Dr Thomas Heege being employed by the commercial company EOMAP, this again does not alter the authors’ adherence to all the PLOS ONE policies on sharing data and materials.
Conceived and designed the experiments: HTK KW LEB TH. Performed the experiments: HTK KW LEB TH. Analyzed the data: HTK KW LEB TH. Contributed reagents/materials/analysis tools: HTK KW LEB TH. Wrote the paper: HTK KW LEB TH.
Research, monitoring and management of large marine protected areas require detailed and up-to-date habitat maps. Ningaloo Marine Park (including the Muiron Islands) in north-western Australia (stretching across three degrees of latitude) was mapped to 20 m depth using HyMap airborne hyperspectral imagery (125 bands) at 3.5 m resolution across the 762 km2 of reef environment between the shoreline and reef slope. The imagery was corrected for atmospheric, air-water interface and water column influences to retrieve bottom reflectance and bathymetry using the physics-based Modular Inversion and Processing System. Using field-validated, image-derived spectra from a representative range of cover types, the classification combined a semi-automated, pixel-based approach with fuzzy logic and derivative techniques. Five thematic classification levels for benthic cover (with probability maps) were generated with varying degrees of detail, ranging from a basic one with three classes (biotic, abiotic and mixed) to the most detailed with 46 classes. The latter consisted of all abiotic and biotic seabed components and hard coral growth forms in dominant or mixed states. The overall accuracy of mapping for the most detailed maps was 70% for the highest classification level. Macro-algal communities formed most of the benthic cover, while hard and soft corals represented only about 7% of the mapped area (58.6 km2). Dense tabulate coral was the largest coral mosaic type (37% of all corals) and the rest of the corals were a mix of tabulate, digitate, massive and soft corals. Our results show that for this shallow, fringing reef environment situated in the arid tropics, hyperspectral remote sensing techniques can offer an efficient and cost-effective approach to mapping and monitoring reef habitats over large, remote and inaccessible areas.
Coral reefs are complex ecosystems which create diverse habitat mosaics and support a wide range of organisms
Understanding the complexity of coral reef ecosystems, their monitoring and management require information which includes bathymetry and habitat maps. The Ningaloo region is extensive, stretching across three degrees of latitude (22°–24°S). Access from the shoreline or with small boats is difficult along much of the coast and logistics for field work are significant as the region is remote and there is very limited support infrastructure. Large areas with clear waters such as those off Ningaloo in Western Australia naturally lend themselves to the application of optical remote sensing as a means of gathering data on coral reef habitats.
Habitat maps derived from imagery collected by various remote sensing instruments have become widely used in marine monitoring and management in the past two decades
The current habitat map of Ningaloo Reef includes only general classes based on visual interpretation of aerial photos
Satellite or airborne remote sensing has increasingly been employed to map coral reef communities worldwide
Only a few studies have attempted mapping of coral reefs using airborne hyperspectral data, such as CASI
Some authors have acknowledged this mixed pixel factor in coral reefs by creating r„mixture groups to represent various realistic scenarios of change in reef communities„ using
In this study, using hyperspectral data, we aimed to develop a marine habitat classification system suitable for the entire Ningaloo Reef and to map the seabed habitats. Within the mapped habitats, special attention was given to percentage cover and coral types present. Objectives of the study were to firstly, acquire and compare field and image derived spectra from cover-forming reef components along the Ningaloo Reef. Secondly, we aimed to develop a classification system applicable for the entire reef. Thirdly, we set out to map and extract summaries for the entire reef using an operational approach with standardised processing. The last objective was to test a range of thematic and spatial generalizations relevant for mapping large areas such as Ningaloo Reef.
Ningaloo is part of the diverse reef system of the Indian Ocean and one of the least anthropogenically disturbed
Outlines of the state and Commonwealth marine park boundaries are also indicated.
The region has a hot and arid climate and the mean annual air temperatures range from 11°C (min) to 38°C (max)
In this study we divided Ningaloo Reef into northern, central and southern regions, based largely on the width of the lagoons and included the Muiron Islands as a separate area since it has not been previously mapped (
HyMap data with 125 spectral bands (450–2500 nm range, 26 bands in the visible range) at 15 nm bandwidths and 3.5 m pixels were acquired over 10 days in April and May 2006 by HyVista under contract to the Australian Institute of Marine Science. The total area of the survey covered 3 400 km2, encompassing Ningaloo Reef to a 20 m depth, as well as the strip of coastal land adjacent to the Ningaloo Marine Park (
The 67 calibrated sensor radiance flight lines (each approximately 30 km long) were individually processed using the physics-based Modular Inversion and Processing System (MIP)
The resulting flight lines of the subsurface reflectance were geo-referenced and mosaicked to generate 17 image data blocks. The correction of water column-related effects was performed using MIP WATCOR module to retrieve the bathymetry and sea floor reflectance
The MIP processing was performed independently of any spectral data collected in the field, such as optical properties of the water constituents, or specific reflectance properties of the sea floor classes. The specific inherent optical properties of the water constituents were first analysed with optical closure calculations in several adjacent deep water areas, and then used as a fixed set of values for the whole reef. All spectra for the sea floor classification were derived by extracting the spectral sea floor characteristics from different sites over the survey area. The statistical variation within each group was analysed for the spectral overlaps between the groups. Class-specific spectral features were used to establish configuration settings for the fuzzy logic discrimination of classes.
Airborne data processing and development of the image classification and validation data sets were supported by ten field trips to different parts of Ningaloo Reef between 2006 and 2009 as it was not possible to collect all field data around the time of HyMap acquisition. There were several reasons for this, namely, logistics and costs, size of the area, and the airborne data collection schedule which depended on weather conditions made simultaneous collection of field data difficult. However, the majority of data points were collected around the same season (April) with similar macro-algal growth conditions. Data were obtained at locations with fairly homogenous cover type, therefore ensuring their representativeness and allowing for any positional errors
Spectral reflectance measurements from reef components including sand, coral and algae were collected
As the field data contained information on percentage cover of up to nine cover types per location (
Name | Code |
Hard coral | HC |
Soft coral (e.g. |
SC |
Branching coral | CB |
“Blue tip” branching coral (Acropora cervicornis) | CBT |
Digitate coral | CD |
Encrusting coral | CE |
Submassive coral | CS |
Tabulate coral | CT |
Massive coral | CM |
Foliaceous coral | CF |
Turfing algae or macro-algae-covered intact dead coral or rubble | TA- or MA-covered IDC or R |
Limestone pavement | LP |
Macro-algae (consisting largely of |
MA |
Rubble | R |
Sand | S |
Turfing algae | TA |
Hierarchy of the habitat classes was based on the type and percentage cover and class labels were made up of a combination of the single cover type in relation to their percentage cover in the sample site. Class selection also incorporated the frequency of occurrence of ground truth points per class in the data set, since a minimum number of points per class was required for training and validation of the classification.
Spectral analysis of
Class label | Description | Example |
Continuous classes | If cover > = 90%, the point was considered as ‘pure’; the remainingcover types were omitted and the class name received a prefix“Continuous” | Component 1 (most dominant) is limestone pavement with 95% cover, component 2 (second most dominant) is hard coral, with 5%, resulting in the label “Continuous limestone pavement” label and component 2 input omitted |
Mixed classes with single dominant category | If the dominant type was biotic and cover was between 50–90% andthe difference between the highest and second highest cover> = 30%, the remaining cover types were incorporated into thename. The name was derived from the dominant componentwith the prefix “Dominant” | Component 1 is limestone pavement with 70% cover, component 2 is hard coral with 25% cover, resulting in the label “Dominant limestone pavement” label and the second label of hard coral for component 2 |
Mixed classes with equal cover | If the (Cover 1– Cover 2) < = 20% they were considered equal;and each received the prefix “equal” if the sum of the equalpercentages > = 90% | Component 1 is limestone pavement with 50% cover, component 2 is hard coral with 45% cover, therefore (Cover1-Cover2< = 20% and component 1 was assigned a label “equal limestone pavement” and component 2 received label “equal hard coral” |
Mixed classes that donot fall into the abovecategories | Mixed classes that did not fit into the above categories remainedin the order they were in and receive the prefix “1”, “2”, “3”, etc.depending on their percentage value | Component 1 is limestone pavement with 60% cover; component 2 is hard coral with 35% cover, component 3 is sand with 5% cover, therefore labels were in that order: “1-limestone pavement”, “2-hard coral” and “3-sand” |
Before the training sites representing final habitat classes could be used for classification, we undertook analysis of image-derived spectra from the 600 field locations. The statistical software package R (R Development Core Team 2008) was used for multivariate spectral analysis to examine class separability, detect outliers and potentially regroup classes. Spectra from the bottom reflectance mosaics were extracted and statistical analyses, including principal component analysis (PCA), hierarchical clustering and Jeffries-Matusita (JM) distance, were performed on the image-derived spectra. A threshold of ≥1.90 was used for JM distance as it indicates good separability
Following the spectral analysis, the 67 habitat classes established by class frequency analysis were reduced to 46 as a result of deleting classes with high spectral similarity and outlier points. The final spectral library set was randomly stratified to 70% for training and 30% for validation.
The habitat mapping was performed on the HyMap mosaics using a supervised classification approach based on the image derived spectral signatures using the MIP software. The classification module incorporated fuzzy logic and first and second order derivatives in addition to reflectance data from the 26 bands which were useable underwater. Rule sets per class included the spectral class ranges as input for the classification. Only one configuration file (spectral signatures) was used in the classification which guaranteed consistent classification results over the whole data set, as well as allowing a more automated and standardised approach. For the final classification, all spectral sub-classes representing sand were merged as an objective of the study was to maximise mapping of classes containing corals. The same rule applied to limestone pavement.
Several post-classification steps were applied to the habitat data, including merging of the image data blocks, masking inconsistencies in deeper areas and generalising the classification at thematic and spatial levels.
To facilitate wider access and use of the data, several hierarchical, thematic levels were created for the classification map by generalising and combining the 46 habitat classes in a look-up table, which could be linked to the classification image using GIS software. After extensive consultation with the range of potential users (e.g., ecologists, biologists, conservation managers and planners), the habitat classes were organised at five levels and sub-levels. The logic of the look-up table was from the simplest (most general) to the most detailed (complex) description in terms of the number of reef components, while also allowing capture of the continuum of cover density from very high (continuous >90%) to very sparse (<20%) cover
Class statistics were calculated to determine the distribution, area and percentage cover of marine habitats. They were calculated from the full resolution data to determine the distribution and proportions of the 46 classes across different geographic domains.
Accuracy assessment of the classification was performed using field validation points. These were randomly stratified for the classes, so that both frequently and less frequently occurring classes were represented in the validation data in similar proportions. As validation data have the inherent issue of geo-location error either in the imagery or the field data, a radius of 10 m was generated around each validation point and the class labels extracted for pixels within that radius. If the same class as the validation class occurred within the 10 m radius, then the accuracy was accepted as correct.
Due to the high spectral similarity and thus “fuzziness” of classes
Type of class fuzziness | Examples |
Similar degree of cover of one class component | Validation area is class “Sparse macro-algae with sand” and classified area is “Patchy macro-algae with sand” |
One or some class component(s) of mixed classes are the sameand other(s) are different | Validation area is class “Patchy macro-algae with sand” and classified area is “Patchy macro-algae with pavement and sand” |
Certain coral growth forms spectrally and texturally similar | Validation area is class “Continuous branching coral” and classified area is “Continuous digitate coral” |
The habitat map of the Ningaloo Marine Park, generated by the Department of Environment and Conservation (DEC), includes eight habitat classes: shoreline reef, coral reef community (subtidal), coral reef community (intertidal), macro-algae, subtidal reef (low relief/lagoonal), subtidal reef (low relief/seaward), sand and pelagic. These were created from visual interpretation of aerial photographs
This project was undertaken under the Department of Environment and Conservation permit to enter the Ningaloo Marine Park for the purpose of undertaking research. No live specimens were removed.
Analysis of
At Ningaloo Reef, only frodose macro-algae occurred at spatial scales detectable by the sensor though coralline and turfing algae also occur.
Compared to biotic, abiotic cover types had higher reflectance; highest for sand and lowest for dead coral and limestone pavement. Two spectra for sand are presented to illustrate the spectral range (
As expected, the image-derived spectral analyses of pure reef components covering a whole pixel showed far less differentiation than the
PCA and hierarchical cluster analysis for the coral classes with more than 90% cover within a pixel showed a high spectral similarity for continuous branching, digitate, massive and soft corals, despite the spectral differences visible in the
Axes created through the PCA process removed correlations evident in untransformed spectra and allowed identification of outliers, trends and groups.
PCA and cluster analyses showed that the continuous tabulate coral class had a relatively large spectral range, partly overlapping with the continuous branching and digitate coral classes. However, on applying the JM distance ( = 2) they were found to be separable. Continuous tabulate coral and the mixed class of dominant tabulate coral with digitate coral were not separable with a JM distance of 1.3.
Continuous macro-algae and several mixed coral classes had a high similarity in PCA and cluster analyses but JM distance results indicated that they were separable. Classes that included the same components with a similar degree of cover (e.g., dominant macro-algae or dominant macro-algae with sand <10%) were found to be spectrally similar. Where the biotic cover was sparse or patchy, it had a low spectral influence on the pixel reflectance and the pixel spectrum was driven by the abiotic cover. For example, limestone pavement and sand had large brightness differences, so classes with sparse hard coral (<20%) with limestone pavement or sparse hard coral (<20%) with sand, were differentiated. These results were used to determine the “fuzziness” between classes during the final accuracy assessment.
The operational approach to classification performed on geo-referenced data mosaics with a single training set and classification parameters for an area that stretched across three degrees of latitude proved very successful. This is especially relevant for large, multi-flight line data sets. Five thematic classification levels and sub-levels were created, ranging from a basic level with three classes (biotic, abiotic and mixed) to the most detailed with 46 habitat classes (consisting of all benthic components and hard coral growth forms in continuous or mixed covers) (
Grey areas on the map do not contain any coral component discernible within a pixel. Legend codes explained in
This study mapped 762 km2 of the reef which included 5.9 km2 (8%) of coral mosaics (sparse to dense cover), 51% of macro-algae and turfing algae and 41% of sand and limestone pavement (
Region | Total area mapped (ha) | Classes with corals (ha) (%) | Classes with macro- or turf algae (ha) (%) | Abiotic classes (ha) (%) |
Muiron Islands | 2419 | 223 (9%) | 1766 (73%) | 430 (18%) |
Northern | 27349 | 2263 (8%) | 14567 (53%) | 10519 (39%) |
Central | 38305 | 2537 (7%) | 17848 (46%) | 17920 (47%) |
Southern | 8090 | 836 (10%) | 4844 (60%) | 2410 (30%) |
The four regions along the Ningaloo Reef (including the Muiron Islands) showed distinct differences influenced by the bathymetry as well as their geographic position (latitude). The northern and central regions had well developed lagoons, features mostly lacking at the Muiron Islands and in the southern area. The width of the mapped area narrowed down from just over 4 km in the north to about 0.5 km in the south. Four maps (
(Ar) Subsurface reflectance of the shallow platform with raised edges and a limestone ridge leading to the Muiron Channel on the left (west), (A) habitats dominated by a limestone platform surrounded by a mix of macro-algae. (Br) Subsurface reflectance of a nearshore area, (B) habitats of the slopes dominated by algae on pavement close to the shore and a zone of dense coral further offshore. (Cr) Subsurface reflectance of a flat limestone platform, (C) habitats with dominant coral cover in the western part and macro-algae with limestone pavement near the shore. Legend from
(Ar) Subsurface reflectance of nearshore, sublittoral pavement along a rocky shore, (A) habitats of extensive macro-algae, limestone pavement and sand. (Br) Subsurface reflectance of outer reef flat, (B) spur and groove structures with coral and macro-algae transitioning to tabulate coral and sand in the deeper lagoon. (Cr) Subsurface reflectance of the littoral alluvial fan off Yardie Creek, (C) habitats with limestone pavement and adjacent macro-algae with sparse coral. (Dr) Subsurface reflectance of the back reef, (D) back reef on the northern edge of the reef pass with clusters of bommies south and east of the reef flats. Legend from
(Ar) Subsurface reflectance of the edge of the reef flat on the southern edge of a reef pass, (A) habitats of the edge of the reef flat with transition from pavement to patchy macro-algae with a number of coral bommies. (Br) Subsurface reflectance of northern part of a large bay, (B) oval patterns of pavement and sparse macro-algae, with a large bommie in the centre (possibly grazing halos). (Cr) Subsurface reflectance of the Five Fingers Reef, (C) linear limestone ridges with dominant hard coral surrounded by macro-algae due west. (Dr) Subsurface reflectance of a nearshore area south of Pelican Point, (D) complex habitat pattern in the nearshore area with the subtidal platform characterized by pavement, sparse coral, macro-algae and sand. Legend from
(Ar) Subsurface reflectance of a nearshore area, (A) limestone pavement habitats with a long ridge parallel to the shore, covered by patchy to sparse coral and macro-algae. (Br) Subsurface reflectance of an area offshore from Cape Farquhar, (B) dense soft and hard coral cover mosaic with some sand and limestone pavement. (Cr) Subsurface reflectance over a cluster of bommies, (C) habitats dominated by coral bommie clusters on limestone pavement and sand with sparse macro-algae. Legend from
The tabulate coral form was the most common, contributing the highest percentage of cover in the Muiron Islands while the lowest percentage cover was in the southern region of the Ningaloo Reef. The coral form which was the second most common varied between the regions; branching and soft corals in the Muiron Islands, digitate in the northern region and soft corals in the central and southern regions.
The islands have no lagoons and only very narrow platform fringes with widths between 50–300 m then dropping off to depths >10 m. The area mapped was about 24 km2 (
The two islands are separated by a narrow (0.3 km) channel reaching depths up to 15 m, with limestone, rubble and macro-algae being the main benthic cover. There were also extensive ridges (50–100 m wide) covered by macro-algae, channels and depressions, especially along the western shores.
The second largest mapped area, the northern region (
Over 60% of the coral cover was made up by dense stands of tabulate coral, either nearer the shore or on reef flats; the central lagoons were characterised by sand and limestone pavement. About 10% of coral was the continuous digitate form and a further 7% comprised “blue-tip” branching coral. Soft coral (mostly as 50–85% cover) with up to 20% of digitate or tabular forms with some macro-algae occurred in the southern parts. Massive, submassive and foliose corals were also present, especially as bommies in deeper parts of the lagoons. Approximately 50% of all abiotic classes were made up by sand and another 20% by a combination of limestone with sand. About 6% of abiotic cover consisted of the mixed class containing rubble, pavement and sand. Some 96% of cover of macro-algae was made up by dominant macro-algae with sand, sparse macro-algae with pavement, and patchy macro-algae with sand, and some pavement.
This region extended from Point Edgar to just south of Pelican Point and covered 383 km2. Macro-algae dominated classes were about equal in cover to abiotic classes (
Over half (60%) of the habitats in the relatively narrow coastal strip of 8.1 km2 were made up by mosaics of patchy, sparse or dominant macro-algae on pavement and sand. Abiotic classes made up 30% of cover and corals 10% (
Macro-algae occurred as mostly patchy or sparse cover (10–45%) on pavement or sand. Abiotic classes were found as either patchy limestone with sand or >90% pavement, with the class containing rubble (with pavement and sand) comprising 18% of cover.
There were 5.9 km2 of coral mosaics mapped along the Ningaloo Reef. The single largest coral mosaic was continuous tabulate coral (2.2 km2 or 36.7% of all corals) (
Individual per-class probability images were generated during data processing. Examples of single class probability images for four spectrally different classes are shown for the Coral Bay area (
The overall accuracy using the 10 m radius and fuzzy logic approach was calculated at 83.81% for level 2a, 70.48% for level 4a with a higher number of fuzzy classes and 63.81% for level 4a with a lower number of fuzzy classes. Validation performed for level 4a showed, as expected, that the higher degree of fuzziness resulted in a higher overall accuracy than the lower degree of fuzziness, with the highest results for level 2a. An example of the confusion matrix at level 4a with high level of fuzziness in class allocation is presented in
Habitat maps created in this study were contrasted with those currently used by managers and researchers
(A) Reef channel west of Sandy Bay is characterized by a mix of macro-algae and pavement in the new map, contrasting with mostly coral cover in the existing map
This study has successfully demonstrated the utility of high resolution airborne optical remote sensing to map shallow marine habitats across three degrees of latitude using operational methods. Data processing used in the study allowed for extraction of highly detailed marine habitat information as well as seamless bathymetry for waters down to 20 m depth. This detection limit is consistent with previous work on coral reefs elsewhere, e.g.
Benthic habitats of the Ningaloo Reef are highly diverse and mixed
Creation of the spectral library allowed for determination of the degree of separability of the dominant spectral cover components. The selection of biotic or abiotic cover types for the final classification was largely based on frequency of occurrence in the field data set as well as spectral separability. Spectral analysis of the image spectra using the PCA and JM distance allowed for exact measures of separability to be determined, thus eliminating subjectivity in the final class selection. This approach provided a sound basis for refining or regrouping classes before the final classification which used bottom reflectance and the first two derivatives.
Field spectra were collected from a wide range of biotic and abiotic covers and represented dominant types, not spectral mixes which typically exist at the pixel level. Some authors
Nearly all discriminating spectral features in coral and macro-algae spectra occurred in narrow wavelength ranges, sometimes as broad as 20 nm, but often of the order of 10 nm
Mapping macro-algae, turfing or coralline algae was not a priority for this project, however, a number of field spectra of commonly occurring species were collected. In the classification scheme, all algae were grouped on the basis of their percentage cover within a quadrat (pixel), rather than using species-specific data. This was mostly because, apart from
Although the spectral library results showed good separation between different macro-algae, turfing algae, live and dead corals, there were very few homogenous pixels in the airborne data to allow such classes to be included. Absence of extensive cover by recently dead coral was supported by the findings of long term monitoring investigations at Ningaloo
The hierarchical classification approach used in this study reflected typical, complex reef mosaics of coral, various algae, sand and pavement, and thus it was logical to classify the images first into basic biotic and abiotic components and then to further organise them at more detailed levels within these broad classes. This approach was similar to the scheme used by Harvey et al.
Spectral analysis of the image derived spectra prior to image classification allowed for refinements in final class definitions, for example, classes with the same biotic or abiotic components requiring merging. While some classes (e.g., “blue-tip” branching
The habitat classifications generated during this study fit into the more complex habitat classifications described by
A hierarchical design with a look-up table for the final habitat maps allows users to create their own maps specific to their needs. This approach also accommodates the fact that the definitions of habitats always have some arbitrary component in class labeling
Many habitat classification schemes based on reef geomorphology have been developed and there appears to be a lot of consistency and standardisation
A number of studies have mapped coral cover using a semi-quantitative approach and described the coral cover as “low”, “medium” and “high” density
The choice of the classification approach is always an important one as the conventional “hard” spectral classification schemes are problematic when applied to mixed pixels because each pixel must be assigned to a single habitat class
Previous studies have used a linear unmixing approach (assumes reflectance of the pixel to have a linear relationship to the sum of the end-member spectra)
Results of the accuracy assessment, while ranging from 64% for the most detailed data set to 84% for the medium detail maps, were in the range of accuracies reported in similar studies elsewhere. Other studies using multi- and hyperspectral sensors which classified habitats to at least eight classes, all reported overall accuracies above 70%
Current airborne or satellite systems do not yet offer spatial or spectral resolutions to map coral reef communities at the species level
Comparison with the existing habitat maps for Ningaloo Marine Park
The two main advantages of using optical remote sensing for this study have been, firstly, the ability to seamlessly map marine habitats across a very large area using a single classification system and secondly, to extract seamless bathymetry (not presented here) across the whole system of lagoons, including very shallow waters over coral normally inaccessible to acoustic surveys. The clear, shallow waters along the Ningaloo coast naturally lend themselves to such optical remote sensing methods.
Findings from this study can be used for management and monitoring. A number of possible indicators include cover of corals, macro-algae, sand, limestone or rubble. Some of the past studies which mapped large scale reef systems focused on geomorphic
Effective management and monitoring of large marine protected areas require detailed data on distribution of benthic habitats. Large areas with complex bathymetry and very clear waters such as at Ningaloo Reef are highly suitable to the application of optical remote sensing as a means of gathering such data. Analysis techniques involved spectral analyses on
The outputs of image analysis contained final classification categories as well as per-pixel probability layers and overall percent cover of corals, macro-algae and sediment. Reef components were classified into abiotic and biotic, and then split further into sand, limestone pavement, several coral cover categories and macro-algae dominated classes. These were organised through a look-up table into five thematic information class levels.
This work demonstrated that it is possible to consistently map coral reef habitats over large areas (spanning three degrees of latitude) with a single processing rule set. We were also able to utilize a hyperspectral sensor to map different coral forms. With the use of a hierarchical classification scheme we offer greater choices in viewing the data, aiming to improve uptake of such data sets for management and monitoring. We have also demonstrated that hyperspectral remote sensing is well suited for automated mapping tasks. These baseline data can be used for ongoing and future monitoring programs using the same or simpler satellite-based multispectral sensors such as QuickBird or WorldView2 to detect change over areas of interest. Hyperspectral sensors provide a non-invasive and cost-effective approach to mapping and monitoring the extent and condition of reefs over large areas because of their capability to identify reef components on the basis of their spectral response.
(TIF)
(TIF)
(DOCX)
(DOCX)
We thank the Australian Institute for Marine Science (Dr Andrew Heyward) for data access and project logistics and HyVista Corporation (Dr Peter Hausknecht) for pre-processing. We also extend our thanks to Dr Mick Haywood and Dr Russ Babcock for providing field validation data and feedback on the results; Murdoch University staff, Dr Nicole Pinnel, Dr Matt Harvey, Dr Mike van Keulen and students, Mark Langdon and Kim Marrs for assistance with the fieldwork. We thank the staff at the Department of Environmental and Conservation office in Exmouth and the regional rangers for advice and assistance with field accommodation.