skip to primary navigation skip to content

Mapping coral reef habitats in the Amirantes and southern Seychelles, western Indian Ocean

Mapping coral reef habitats in the Amirantes and southern Seychelles, western Indian Ocean

The presence of many drowned reefs and shallow banks, rather than sea level reefs, on the Seychelles Bank, the Amirantes and the Mascarene Plateau in the western Indian Ocean suggests environmental controls may prevent coral reefs reaching their growth potential in this region (Spencer and Turner, 2001; Spencer et al., 2005). Following the catastrophic impact on corals of the 1997-1998 El Niño Southern Oscillation ocean warming event in this region (Spencer et al., 2000; Spalding et al., 2001; Teleki and Spencer, 2002a, 2002b; Hagan and Spencer, in press), it seems possible that periodic high sea surface temperatures are an important control on reef growth and maintenance.

Some predictions suggest an increase in El Niño frequency and severity with global environmental change in the near future; if so, this may have important consequences for the health of reefs and associated ecosystems in this region and the strongly marine-based economies of countries such as the Seychelles.

It is important, therefore, to better define the shallow marine ecosystems of this region, both in terms of current resource and as a baseline against which to monitor future changes in coral, seagrass and mangrove extent.

index00.jpg index01.jpg

The banks, atolls and islands of this region are often large (the Providence Cerf - Bank is 40 km in length and 1 - 10 km in width) and not amenable to detailed ground mapping over relatively short field visits. Such sites are however ideally suited to extensive airborne mapping. This was the approach taken in a collaborative expedition between the Khaled bin Sultan Living Oceans Foundation, the Seychelles Centre for Marine Research and Technology - Marine Parks Authority (SCMRT - MPA) and the Cambridge Coastal Research Unit in January 2005 (Ashworth et al., 2005). Overall science co-ordination in Cambridge was by Dr Spencer with the field team being led by Dr Annelise Hagan. The expedition also made immediate post-tsunami observations throughout the Seychelles (Spencer, 2006; Hagan et al., 2006).

index02.jpg index03.jpg
Figure 2A: Figure 2B:
Support vessel,
M.Y. Golden Shadow
Seaplane Golden Eye
on board the Golden Shadow

A CASI sensor, onboard the seaplane 'Golden Eye' (Figure 2B), was flown to collect reflectance data for the islands of Providence - Cerf, St. Pierre and all the islands of the Amirantes archipelago, covering a total area of over 800 sq km at a resolution of 1m2 pixels (Figure 3A, 3B). A total of 10 bandwidths were selected in the blue (400-500nm) and green ranges (500-600nm), 2 bandwidths in the red range (600-700nm) and 1 bandwidth in the infra-red range (800-900nm). Ground truthing and quantification of airborne mapping by local quadrat-based survey of shallow marine communities was carried out from small boat deployments (for both cover characteristics and biomass (where appropriate)) from the research vessel 'Golden Shadow' (Figure 2A). Local SCUBA diving, with underwater videography, allowed assessments of coral condition, including assessments of coral recovery from coral bleaching associated with the 1997-98 warming.

index04.jpg index05.jpg
Figure 3A: Figure 3B:
CASI flight lines at Alphonse (top) and
St. François (bottom)
Alphonse Atoll, southern Amirantes taken from the International Space Station (Earth Sciences and Image Analysis Laboratory; NASA-Johnson Space Centre, 2004)

Key image processing steps for producing the habitat maps from the raw imagery were geocorrection, the collection of training signatures for the maximum likelihood classifier, supervised classification and the subsequent addition of external information, such as field data, aerial photographs and expert knowledge of the area, see figure 4.


Figure 4: Key image processing steps

The geocorrection process corrected the position and orientation of one strip to its neighbour. Neighbouring strips were subsequently corrected until all strips for the site were corrected to each other. Common points on adjacent strips, known as ground control points (GCPs) were identified and the differences between them were modelled. Once all strips for a site were corrected, they were mosaiced into a single image covering the whole island.

A classification was then performed to convert the observed intensity values in the seventeen bands of the overall image into user defined habitat classes. Training for a supervised classification requires, for each class, the identification of multiple homogenous regions. Spectral signatures of training regions were then compared and, if appropriate, merged. Signatures defined populations of different habitat classes in feature space based on the seventeen dimensions represented by the individual bands. A team of experts held, at least, one meeting per site to interpret classified images using a combination of additional data sources, including photographs and groundtruth data. Pixels clearly located in the wrong position based on their contextual surroundings were recoded. This process was repeated for 12 islands, spanning 910 training areas across 110 flight lines, see table 1.

Island Flight Lines Training Areas Pixels classified
African Banks 8 71 27345198
Remire 12 87 19464849
D'Arros 6 58 21815442
St Joseph 11 84 30097365
Sand Cay 5 77 8485542
Poivre 10 92 20239183
Etoile 5 65 15380358
Boudeuse 6 54 9654526
Marie Louise 7 78 7838570
Desnoeufs 7 45 7806643
Alphonse 16 87 22548426
Bijoutier/St Francois 17 112 77691652

Table 1 Number of flight lines, training areas and pixels classified for each Island

A classification scheme was developed that was guided by the data collected in the field, but ultimately led by the discriminating ability of the CASI sensor. This covered both terrestrial and benthic assemblages and reflected the diverse nature of the various islands on the ridge, see figure 5.


Figure 5: Amirantes Habitat Classification scheme

Absolute correction of the imagery was achieved using large plastic sheets within the region covered by the sensor, which were located underwater at depths of 5, 10 15 and 20 m and on land. The location of these targets was taken using a handheld GPS. Targets were identified on the imagery and the underlying map model was changed to ensure that the field GPS coordinates matched the location of the same points on the imagery. This overall process flowline was repeated for 12 islands to develop a range of maps for each island on the Ridge, see figure 6.


Figure 6: Habitat maps created of the Amirantes Ridge

Hardcopy maps of each island were supplied to the Government of the Republic of the Seychelles, see figure 7. Having provided the first modern maps of all these locations, it is anticipated that the findings will be incorporated into the wider plans for the conservation of marine biodiversity in the western Indian Ocean region.


Figure 7: Hardcopy Map of the Poivre Island habitat