cool. The weight loss was calculated by re-weighing the sample and the percentage weight of water determined. The same samples were then heated to 450 C for 4 hours and 23

36 transferred to a desiccator to cool. The weight loss was calculated by re-weighing the sample and the percentage weight of organic matter determined. A further heating step to determine the weight loss of biogenic carbonate involved exposing the same sample to a temperature of 850 C for 30 minutes. 3.4 Satellite and air photography Given the extensive nature of Scotland s marine environment techniques are required which facilitate the broad scale mapping of seabed habitats and meet the requirements for monitoring on a routine basis. Remote sensing from satellites or aircraft offers a non-invasive technique with which to rapidly monitor changes in the cover and health of submerged habitats and which might be of significant complementary benefit as a tool for monitoring shallower water environments around Scotland's coastline which might be more difficult to survey by other techniques. The principal advantage of remote sensing in general is that it collects certain information more uniformly in both time and space, and over large areas. As the data are digital in nature they can be calibrated using a small amount of fieldwork. Optical systems may prove highly cost-effective for mapping shallow habitats up to 30m depth in clear waters over large areas (Mumby et al., 1998). However, the full potential of remote sensing is still to be exploited in sublittoral environments, where under certain situations, the strong attenuating influence of the water column has been a limiting factor. In order to accurately map marine sublittoral zones data are required from remote sensors which offer the following key characteristics: high spatial resolution, to match the scale of the variations in subsurface habitat types high radiometric resolution, to offer greater ability for discriminating habitat types in deeper areas high spectral resolution, particularly with respect to bands in the blue-green region of the visible spectrum which penetrate furthest through the water column. Conventional space-borne optical sensors (eg Landsat TM, SPOT) can only offer coarse spatial resolution (20 30m), poor radiometric resolution (256 measured radiance levels) and, in the case of SPOT data, limited spectral resolution (lacking a blue band). Data from these sensors are thus limited for mapping and monitoring coastal habitats with any reliability. However, recent advances in optical satellite technology offer considerable promise for mapping marine habitats at scales acceptable to conservation agencies. For example, the unique data that are now provided by the IKONOS and QuickBird sensors (Table 3.3), at both high spatial and radiometric resolutions, will finally allow for the routine monitoring of shallow marine habitats by remote sensing to become a reality. All remotely sensed measurements of reflected radiance over submerged substrate are influenced by water column effects (depth and attenuation) which ultimately affect the accuracy with which spectral classifications of individual species can be performed (Mumby et al., 1998). Over the last two decades a number of equations have been proposed which have allowed for either the calculation of water column depth, or in revised form the calculation of bottom reflectance after accounting for water column depth and attenuation (eg Lyzenga, 1981; Moussa et al., 1989; Bierwirth et al., 1993). Although uptake in the use of these approaches has been slow, the use of water column correction is now considered standard for the objective measurement of habitat change using remote sensing for routine monitoring of bottom habitats. Such a process has been shown to significantly improve the accuracy of classification of these habitats (Mumby et al., 1998). 24

37 Testing of these algorithms in UK coastal waters is required; we are not aware of any tests of these column correction algorithms in European coastal waters. Work by Karpouzli et al. (2002) has shown that spatial variation in attenuation may vary considerably which suggests direct measurements of water attenuation are required. Table 3.3 Characteristics of the IKONOS and QuickBird satellites and sensors Characteristic IKONOS QuickBird Launch: September 1999 October 18, 2001 Altitude: 681km 450km Orbit: 98.1 degrees, sun-synchronous 98 degrees, sun-synchronous Imaging modes: Panchromatic and Multispectral Panchromatic and Multispectral Spatial resolution: 1m (Panchromatic) 4m (Multispectral) 0.6m (Panchromatic) 2.6m (Multispectral) Spectral resolution: Four bands (Blue, Green, Red, Four bands (Blue, Green, Red, Near infrared) Near infrared) Revisit Frequency: 3 days n/a Radiometric resolution: 11 bit (2048 radiance levels) 11 bit (2048 radiance levels) Swath: 30km 16.5km Other features: Pointability Pointability Aims and objectives As part of the wider marine mapping project undertaken in selected Scottish marine areas during 2001 using underwater acoustic techniques and in situ observational and seabed sampling techniques for groundtruthing, this work evaluated IKONOS sensor and other optical data from the Sound of Eriskay (within the northern section of the Sound of Barra psac) to develop techniques leading to the operational application of such data. The specific objectives of this section of the research project were: To investigate IKONOS 1m panchromatic and aerial photographic datasets for deriving accurate bathymetric maps To evaluate suitable water column correction techniques on the multispectral IKONOS data for Scottish coastal waters Subsequently, to evaluate processed data for discriminating typical shallow water habitat types To infer the depth limits of water column attenuation properties on levels of discrimination of bottom habitat types To evaluate the complementarity between high spectral resolution remote sensing and other techniques (eg aerial photography and subsurface acoustic methods) 25

38 3.4.2 Field methods All field-based measurements were undertaken during a targeted field campaign in the Barra region from the 8th 13th August Establishment of ground control points For the accurate geo-correction of the IKONOS imagery and photographic datasets, 24 prominent and permanently located sites which were discernable on the imagery were measured for their precise location using the Global Positioning System (GPS). A 12 channel Garmin III plus GPS unit was used to record the positions for at least 10 minutes, which, with Selective Availability discontinued, gave a positional accuracy of approximately 2m. This GPS unit was used to accurately locate all other land and water-based measuring sites. Measurements of land field targets for atmospheric correction The empirical line method has been shown to provide an accurate method for atmospherically correcting IKONOS imagery (Karpouzli and Malthus, 2002). To apply this method to the Barra IKONOS data, five large and relatively homogeneous land targets of varying brightness were measured for their spectral reflectance properties. Measurements were performed using a GER 1500 spectroradiometer obtained on loan for this research from the NERC Equipment Pool for Field Spectroscopy. This rapid scanning instrument uses a linear photodiode array to measure radiance over the visible to near infrared wavelength range ( nm) with a nominal dispersion of 1.5nm and resolution of 3nm. The targets selected ranged from dark tarmac, a new concrete slipway, a football field, a large lawn and a sandy beach. A single spectroradiometer sensor head was used fitted with a 4 lens operated from a height of approximately 1.5m giving a ground footprint of approximately 16cm diameter. Between 10 and 25 spectra were taken of each target depending on the apparent variation visible within the target itself. References to incident irradiance over a calibrated Spectralon panel were obtained for every 3 10 target measurements, depending on the degree and change of cloud cover at the time of the measurement. The spectral data were processed to absolute reflectance. From the averaged reflectance spectra for each target the reflectance values for the IKONOS bandwidths were calculated using filter functions based on the sensor response curves provided by Space Imaging Inc. Measurements of within-water broad band attenuation Measurements of gross spatial variations in downwelling Photosynthetically Active Radiation (PAR) and broad blue band attenuation were made at six selected stations to the west of the causeway. The blue light sensor closely matched the visible blue band width of the Landsat TM sensor. The sensors were lowered on a weighted lowering frame and measurements were made at half metre intervals to a maximum of approximately 10m depth depending on overall water column depth (Figure 3.6). Measurements were referenced to above-surface incident irradiance using a continuously logging, gimble-mounted pyranometer to monitor variations in incident flux due to drifting clouds (Figure 3.7). All measurements were made between 10:00 and 15:00 hours local time each day, to minimise solar angle effects. Site positions were located to an accuracy of less than 4m using GPS. Diffuse vertical attenuation coefficients (K d ) were calculated as the slope of the logarithm of downwelling irradiance with respect to depth (Kirk, 1994). 26

39 Figure 3.6 Measurements of broad-band attenuation being undertaken Figure 3.7 Gimble-mounted pyranometer recording fluctuations in incident irradiance Measurements of in situ spectral attenuation In order to characterise the spectral nature of light attenuation in the region, high spectral resolution attenuation measurements were made at two sites in the Sound of Eriskay using the GER 1500 spectroradiometer. The radiometer was fitted with a 3m fibre optic probe and cosine-corrected sensor and used to make measurements of incident downwelling irradiance. The cosine-corrected sensor was fixed on a lowering frame and lowered for measurements at 0.5m intervals down to 2.5m below the water surface (Figure 3.8). 27

40 Figure 3.8 Measurements of in situ spectral attenuation using the spectroradiometer fitted with fibre optic probe Measurement of substrate reflectance levels A number of spectral reflectance measurements were made of typical substrate types for the area, in order to characterise their variations in reflectance properties such that they may inform the habitat classification process. Measurements were made either on samples extracted from their habitat for measurement on-board Figure 3.9 Measurement of reflectance properties of a Zostera marina sample 28

41 (eg Figure 3.9) or were measured in situ. In situ measurements consisted of either measurements of exposed intertidal species (Figure 3.10), direct measurements made over shallow targets through the water column from the water surface, or direct measurements immediately above the submerged species made using the fibre optic extension. Figure 3.10 Measurement of reflectance properties of attached seaweeds in situ Measurement of transects of water depths As an independent means of calibrating depth estimates calculated using the remotely sensed data, measurements of four transects of water depth were undertaken using a hand held echosounder. Each transect was approximately 1 2km in length, centred in the region bounded by Stag Rock, Gorston Rock and Calvay (Figure 3.11). In total 174 estimates of depth were recorded, geolocated using GPS, over water columns ranging from m in depth Aerial photography Acquisition A set of 1:6000 scale aerial photographs of the Sound of Eriskay were obtained for SNH by PhotoAir Ltd in 1999, prior to the construction of the causeway. These were collected on 70mm film using a Vinton/Zeiss camera fitted with a 50mm focal length Distagon lens, but lacked calibrated fiducial marks. Fifteen photographs covering the site of the proposed causeway and narrowest portion of the Sound to the east, were selected for orthorectification and further processing. The order of the photo acquisition over the area is shown in Figure The photo-negatives were scanned at 2400 dpi (giving an approximate ground resolution of 0.07m per pixel) using a Polaroid SprintScan 45 digital scanner. Each photo represented 108 Mbytes in size. All processing of the digital aerial photography was undertaken using ERDAS Imagine version

42 Figure 3.11 Location of echosounder based depth transects Figure 3.12 Sequence of air-photo acquisition over the Sound of Eriskay Third flight line Second flight line First flight line Orthorectification Orthorectification of the imagery was undertaken using ERDAS Imagine OrthoBase software. In the absence of fiducial and camera calibration information, the dimensions of one negative were measured on a binocular microscope with a vernier scale objective stage. Thirty ground control points (GCPs) were identified on the overlapping sections of all images to inform the geo-rectification of the photo block. Control in the x and y directions was provided by coordinate positions identified from the geo-corrected panchromatic IKONOS image for the area (see below). Control in the z dimension was provided through the use of a 50m resolution Digital Elevation Model (DEM) for the area obtained from the OS Digimap service. An additional 180 tie-points, particularly located around the coastline, were used to tie features on overlapping images. The locations of GCPs and tie-points on the photo block are shown in Figure These show that the points are well distributed. 30

43 Figure 3.13 Locations of GCP s (triangles) and tie-points (squares) over the Barra photo image block All photos were then orthorectified to nominal accuracies to within 3 5 pixels ( m). Visual inspection indicated that the coastline and littoral areas were well controlled in the process with greatest errors occurring at the edges of the scanned photos and associated with areas of higher terrain. Mosaicking The individual trimmed and orthorectified aerial photos were then mosaicked into a single image dataset. Non-linear cutlines were used on all land areas and some water regions to better hide the joins and to overcome positional errors arising after the orthorectification. The quality of the final product is largely affected by the quality of the orthorectification and the degree of colour imbalancing between images. Colour imbalances were evident in the Barra datasets with the middle flight line showing generally brighter images than the two outer flight lines. This may be the result of this flight line being acquired in a west to east direction and being located largely over water targets, where the other flight lines were acquired east to west and contained significant and more brightly reflecting land areas. Although a number of approaches offered by the software for automatic image matching were tested, manual colour balancing was undertaken prior to mosaicking using the individual image look-up-tables. Full resolution (0.26m) and 1m resolution mosaic datasets were produced. The results of the image mosaicking process is shown in Figure These show some degree of colour imbalancing still remaining, particularly over the water areas. However, positional accuracy is well maintained. 31

44 Figure 3.14 Orthorectified mosaicked photographic dataset (British National Grid) Figure 3.15 Mosaicked and masked aerial photographic dataset (British National Grid) 32

45 Masking The mosaicked photographic datasets were masked to eliminate all land areas from the image using the digitised Sound of Barra psac boundary supplied by SNH. Masking is a useful technique in coastal studies to eliminate bright pixels from land such that interpretation can be focussed on the generally darker water features. Eliminating land pixels also allows for any automated classification to focus solely on aquatic features without the need for additional categories to account for land pixels. Prior to masking being undertaken, nodes on this boundary dataset were manually adjusted to better fit the coastline evident in the photographs. The resultant masked mosaicked dataset is shown in Figure Satellite Imagery IKONOS data acquisition and quality review Initial panchromatic and multispectral images over the Eriskay region were acquired on , within seven days of the order for the imagery being placed. Whilst meeting the acquisition requirements of less than 20% cloud cover, these images showed one large cloud lying directly over the causeway region. The data were returned to Space Imaging with a request for a retake of the data. A second image was acquired on 27th of September 2001 at 11:32 GMT which was entirely cloud free. Visual inspection of the data revealed no radiometric problems or other flaws such as missing scan lines etc. This dataset was used in all subsequent processing. Acquisition characteristics for the image are given in Table 3.4. Table 3.4 IKONOS image data acquisition characteristics Acquisition Date: Acquisition Time: 11:32 GMT Platform altitude: 680 kilometres Orbit: 98.1 degrees, sun-synchronous Image Coordinates: Top Left: N, W Bottom Right: N, W Geometric Processing Level: Standard Interpolation Method: Nearest Neighbour Bits per Pixel: 11 (2048 brightness levels) Scan Azimuth: Nominal Collection Azimuth: Nominal Collection Elevation: Sun Angle Azimuth: Sun Angle Elevation: Panchromatic resolution: 1m Panchromatic bandwidth: nm Multispectral data spatial resolution: 4m Multispectral bands: Blue nm Green nm Red nm Near Infrared nm 33

46 Geocorrection The IKONOS datasets were initially supplied in geocorrected form to a guaranteed 25m accuracy by Space Imaging. To further improve the accuracy for mapping purposes, the 24 prominent sites for GCPs measured in the field using GPS were found and located on both the panchromatic and multispectral IKONOS datasets. Two points were excluded from the geocorrection process with high root mean square (RMS) errors (>9m) leaving 22 used for the geocorrection itself (with RMS errors ranging from m). Thus, GCPs were accurate to less than one pixel of the multispectral image. The locations of the GCPs are shown on the multispectral image in Figure A linear first order polynomial transformation in conjunction with nearest neighbour resampling was used to correct both image datasets. The projection used was the Transverse Mercator British National Grid. Figure 3.16 Distribution of ground control points used for geocorrecting the Sound of Barra IKONOS image datasets Qualitative comparison of our secondary geocorrected image with that of the original image supplied by Space Image indicated that the correction performed by Space Imaging was accurate to within 1 or 2 pixels (approximately 8m) of the true location on the second image. There is no reason to suggest that this level of accuracy would not be maintained in any subsequent images obtained from Space Imaging. It may be felt, therefore, that the level of accuracy of the data originally supplied is acceptable to work with, without the need for further geocorrection. Although it is unknown what system or map-base Space Imaging use for their original corrections, it is likely that this level of accuracy would be similar for any images obtained around the Scottish coastline. 34

47 Atmospheric correction Using the five sets of homogeneous target measurements measured in the field with the spectroradiometer, relationships were developed between the calculated ground-based IKONOS reflectance values and related values extracted from the corresponding pixel locations in the multispectral IKONOS dataset (Figure 3.17). Although the data are somewhat bimodally distributed because of low absorption in the visible region, these relationships show considerable linearity. Previous research shows that the response of the IKONOS sensor can be considered to be linear across the range of typical earth surface reflectances. The points where the lines would intersect the X-axis indicate the contribution to radiance from background reflectance of light in the atmosphere. This is highest in the blue region where atmospheric scattering is considerable and generally decreases with increasing wavelength. These empirically derived relationships were used to atmospherically correct the IKONOS data to percent reflectance. The panchromatic dataset was not atmospherically corrected. Figure 3.17 Empirical line relationships for the individual IKONOS wavebands 50 Blue Waveband 50 Green Waveband Ground reflectance (%) Cement Sand 10 Tarmac y = x Grass R 2 = Lawn IKONOS Radiance Ground reflectance (%) Cement Sand 10 Lawn Grass Tarmac y = x R 2 = IKONOS Radiance Ground reflectance (%) Red Waveband Cement Sand 10 Tarmac Grass Lawn y = x R 2 = Ground reflectance (%) NIR Waveband Cement Grass Sand Lawn y = 0.096x R 2 = Tarmac Masking The corrected IKONOS image datasets were masked to the SAC boundary to eliminate land areas using the digitised Sound of Barra psac boundary manually adjusted to better fit the coastline evident on the images. The resultant masked images are shown in a later section in Figures 4.16 (panchromatic) and 4.20 (multispectral). 35

48 4 RESULTS 4.1 Acoustic Maps The final acoustic based maps were produced using the sequence described in Section Figures 4.1 and 4.2 show the bathymetry of the north and south areas respectively rendered using a TIN of the Submetrix Bathymetric sidescan results. The maps are colour coded by depth in 5m intervals and show a series of northwest-southeast extending ridges or reefs. These areas of elevated bathymetry are a result of differential erosion of the bedrock in the area with the dominant structural grain oriented in this direction and had been identified on previous surveys. The deepest part of the acoustic survey area can be seen in the south with depths to greater than 40m, however, much of the area is relatively shallow at between 10 25m. A significant part of the area to the north is less than 5m deep with rapid shallowing to depths of less than 3m into the Sound of Eriskay. A single track of data was acquired through the Sound of Eriskay, however, this proved of limited use due to the shallow depths and therefore lack of range for the acoustic transducers. Figure 4.2 shows a maximum depth of 40m to the east side of the southern area. The bathymetric chart shows a diverse sea floor with the eastern margin displaying a complex bathymetry consisting of a number of isolated reefs. These reefs range in size from a few metres across to 50m with an elevation gain of 1 5m above the surrounding sea floor. No survey data were obtained from the area to the northwest of Hellisay as the average depth here was less than 5m with significant areas at less than 3m. In these areas, the acoustic seafloor coverage is minimal and error in the bathymetry is higher than acceptable. Figures 4.3 and 4.4 show three dimensional views of the north and south areas respectively with the sidescan images and AGDS track plots draped onto the bathymetric models. The images show the contrasting nature of reflectivity between the finer grained and mixed sediment areas and the rocky reefs. The range in amplitude contrast and texture pattern were associated with different biotope complexes encountered across the survey areas. A series of additional acoustic maps from north and south areas (slope map, mosaicked sidescan data and the Echoplus AGDS line track data) are given in Appendix B. 4.2 Infaunal analysis Of the twenty-three stations sampled by grab or core for infaunal analysis (Figures 4.5 and 4.6 for the north and south areas respectively), nineteen provided adequate samples for quantification of the fauna. The full quantitative results are given in Table A3 (Appendix A). Diversity and total abundance data for these stations are given in Table 4.1, while similarities and trends amongst the species assemblages revealed by cluster and ordination analyses are shown by dendrogram (Figure 4.7) and 3-dimensional ordination plot (Figure 4.8). In the following descriptions abundances, given in brackets, follow the SACFOR scale (Connor et al., 1997). Species (or more precisely taxon) richness, Shannon-Wiener diversity (employing log 2 ) and abundance are abbreviated as S, H and A respectively. 36

49 Figure 4.1 Bathymetric model of northern area using Submetrix System 2000 Figure 4.2 Bathymetric model of southern area using Submetrix System

50 Figure 4.3 3D view of sidescan and AGDS draped on the bathymetric model for the north area Figure 4.4 3D view of sidescan and AGDS draped on the bathymetric model for the south area 38

51 Table 4.1 Diversity and abundance for infaunal grab and core samples collected in the northern and southern regions of the Sound of Barra psac Station Shannon-Wiener Shannon-Wiener Pielou Evenness No. taxa Abundance Function (log 2 ) Function (log e ) (No./0.1 m 2 ) S S S S S S N N N N N N N N N N N N N Explanation of survey coding system Survey stations are referred to by a station number prefixed by an S for stations in the southern region or an N for stations in the northern region (eg S1 or N1). An additional prefix of a D is used to denote the site was surveyed by diver (eg S1 = DS1) or an R is used where sites were surveyed by ROV (eg N1 = RN1). In the text and when referring to sampled infauna or sediments the shorter version of the code is used without the D or R prefix. A separate suite of stations in the north are coded by a number with a prefix V or O. The former refers to sites examined by drop down video (eg V1) and the latter refers to sites examined by snorkel diver or glass bucket observations (eg O1). See Appendix A, table A1 for full details of sample stations. Stations S1 and S7 in Outer Oitir Mhór consisted of slightly muddy fine sand (9 10% silt/clay content) in m. The infauna were very diverse and abundant with S = , H = and A = ind./0.1m 2. The community was a good fit to the biotope CMS.AfilEcor, despite Echinocardium cordatum being unrecorded. The following characterising species were present: Virgularia mirabilis (O), Pholoe spp. (C), Nephtys hombergi (P), Eudorella truncatula (F), Turritella communis (C A), 39

52 Figure 4.5 Ground truth stations in the northern region of the Sound of Barra psac 40

53 Figure 4.6 Ground truth stations in the southern region of the Sound of Barra psac Nucula nitidosa (P), Mysella bidentata (C), Arctica islandica (O), Clausinella fasciata (F C) and Amphiura filiformis (A). Station S10 to the northeast of Gighay consisted of very slightly muddy sand (4% silt/clay content) with pebbles at 28m. The fauna was a modified version of CMS.AfilEcor, differing from the above stations by the absence of V. mirabilis, N. nitidosa and E. truncatula. Nepthys hombergi was unrecorded, although juvenile Nepthys spp. were common. Hydroids were present on the pebbles. The fauna was diverse (S = 91, H = 5.44) and moderately abundant (A = 560 ind./0.1m 2 ). Stations S3 and S6 were located in Inner Oitir Mhór at m. The shelter provided by the surrounding islands resulted in a distinctly muddier sand than in Outer Oitir Mhór (24 31% silt/clay content), with a scatter of pebbles and shells. The infauna was fairly similar to CMS.AfilEcor, with Amphiura filiformis 41

54 Figure 4.7 Cluster analysis of species abundance data from infaunal samples from the Sound of Barra psac (at S3), Nucula nitidosa, Mysella bidentata, Turritella communis (at S3), Nepthys hombergi, Pholoe spp. and Eudorella truncatula, but Arctica islandica and Clausiella fasciata were unrecorded. The diversity and abundance was similar to the CMS.AfilEcor stations above (S = , H = , A = ind./0.1m 2 ). Diving observations recorded Cerianthus lloydi (O), Virgularia mirabilis (F) in softer areas (S3) and Arenicola marina (O C). Diving also revealed a scatter of drift weed over this area (especially Laminaria saccharina and Sacchoriza polyschides), as well as sparse live kelp, particularly the cape form of L. saccharina. In view of this, the area has been ascribed to the IMX.LsacX biotope. A recent grab survey of Inner Oitir Mhór by Cordah (2000) recorded a broadly similar infauna, although their stations close to S3 and S6 were muds with around 80% silt/clay content. This may indicate marked short-term temporal changes in sediment characteristics in this area or high spatial variability. Samples from the very shallow sandy area in the Sound of Eriskay (N21 at 0.7m), from the deep channel in the Sound of Eriskay (N12 at 7.7m) and from just outside the eastern entrance to the Sound (N1 at 8.8m) were all of medium sand and cluster together on the dendrogram (Figure 4.7) and are close on the ordination plot (Figure 4.8). This provides some evidence that the infauna of much of the Sound of Eriskay can be referred to the same biotope. Amphipods were the dominant taxa at all stations, especially Perioculodes longimanus and Pontocrates arenarius. The stations have been allocated to IMS.EcorEns, being the closest biotope, although they appear to lie between IMS.EcorEns and IGS.NcirBat. Of the characterising species, Echinocardium cordatum was found to be abundant at N1. Diver and video observations in the area revealed the presence of low densities of Lanice conchilega, Arenicola marina and Zostera marina, as well as the dead shells of Ensis spp. Comhairle nan Eilean Siar (1999b) analysed the fauna from sixteen grab samples from shallow sand, a Zostera marina bed and the deep channel of the Sound of Eriskay and also found their shallow sand 42

55 Figure 4.8 Detrended correspondence analysis of species abundance data from infaunal samples from the Sound of Barra psac Sediment coarseness S7 S1 S10 N20 N12 N1 N18 N25 N23 N21 Depth N17 N13 N Axis S3 S Axis S8 N5 2.5 N N Axis 2 slightly muddy fine sand (CMS.AfilEcor) muddy fine sand (IMX.LsacX) shallow medium sand (IMS.EcorEns) Zostera bed on fine sand (IMS.Zmar) shallow maerl bed (IGS.Phy.R) shallow coarse sand (IGS.Sell) deep coarse sand (CGS.Ven.Neo) deep maerl bed (IGS.Phy.HEc) samples clustering together and with the non-maerl sample from the deeper channel. They recorded a similar but comparatively impoverished sand fauna, with samples containing taxa and ind./0.1m 2, whereas the current survey recorded taxa and ind./0.1m 2. Station N20 was also located in the shallow sandy area (0.5m) of the Sound of Eriskay but within a dense bed of Zostera marina, where diving observations recorded abundant Z. marina, Occasional Arenicola marina and Gibbula sp., as well as the presence of Obelia sp., Anemonia viridis and balls of Ectocarpus sp. on the Z. marina leaves. The infauna was fairly similar to the IMS.EcorEns stations described above but was more speciose and abundant (S = 71, A = 2937 ind./0.1m 2 ). The diversity (H = 4.0) did not reflect the species richness due to strong dominance by annelids, especially species characteristic of organicallyenriched sediments. Capitella capitata and Tubificoides benedii made up 35% of the fauna. The station has been allocated to IMS.Zmar. Comhairle nan Eilean Siar (1999b) recorded a very similar fauna (S = 60 74, A = ind./0.1m 2 ). 43

56 Stations N14, N17, N18, N23 and N25 were all located to the west of the Sound of Eriskay at depths of 1.2 8m. The grab samples consisted of coarse sand, with an admixture of pebbles in the case of N14. The samples are close on the ordination plot (Figure 4.8), with N14 slightly set apart. The infauna is a reasonable fit to IGS.Sell. Despite the dynamic nature of the sediments in this area, as evidenced by the presence of sand waves, the infauna is generally moderately diverse and abundant (S = 37 95, H = , A = ind./0.1m 2 ), except for station N25 (S = 17, H = 3.6, A = 53 ind./0.1m 2 ). Of the characterising taxa, Spisula elliptica and Dosinia exoleta were abundant (except at N25), Nepthys cirrosa abundant (except N23) and Spiophanes bombyx, Spio spp. and Bathyporeia spp. present at most stations. Comhairle nan Eilean Siar (1999c) carried out a grab survey just to the south of this area, and their northernmost station (Site 3) lies close to N23. Their southern stations were on medium sand and were considered to be closest to the IGS.NcirBat biotope. Site 3 was in an area of coarse sand and was placed between IGS.Sell and IGS.NcirBat. Station N13 was also located to the west of the Sound of Eriskay but in the centre of a relatively deep pool (7.2m), encircled by shallower ground. The sediment consisted of coarse sand and maerl gravel, with abundant live Phymatolithon calcareum. As shown by the ordination plot (Figure 4.8), the infauna was close to the other coarse sand stations described above, but much more abundant and diverse (S = 125, H = 5.0, A = 4625 ind./0.1m 2 ). The site has been allocated the biotope IGS.Phy.R. Although a rich red algal flora was not recorded for this site, as it was for maerl patches at similar depths further east along the channel, diver observations included other algae such as Laminaria saccharina (O), Sacchoriza polyschides (O), Halidrys siliquosa (F), Dictyota dichotoma (F) and Chorda filum (O). Samples from Stations S8 and N2 were from coarse sand and shell gravel at 18 20m, S8 being located in the Sound of Hellisay and N2 to the east of Eriskay. The infauna contained several of the characteristic species of CGS.Ven, such as the bivalves, Gari tellinella, Clausinella fasciata and Timoclea ovata, and Amphioxus Branchiostoma lanceolatum. Diver and ROV observations also revealed the presence of Lanice conchilega, Pecten maximus and algae, including small amounts of Phymatolithon calcareum at N2. These stations have been ascribed to CGS.Ven.Neo, although no holothurians were recorded. Grab samples from N3 and N5 to the east of Eriskay were from maerl beds and contained live Phymatolithon calcareum. The sediment consisted of gravel and pebbles (strongly rippled in the case of N3), with 20 60% maerl cover at N5 and 50 80% cover at N3. This coarse material overlay muddy sand. The infauna showed affinities with the CGS.Ven.Neo stations (S8, N2), with Gari tellinella, Clausinella fasciata and Timoclea ovata in moderate to high numbers. The fauna showed low affinity with the shallow maerl beds to the west of Eriskay causeway and in the Sound of Eriskay, which also had a far richer algal flora. These deeper stations have been assigned to IGS.Phy.Hec. The infauna is extremely diverse (S = , H = , A = ind./0.1m 2 ). Trends in the faunal data revealed by ordination (Figure 4.8) implicate sediment coarseness and water depth as major factors correlating with community composition in the Sound of Barra. The first four ordination axes account for 33% of the variance in the data. The main trend (13%) orders the samples from muddy sands with low axis 1 scores, through medium sands, to coarse sands and gravels with the highest scores. Axis 2 (9%) separates the shallow coarse sediments at depths of 1 8m (with high scores) from the deeper coarse sediments from 18 22m (with low scores). 44

57 4.3 Sediment particle size and loss on ignition results The summary data are presented in Table 4.2. Table 4.2 Granulometric parameters associated with infaunal grab samples collected in the northern and southern regions of the Sound of Barra psac Sample Mean particle Median particle % Silt/Clay % Sand % Gravel Number diameter (µm) diameter (µm) (<63µm) (63µm 2mm) (>2mm) S S S S S S S N N N N N N N N N N N N N Satellite and air photography results Field methods In situ attenuation measurements Stations where PAR and spectral attenuation measurements were taken are shown in Figure 4.9. Calculated PAR attenuation coefficients (K d,par ) are given in Table 4.3. The lowest K d,par was recorded at Station 2 to the east of the causeway, although this was also the deepest station. Instrumental problems prevented further measurements being made to the east of Station 2. There is an apparent slight increase in attenuation towards the west (from m -1 ). Highest attenuation was recorded at the shallowest Station 4 to the south of the psac. Blue light attenuation was lower than PAR attenuation as this light is the most penetrating visible wavelength. Patterns in variation in blue light attenuation (K d,blue ) were generally similar to K d,par. 45

58 Table 4.3 Calculated broad-band Blue and PAR attenuation coefficients Station Depth (approx. m) K d,par (m -1 ) K d,blue (m -1 ) Downwelling spectral measurements, made using the spectroradiometer fitted with the fibre optic probe, are presented in Figures 4.10a and b. Station A was over a Zostera marina bed and Station B over bare sand. Greater variation was recorded at Station A, which may be the result of wave action at this site coupled with much lower light levels during this set of measurements. However, both sets of spectra show relatively little variation in spectral quality of penetrating light, at least to the 2.5m depth limit to which the measurements were made. Calculated spectral attenuation coefficients are presented in Figure Perhaps unsurprisingly, given the proximity of the two stations, calculated attenuation is remarkably similar. Attenuation is lowest over the visible spectrum; increases of attenuation at approximately 600, 650 and 700nm are typical of normal light absorption by water. Higher attenuation of blue light, compared to other visible light wavelengths, could be the result of increased scattering and possible absorption by dissolved aquatic humus, the latter potentially arising from runoff from land. The spectra show no visible signs of absorption due to pigments in phytoplankton, suggesting that phytoplankton concentrations in the region, at least at the time of measurement, were low. Summary of light conditions in the Sound of Eriskay Overall, attenuation of light in the region of the Sound of Eriskay is low, in comparison to other sites around the world so close to land (eg Kirk, 1994, p138). This indicates optically clear waters of relatively high quality with little in the way of aquatic colour and phytoplankton evident. Although there was some variation recorded in the limited attenuation measurements made, the differences are not marked. There is some evidence to suggest that K d values are lower to the east of the causeway, although more measurements would be needed to confirm this. Variation could also potentially be explained by variation in depth, with shallow stations perhaps having higher attenuation as a result of proximity to land and from greater risk of wind-induced re-suspension of bottom sediment. The results suggest that a single K d value could reasonably be assumed for correction of the remotely sensed data. The most penetrating light is in the region of nm, which indicates that the green waveband of IKONOS data may be the most useful for bathymetric calculations. 46

59 Figure 4.9 Locations of broad-band Blue and PAR (red) and spectral (green) attenuation measurements 47

60 Figure 4.10a Measured spectral variation in downwelling light at 0.5m intervals at Station A Figure 4.10b Measured spectral variation in downwelling light at 0.5m intervals at Station B Irradiance m 1.0 m 1.5 m 2.0 m 2.5 m Irradiance m 0.5 m 1.0 m 1.5 m 2.0 m 2.5 m Wavelength (nm) Wavelength (nm) Figure 4.11 coefficients Calculated spectral attenuation Over sand Over Zostera Attenuation (m -1 ) Wavelength (nm) Substrate reflectance levels Spectral reflectance levels, measured over a range of substrate types either extracted from depth, or in situ either intertidally exposed or submerged, are shown in Figures Reflectance levels are generally low across the visible spectrum for Z. marina and algal species whether exposed or in situ (Figures 4.12 and 4.13). All reflectance levels from exposed surfaces generally show an increase in the near infrared region as a result of scattering within the leaves and fronds. However, this region is also strongly absorbed by water itself such that very little near infrared light exits the water column as reflected light (eg Figure 4.13b). Cladophora spp. and Z. marina show peaks in reflectance in the green region of the spectrum (ca 530nm) while brown algal species show highest reflectance levels in the brown/red part of the visible spectrum (>550nm). Pelvetia canaliculata shows highest reflectance levels in the visible, indicative of bright yellowish colour of the upper fronds. Measures of reflectance of specific Zostera marina beds were made using both the fibre optic and from just below the water surface (Figures 4.14a and b). These are generally similar in shape (measurements from the water surface are lower as the reference panel is measured above surface). The shape is largely dictated by absorption by water itself, particularly from above approximately 570nm. 48

61 Figure 4.12a Measured spectral reflectance s from exposed kelp, seagrass and Cladophora sp Figure 4.12b Measured spectral reflectance s from exposed brown intertidal algal species Reflectance (%) Laminaria digitata Cladophora sp. Zostera marina Reflectance (%) Fucus serratus Fucus spiralis Pelvetia sp. Ascophyllum nodosum Wavelength (nm) Wavelength (nm) Figure 4.13a Measured in situ spectral reflectance s over algal species, using the fibre optic Figure 4.13b Measured in situ spectral reflectance s over algal species, using the fibre optic Reflectance (%) Fucus serratus Leathesia, Cladophora Laminaria digitata Reflectance (%) Fucus serratus Laminaria digitata Cladophora sp Wavelength (nm) W avelength (nm) Figure 4.14a Measured in situ spectral reflectance s over Zostera, using the fibre optic Figure 4.14b Measured in situ spectral reflectance s over Zostera, from the water surface Reflectance (%) Reflectance (% ) Wavelength (nm) W avelength (nm) 49

62 Bare surfaces show less variation in reflectance across the visible spectrum (Figures 4.15a and b). Dry and wet bare rock surfaces showing the least variation with only slightly increasing reflectance with increasing wavelength. The barnacle covered rock surfaces show similar flat-ish reflectance but typical of algal covered surfaces with minima in reflectance in the visible blue and red (670nm) regions where absorption by photosynthetic pigments is strongest. Beach sand is brightly reflecting and slightly increasing across all wavelengths. Sand measured in situ is also brightly reflecting in comparison to reflectance from algal species in situ (Figure 4.15b). The shape of the reflectance spectrum of sand under water is largely driven by the shape of absorption by the overlying water. Figure 4.15a Measured spectral reflectance s from exposed bare substrate surfaces Figure 4.15b Measured in situ reflectance from sand, made using the fibre optic Rock with barnacles Bare dry rock Beach sand Rock with barnacles Bare wet rock 6 5 Reflectance Reflectance (% ) Wavelength (nm) Wavelength (nm) In summary, apart from sand, reflectance levels for all substrate types across the visible spectrum are low and show little variation. This has implications for their potential differentiation using the IKONOS multispectral data. Bearing in mind that near infrared reflectance measures are not available in submerged regions due to strong absorption by the water column and that three wavebands only are available (Blue, Green and Red) the results suggest that discrimination between the types may be limited Analysis of image quality Panchromatic imagery After an enhanced stretch, the geocorrected and masked panchromatic data are presented in Figure The data range of the entire image varies between 72 and 648 radiance units (median ), with the main pixels over water lying in the range of This represents a much narrower radiometric range than the individual multispectral bands, particularly the blue and green bands (see below). Furthermore, the image shows signs of banding in parts. It is unclear why this occurs but could be to do with the way in which the IKONOS sensor (or sensors) acquires the image and/or potential calibration problems. Such problems are frequently observable over regions where signal to noise is likely to be low, particularly dark water targets. Closer inspection highlights considerable noise in the data and the presence of other striping (Figure 4.17). A median spatial filter can be used to overcome this problem to some extent, although there is some loss of detail (Figure 4.18). 50

63 Figure 4.16 Geocorrected and masked panchromatic IKONOS image, stretched to enhance differences in the water areas 51

64 Figure 4.17 Close-up of panchromatic data to show noise and striping problems in the raw data, at the scale of individual pixels. Each individual pixel represents 1m Figure 4.18 Close-up of same region as above after a median filter has been applied 52

65 Although the bottom of the Sound is generally evident when a suitable data stretch is applied, differences in deeper parts of the channel are not particularly clear, especially in the channel. The relative lack of detail is probably due in large part to the broad spectral sensitivity of this band (Figure 4.19), which is broadly sensitive across the nm region. Sensitivity in the red and near infrared regions is not particularly useful in the context of bathymetric profiling, with water itself absorbing strongly in these regions. Figure 4.19 Relative spectral sensitivity of the IKONOS panchromatic band (courtesy of Space Imaging Inc.) 1.0 Relative Spectral Responsivity Wavelength (nm ) Multispectral data image quality The data ranges for the multispectral image are much greater than the panchromatic band, particularly for the blue and green spectral bands where there is good penetration into the water column (Figure 4.20). Mean reflectance values (x 10 2 ) were , and for the blue, green and red spectral bands, respectively. Banding is not evident in any visible band. However, closer inspection of individual pixels over deep water does reveal considerable variation between neighbouring pixels (Table 4.4). Analysis of the variation in pixel values along a transect over deep water reveals relatively high standard deviations. Reflectance levels here are very low and the instrument may be working near to its limit of detection, but variation between pixels is of the order of up to 30% in the blue and green bands and 50% for the red and near infrared bands. Table 4.4 Variation in reflectance (x 10 2 ) along a transect over deepwater Band Mean Standard deviation Blue Green Red Near Infrared In shallower regions the range of reflectance is greater leading generally to less variation between pixels (Figure 4.21). All areas of the Sound are clearly discernible, although qualitatively it is difficult to determine if parts of the channel bottom are not visible or whether or not they have darker bottom reflectance. 53

66 Figure 4.20 Geocorrected and masked and atmospherically corrected multispectral IKONOS image 54

67 Figure 4.21 Close-up of IKONOS multispectral data to show level of noise in the raw data, at the scale of individual pixels. Each individual pixel represents 4m Approaches to water depth correction Classification of submerged habitats and estimation of depths in shallow water environments is complicated, as above-surface reflectance over such sites can be regarded as a function of: Substrate reflectance Water depth Scattering and absorption in the water column Surface reflectance effects. The first two of these factors might be regarded as parameters of interest to be retrieved from remotely sensed optical images, the first for images to classify bottom type, the second to produce a bathymetric model for the area of interest. While surface effects can probably be regarded as being fairly constant for a given resolution, the other effects may vary spatially from area to area. The retrieval of water depth has therefore to account for varying scattering and absorption (turbidity) in the water column and varying bottom reflectance. Conversely, the retrieval of bottom reflectance has therefore to account for varying water column turbidity and varying depth. For the latter, it has been shown that correcting water depth and turbidity improves subsequent identification of submerged habitats. The first and most commonly used technique to address this problem is that offered by Lyzenga (1978, 1981) who proposed a useful approach for deriving a unit related to bottom reflectance, which first 55

68 involves the linearisation of reflectance s in all available bands for the attenuating effects of water depth: [1] where L i is the measured radiance in band i, and L si is the measured radiance over deep water. For a homogeneous bottom region, the linearised X i values are then plotted against each other to determine K i /K j, the ratio of water attenuation coefficients of the two bands plotted. Calculated bottom type indices (Y i ) are then determined by: [2] where it should be noted that Yi is related to, but is not, bottom reflectance. An alternative, but related approach is that of Bierwirth et al. (1993) who proposed a model for water depth (Z) in areas of uniform bottom reflectance: [3] Where N is the total number of bands available, Ri is surface reflectance, ki is attenuation coefficient in band i. Bierwirth et al. also proposed a model for calculating bottom reflectance (RBi): [4] Note that the Bierwirth et al. approach (equation [4]) requires knowledge of both water column attenuation (k i ) and depth (Z). The advantage of equation [4] over Lyzenga s approach [2] is that bottom reflectance is expressed in reflectance units. A combination of both these approaches has been attempted in this study to retrieve water depth and bottom reflectance. Originally it had been intended that the panchromatic image would be used to calculate water depth, a map of which would subsequently be used to correct the multispectral image to bottom reflectance. However, because the panchromatic image was found to be less than ideal for its water penetration capabilities, a second perhaps less satisfactory approach was adopted. Firstly, water depth was independently estimated by digitising the hydrographic map for the area and interpolating the values to produce a bathymetric model. Secondly, the interpolated depth map was used to correct surface reflectance for the map of bottom reflectance which was subsequently classified for bottom habitat. Bathymetry To test the Bierwirth et al. (1993) algorithms for the calculation of bottom reflectance requires an estimate of water column depth. Depth data for the Sound of Barra were derived from manual digitisation of the 1:30000 Hydrographic Chart for the region (Hydrographic Office, 1978). For the wider area of the northern psac some points were digitised (Figure 4.22) and interpolated using two approaches: 56

69 1) Triangulation, an exact interpolation where the depth at a location is estimated as a linear interpolation of the plane defined by the three nearest points. 2) Radial basis function employing a multi-quadratic kernel type, undertaken using Surfer version 7.0 using a search radius of 1000m. This is another exact interpolator, where the depth is weighted as a function of the distance of measured points to the location to be interpolated. Results of the two interpolations are presented in Figures 4.23 and Overall, the results are similar and produce sensible depth surfaces for the Sound; differences are largely in minor details. Triangulation gives a more blocky appearance, particularly where the method is applied in areas of sparse data points (this interpolation is ideally suited to evenly spaced data points). The radial basis interpolation gives an apparently smoother image. Subtle differences are evident in shallow regions where data points are densest. To most accurately assess the quality of the interpolated bathymetric maps, the data were compared to the echosounder transect and sonar data measured in the field (transect locations in Figure 4.25, results of comparisons in Figures 4.26 and 4.27). For echosounder Transects E1, E2 and E4, reasonable estimates of depth were obtained using the interpolated data with accuracies of the order of 1 2m. Greatest differences were observed with Transect E3 where considerable error between measured and interpolated depths can be observed. Although the interpolated map indicates similar levels of depth of the channel, it is positionally offset compared to the transect data. Whilst this may be due to errors in the interpolation, this could also be the result of changes in channel morphology between when the depth soundings on which the map is based were made and the present day measurements. In comparison to along-channel transects of sonar data the interpolation shows similar form but the sonar data indicates a deeper channel. Errors are in the region of 2 3m and in some cases up to 5m. Overall, the comparisons suggest that while the interpolated hydrographic map may provide a potentially acceptable (albeit approximate) map of bathymetry, finer morphological detail is missing. The interpolated hydrographic data generally overestimated depth in both comparisons (to echosounder transects and sonar data). This may suggest that some part of the difference is due to a general shallowing of the channel over time and other changes in form since the map was produced. 57

70 Figure 4.22 Locations of digitised hydrographic soundings for interpolation to produce the interpolated bathymetric map 58

71 Figure 4.23 Depth contours for Sound of Eriskay using the triangulation interpolation 59

72 Figure 4.24 Depth contours for Sound of Eriskay using the radial basis function interpolation 60

73 Figure 4.25 Location of the transects used for the depth comparisons (E1 to E4, echosounder transects; S1, S2 sonar data transects) 61

74 Figure 4.26 Comparison of field-based echosounder determinations of water depths to the interpolated (radial basis function) hydrographic data. Depths are from Chart Datum Echosounder transect 1 Interpolated Hydrographic data Depth (m) ID no Echosounder transect 2 Interpolated Hydrographic data Depth (m) ID no Depth (m) Echosounder transect 3 Interpolated Hydrographic data ID no Echosounder transect 4 Interpolated Hydrographic data Depth (m) ID no 62

75 Figure 4.27 Comparison of depth estimates from two transects from the channel obtained using the sonar to the interpolated (radial basis function) hydrographic data. Depths are from Chart Datum Sonar Transect 1 Sonar data Interpolated Hydrographic data Sonar data Interpolated Hydrographic data Sonar Transect

76 4.4.4 Bottom classification A preliminary supervised classification has been performed on the depth-corrected image of bottom reflectance shown in Figures 4.28 and Training areas of known surface types were defined, informed by the biological surveys conducted by Heriot-Watt University. It was found that for a number of surfaces (eg seagrass and intertidal areas) several training zones were required to be defined. Three sand types were differentiated: exposed beach sand, submerged medium-coarse sand and submerged fine-medium grained sand. Rock surfaces were differentiated as exposed algae and bedrock, submerged bedrock and rock ridges. Submerged plant coverage was separated into algae and seagrass types. A maximum likelihood classification was used to classify the image (Figure 4.30). Of the exposed surfaces, beaches (white) and exposed rock and algal covered surfaces (dark green) are quite well delineated. At the western end of the psac the submerged bedrock (dark blue) and rock ridges (blue) are also well delineated. There is some confusion, however, in deeper portions of the channel (areas A and B) where the depth correction has not performed well and where these regions have classified as bedrock or intertidal zones. The eastern, deeper portion of the psac has also been classified as bedrock for similar reasons. Areas of seagrass (light green) and submerged algae (burgundy) show some confusion; in particular, seagrass distribution in some areas is overestimated (in regions which are probably algal dominated) and in some areas underestimated (where it is confused with fine sand). Areas of fine and medium-coarse sand are probably well delineated, but overclassified in regions of patchy seagrass distribution. The main factor which contributes to the relatively poor classification achieved on the basis of spectral data alone, is the similarity in spectral signatures between many of the habitat surfaces in the three IKONOS wavebands. This is exacerbated by the effects of water column depth, where differences become harder to discern in areas where the water column is deepest (ie approaching the depth limit of penetration), even after the data have been depth corrected. 64

77 Figure 4.28 Image of bottom reflectance calculated using interpolated (radial basis) depth surface 65

78 Figure 4.29 Image of bottom reflectance calculated using Lyzenga approach 66

79 Figure 4.30 Spectrally based classification of bottom habitat types performed on the bottom reflectance (depth corrected) image B A 67

80 4.4.5 Detecting change comparison of aerial photography and IKONOS data Comparisons of the three remotely sensed image data sets (1999 aerial photography and 2001 panchromatic and multispectral IKONOS images) for the Sound of Eriskay are shown in Figure The panchromatic image shows lowest contrast in water so that differences between it and the photomosaic image are clearest on land. In particular, the panchromatic data shows clearly the scarring caused by the construction of the road to the new causeway, both around the road itself and a construction track leading down from the area at the very top of the image. Differences on land caused by the causeway s construction are less evident in the comparison of the multispectral IKONOS data and the photomosaic. However, water column features are more evident in both images such that the effects of construction on benthic habitats may be compared. Qualitatively there are few changes evident from one image to the other. The channel is less clear in the multispectral image although the tide was slightly higher when this image was acquired. A quantitative comparison of the two colour datasets was undertaken by comparing the green bands from each set. Subtraction of the IKONOS green band data from the green channel photographic data reveals change in positive (increase in reflectance) and negative (decrease in reflectance) directions. These are displayed in Figure 4.32 (the blockiness in this image reflects the differences in resolution between the two datasets). It shows clear increase in reflectance (highlighted in green) caused by the presence of the causeway. Red indicates regions of a decreased reflectance change between the two images. Significant decrease changes are evident on the southern side of the channel near to the causeway and surrounding the seagrass bed to the east of the causeway. Changes here are possibly the result of differences in either bottom reflectance and/or bottom morphology induced as a result of causeway construction. In particular, the area directly around the causeway (compared in close-up in Figure 4.33) shows a definite darkening of the bottom immediately around the causeway and a potential scouring of sand in a semicircle east to Stag Rock. Other areas of increased and decreased reflectances highlighted as change in Figure 4.32 probably reflect minor differences between the two datasets in geometric correction, tidal state and illumination differences. 68

81 Figure 4.31 Direct comparison of the three image datasets (left, 1999 aerial photography; centre, 2001 panchromatic IKONOS image; right, 2001 multispectral IKONOS image) 69

82 Figure 4.32 Detected increased reflectance change (green) and decreased reflectance change (red) after comparison of the photographic and multispectral datasets 70

83 Figure 4.33 Close-up comparison of causeway on the south side of the channel to show changes in bottom reflectance and bottom morphology as a direct result of causeway construction 71

84 5 DATA INTEGRATION 5.1 Distribution patterns of biotopes Northern area The northern region of the psac extends from Hartamul Island just inside the eastern boundary (Figure 5.2), through the Sound of Eriskay to Lingay which is just inside the western boundary (Figure 5.1). West of the eastern entrance to the Sound of Eriskay the sea bed is rarely deeper than 5m and subject to strong tidal currents and, west of the Eriskay causeway, strong wave action. On the most exposed, northwestern limit of the psac, north of the point at Pollachar, most of the sea bed consists of bedrock and boulders, with some pockets and channels of medium-coarse sand (Figure 5.1). Kelp forest, with an understorey of scour-tolerant sponges and algae, dominates the area, with both Laminaria hyperborea and scoured kelp biotopes being recorded (MIR.Lhyp.Ft, MIR.HalXK). The sediment biotope has been referred to IGS.Sell (Spisula elliptica and venerid bivalves in infralittoral clean sand or shell gravel). South of the point at Pollachar and west of a line passing from East Kilbride, via The Witches, to Narrow Point on Eriskay, the sea bed forms two major physiognomic types. Approximately half of the sublittoral here is floored by extensive areas of rock ridges trending NW SE, with narrow runnels of medium-coarse sands between the ridges. The other major seabed type consists of broad channels of medium-coarse sand formed into waves. The areas of rock ridges support similar kelp biotopes to those found north of Pollachar, with MIR.Lhyp.Ft, MIR.LhypGz.Ft and MIR.HalXK on bedrock and boulders and IGS.Sell forming the sedimentary habitat. The satellite imagery reveals geographical variation in the topography of the broad sand wave channels, with the wavelength varying from around 50m in the northwest of the area, to c.160m in the southeast. Despite the high mobility of sediments in this region, the IGS.Sell biotope recorded here is by no means impoverished (see description of infaunal samples in Section 4.2). Moving eastwards towards the Sound of Eriskay, increasing shelter from wave action is accompanied by a change in the dominant biotopes. East of the line passing from East Kilbride, via The Witches, to Narrow Point, and including most of the Sound of Eriskay, the principal subtidal biotope is IMS.EcorEns in an area of fine-medium sands. Scattered rock outcrops and coastal rock support a narrow band of Laminaria digitata (MIR.Ldig, SIR.LsacLdig) in places in the sublittoral fringe and down to 1.4m (see also Comhairle nan Eilean Siar, 1999a). Elsewhere tideswept scoured mixed kelp biotopes predominate (MIR.HalXK, MIR.XKScrR) (see also Comhairle nan Eilean Siar, 1999b). Several Zostera marina beds (IMS.Zmar) are present on fine-medium sands within the area delimited by IMS.EcorEns. Z. marina is widely distributed throughout the area, being recorded as frequent down to 3.5m in places, but the principal beds with abundant Z. marina have been recorded in depths of 0.5m above to 1.7m below CD. One Z. marina bed is present on medium-coarse sand in exposed conditions off the northwest of Lingay. Here, patches of abundant Z. marina were recorded down to 4.4m. Based on measurement of the delimited areas of Z. marina beds within ArcView, the current coverage of eelgrass beds is estimated to be approximately 167ha. 72

85 Sediments in the deep channel passing through the Sound of Eriskay and continuing beyond the causeway in a southwesterly direction vary from fine sands with a microalgal surface film to coarse sands and gravel, although the principal biotope is IMS.EcorEns. The channel acts as a sink for shell and dead plant material. Prior to the establishment of the causeway, Comhairle nan Eilean Siar (1999b) recorded the presence of several small maerl beds (IGS.Phy.R) of 5 35m 2 in area at depths of 6 10m in this channel. During the current survey dense living Phymatolithon calcareum was recorded at three locations in the channel, generally in association with a dense binding algal flora, especially of filamentous reds. At each location the maerl beds are patchily distributed so that delineation of the extent of maerl-covered sediment at each location was not achieved. However, in view of the high density of survey stations worked within the channel, it appears that maerl beds are highly localised here. The eastern extremity of the Eriskay channel is floored by a mixed substrate of sand, gravel, pebbles and shells, with drift and attached algae, such as Laminaria saccharina, Chorda filum and Ulva lactuca (IMX.LsacX). A tongue of fine-medium sand extends southeasterly from the eastern entrance to the Sound of Eriskay (Figure 5.2) supporting a similar community to the sands within the Sound (IMS.EcorEns). Beyond depths of 10 15m the sea bed in this area consists predominantly of rock and coarse sediments. Outcropping bedrock and boulders support communities tolerant of sediment scour, with mixed kelp forest extending to around 20m as MIR.XKScrR, beyond which it is replaced by scour-tolerant faunal turfs (MCR.Flu.HbyS). The coarse, often rippled, sediments in this region are mixtures of coarse sand and gravel, with the addition of pebbles in some pockets. Living maerl Phymatolithon calcareum, is common to superabundant over much of this region and overlies muddy sand in the deeper parts of the surveyed area. The maerl biotope (IGS.Phy.Hec) is extremely diverse, with up to 153 taxa being recorded in a single grab sample. Coarse sediment in this area not supporting living maerl displays the characteristics of CGS.Ven.Neo. As a result of the confusion evident in the multispectral classification (Figure 4.30) a revised classification was undertaken to add more detail on the biotope distribution map derived from a manual interpretation of direct field interpretation and satellite images (Figure 5.1). The revised classification was undertaken using the Knowledge Engineer expert system within the Erdas Imagine software package. This component allows for the development of a rule-based approach to multispectral image classification. A hierarchy of rules was developed which allowed for the boundaries of the manual interpretation (Figure 5.1) to guide the spectral discrimination of bright and dark targets within them. The resulting classification is shown in Figure 5.3. The classification allowed for the addition of greater detail in many of the demarcated biotopes. For example, in the bedrock and rock ridge categories, discrimination between the hard surfaces themselves and the coarse sand patches within them has been achieved. Similarly, in the intertidal unclassified surfaces, darker hard rock surfaces have been discriminated from brighter sand surfaces. Dense patches of seagrass have been highlighted within the delineated seagrass zones which displays their distribution within patches in much greater detail. As assessment of the area of seagrass so delineated gives an area of 75ha., much less than that estimated on the basis of delineation of polygons. Whilst this figure is likely to be an underestimate, as sparse canopies may not have been detected, it is likely to be a more realistic estimate of seagrass coverage than the earlier polygon based prediction. 73

86 Figure 5.1 Distribution of biotopes in the shallow western sector of the northern region of the Sound of Barra psac 74

87 Figure 5.2 Distribution of biotopes in the eastern sector of the northern region of the Sound of Barra psac Moderately exposed infralittoral rock (<20m) with scoured kelp and L. hyperborea biotopes (MIR.XKScrR, MIR.Lhyp.Ft&Pk) imoderately exposed cira c littoral rock (>20m) with scour tolerant faunal turf (MCR.Flu.HbyS) Fine to medium sand with ripples (IMS.EcorEns) Gravel and pebbles with maerl (CGS.Ven.Neo, IGS.Phy.HEc) Gravel and pebbles (ripples) with maerl (CGS.Ven.Neo, IGS.Phy.HEc) Land 75

88 Figure 5.3 Classification of biotopes in the shallow western sector of the northern region of the Sound of Eriskay, performed on the basis of manual interpretation and depth corrected bottom reflectance 76

89 5.2 Distribution patterns of biotopes Southern area A summary of biotopes for the southern area is given in Figure 5.4. Much of the exposed eastern flank of the region displays similar biotopes to those described for the area east of Eriskay. The sea bed is a mosaic of coarse gravelly sands comprising IGS.Phy.Hec and CGS.Ven.Neo, with outcropping bedrock and boulders. The circalittoral rock beyond 19m displays a scour-tolerant faunal turf (MCR.Flu.HbyS), which extends into the lower infralittoral beneath the kelp canopy. The infralittoral rock (<19m) throughout much of the southern region, is largely composed of scour-tolerant mixed kelps (MIR.XKScrR, EIR.LsacSac) and Laminaria hyperborea biotopes (MIR.Lhyp.Ft, MIR.LhypGz.Ft & Pk), depending upon the proximity of mobile sediments. Steep and vertical rock faces, much more extensive in the southern region than the northern, commonly have rich communities dominated by Alcyonium digitatum, Corynactis viridis and often large patches of the boring sponge Cliona celata (IR.CorMetAlc). Outer Oitir Mhór is situated in more sheltered conditions off the north of Gighay and Hellisay, where the extensive plain of slightly muddy fine sand at depths of 10 20m supports a diverse and abundant fauna (CMS.AfilEcor). This biotope also seems to predominate in North Bay. The increased shelter in Inner Oitir Mhór has promoted the establishment of soft muddy sand sediments in this area. The infauna displays similarities to CMS.AfilEcor but is overlain with scattered kelp (especially Laminaria saccharina) and, in places, shells (IMX.LsacX). Cordah (2000) recorded maerl at one site near the southeastern channel entrance to Inner Oitir Mhór, where tidal currents may be accelerated. Cordah (2000) also surveyed the rocky coastline of Aird Mhór along the southern margin of Inner Oitir Mhór. Here, bedrock and boulders extend to about 3m depth and support scoured seaweed communities (MIR.HalXK, MIR.XKScrR, MIR.PolAhn), as well as Laminaria hyperborea forests (MIR.Lhyp.Ft, SIR.LhypLsac.Ft). Very sheltered habitats are rare in this psac, the major occurrences being amongst the islands northwest of Fuiay and in the anchorage between Gighay and Hellisay. An extremely sheltered site in the inner region of the Gighay anchorage was examined. The sublittoral fringe here is dominated by a thick green algal mat of Enteromorpha prolifera overlying cobbles and boulders (IMX.FiG), above a Laminaria saccharina forest on boulders (SIR.Lsac.Ft) in the upper infralittoral. Soft muddy sands continue below 4m, covered with a mat of Rhodothamniella (=Audouinella) floridula to a depth of 11m (IMX.Tra), below which a diatomaceous film covers a heavily mounded sediment containing common Arenicola marina and megafaunal burrows (IMU.AreSyn). 77

90 Figure 5.4 Distribution of biotopes in the southern region of the Sound of Barra psac 78