Reef Cover, a coral reef classification for global habitat mapping from remote sensing

Reef Cover was specifically developed to support the process used to produce and deliver globally applicable coral reef mapping products from remotely sensed data¹⁶. The typology acts as a key to bridge historic and contemporary knowledge, plot-scale and aerial viewpoints, and pixel data with natural history to convert pixel data into information in a form suitable for reef management decisions. Accompanying case-studies^18,19 we use to demonstrate its application are also publicly available. The mapping products described in each case study were developed specifically using Reef Cover to support science and conservation of coral reef ecosystems.

We sought to develop a robust system which balances the geomorphic complexity of reefs with the need to develop high accuracy maps of each class in the system. The result is a 17-class system that can be (i) applied to remote sensing datasets for future mapping, (ii) used to interpret coral reef maps (iii) effectively disseminated to users – mainly in coral reef ecology and conservation space – in a way that promotes research and conservation.

Three steps were used in the development of the classification scheme.

1. 1.

   Step 1. Review. Existing coral reef geomorphic classification schemes (expert-led classifications from Darwin’s 1842 reef classification²⁰ to the Millennium Coral Reef Mapping Project classification²¹) were carefully reviewed²² to identify synergies in terminology and definitions for reef features, and evaluate how well common features can be described in terms of remote sensing biophysical data. The review allowed us to develop a set of classes that build constructively on previous foundational knowledge on coral reef geomorphology and are relatable to existing mapping and classification efforts, and addresses the challenge of relevance.

2. 2.

   Step 2. Development. Reef Cover classes were then derived from attributes data, building on established machine-led reef mapping theory¹³. Physical attributes datasets commonly available to remote sensing scientists were examined to refine a set of 17 meaningful internal reef classes that relate to broader interpretation from a natural history point of view, gathered in Step 1. A workshop was organised to gather feedback on classes. Clarity around how each class relates to attribute data addresses the challenge of transparency.

3. 3.

   Step 3. Dissemination. Reef Cover classes were then documented²³ in a way to promote re-use and cross-walking, with a strong focus on needs of the users, to address the challenges of clarity and accessibility. Development of the Reef Cover document considers and details the 1) relevance e.g. rationale behind why it was important to map this class, but also broader global applicability of the class, 2) simplicity e.g., promoting user-uptake by employing plain language, not over-complicating descriptors and limiting the number of classes to manageable amount, 3) transparency supplying methodological basis behind each class, and exploring caveats and ambiguities in interpretation, 4) accessibility including discoverability, open access and language translations to support users, and 5) flexibility allowing for flexible use of the scheme depending on user needs, allowing for flexible interpretation of classes by providing cross-walk to other schemes and existing maps, and making the classification adaptable, and open to user feedback (versioning).

Finally, as a proof-of-concept the Reef Cover classification was tested in two large scale coral reef mapping exercises: one in the Great Barrier Reef ²⁴ (Case Study 1) and one across Micronesia¹⁹ (Case Study 2, Technical Validation section). During this process, the Reef Cover dataset was reviewed to assess how useful it was for both a) producers using Reef Cover to map large coral reef areas from satellite data, and b) consumers using Reef Cover to interpret map products for application to real world problems.

Step 1. Review. Building global classes on foundational reef mapping and classification work

Global reef mapping: the need for a geomorphic classification to map coral reefs at scale

Coral reefs represent pockets of biodiversity that are widely dispersed, often remote/inaccessible and globally threatened^2,3. Communities and economies are highly dependent on the ecosystem services they provide^25,26,27. This combination of vulnerability, value and a broad and dispersed global distribution mean global strategies are needed for reef conservation, for which maps (and the classifications that underpin them) play a supporting role. Global coral reef maps have been fundamental to geo-political resource mapping and understanding inequalities^28,29, the valuation of reef ecosystem services²⁶, understanding the past³⁰, present³¹, and future threats to reefs³², supporting more effective conservation^33,34 and reef restoration strategies^35,36, and facilitating scientific collaborations and research outcomes³⁷. Reef conservation science and practice may particularly benefit from technological advancements that allow delivery of more appropriate map-based information, particularly across broader, more detailed spatial scales and in a consistent manner^34,36,38.

Existing expert-led reef classifications

Traditionally, coral reef features have been grouped based on observations of morphological structure, distributions of biota and theories on reef development, gleaned from aerial imagery, bathymetric surveys, geological cores and biological field censuses by natural scientists^7,39 (Fig. 1). Natural scientists were struck by both the uniformity and predictability of much of the large-scale three-dimensional geomorphic structure of reefs and biological partitioning across that structure, and how consistent these characteristic geologic and ecological zones were across large biogeographic regions^20,40). Technological developments of the 20^th century, such as SCUBA demand regulators and compressed air tanks (commercially available in the 1940s⁴¹), acoustic imaging for determining seafloor bathymetry (e.g., side-scan sonar developed in the 1950s), light aircraft for aerial photography (first applied in the 1950s⁴²) and lightweight submersible drilling rigs for coring (applied in the 1970s⁴³), allowed reef structure to be viewed from fresh perspectives. New aerial, underwater and internal assessments of reef structure expanded the diversity of external and internal classes, with hundreds of new terms for features defined^7,8,10 (Online-Only Table 1). However, the localised nature of most of these applications (Fig. 1) meant that many of the classes developed using these tools were region-specific, leading to experts warning against too heavy a reliance on “the imperfect and perhaps biased existing field knowledge on reefs” for developing global classifications⁴⁴.

Existing reef classifications derived from satellite data

Shallow water tropical coral reefs are particularly amenable to global mapping from above¹². They develop in clear, oligotrophic tropical waters, so many features are detectable from space⁴⁵. Satellite technology has spawned a wealth of data on reefs, enabling large area coverage, with resolution of within reef variations. Initial approaches to reef mapping in the 1980s expanded our traditional viewpoint from single reef mapping and extent mapping to detailed habitat mapping of whole reef systems⁴⁶. Through the 1990s and early 2000s evolving field survey techniques described above enabled more effective linkage of ecological surveys to remote sensing data^47,48. Accessibility to higher spatial resolution images over larger areas in combination with detailed field data, physical attributes and object-based analysis resulted in large reef area mapping^13,21,49,50. In the last five years, the increase in daily to weekly global coverage of this type of imagery, in combination with cloud-based processing capability has expanded to a global capability for reef mapping³⁸. This is a new type of global information that requires a different approach to classification to make sense of complex natural systems at ocean scales.

One of the first steps in creating the Reef Cover classification was reconciling existing classification schemes across the nomenclature driven by disciplinary, linguistic and regional biases. To do this we conducted a review of reef geomorphic classifications, looking for consistencies and usage of terms that transcended divides in discipline²² (see summary in Online-Only Table 1).

Scaling and consistency: choosing an appropriate level for Reef Cover

Remote sensing scientists have been developing automated methods to make sense of the increasing availability of earth observation data over coral reefs, yielding information on ecosystem zones derived from data sources such as spectral reflectance and bathymetry at increasingly larger scales^12,51. As more data increasingly reveal the diversity and complexity of reefs, selecting an appropriate level at which to map reefs on the global scale requires balancing the need for a limited number of classes that can be mapped consistently based on available earth observation data, with user need for comparable information.

Reef type classification

Morphological diversity can make global geomorphic classification – particularly between reefs (at the “reef type” level, e.g., Fringing, Atoll reefs) – challenging. Divergent regional morphologies (e.g., Pacific atolls vs Caribbean fringing reefs) and endemic local features (e.g., Bahamian shallow carbonate banks, Maldivian farus) are created by underlying tectonics, antecedent topography, eustatics, climate and reef accretion rates which can all vary geographically⁵². The diversity of reef types is reflected in the large number of classes defined in the impressive Millennium Coral Reef Mapping Project (68 classes at the between-reef geomorphic level L3), the most comprehensive globally applicable coral reef classification system to date²¹.

Geomorphic zone classification

Internally, reef morphology becomes a lot more consistent. Physical boundaries in the depth, slope angle and exposure of the reef surface create partitioning into “geomorphic zones” (e.g., Reef Flat, Reef Crest), developed in parallel to the reef edge and coastlines and generally with a distinct ecology^17,39,53. These internal patterns of three-dimensional geomorphic structure can be remarkably predictable, even between oceans. This makes geomorphic zonation a good basis for consistent and comparable mapping at regional to global scales⁵⁴. Moreover, congruence between geomorphic zones and ecological partitioning means that ecological understanding can be derived from geomorphic habitat classes, making geomorphic mapping valuable to conservation practitioners⁵⁵.

Benthic classification

Many classifications developed for reef mapping (e.g., Living Oceans Foundation⁴⁹, NOAA Biogeography Reef Mapping Program⁵⁰), monitoring (Atlantic and Gulf Rapid Reef Assessment⁵⁶, Reef Cover Classification System⁵⁷, Reef Check^58,59) and management (Marine Ecosystem Benthic Classification⁶⁰) have included an ecological component. Classifying reef benthos is important as associated metrics, such as abundance of living coral and algae, are widely used indicators of ecosystem change. However, most classifications that consider benthic cover are operational at reef⁶¹ to regional scales, due to the need for very high-resolution remote sensing data¹¹ (e.g. from UAVs and CASI⁶², or high-resolution satellites like QuickBird and WorldView 1 m) to be able to reliably determine classes such as coral cover and type, soft coral, turf, coralline algae, rubble and sand. A comprehensive benthic coral reef classification²³ that met the Reef Cover objective of being globally scalable (both in terms of remote sensing biophysical data availability and processing capabilities) but that also fully recognises and includes the rich benthic detail required to address ecological questions at sub-metre scales is beyond the scope of this classification. In the coming years it is likely that further advancements in technology – both downscaling of remote-sensing and up-scaling of field observations⁶³ – will enable us to address this spatial mismatch.

The challenge of creating the Reef Cover classification was to create a set of classes that related to natural science observations, despite using data pulled from remote sensing. Intra-reef zones defined by natural scientists often represent different biophysical /ecological communities that in turn reflect environmental gradients (e.g., in light, water flow) and geo-ecological processes (sediment deposition, reef vertical accretion) below the water that led to the arrangement¹⁷. However, these classes frequently also can be related to biophysical information on slope, depth and aspect that can be determined remotely. A thoughtfully prepared classification – that adheres to Stoddart’s (1978) classification principles, which state that classes should be explicit, unique, comprehensible, and should follow the language of prior schemes – can support production of maps and other science (monitoring, management) that are still relevant to historic work but that can go forward with consistent definitions²¹.

Step 2. Development. Creator requirements – relating Reef Cover classes to remote sensing data

Development of appropriate mapping classes requires a sensitive trade-off between the needs of users (in terms of the level of detail needed, appropriate for scaling, consistent across regions, simple enough to be manageable but detailed enough to be understandable), and the input data available and quality of the globally repeatable mapping methods of the map producers.

While vast in terms of scalability, data producers are more constrained in terms of sensor capabilities such as spatial resolution (limited to pixels) and depth detection limits (limited by light penetration), and processing power (high numbers of map classes becoming more computationally expensive). Physical conditions and colour derived from remote sensing, along with their textural and spatial relationships, can be linked to reef zonation¹³, with depth and wave exposure being particularly important information to explaining geomorphology⁶⁴.

To select a set of Reef Cover classes that could be defined by attributes available from most commonly available public access or commercial satellite data, but that also corresponded to common classes found in the classification literature, and relevant from a user perspective, we looked for intersectionality between physical attribute data that can be derived from satellites but also help shape and define reef morphology.

Physical attributes

The physical environment – light, waves and depth – plays a deterministic role in reef structural development and the ecological patterning across zones³⁹. Underlying geomorphic structural features can almost always be characterised in terms of three core characteristics: i) depth, ii) slope angle and iii) exposure to waves (Fig. 2).

Depth

Depth is a useful attribute for bridging human and machine-led classifications. Bathymetry can be derived from spectral information from satellites since the absorption of light at specific wavelengths also has known relationships with water column depth⁶⁵ but also relates to reef geomorphology (due to role of primary production in powering biogenic calcification)⁶⁶. Bathymetric data also provides the basis for other critical depth-derived products, slope and aspect, which are used to distinguish geomorphic classes and reef environmental parameters, e.g., exposure to breaking wave energy (Fig. 2).

Reef Crest, for example, is often described as the shallowest part of the reef ⁶², while Lagoon represents a deep depression in the reef structure^57,67. Depth thresholds are sometimes defined: a threshold of 10 m was suggested to differentiate true lagoons from shallow water areas⁶⁸, and an 18 m threshold has been used to distinguish Reef Front from Reef Slope¹⁷. In the Reef Cover classification, depth was particularly important for distinguishing Fore Reef classes (e.g. Reef Slope, Terrace) from Reef Crest and Reef Flat classes (Table 1, Fig. 2). Generally, tides and variability in water clarity and regional eustatic discrepancies in reef top depth (e.g., Reef Flat in Atlantic systems generally lie much lower with respect to tides than in the Indo-Pacific⁶⁷) mean relative depths are more appropriate, which is why absolute numbers were not used in Reef Cover definitions.

Slope

Slope angle, either absolute angle or discontinuities in angle acting as a break between zones, is an important differentiator of reef zones. Reef Flats are defined as being horizontal ‘flattened”⁶⁹ “flat-topped”⁷⁰; Fore Reef slope zones often include references to slope angle (e.g., in one classification Fore Reef has been defined as “any area of the reef with an incline of between 0 and 45 degrees”⁶²), and Walls– common on atolls – are “near vertical” features. Variability in slope continuity can also be an important way to demarcate zones⁷¹. Montaggioni illustrated a range of representative profiles across atolls and barrier reefs, with convoluted profiles allowing subdivisions of reef slope, particularly across fringing reefs which are less likely to show a uniform reef slope than an atoll⁵³, and Reef Crest is sometimes defined as a demarcation point separating the Fore Reef from the Reef Flat^53,62,72. Where water depth can be derived from remotely sensed spectral data, bathymetry can be used to directly calculate slope (i.e., by calculating the slope angle between a pixel and its neighbours) or by considering the local variance in depth (e.g., the standard deviation in depth values within some radius of each pixel).

In the Reef Cover classification, slope angle data were used to distinguishing Fore Reef classes such as Reef Slope and Reef Front, from horizontal classes such as Outer and Inner Reef Flat and Lagoons (Table 1, Fig. 2).

Exposure

Physical exposure of reefs is a key driver of zonation. Reef Crests – linked to wave breaking – are often described as “an area of maximum wave shoaling”, i.e. a zone that absorbs the greatest wave energy^62,69. Fore Reefs are frequently sub-divided based on relative exposure (e.g. exposed vs sheltered slope, or windward vs leeward⁵⁷). Exposure influences profile shape and importantly the communities growing in the zone, so that slopes with identical profiles could have very different communities^57,73. Sometimes these zones are related to the communities found there. Meanwhile, exposure across the reef means back-reef zones contain sheltered water bodies. Together with data on water depth and bathymetry, wave energy data was key for distinguishing key Reef Cover classes^74,75.

Colour and texture

Sub-surface spectral reflectance data can provide measurements of reef colour and texture over large areas. Concentrations of photosynthetic pigments in coral, algae and seagrass as well as light scattering by inorganic materials means spectral reflectance can also be used to determine biophysical properties of the reef ⁶⁵. Colour and texture information derived from satellites can be used to manually draw polygons around similar geomorphologic units or habitats but provide the basis to drive image-based thematic mapping (such as digital number, radiance, reflectance) and texture, through spectral processing⁶⁴. Texture measures are also used to improve classification by allowing spectrally similar substrates like corals and macroalgae to be distinguished. Reef Flats, for example, having a single driver of zonation, in contrast to several drivers on most other zones, makes benthic zonation particularly distinct³⁹, and easily detectable as coloured bands in aerial images of reef flats. This allows colour and texture to be used to distinguish Outer Reef Flats, which have a greater component of photosynthetically active corals and algae, from Inner Reef Flats which appear brighter due to a higher proportion of sand build up in this depositional area (Fig. 3).

Spatial relationships

Size and shape

The size and shape of reef features can help determine Reef Cover class. Many large-scale reef structural features appear elongate as the shelf constrains shape – and reef morphology can even help predict shape as they constrain accommodation space and influence deposition⁷⁶. Reef Flats, for example, boast the broadest horizontal extent of any geomorphic zone, typically 500 to 1000 m across, but reaching several kilometres in width across some Pacific atolls⁷¹. Lagoons also tend to be broad in width although width and shape can be variable depending on reef type. Understanding some of these characteristics can help determine classes, although these are usually defined relationally rather than by application of size thresholds.

Neighbourhood and enclosure

Natural scientists agree that reefs feature three major geomorphic elements: a Fore Reef, a Reef Crest and a Back Reef (although subdivisions and complexities exist around these). Because of the influence of large-scale processes on reef development, these zones occur in order^17,39,53. Reef Crest is arguably the most defining characteristic of any reef – the break point at which a sharply defined edge divides the shallower platform from a more steeply shelving reef front⁷¹, around which other geomorphic zones arranged in parallel⁷⁷. As a result, spatial arrangement of zones can be informative for mapping (Table 2). For example, Back Reef is often defined as being contiguous to the Reef Crest (Back Reef is often defined as any reef feature found landward of the crest).

Enclosure to semi-enclosure within a bordering reef construction (e.g., in lagoons⁶⁸) is another feature used classically to define reef zones, but that could also be derived from satellite imagery.

The Reef Cover typology presented is derived from earth observation data, but attempts to link classes to genetic process, social, ecological and geological importance²³. By focussing on the attributes of depth, light, exposure, colour and texture and spatial relationships that are common to both domains, our traditional biophysical knowledge of reefs can be integrated with remote-sensing capabilities. Attributes can be combined to make decision trees (Fig. 4) to help use satellite data to map reefs at the global scale. The Reef Cover list of classes can all be distinguished from these physical attributes alone, supporting production of maps that are still relevant to existing work but that can allow computationally inexpensive determination of mapping classes to beyond what was previously possible³⁸.

Step 3. Dissemination. Providing user friendly Reef Cover class descriptors that facilitate uptake and use

Computers have revolutionised our ability to classify multidimensional data sources, which allows mapping and modelling at far larger scales for the same effort compared to a human taxonomist. However, without proper consideration of the needs of the end user, classified data may not be effectively applied to conservation challenges⁴. The Reef Cover classification was developed with five user-needs in mind: relevance, simplicity, transparency, accessibility and flexibility.

Relevance

Different habitats within reefs contribute differently to biological and physical processes. For example, Reef Crests play a disproportionate role in coastal protection, dissipating on average 85% of the incoming wave energy and 70% of the swell energy^78,79; Reef Slopes supply an order of magnitude more material to maintain island stability^61,80; shallow Reef Front areas often host more coral biodiversity⁸¹; Reef Flats support herbivorous fish biomass⁸² and accessibility of Lagoons often affords them cultural importance as places important for artisanal harvesting⁸³. A classification that effectively captures the appropriate diversity of these habitats can therefore better inform social, biological and physical studies³⁶, such as global conservation planning to safeguard reefs, for example, in order to meet the Convention on Biodiversity Aichi targets³⁴. Map classes need to reflect differences of interest to a wide range of reef scientists, from oceanographers to paleoecologists and fisheries scientists – so careful consideration of natural history is important. Global mapping is usually to enable spatial comparisons, so a classification that is globally applicable was also important.

To explore relevancy, a crosswalk was performed between Reef Cover and a selection of major regional to global coral reef classification^10,60,84, mapping^22,36,85 and monitoring efforts^7,8,56, to make sure important classes from established classifications had not been missed²³ (Online-Only Table 2).

Simplicity

Simplicity was achieved by (1) choosing an appropriate mapping scale (internal geomorphic classes), (2) limiting the number of geomorphic classes (17 classes), and 3) providing clear (1 line) descriptors with additional information to address issues of semantic interoperability.

1. 1.

   Reef Cover was developed to provide consistent mapping of reefs across very large areas: classification of geologic and ecological zones is much more amenable to mapping using remote sensing, given greater consistency in geomorphology across large biogeographic regions³². Satellite data has supported the development of several detailed regional “reef type” classifications, such as nine reef classes for the Great Barrier Reef from Landsat imagery⁷³, six reef classes from the Torres Straight⁷⁴ and 16 classes for the Red Sea from Quickbird⁶. However, local reef type classifications are not always applicable globally due to large regional discrepancies in Reef Type. As a result, detailed reef type typologies are more suitable for local to regional classifications^6,35. For global mapping, an internal geomorphic approach is better. Finer spatial scale classifications from satellite data are also challenging, due to differences in the spatial scale at which spectral data can be generated (metres) and which benthic assemblages display heterogeneity (sub-metres)³². Medium spatial resolution multispectral data (5 to 30 m) is the most commonly used satellite information used for coral reef habitat mapping⁴², and classification of internal geomorphic structures may be best suited to this kind of data.

2. 2.

   Reviews of habitat mapping from remote sensing found the number of map classes averages 18 at continental and global scales¹³. More than this can become overwhelming for users and computationally expensive for developers at this point in time. Many coral reef classifications contain four or five hierarchical levels and high numbers of classes: the Millennium Coral Reef Mapping Project (MCRMP) was ambitious in developing a standardised typology that captured much of the reef type diversity, but despite defining over 800 reef classes defined at the finest (level 5, essential for local reef mapping) scale³⁶, level 3 (68 classes) continue to be more popularly adopted in publications using this dataset. To keep the classification simple, Reef Cover was limited to 17 geomorphic classes, with simple one line definition provided. A limited class was needed 1) to make it manageable for users, 2) to make it computationally manageable for very large (regional and global) data processing and 3) reduction in classes compared to MCRMP allowed for consistent automated mapping at the global scale – so that whole regions could be directly compared for monitoring and management.

3. 3.

   Short definitions were provided in plain language for simplicity. To address additional uses issues of semantic interoperability each Reef Cover class definition also outlines other commonly used terms for concepts (synonymy) and explains different interpretations of the same meanings and understanding of the relations between concepts.

Transparency

One barrier to the use of analysing and interpreting big data is user-friendliness. Of 79 coral reef mapping attempts reviewed (62 benthic coral reef maps, 6 geomorphic coral reef maps and 11 mixed), only 13% were accompanied by a clear classification that defined the meaning of map classes¹⁴. Describing how the classification relates to data (Step 2) and producing a detailed descriptor (Step 3) along with a diagram allows classification to be understood and also adopted for different projects. We also attempted to address transparency by relating Reef Cover classes to other major global mapping and monitoring efforts (Online-Only Table 2) and providing a decision support tree for users (Fig. 4)⁶⁹.

Accessibility

Another barrier to the use of analysing and interpreting big data is access⁷⁵. Much information remains locked behind paywalls, and additional barriers exist including discoverability. To promote accessibility and encourage use, all data were made publicly available (see Data Records section for access). Terms were translated into different languages, as science published in just one language has been shown to hinder knowledge transfer and new findings getting through to practitioners in the field⁸⁶.

Flexibility

One criticism of thematic habitat maps derived from remote sensing is a lack of flexibility: categorical descriptions of habitats are by design a discrete simplification of the ecological continua, thus classifications limit the interpretation and questions that can be asked⁷⁶. Flexibility issues were addressed by 1) not prescribing absolute thresholds to each class, instead providing information on how classes relate to each other (Tables 1–3) allowing a) map producers to adapt application of Reef Cover to their own needs, and b) users to interpret with flexibility, 2) providing additional information (Standard Descriptors) including main features, exceptions to rules and broadness as to provide users with a broader understanding of hidden complexities when interpreting class meaning, 3) remaining open to feedback, we hope this Reef Cover version 1 can be improved upon with feedback from the community.

Source: Ecology - nature.com