INTEGRATION OF SATELLITE REMOTE-SENSING OF SUBTIDAL HABITATS WITH VESSELBASED VIDEO SURVEY (DUBAI, UAE, ARABIAN GULF)* B. Riegl1, S. Andrefouet2, R.P. Moyer1, B.K. Walker1 1 National Coral Reef Institute, Nova Southeastern University Oceanographic Center, 8000 N Ocean Drive, Dania FL 33004, USA 2 College of Marine Science, University of South Florida, 140 7th Avenue S, St. Petersburg FL 33701, USA ABSTRACT Satellite remote sensing is convenient, rapid and cost-efficient for obtaining seafloor data over wide areas, but ground-truthing is needed. We compared results from a vesselbased video-survey and classified Ikonos imagery. Two datasets with 100% space cover were compared. Ikonos detected 8 bottom types, the video survey detected 7, which were compatible. Some differences were found which could be attributed to algal encroachment onto coral areas that had died in between the surveys. The satellite image classification suggested that the video-survey might have missed some coral areas due to poor visibility. The satellite data did not provide benthic health-status information, but clearly indicated that most dense coral areas were dead (“algae” pixels mixed among the “coral” pixels reflected the observed algal overgrowth of dead coral colonies). For areas less than 10m in depth, the results of the satellite-remote sensing and the vessel-based video-survey compared well. 1.0 INTRODUCTION The imagery produced by increasingly sophisticated satellite sensors provides dramatically improved remote-sensing capabilities and allows the production of very fine detail maps. To the earth-based observer, this provides new challenges in ground-truthing. The increased amount of high-quality remotely sensed information also requires techniques that allow large-scale and relatively rapid ground-truthing. Particularly in the marine environment this can be a challenge. This short contribution presents a study in Dubai, where the classification results of remotely sensed imagery obtained by Ikonos satellite was ground-truthed and verified by a video survey that had been conducted in the same areas several years previously. 2.0 METHODS 2.1 VIDEO SURVEY The video survey was conducted in fall 1995 from a 41 foot research vessel that was fitted with sidebooms fitted with downward-facing cameras with auto focus, auto exposure control, electronic image stabilizer, and a x24 digital zoom (Riegl et al 2001). The cameras were fitted with wide-conversion lenses that gave a roughly 1:1 height to field-of-view ratio. The width was about 15 cm higher, which allowed for overlap of adjacent images without interfering with image quality. The cameras were fixed on vertical extension tubes that could be raised and lowered, which was important for keeping a constant distance to the seafloor in variable water-depth. During the survey, average depth was 6.4m, which provided a 24m *

Presented at the 7th International Conference on Remote Sensing for Marine and Coastal Environments, Miami, Florida, 20-22 May 2002

wide swath of continuous video coverage. Survey lines were set at 20m spacing, which ensured overlap of adjacent lines and therefore 100% coverage of the seafloor. The video images were geo-referenced by means of a sound signal that was recorded at predetermined waypoints. These were so designed that at constant vessel speed a constant number of video frames would be recorded between waypoints. Video footage was then viewed and observers identified bottom types. After viewing several hours of tape, a list of bottom types that could be visibly discerned was drawn up. These were then identified and their distribution recorded along the survey lines.

Figure 1: Principle of the video survey and vessel setup. 2.2 IKONOS SATELLITE IMAGERY An Ikonos satellite imagery scene was obtained for the area between Ras Ghantoot and Jebel Ali in the United Arab Emirates (Emirate of Dubai) in July 2001, concentrating on coverage of the immediately offshore marine area. Ikonos imagery was delivered in two resolutions: 4m pixel resolution in the red (632698nm), green (506-595nm), blue (445-515nm) and near infrared bands (757-853nm), and 1m pixel resolution in the panchromatic band. The actual ground sampling distance of the sensor is 3.2m to 4.8m,

which is resampled for a 4.0m map increment. Only the RGB bands were used for classification of marine bottom types. Imagery was imported into ERMapper, where it was processed for maximum contrast (99% clip) and was then subjected to supervised classification with operator-specified training regions. 2.3 GROUNDTRUTHING For groundtruthing purposes the survey area was checked and quantitatively sampled by divers on a grid of equally spaced sample sites. In these areas, the bottom type and the depth were recorded. Since coral areas were considered to be of highest interest, these were sampled by means of line transects (see Riegl 1999). 3.0 RESULTS 3.1 VIDEO SURVEY The video-survey resolved 7 habitat types that could be mapped: - 1) Three Acropora branching coral dominated coral assemblages of increasing density (sparse=living coral cover of the substratum<25%, medium dense=living cover<50%, dense=living cover>50%). - 2) A dense assemblage of columnar Porites harrisoni corals. - 3) An assemblage of widely spaced (1-4m) individual, large Porites boulder corals. - 4) Seagrass (mainly Halodule and Halophila) and algae beds (Sargassum and Hormophysa cuneiformis, which were not differentiated but lumped into one category. - 5) Barren sand

Figure 2: Classification (maximum likelihood algorithm in ERMapper) and illustration of detected bottom types between Ras Ghantoot and Jebel Ali in Dubai (UAE), Arabian Gulf.

These habitat types were mapped in their outer boundaries by identifying on video the transition from one habitat type into the next along the survey lines. This was possible since each video frame was georeferenced. These points were then connected to form the outline. On the Ikonos image, seabed features were resolved best by using the RGB bands. Near IR and panchromatic were not used. Supervised classification was used to separate the different habitats. Eight habitats were differentiated: - 1) Dense coral growth (including Acropora and Porites areas) - 2) Sparse, widely spaced Porites coral growth - 3) Seagrasses - 4) Shallow algae beds (mainly Sargassum) - 5) Deep algae beds (Sargassum and Hormophysa) - 6) Shallow barren sand - 7) Deep barren sand - 8) Bare hardground We used the enhanced maximum likelihood algorithm provided by ERMapper, which assigns a cell to a class by taking into account the distance weighted by the covariance matrix of means and the prior probability of the cell belonging to it. Classification using Mahalanobis distance provided very similar results. The boundaries of the habitat types obtained by video survey were then superimposed over the classified Ikonos image. It was found that the habitat boundaries obtained by image classification coincided well with the habitat boundaries obtained by the video survey. Due to the coarser resolution on the Ikonos image (4m pixel size), it was not possible to resolve coral assemblages as finely as in the video survey. Thus, where the video survey had been able to differentiate four differently dense coral assemblages of branching and columnar corals and a fifth assemblage of widely spaced boulder corals, the classification of the satellite image was only able to differentiate between dense coral areas (lumping the four branching and columnar coral assemblages) and sparse coral areas. Habitat type video survey 1) <25% cover branching coral assemblage 2) <50% cover branching coral assemblage 3) >50% cover branching coral assemblage 4) Dense columnar coral assemblage 5) Widely spaced boulder corals 6) Algae and seagrasses 7) Barren sand

Habitat type Ikonos image 1) Dense coral assemblage 1) Dense coral assemblage 1) Dense coral assemblage 1) Dense coral assemblage 2) Widely spaced boulder corals 3) Seagrasses 4) Shallow algae 5) Deep algae 6) Shallow barren sand 7) Deep barren sand

The correctness of the classifications both qualitatively and in space was proven by diver surveys. 4.0 DISCUSSION Overall the coincidence of the habitat delineation obtained by vessel-based video survey and Ikonos satellite imagery was highly satisfactory. As a vehicle for large-scale data-acquistion, Ikonos imagery, even if purchased commercially, proved cheaper and easier. While having the apparent advantages, the correct interpretation of the classified satellite image remained tricky and required repeated ground-truthing.

Figure 3: Overlay of benthic habitat boundaries obtained by video survey in 1995 over the classified Ikonos image of 2001. Note the good correlation in the right hand portion (between Ras Hasyan and Jebel Ali) where the video survey was conducted with greater intensity than in the left hand section (Ras Ghantoot to Ras Hasyan) and apparently missed some coral areas due to bad visibility that are seen in the Ikonos image.

Some apparent misclassifications existed, where pixels inside the dense coral areas were classed as “dense algae” pixels. Groundtruthing by diving revealed that these pixels were actually correctly identified: many corals within the dense coral areas had died during a positive sea-surface temperature anomaly in 1996 and were now overgrown by a turf of fleshy brown algae (Lobophora). These algae have a very similar spectral signature to the other brown algae making up the deep algal areas (mainly Hormophysa and Sargassum) and were therefore assigned to the same class.In other areas where algae pixels were seen in areas identified by the video survey as dense coral, almost always situated on the fringes of the dense coral areas, groundtruthing by divers showed that Sargassum algae had started to encroach onto the dead corals. Thus, indirectly, via differentiating between algae and corals, we were also able to differentiate between live and dead corals – live corals having had a coral spectral signature and grouping as “corals”, dead corals being overgrown by brown algae and therefore grouping as “deep algae”. Ikonos imagery allowed to detect shifts in composition within defined coral assemblages at an even higher level of spatial resolution than was previously possible (Dustan et al. 2001). The detected changes were not only on a “whole reef” scale, which showed live and dead areas relative to the whole structure, but

it was possible to investigate individual areas of dense/sparse growth and differentiate within these the live and dead corals. While still short of being able to identify individual corals (average coral diameter in dense Acropora areas was 1-1.5m, pixel size resolution was 4m), Ikonos imagery comes very close to this goal.

Figure 4: Explanation for apparent misclassifications in the dense coral areas: dead corals are overgrown by brown algae (Lobophora) and therefore are classed as algae (grey pixels) and not as corals (red and orange pixels). The sides of the coral area are encroached by Sargassum brown algae (green pixels), which is shown by green pixels encroaching the red area. 5.0 ACKNOWLEDGEMENTS Ikonos satellite imagery was made available via the NASA scientific data purchasing program and was handled by F. Muller-Karger and S. Andrefouet. Groundtruthing and video surveys were supported and partly developed by Martin Mid East Ltd (Abu Dhabi and Dubai) and also supported by the Dubai Municipality. BR was additionally funded through Austrian Science Foundation Grant P-13165-GEO. 6.0 REFERENCES P. Dustan, E. Dobson, and G. Nelson. “Landsat Thematic Mapper: Detection of shifts in community composition of coral reefs”. Conservation Biology 15(4), Pp. 892-902, 2001. B. Riegl. “Corals in a non-reef setting in the southern Arabian Gulf (Dubai, UAE): fauna and community structure in response to recurring mass mortality”. Coral Reefs 18(1). pp. 63-73, 1999. B. Riegl, J. L. Korrubel, and C. Martin. “Mapping and monitoring of coral communities and their spatial patterns using a surface-based video method from a vessel”. Bulletin of Marine Science 69(2). Pp. 869880. 2001.