in coastal regions and in small islands
Coastal management sourcebooks
Part 3 Habitat Classification and Mapping
|9||Methodologies for Defining Habitats|
Summary What is understood by ‘habitat mapping’ may vary from person to person so it is important to make a clear definition of habitats prior to undertaking the study. The objectives of most habitat mapping exercises can be classified into five groups: (i) ad hoc definition of habitats without field data, (ii) application-specific studies concerning only a few habitats, (iii) geomorphological studies, ( iv) ecological studies, and (v) studies which combine more than one type of information (e.g. geomorphology and biotic assemblages).
An ad hoc approach to defining habitats is relatively cheap but is only recommended in cases where the accuracy of habitat maps may not be important. Habitat-specific studies should include the habitat of interest and those additional habitats which are most likely to be confused with it, thus permitting the accuracy of mapping to be evaluated. Geomorphological classifications can be assigned to imagery with little or no field work and some examples of such classifications are given. The establishment of ecological habitat classifications (e.g. assemblages of bottom-dwelling species and substrata) requires multivariate analysis of field data, such as hierarchical cluster analysis. These methods simplify the dataset and provide the classification scheme with an objective basis. The characteristic and discriminating features of each habitat can be identified and used to make a clear and unambiguous description of the classification scheme, thus facilitating its use and interpretation. Combined geomorphological and ecological classifications may be the most appropriate for remote sensing of tropical coastal areas.
It is vital to have a clear idea of the study objective (s) before conducting field work or selecting imagery. For example, stating that the study aims to map ‘coral reef habitats’ is not necessarily adequate because the definition of ‘coral reef habitat’ is likely to differ from person to person. In this example, the definition could embody reef geomorphology, assemblages of reef-dwelling organisms, the reef’s physical environments, or a combination of each .
The mapping objectives of most remote sensing exercises fall into one of five categories:
studies using an ad hoc definition of habitats,
studies which focus on a particular habitat type for a specific application,
studies which are principally concerned with mapping geomorphology,
ecological studies which define habitats through quantification of biotic assemblages,
studies which combine more than one type of information (e.g. geomorphology and biotic assemblages).
This chapter outlines the methods and pitfalls of these five approaches.
1. Ad hoc definition of habitats
Habitats can be defined in an ad hoc fashion if the analyst is familiar with the area concerned or if a comparable habitat classification scheme is available. This approach does not usually include the collection of new field data and is often favoured because of its relatively low cost. However, there are several important drawbacks to making habitat maps without reference to field data.
The fundamental definition of classes may be incorrect and not applicable to the area concerned.
Even if the habitat classification scheme is appropriate, habitats may be identified incorrectly on the imagery (i.e. inaccuracies during image interpretation).
Maps tend to have a vague definition of habitats.
Accuracy assessment is not possible without independent field data so the reliability of the map is uncertain.
The resulting maps can be difficult to interpret meaningfully, particularly by individuals who are not familiar with the interpreter’s concept of a habitat. By way of example, Table 9.1 outlines a marine habitat classification scheme that we created prior to conducting marine surveys of the Turks and Caicos Islands explicitly for habitat mapping. Categories were based on one person’s existing familiarity with the area (two years teaching marine science on the Caicos Bank) and another person’s experience from surveying similar ecosystems elsewhere in the Caribbean.
Table 9.1 Preliminary marine classification for habitats of the Turks and Caicos Islands. Note: the scheme was created without field data and was later found to be unsuitable for habitat mapping.
|General habitat type||Specific habitat classes|
Spur and groove
Acropora palmata zone
(i.e. branching corals)
Mixed back reef community
(seagrass / corals)
Algal rubble, Porites spp. zone
Thalassia dominated (low density)
Syringodium dominated (high density)
Syringodium dominated (low density)
Mixed seagrasses (high density)
Mixed seagrasses (low density)
|Algal dominated||Calcareous green algae
Fleshy brown algae
Field data were collected soon after the scheme was created and in the light of these data, the preliminary scheme was found to be unsatisfactory. Firstly, the scheme presented in Table 9.1 was inconsistent and combined geomorphological categories such as ‘spur and groove’ with ecological categories such as ‘Thalassia dominated (high density)’. The revised scheme (Mumby et al. 1998) was confined to ecological categories. Secondly, the scope of the scheme was fairly limited and failed to include many of the ecological classes that were only discovered during field survey (e.g. assemblages of encrusting sponge, seagrass and calcareous red algae). Thirdly, and perhaps most importantly, some of the predicted classes were not actually found at the study site. For example, whilst an ‘algal rubble/Porites spp. zone’ is common in Belize, it was not found as a distinct habitat on the Caicos Bank.
In conclusion, unless the accuracy of habitat maps is not deemed to be particularly important, an ad hoc approach to defining habitats is not recommended.
2. Application-specific studies
Some remote sensing studies may be highly focused on specific surface features and, therefore, not concerned with mapping all habitats in an area. For example, a manager may be interested to know the extent of fringing red mangrove (Rhizophora spp.) along a particular area of coast. Provided that the manager’s definition of fringing red mangrove is clear (e.g. areas where the mangrove species composition exceeds 80% for Rhizophora spp.), a comprehensive habitat classification scheme is not required. However, it is sensible to extend the scope of the mapping exercise to include those habitats which most strongly resemble the habitat of interest (e.g. stands of black mangrove, Avicennia spp.). By incorporating those habitats which are most likely to create misclassifications and errors in the habitat maps, the accuracy of the mapping objective can be determined. In this example, an accuracy assessment may show that fringing red mangrove and black mangrove are often confused and, therefore, the manager must attribute less confidence to the estimate of red mangrove cover.
3. Geomorphological classifications
Most remote sensing studies of coral reefs have focused on mapping geomorphological classes (Green et al. 1996). Labelling such classes is relatively straightforward because several geomorphological classification schemes exist (Hopley 1982, Kuchler 1986, Holthus and Maragos 1995). Examples of some geomorphological classes are given in this chapter to provide guidance (Plate 4) and encourage standardisation.
Geomorphological classes for coral reefs
Holthus and Maragos (1995) provide a comprehensive and nicely illustrated guide to reef geomorphology; readers are directed to this publication for detailed information. In addition, several appropriate geomorphological terms for reef flat, forereef and lagoon habitats are illustrated in Figures 9.1 to 9.3.
Mangrove community types
Odum et al. (1982) defined several mangrove communities based on their morphology and environment (Figure 9.4). This classification is useful for field studies and can be related to mangrove function (Odum et al. 1982).
4. Ecological classification of habitats
Unlike geomorphological classifications, ecological assemblages do not lend themselves easily to standard classifications. ‘Ecological’ definitions of habitat may be limited to assemblages (communities) of plant and animal species or widened to include species (or higher taxonomic or functional descriptors) and the substrata which collectively comprise the upper layer of the seabed (benthos). Assemblages of species and/or substrata often exhibit considerable variability and several distinct assemblages may inhabit each geomorphological zone (see Fagerstrom 1987). As such, it is often more difficult to distinguish ecological assemblages whereas geomorphology can usually be interpreted straight from an image (i.e. in the absence of field survey). As we have seen already, to infer ecological assemblages without field survey is potentially misleading.
It must also be borne in mind that geomorphological zones tend to have more distinct boundaries than ecological habitats (assemblages) which tend to exhibit change along gradients (e.g. progressive changes in species composition with changing depth). This makes the classification of ecological habitats somewhat inexact and one might ask how different two habitats must be before they are considered separate. In fact, many classification schemes have a hierarchical structure to reflect this uncertainty. At one end of the hierarchy, habitats are clearly distinct with little in common (e.g. coral reefs and seagrasses). Whereas at the other end of the hierarchy, habitat types might share a similar complement of species and are only separated because their dominant species differ (e.g. reefs dominated by the macroalgae Lobophora spp. or Microdictyon spp. – Plate 8).
There is no absolutely correct method of categorising (classifying) ecological habitats so the choice of methods depends on the objective(s) of the study. Generally, the classification aims to reflect the major habitat types found in the area of interest as faithfully as possible. Regardless of which method is eventually selected, the definition of habitats will always be slightly arbitrary, particularly where gradients of assemblages are involved. It follows that the imposition of habitat boundaries on a map will also be somewhat arbitrary which, practically speaking, means that habitat maps can never be 100% accurate.
A good habitat classification scheme must be interpreted easily and be unambiguous. To fulfil these criteria, habitat types should be determined objectively and have semi-quantitative, or preferably quantitative, descriptors that characterise habitats and discriminate between them. Linking habitat maps to reality is important for several reasons:
Maps can be related explicitly to the species/life-forms/ substrata in each habitat.
Quantitative descriptors of the habitat type facilitate the recognition of habitats in situ. This is important for further field survey (e.g. accuracy assessment) and enables other surveyors to interpret the scheme and adopt it in other areas.
Each habitat has an objective basis and it may, therefore, be thought of as a unit of assemblage (or community) diversity. Planners may then identify representative habitats (McNeill 1994) or assess patterns and hotspots of habitat diversity.
Ideally, the scheme should have a hierarchical structure to encompass a broad range of user needs, technical ability and image types. For example, it is easy to visualise a coastal mapping strategy which uses Landsat TM to make a national marine habitat map with coarse descriptive resolution (i.e. few classes; see Chapter 11), and that this would be augmented using airborne digital imagery with finer descriptive resolution (a greater number of habitats), at specific sites of interest.
An ecological dataset may include; species cover (abundance), substratum cover, tree height, canopy cover, biomass, and so on. Each of these variables describe part of the ecological and physical characteristics of a habitat. After field work , the data analyst may face a complex data set which includes measurements for multiple variables at each site. To extract the characteristics of each habitat (i.e. natural groupings of data), some form of multivariate analysis is required to simplify the data. Multivariate statistics have been developed for over a century and aim to simplify and describe complex data sets. The rest of this chapter focuses on the use of multivariate statistics for defining habitats. A broader discussion of the classification of marine habitats is given in Mumby and Harborne (1999).
Multivariate classification of field data
The definition of habitat types from field data is not difficult but several important decisions must be made during the analysis (Figure 9.5). An overview of these decisions is provided here, but readers are referred elsewhere for a more detailed discussion (e.g. Digby and Kempton 1987, Hand 1981, Clarke 1993, Clarke and Warwick 1994). To identify groups (habitats) in the data, the most appropriate suite of statistical tools are known as multivariate classification or cluster analysis. A biological example will be used to explain the concept of cluster analysis. Imagine that the composition of seagrass and algal species had been measured at each of (say) 70 sites. The similarity (or dissimilarity) between each pair of sites can be determined using a similarity coefficient. For every pair of sites, the coefficient evaluates the similarity in abundance of each species. The result is usually the algebraic sum of similarities for each species and ranges from 0–1. Two sites with a similarity of 0, have no species in common whereas a similarity of 1 constitutes an identical complement of species with identical abundances at each site. Simulation experiments with various similarity measures have found the Bray-Curtis similarity coefficient (Bray and Curtis 1957) to be a particularly robust measure of ecological distance (Faith et al. 1987). The similarity coefficient essentially exchanges the multivariate data set (e.g. abundances of, say, 20 species) for a single measure of distance/similarity between each pair of sites. A classification algorithm is then used to group sites according to their relative similarities. Those sites that are most alike will cluster (group) together whereas those sites that are more dissimilar are unlikely to join the same cluster. The clusters are represented in a tree-like diagram called a dendrogram (see Figure 9.6).
There are dozens of methods available for generating clusters and a discussion of their relative merits is beyond the scope of this Handbook. The key point to bear in mind is that no method is perfect: there is no absolute way to describe the grouping of sites. This is because the dissimilarity of sites varies semi-continuously (e.g. along gradients) and the cluster analysis frequently has to make quite arbitrary decisions over which sites should cluster with which others. If specialist software is available (e.g. PATN: Belbin 1995), a flexible cluster analysis can be used: in this case, classes are ‘believed’ if they remain stable using several different approaches. Alternatively, one of the most popular and widely available algorithms is hierarchical classification with group-average sorting (Clarke and Warwick 1994).
Hierarchical methods will result in subgroups forming from groups and so on. A hierarchical breakdown of habitat types is useful for defining different descriptive resolutions. Large dissimilar groupings (at the left of Figure 9.6) are more different from one another than the smaller groups toward the right of the dendrogram. Three hierarchical levels of descriptive resolution are illustrated in Figure 9.6. Since the similarity between groups will be partly reflected in different reflectance characteristics, it follows that remote sensing will distinguish the larger groups more easily than the smaller (more similar) groups.
Before data are analysed, pre-processing options should be considered (Figure 9.5). It might be necessary to remove rare species as inclusion may create unnecessary ‘noise’ and obscure the groupings of sites. If different types of data are being combined such as seagrass standing crop and percent algal cover, each variable can be standardised using the methods in Figure 9.5. If data are not transformed, larger values will exert a greater influence on the cluster analysis. The extent of this effect can be altered using transformations of varying severity. For example, when clustering percentage cover data from coral reefs, we elected not to transform the data, thus allowing the dominant species or substrata to exert an appropriately large influence on the habitat grouping. This was deemed necessary because remote sensing is more likely to discriminate habitats on the basis of their major benthic components rather than more cryptic species or substrata. Finally, Mumby et al. (1996) present a method of weighting individual species during the calculation of similarity matrices. This might be appropriate if some species are considered to be more important from a conservation perspective but could equally well apply to different substrata, canopy cover values and so on.
Describing characteristics of habitats
The characteristics of each habitat are determined by examining the data in each cluster. In the absence of specialised software such as PRIMER (Clarke and Warwick 1994), the easiest approach is to calculate the mean and standard deviation of each variable in the cluster. For example, if 30 sites had formed a single cluster (habitat A), the average algal cover, seagrass species composition (etc.) could be calculated for this habitat. Those species and substrata with the highest mean covers/densities are the dominant features of the habitat. It is also worth calculating the coefficient of variation (COV) which is the ratio of the mean to standard deviation. Those species with the highest COVs have the most consistent cover/density in the habitat , although they may not necessarily be dominant. Dominant and consistent features (species, substrata) may be considered characteristic of the habitat. However, different habitats can share the same characteristic features. To identify discriminating features, comparisons must be drawn between pairs of habitats.
A more objective method of identifying characteristic and discriminating features, is Similarity Percentage analysis (SIMPER), described by Clarke (1993) and available in the software, PRIMER (Plymouth Routines in Multivariate Ecological Research). To identify discriminating features, SIMPER calculates the average Bray-Curtis dissimilarity between all pairs of inter-group samples (i.e. all sites of habitat 1 against all sites of habitat 2). Because the Bray-Curtis dissimilarity measure incorporates the contribution of each feature (e.g. each species), the average dissimilarity between sites of habitat 1 and 2 can be expressed in terms of the average contribution from each species. The standard deviation provides a measure of how consistently a given species will contribute to the dissimilarity between habitats 1 and 2. A good discriminating species contributes heavily to interhabitat dissimilarity and has a small standard deviation.
Characteristic features can be identified using the same principle; the main difference is that average similarity is calculated between all sites of each habitat. The species which consistently contribute greatly to the average similarity between sites are considered characteristic of the habitat. The following example illustrates this procedure.
The following SIMPER result (Table 9.2) was obtained from the classification of seagrass habitats (Figure 9.6). Clusters 8 and 9 were compared to identify discriminating features. Penicillus spp. is clearly the best discriminating genus, accounting for over 40% of the dissimilarity between habitats. Halimeda spp. also offer a fair means of discrimination. Standing crop (of seagrass) would be a better discriminator than Laurencia spp. because its ratio of average contribution/standard deviation is higher.
|Table 9.2 SIMPER analysis of dissimilarity between seagrass clusters (habitats) 8 and 9 (Figure 9.6 and Table 9.3). The term ‘average abundance’ represents the average abundance, biomass, density (etc.) of each feature. ‘Average contribution’ represents the average contribution of feature i to the average dissimilarity between habitats (overall average = 33.7%). Ratio = contribution average/standard deviation. Percentage contribution = average contribution/average dissimilarity between habitats (33.7). The list of features is not complete so percent values do not sum to 100%.|
An example of a habitat classification scheme
A summary of each habitat type can be presented which broadly follows the structure of the dendrogram. Table 9.3 gives such a summary for habitats categorised by the dendrogram in Figure 9.2.
Table 9.3 Summary of seagrass habitat classes showing cluster numbers from dendrogram (Figure 9.2). Note: the first three clusters (1, 2 and 3) were excluded from the scheme below because it became apparent, through further field survey, that these habitats were rare and thus could not feasibly be mapped.
|Halodule wrightii of low standing crop (5 g.m-2)||1|
|Syringodium filiforme of low standing crop (5 g.m-2)||2|
|Thalassia testudinum, Syringodium filiforme, and Halodule wrightii of low to medium standing crop (< 10 g.m-2)||3|
|Thalassia and Syringodium of medium to high standing crop||Thalassia testudinum and Syringodium filiforme of standing crop (5-80 g.m-2)||4|
|Thalassia testudinum and Syringodium filiforme of standing crop (80-280 g.m-2)||5|
6, 7, 8
low standing crop Thalassia and sparse algae
Thalassia testudinum of low standing crop (5 g.m-2) and Batophora sp. (33%)
|Thalassia testudinum of low standing crop (5 g.m-2) and sand||7|
|low to medium standing crop||
medium dense colonies of calcareous algae
– principally Halimeda spp. (25 m-2)
|Thalassia and dense calcareous algae||
dense colonies of calcareous algae – principally
Penicillus spp. (55 m-2)
and Halimeda spp. (100 m-2)
Thalassia testudinum of medium standing crop (~80 g. m-2)
Ignoring the first three clusters, which were discarded as rep resenting rare habitats, descriptive resolution was partitioned to give 6 fine-level habitats (clusters 4–9) which we re arranged into 4 medium-level habitats. Medium descriptive resolution included sites of low sea-grass standing crop (clusters 6 and 7), low to medium standing crop (clusters 8 and 9) and medium to high standing crop (clusters 4 and 5; although two of the sites had low standing crop). At coarse descriptive resolution, the low standing crop classes were merged with sand-dominated habitats of the reef classification. This decision was taken because the lower end of the visual assessment scale for seagrass standing crop includes extremely low biomass values (< 5 g.m-2) and the distinction between seagrass of low standing crop and sand-dominated habitats is some-what arbitrary. Medium to high standing crop classes were merged to give a coarse descriptive resolution labelled seagrass.
Most general purpose statistical software offers hierarchical cluster analysis (e.g. SPSS, Minitab). The following also offer more specialised routines:
PRIMER contains objective methods for describing the characteristics of each cluster (group). It costs approximately £330 and is available from: M. Carr, Plymouth Marine Laboratory, Prospect Place, West Hoe, Plymouth PL1 3DH, UK;
PATN contains a comprehensive range of multivariate tools. It costs approx. £500 and is available from Fiona Vogt, CSIRO Division of Wildlife and Ecology, PO Box 84, Lyneham, ACT 2602, Canberra, Australia.
Given that benthic geomorphology and biotic/substratum cover strongly influence the radiance recorded by a remote sensing instrument, habitat mapping may subsume both geomorphological and ecological habitat classifications (sections 3 and 4, above) into a single scheme (Mumby and Harborne 1999). A merged habitat classification scheme provides a geomorphological and ecological component to each polygon on a habitat map (e.g. use of ‘shallow lagoon floor (< 12 m) + Thalassia testudinum and Syringodium filiforme of standing crop 5–80 g.m-2, in a legend). Both components can have a hierarchical structure (e.g. patch reef versus dense patch reef or diffuse patch reef; coral versus branching corals or sheet corals). The structure of a combined habitat classification is systematic in that geomorphological and ecological classes are not mixed or used interchangeably, and it also provides flexibility. For example, the geomorphological class ‘shallow lagoon floor (<12 m)’ might also be coupled with the ecological class ‘medium dense colonies of calcareous algae’.
Providing that supporting documentation is clear, use of a hierarchical classification scheme allows some areas to be mapped in greater detail than others without confusing interpretation. Where assignment of a label is uncertain, the designation should reflect this. For example, if the depth of the lagoon is unknown, the geomorphological component should be labelled at a coarser level of the hierarchy (i.e. lagoon). Similarly, the ecological component ‘Thalassia testudinum and Syringodium filiforme of standing crop 5–80 g.m-2, may be used in areas which are data rich but ‘seagrass’ might be more appropriate elsewhere.
In practice, a coastal mapping strategy is envisaged which uses Landsat TM data to make a regional marine habitat map of coarse descriptive resolution, and that this would be augmented using airborne digital imagery (e.g. CASI) or possibly colour aerial photography, with finer descriptive resolution, at specific sites of interest. A hierarchical habitat classification scheme helps integrate such hierarchical mapping activities.
A total of 170 survey sites were visited in the Turks and Caicos Islands mapping campaign. Percentage cover/density data were collected from six 1 m2 quadrats at each site, in addition to date, time, water depth, GPS position and a visual estimate of the size of the habitat patch. These data took 4 person-days to enter into a database and a further two days were necessary to produce the ecological classification used throughout the Handbook.
The meaning of ‘habitat’ should be made explicit at the outset of any habitat mapping study. Ad hoc habitat definitions are not recommended because they are prone to being vague and the accuracy of habitat maps cannot be determined in the absence of field survey. Highly focused studies should embody the habitat of interest and those additional habitats which are most likely to be confused with it, thus permitting the accuracy of mapping to be evaluated. Geomorphological classifications can be assigned to imagery with little or no fieldwork. The derivation of ecological habitat classifications should usually involve objective multivariate cluster analyses of field data. Cluster analysis can be tailored to study objectives by using appropriate pre-processing of field data (i.e. to place emphasis on specific elements of the dataset). Habitat classes should be described quantitatively to facilitate use of the classification scheme by other surveyors and improve its interpretation. A hierarchical approach to habitat classification is useful where coastal areas are mapped with varying detail (e.g. where more than one sensor is used) and to satisfy a broad range of user needs and expertise.
Belbin, L., 1995, PATN Pattern Analysis Package (Canberra: CSIRO Division of Wildlife and Ecology).
Bray, J.R., and Curtis, J.T., 1957, An ordination of the upland forest communities of Southern Wisconsin. Ecological Monographs, 27, 325–349.
Clarke, K.R., 1993, Non-parametric multivariate analyses of changes in community structure. Australian Journal of Ecology, 18, 117–143.
Clarke, K.R., and Warwick, R.M., 1994, Change in Marine Communities: an Approach to Statistical Analysis and Interpretation (Plymouth: Natural Environment Research Council).
Digby, P.G.N.,and Kempton,R.A.,1987, Multivariate Analysis of Ecological Communities (London: Chapman and Hall).
Fagerstrom, J.A., 1987, The Evolution of Reef Communities (New York: John Wiley & Sons).
Faith, D.P., Minchin, P.R., and Belbin, L., 1987, Compositional dissimilarity as a robust measure of ecological distance. Vegetatio, 69, 57–68.
Green, E.P., Mumby, P.J., Edwards, A.J., and Clark, C.D., 1996, A review of remote sensing for the assessment and management of tropical coastal resources. Coastal Management, 24, 1–40.
Holthus, P.F.,and Maragos, J.E.,1995, Marine ecosystem classification for the tropical island Pacific. In Marine and Coastal Biodiversity in the Tropical Island Pacific Region, edited by J.E. Maragos, M.N.A. Peterson, L.G. Eldredge, J.E. Bardach and H.F. Takeuchi (Honolulu: East-West Center),pp. 239–278.
Hand, D.J., 1981, Discrimination and Classification (Chichester: John Wiley & Sons).
Kuchler, D.,1986, Geomorphological Nomenclature: Reef Cover and Zonation on the Great Barrier Reef (Townsville: Great Barrier Reef Marine Park Authority Technical Memorandum 8).
McNeill, S.E.,1994, The selection and design of marine protected areas: Australia as a case study. Biodiversity and Conservation, 3, 586–605.
Mumby, P.J., Clarke, K.R., and Harborne, A.R., 1996, Weighting species abundance estimates for marine resource assessment. Aquatic Conservation: Marine and Freshwater Ecosystems, 6, 115–120.
Mumby, P.J., Green, E.P., Clark, C.D., and Edwards, A.J., 1998, Digital analysis of multispectral airborne imagery of coral reefs. Coral Reefs, 17, 59–69.
Mumby, P.J., and Harborne, A.R.,1999, Development of a systematic classification scheme of marine habitats to facilitate regional management and mapping of Caribbean coral reefs. Biological Conservation, 88, 155–163.
Odum, W.E., McIvor, C.C., and Smith III, T.J., 1982, The Ecology of the Mangroves of South Florida: A Community Profile (Washington DC: US Fish and Wildlife Service, Office of Biological Services FWS/OBS – 81/24).