Development of a Regional Habitat Classification Scheme for the Amirante Islands, Seychelles

University of Wollongong Research Online

Faculty of Science - Papers (Archive) Faculty of Science, Medicine and Health

1-1-2010

Development of a regional habitat classification scheme for the Amirante Islands, Seychelles

Sarah Hamylton University of Wollongong, [email protected]

Tom Spencer University of Cambridge, [email protected]

Annelise Hagan University of Cambridge

Follow this and additional works at: https://ro.uow.edu.au/scipapers

Part of the Life Sciences Commons, Physical Sciences and Mathematics Commons, and the Social and Behavioral Sciences Commons

Recommended Citation Hamylton, Sarah; Spencer, Tom; and Hagan, Annelise: Development of a regional habitat classification scheme for the Amirante Islands, Seychelles 2010, 43-55. https://ro.uow.edu.au/scipapers/995

Research Online is the open access institutional repository for the University of Wollongong. For further information contact the UOW Library: [email protected] Development of a regional habitat classification scheme for the Amirante Islands, Seychelles

Abstract A collaborative expedition between Khaled bin Sultan Living Oceans Foundation, Cambridge Coastal Research Unit and Seychelles Centre for Marine Research and Technology – Marine Parks Authority (SCMRT-MPA) was conducted to the southern Seychelles, western Indian Ocean, in January 2005. This resulted in a series of habitat maps of the reefs and reef islands of the Amirantes Archipelago, derived from remotely-sensed Compact Airborne Spectrographic Imager (CASI) data. The procedures used in map development, image processing techniques and field survey methods are outlined. Habitat classification, and egional-scaler comparisons of relative habitat composition are described. The study demonstrates the use of remote sensing data to construct digital habitat maps for the comparison of regional habitat coverage, a key function for coastal management.

Keywords seychelles, scheme, amirante, islands, island, habitat, development, classification, GeoQUEST

Disciplines Life Sciences | Physical Sciences and Mathematics | Social and Behavioral Sciences

Publication Details Hamylton, S., Spencer, T. & Hagan, A. (2010). Development of a regional habitat classification scheme for the Amirante Islands, Seychelles. Western Indian Ocean Journal of Marine Science, 9 (1), 43-55.

Development of a Regional Habitat Classification Scheme for the Amirante Islands, Seychelles

Sarah Hamylton, Annelise Hagan and Tom Spencer Cambridge Coastal Research Unit, Department of Geography, University of Cambridge, Downing Place, Cambridge, CB2 3EN, UK.

Keywords: Remote sensing, Geographical Information Systems, coastal management, habitat.

Abstract—A collaborative expedition between Khaled bin Sultan Living Oceans Foundation, Cambridge Coastal Research Unit and Seychelles Centre for Marine Research and Technology – Marine Parks Authority (SCMRT-MPA) was conducted to the southern Seychelles, western Indian Ocean, in January 2005. This resulted in a series of habitat maps of the reefs and reef islands of the Amirantes Archipelago, derived from remotely-sensed Compact Airborne Spectrographic Imager (CASI) data. The procedures used in map development, image processing techniques and field survey methods are outlined. Habitat classification, and regional-scale comparisons of relative habitat composition are described. The study demonstrates the use of remote sensing data to construct digital habitat maps for the comparison of regional habitat coverage, a key function for coastal management.

INTRODUCTION diversity and ecological status, allow comparison of the status within and Surveys of coastal environments between ecoregions, and facilitate provide information on the the detection of changes in coastal distribution and abundance of shallow ecosystems relative to established water benthic communities, as well baselines (English et al., 1997). In the as local topography and bathymetry. shallow water environments of the Standardised survey protocols permit tropics and sub-tropics, standardised assessment of regional biological surveys can be applied to ecosystems

Corresponding author: SH E-mail: [email protected] 44 S. HAMYLTON ET AL. such as coral reefs, seagrasses and temporally dynamic benthic surfaces mangroves, which collectively provide as discrete units. This requires valuable ecological functions and some form of classification scheme. services in the marine environment. Habitat classification schemes play Remote sensing instruments an important role in standardising provide a synoptic portrait of the the thematic content of regional Earth’s surface by recording numerical benthic maps, facilitating their use information on the radiance measured as a common baseline against which in a series of picture elements (pixels) regional subsets can be interpreted. In across a number of spectral bands a spatial context, classes are assigned (Green et al., 2000). When mounted to homogenous patches of surface on an airborne platform, they sample that differ in appearance from their habitat cover over extensive spatial surroundings but may vary widely in ranges. The ability to survey detailed size, shape, type, relative heterogeneity information from relatively large areas and boundary characteristics (Forman is particularly valuable at the coast and Godron, 1986). Ecological because changing weather, locations units can be divided hierarchically which are inaccessible or difficult to to accommodate user requirements. access, and the logistical challenges Top-level descriptions often identify of carrying out underwater fieldwork geomorphological context, while all compromise the ability to survey lower tiers describe the relative cover a representative sample of these of benthic organisms discriminated shallow water environments directly. from ground-verified data (e.g. Remote sensing data benefit the Mumby and Harborne, 1999). A mapping process by enabling accurate hierarchical classification scheme that extrapolation of information to broader subsumes both geomorphological and scales, providing information on areas ecological characteristics is systematic not accessible in the field. In this way, in structure with tier-level descriptors cost savings can be achieved. Mumby that cannot be used interchangeably. et al., (1999) found the combined use Geomorphological classes (e.g. fore- of remote sensing and field surveys to reef slope coral spur) can, therefore, be a cost-effective means of acquiring be coupled with ecological ones (e.g. meaningful survey data on the Caicos high cover of calcareous algae) within Bank when compared to a solely field- the same tier level. based approach. Many coastal mapping strategies employ coarse mapping at the Hierarchical classification schemes regional scale, augmented by finer for coastal zone management resolution maps at locations of To map coastal habitats, it is necessary specific interest (e.g. Borstadt et to represent spatially continuous and al., 1997). Despite the potential use DEVELOPMENT OF A REGIONAL HABITAT CLASSIFICATION SCHEME 45 of habitat classification schemes in discrete data structure on communities regional comparisons, most coastal that often present themselves as a habitat mapping has been conducted continuum of changing densities, on an ad hoc basis and displays such data formats lend themselves little consistency in terminology. well to analysis within Geographical This limits the interpretation of map Information Systems (GIS) (Burrough products, particularly where regional and McDonnell, 1998). Such an comparisons would be of value. The approach to data storage provides a use of digital habitat maps derived computationally efficient means by from remotely-sensed imagery is also which to carry out statistical analyses limited by these difficulties as they on landscape-scale datasets, e.g. range considerably in resolution. extracting coverage values for the Habitat maps are commonly different habitats. compiled from vector data, which partition an area in such a way that STUDY AREA each location falls into a polygon The Amirantes Archipelago, which assigned a value that is assumed to be lies SW of the extensive, shallow homogenous for all locations within water Seychelles Bank in the western that polygon. Although this imposes a Indian Ocean, comprises a group of (i.)

Seychelles

Figure 1. (i.) Location of the Seychelles Islands, western Indian Ocean. 46 S. HAMYLTON ET AL.

(ii.)

53º00 E

African Banks

Remire

D’Arros

St. Joseph

Sand Cay

Desroches Poivre Etoile

06º00 S Boudeuse Marie Louise Desnoeufs

Alphonse 07º00 S Bijoutiere and St. Francois

Figure 1. (ii.) The Islands of the Amirantes Bank, Seychelles. carbonate islands and islets extending at sea level with varying degrees of over a distance of ~152 km, from subaerial sand cay and coral island o 4°52’S (African Banks) to 6 14’S development. They have evolved (Desnoeufs) (Fig. 1). The majority over the last 6,000 years since the of the islands are coral reef platforms post-glacial sea level approached its DEVELOPMENT OF A REGIONAL HABITAT CLASSIFICATION SCHEME 47

Table 1. Aerial coverage of CASI data. cyclonic gyre to the south (between 40–15°S) and reversing monsoon Number of islands surveyed 13 gyres north of 10°S. The northern Area flown 268 sq kms boundary of the subtropical gyre Flight lines flown 110 lines is formed by the South Equatorial Current (SEC). During the northern Data volume (raw) 65 Gbytes hemisphere summer the SEC is Data volume processed (estimated) 150 Gbytes displaced northwards as far as present level on the Amirantes Bank 6°S, and thus into the region of the (Stoddart, 1984). The Amirantes Bank Amirantes Archipelago, with typical -1 is an elongate structure, measuring current speeds of 0.25 m s and an approximately 180 km by 35 km, estimated transport rate of 50 Sv 6 3 -1 deepest in its centre (up to ~70 m) with (where 1 sverdrup (Sv) = 10 m s ). a marginal rim 11-27 m deep. With These climatic patterns influence the exception of the islands of Etoile both the structural form and surface and Boudeuse, the reef platforms lie communities of the reefs of the towards the eastern margin of the Amirantes, which display notable Bank. The atolls of Alphonse and leeward-windward contrasts in reef Bijoutier/St François which form the platform development (Spencer et Alphonse Group are approximately 95 al., 2009). km further south. Of the 14 islands, 13 were mapped, the exception being METHODS Desroches, a shallow submerged atoll, The primary aim of this collaborative 19-21 km in diameter, lying 16 km to expedition was to use a Compact the east of the Amirantes Bank. Airborne Spectrographic Imager The climate of the western Indian (CASI) remote sensor onboard a Ocean is humid and tropical, with seaplane to conduct large-scale mean monthly temperatures always >20°C and an annual rainfall >700 mapping of the reefs and islands of the mm. Seasonal and inter-annual Amirantes Bank. The sites comprised climatic variability is determined laterally extensive shallow water by i) the SE Asian Monsoon and the landforms, which were ideally suited associated seasonal reversal of winds; to airborne mapping. Concurrent field ii) monsoon-related movements of surveys were conducted alongside the Inter-Tropical Convergence Zone (ITCZ); iii) changes in the position the airborne surveys. Data were and intensity of the South Indian collected on the terrestrial and marine Ocean subtropical high pressure; and habitats. Results from the CASI image iv) variations in ocean circulation processing provided the first detailed and sea surface temperature. maps of the distribution of shallow The surface ocean circulation marine habitats for each of these of the western Indian Ocean is characterised by a subtropical, anti- locations. 48 S. HAMYLTON ET AL.

Airborne remote sensing surveys between 0.005 and 10 km, in spatial Airborne remote sensing data were extent was employed in the analyses, acquired over 13 islands in the encompassing both intra-reef and reef Amirantes (Table 1). Reflectance data system landforms (Perry et al., 2008). were recorded between 430-850 nm Ground-referencing using a CASI sensor. The sensor was calibrated to measure radiance in 19 Ground-referencing was conducted spectral bands at a pixel size of 1 m2, in the terrestrial and marine yielding continuous data layers for the environments at eight of the islands. area. Synoptic coverage was acquired Over 1,500 ground-reference points by following a predetermined flight were recorded; locations were marked pattern over each island at an altitude with hand-held GPS units (horizontal of 1,000 m. This generated a series of accuracy of ±10 m). In shallow water, parallel flight lines, each 512 m wide, ground-referencing was conducted which could be geo-corrected and from the surface using a glass- processed. A landscape scale, ranging bottomed bucket lowered over the side

Table 2. Two-tier classification scheme for the marine habitats of the Amirantes Islands. First tier Second tier 1. Terrestrial vegetation: Trees and shrubs 1.1 Coconut woodland 1.2 Other trees and shrubs 2. Herbs and grasses 3. Saline ponds 4. Cleared/bare ground 5. Littoral hedge 6. Mangrove woodland 7. Coarse beach material & rocks 7.1 Coral sandstone/raised reef 7.2 Coral boulders 7.3 Beachrock 8. Beach sand 9. Rock pavement 10. Reef-flat sand 11. Seagrass 11.1 Low density seagrass/macroalgae 11.2 Medium density seagrass 11.3 High density seagrass 12. Lagoon patch reef 13. Lagoon sand 14. Fore-reef slope (not sand) 14.1 Coral rubble with coralline algae 14.2 Fore-reef slope coral spurs with coralline algae 14.3 Rocky fore-reef slope 14.4 Fore-reef slope (rubble and sand) 14.5 Fore-reef slope with coral 15. Fore-reef slope sand DEVELOPMENT OF A REGIONAL HABITAT CLASSIFICATION SCHEME 49

i ii iii

vii vi viii xi

v x ix

Figure 2. Habitat maps of the Amirantes Islands: i. African Banks, ii. Alphonse, iii. Marie-Louise, iv. Boudeuse, v. Bijoutier & St François, vi. Desnoeufs, vii. D’Arros & St Joseph, viii. Etoile, ix. Remire, x. Sand Cay, xi. Poivre. See Figure 1(ii.) for island locations. of a small boat. The boat was driven Correction Now (ACORN) algorithm at a constant speed along a consistent to retrieve radiance values at the line of bearing perpendicular to the water surface (ACORN, 2001). beach from a depth of ca. 20 m to Raw data were geo-corrected using the shallow (ca. 3 m), stopping at ground control points and a first order one minute intervals to record the polynomial model was applied to substratum type and the GPS position. correct for linear offset, with nearest Image pre-processing neighbour resampling (Erdas inc., 1997). Strips were mosaiced and a Prior to processing of the remote band-wise linear colour balancing sensing data, the effects of scattering model was applied to minimise and absorption in the atmosphere across-track variance, with histogram were corrected using the Atmospheric matching to adjust for radiance offset (Rees, 1990). 50 S. HAMYLTON ET AL.

100% 90%

80% 70%

60%

50%

40%

30%

20%

10%

0% African Remire D’Arros St. Joseph Sandy Cay Poivre Etoile Boudeuse Marie Desnoeufs Alphonse Bijoutier Banks Louise and St. Francois Terrestrial vegetation; trees and shrubs Herbs and grasses Buildings and other structures Littoral hedge Beach sand Rock pavement High density seagrass Lagoon patch reef Fore-reef slope sand Cleared/bare ground Saline pond Coarse beach material and rocks Mangrove woodland Seagrass Reef-flat sand Fore-reef slope material Lagoon sand Figure 3. Tier 1 breakdown of habitat coverage in the Amirantes Islands, as mapped by the Compact Airborne Spectrographic Imager (CASI).

Water column correction was where the radiance just above the

performed using the method outlined water surface Rrs (z=a) was provided by Purkis and Pasterkamp (2004), by atmospherically-corrected in which the formulae derived by imagery, the water column reflectance

Bierwirth et al., (1993) are adjusted of optically deep water (Rw) was to account for the refractive influence estimated from image statistics, of the water surface at the air/water and the depth (z) of each pixel was boundary and rearranged to solve for extracted from a bathymetric map

substratum reflectance, bR : generated using the Stupmf et al., band ratio model (2003). For each band, R = 1/0.54 x R –(1 – e-2kjz) R Equation 1 b z w the diffuse attenuation coefficient e-2kjz (k) was estimated through regression DEVELOPMENT OF A REGIONAL HABITAT CLASSIFICATION SCHEME 51

53º0053º00 E E 1 Terrestial vegetation:trees and shrubs Littoral hedge African Banks Rock pavement 05º00 S Beach sand Remire Herbs and grasses High density seagrass 10 Reef flat sand 5 Lagoon sand D’Arros Coarse beachmaterial and rocks 11 St. Joseph Seagrass Fore-reef slope material Sand Cay Fore-reef slope sand 7 Desroches Poivre Image classification Etoile 9 and development of 06º00 S the Amirantes Islands 4 Boudeuse classification scheme Marie Louise Desnoeufs 8 A maximum likelihood classification was used to

6 assign each pixel of the image to the most likely thematic class on the basis of statistical probability 2 (Mather, 2004). This supervised classification Alphonse 07º00 S required the user to define Bijoutiere and St. 3 substratum and vegetation Francois type in a number of pixels where the content had Figure 4. Tier 1 breakdown of habitat coverage in the Amirantes Islands. Pie chart key: 1. African been verified in the field. Banks, 2. Alphonse, 3. Bijoutier & St François, 4. Training signatures were Boudeuse, 5. D’Arros & St Joseph, 6. Desnoeufs, 7. thus established to analyse Etoile, 8. Marie-Louise, 9. Poivre, 10. Remire, 11. reflectance over each strip Sand Cay. of imagery across a number of spectral subclasses. In of digital data taken from a series of this way, a statistical population of white targets deployed at known and reflectance values was built up for varying depths. The gradient of the each island habitat. For each individual regression line of log-transformed flight strip, as much heterogeneity was data plotted against depth yielded an captured in the images as possible. estimation of k. A total of 2,018 signatures were 52 S. HAMYLTON ET AL. collected and evaluated before being loaded into a hand-held GPS. These merged into a subset of 720 signatures. locations were visited in the field This merging was undertaken on the and the habitat type at each location basis of spectral similarity, using was recorded in terms of the overall the Erdas Imagine signature editor classification scheme. Validation data (Spencer et al., 2009). As a parametric were compared against the habitat class classifier, the maximum likelihood assigned by the mapping process and function assumes that each thematic scored as being correct or incorrect. The sample category can be represented overall accuracy was expressed as the by a Gaussian probability density proportion of patches that were correct function, derived from the varying (Congleton, 1991). This approach image radiance across each spectral encompassed both the locational band (Thomas et al., 1987). Given the and thematic aspects of accuracy. user-specified training signatures, it Where islands could not be visited, a is possible to compute the statistical similar approach was undertaken by probability of a pixel belonging to comparing the maps with oblique aerial each thematic spectral class. photographs of the islands. Habitats At the regional scale, individual were identified in these photographs island habitat map keys were by three independent validators with combined to produce an overall expert knowledge of western Indian scheme for the Amirantes Islands. A Ocean shallow water geomorphology habitat classification scheme with a and ecology. hierarchical structure was developed Using the conversion tools in to accommodate user requirements, ArcMap (ArcGIS 9), a raster to vector field data availability and the spatial conversion was carried out on the and spectral resolution of the CASI resultant habitat maps and the cover sensor. The classification process of the different habitat types was yielded 15 classes at Tier 1; four of computed with respect to both the first these Tier 1 classes were subsequently and second tiers of the classification sub-divided into a series of Tier 2 scheme. classes, totalling 24 habitat classes (Table 2). RESULTS Map Validation Habitat classification scheme Classification accuracy was assessed Image classification provided a clear with field data collected in situ on the and accurate representation of the eight islands that were visited. Large, heterogeneity apparent in the raw heterogeneous patches of the island images (Fig. 2). The overall accuracy habitat maps were randomly selected of the maps when ground-truthed and their central coordinates were ranged from 67-77%. DEVELOPMENT OF A REGIONAL HABITAT CLASSIFICATION SCHEME 53

Seagrasses constituted the most structure and benthic characteristics widely represented shallow marine as class descriptors. In doing so, it habitat class (13-84% cover), drew on both of these aspects of the encompassing low, medium and remotely-sensed image. Langrebe high density communities. The (1998) describes three domains in most abundant seagrass species which spectral datasets can be viewed: were Thalassodendron ciliatum and image space, spectral space and feature Thalassia hemprichii. Fore-reef slope space. Geomorphological zones have material, reef-flat sand and lagoon more distinct contextual boundaries sand were also abundant (Fig. 3). than benthic assemblages and, as Habitat types particularly in such, the position of a pixel in typical subaerial conditions (Fig. 4) on the reef zones (Done, 1983), determined islands on the western margin of the by viewing the image space, can Amirantes Bank were characterised facilitated contextual editing. The by a restricted range of terrestrial classification algorithm on the other and littoral habitats, whereas those hand, operated in feature space to on the eastern side of the Bank were fully exploit the spectral information more diverse in habitat, particularly in of the image. subaerial conditions. The inclusion of a substantial set of training signatures facilitated DISCUSSION AND the application of image statistics CONCLUSIONS in the development of the habitat classification scheme. A top-down The map classification scheme approach in resolving the different incorporated the variable habitats from the CASI data was geomorphological nature of the islands found to be an appropriate technique within a consistent, standardised in devising the habitat classification approach. This was largely made scheme for the Amirantes Islands. possible by the use of the CASI Seagrasses were the most widely sensor. The image-based classification represented class in the maps as the employed data from an extensive set of extensive reef-flats provided the habitat training signatures, which represented needs of the genera Thalassodendron a combination of user and computer and Thalassia. Their requirements input into the classification process. include an adequate rooting The hierarchical structure of the substratum, sufficient immersion in classification scheme reflected the seawater and adequate illumination capability of the CASI sensor to resolve to maintain growth (Hemminga the various tiers at a comparably and Duarte, 2000). Seagrasses have high spectral resolution. The scheme broad, splayed leaves, a growth form employed both geomorphological that maximises light capture. These 54 S. HAMYLTON ET AL. appeared extensive when viewed from such comparisons can be extended above and remote sensing thus lends to the temporal domain, adding to its itself well to mapping their spatial usefulness in coastal management. coverage. They were, therefore, Acknowledgments: The following well represented in habitat coverage organisations are thanked for funding statistics, alongside the equally and logistical support in the field: extensive cover of reef-flat and Khaled bin Sultan Living Oceans lagoonal sands (Fig. 3). Conversely, Foundation, Seychelles Centre for some habitats, such as beach sand and Marine Research and Technology the fore-reef categories, appeared to - MPA, Island Development be limited in their spatial extent being Company, Seychelles and Great characterised by relatively steeply Plains Seychelles. Dr Chris Banks is sloping surfaces. thanked for assistance with image data There was a clear distinction in the processing. habitat maps between stable, vegetated islands located on a rock or reef REFERENCES platform (e.g. Alphonse) and islands ACORNTM (2001) Atmospheric Correction characterised by ephemeral sand cays Now: Analytical Imaging and surrounded by extensive seagrass Geophysics LLC, Version 3.12 beds (e.g. Etoile). These differences Bierwirth, P.N., Lee, T.J. & Burne, R.V. relate to the presence or absence (1993) Shallow Sea-Floor Reflectance of suitable antecendent topography and Water Depth Derived by Unmixing for postglacial coral growth and the Multispectral Imagery. Photogramm. variation in environmental parameters Eng. Remote Sens. 59: 331–338. such as wave and current fields across Borstad, G., Brown, L., Cross, W., Nallee, the Amirantes Bank. M. & Wainwright, P. (1997) Towards a Management Plan for a Tropical Comparisons of habitat coverage Reef-lagoon System Using Airborne between the 13 islands of the Multispectral Imaging and GIS. Fourth Amirantes revealed biological and International Conference on Remote geomorphological trends that would Sensing for Marine and Coastal not have been apparent without a Environments, Florida, March, 1997. standardised classification scheme Burrough, P.A. & McDonnell, R.A. (1998) applied at the regional scale. This study Principles of Geographical Information demonstrates how remotely-sensed Systems. Oxford University Press, Oxford. 327pp. imagery can be used for resource assessment and regional comparison Congleton, R. (1991) A Review of Assessing the Accuracy of Classifications of of habitat coverage. Such an approach Remotely Sensed Data. Remote Sens. would not have been possible in a Env. 37: 35-46. cost-efficient manner through field surveys alone. With repeat surveys, DEVELOPMENT OF A REGIONAL HABITAT CLASSIFICATION SCHEME 55

Done, T.J. (1983) Coral Zonation: Its Nature Perry, C.T., Spencer, T. & Kench, P.S. (2008) and Significance. In: D.J. Barnes Carbonate Budgets and Reef Production (ed.) Perspectives on Coral Reefs. States: A Geomorphic Perspective on Australian Institute of Marine Science, the Ecological Phase-Shift Concept. Townesville. pp. 107-147. Coral Reefs 27: 853–866. English, S., Wilkinson, C.R. & Baker, V. Purkis, S.J. & Pasterkamp, R. (2004) (1997) Survey Manual for Tropical Integrating in situ Reef-top Reflectance Marine Resources. Australian Institute Spectra with Landsat Tm Imagery to of Marine Sciences, Townesville. Aid Shallow-Tropical Benthic Habitat 390pp. Mapping. Coral Reefs 23: 5–20. Erdas Inc. (1997) Erdas Field Guide, Fourth Rees, W.G. (1990) Physical Principles of edition, Revised and Expanded. Atlanta, Remote Sensing. Cambridge University Georgia. 394pp. Press, Cambridge. 247pp. Forman, R.T. & Godron, M. (1986) Landscape Spencer, T., Hagan, A.B., Hamylton, S.M. Ecology. John Wiley & Sons, New York. & Renaud, P. (2009) Atlas of the 379pp. Amirantes. Cambridge Coastal Research Green, E.P, Mumby, P.J., Edwards, A.J. & Unit, University of Cambridge, UK. vi Clark, C.D. (2000) Remote Sensing + 66pp. Handbook for Tropical Coastal Management. UNESCO Publishing, Stoddart, D.R. (1984) Coral Reefs of the Paris. 316pp. Seychelles and Adjacent Regions. In: Stoddart, D.R. (ed.) Biogeography Hemminga, M.A. & Duarte, C.M. (2000) and Ecology of the Seychelles Islands. Seagrass Ecology. Cambridge W.Junk, The Hague. pp. 63-81. University Press, Cambridge. 292pp. Langrebe, D. (1999) Information Extraction Stumpf, R.P., Holdereid, K & Sinclair, M Principles and Methods for Multispectral (2003) Determination of Water Depth and Hyperspectral Image Data. In: Chen, with High-Resolution Satellite Imagery L. (ed.) (1999) Information Processing over Variable Bottom Types. Limn & for Remote Sensing. World Scientific Oceanogr. 48: 547–556. Publishing Company, Singapore. 30pp. Thomas, I.L., Benning, V.M. & Ching, N.P. Mather, P.M. (2004) Computer Processing (1987) Classification of Remotely of Remotely Sensed Images: An Sensed Images. A. Hilger Publishing, introduction. Second edition, John Bristol, UK. 268pp. Wiley & Sons, New York. 324pp. Mumby, P.J. & Harborne A.R. (1999) Development of a Systematic Classification Scheme of Marine Habitats to Facilitate Regional Management and Mapping of Caribbean Coral Reefs. Biol. Conserv. 88: 155-163. Mumby, P.J., Green, E.P., Edwards, A.J., & Clark, C.D. (1999) The Cost- effectiveness of Remote Sensing for Tropical Coastal Resources Assessment and Management. J. Env. Man. 55: 157- 166.