3.1 Development process
The development of the Marine Environment Classification began in 1999 with a series of consultative workshops that established the need for a spatial framework of New Zealand's marine environments. A steering group was assembled by MfE to oversee the development of the Marine Environment Classification. The purpose of the steering group was to ensure that the classification would provide a suitable management tool. The steering group membership was made up of people from the Department of Conservation, SeaFIC, NIWA, regional councils and MFish. Specifically the steering group was required to:
- discuss and define the needs of users and the scope of the Marine Environment Classification and decide on the approach for its development
- agree on the processes and techniques for the development of the classification system
- review the outputs at various stages of development of the classification and where necessary choose from among options for subsequent development stages.
A second group of experts was involved in the detailed design and technical development of the classification system. The experts contributed to each of the following development phases:
- choice of approach to design and development of the marine classification system and spatial resolution (mapping scale)
- candidate environmental variable selection
- development of environmental variables
- validation of environmental variables
- classification definition and tuning
The purpose of the development phases are outlined in general terms below.
3.2 Development phases
3.2.1 Choice of approach and spatial resolution
Various approaches to developing the classification were considered. One of the most desirable attributes of a spatial framework is the ability to resolve differing characteristics at a range of levels of detail and spatial scales. Regionalisation was rejected as an appropriate methodology at the outset because of its very limited ability to meet this requirement. Instead we concluded that an automated numerical classification of individual cells in a grid that are described by multiple variables was the most easily defended approach. Numerical methods are ideally suited to the production of classifications that are hierarchical. Hierarchic classifications can seamlessly expand and contract their resolution of character and are, therefore, suitable for use across a range of spatial scales. In addition, classes are defined in this approach solely on the basis of their environmental or biological similarity (i.e. independent of their geographic location). The geographic independence of such methods allows them to more accurately describe the inherent geographic configuration of variation in ecological character. Finally, the explicit measurement of similarity between geographic units that are produced by numerical methods has benefits for specific applications of the spatial framework, particularly in conservation applications that are considering trade-offs between locations (e.g. Belbin 1993, Leathwick et al.. 2003b).
The approach taken is similar to that used for the Land Environments of New Zealand (LENZ) framework (Leathwick et al. a and b). The multivariate approach of LENZ and Marine Environment Classification is, however, different to another spatial framework that has been developed for New Zealand - the River Environment Classification (REC) (Snelder and Biggs 2002). The REC is a 'controlling factor' approach. In this approach rules are used to sequentially subdivide the environmental domain according to differences in a set of environmental factors. The rules are based on a hierarchical model which proposes that variation in a single factor (e.g. climate, topography, geology) is the cause of ecological pattern at a series of spatial scales. While the controlling factor approach is appealing, we considered that its application to marine ecosystems was problematic because a robust hierarchy of factors is not easily defined and may be spatially unstable.
The next consideration was whether environmental or biological attributes should be used to define the classification. Biological data is limited for New Zealand's marine area. Indeed, a primary reason for developing a spatial framework is to make inferences about biological distributions for locations for which minimal or no biotic data are available. By contrast, a range of data describing the physical environment was either already available or could be modelled reasonably robustly for the entire Exclusive Economic Zone (EEZ). For this reason, plus the need to consider environmental and biological factors in the integrated management of marine resources, an environmental classification (i.e. based on attributes of the physical environment) was chosen.
In this context, the key assumption of an environmental classification is that the pattern of the physical environment can be used as a surrogate for biological pattern. This assumption is most plausible at extensive spatial scales where the broad distributions of many individual species and communities are determined largely by physiological limitations imposed by the environment. At more local scales there is an increasing likelihood that biological interactions (e.g. predation) and processes (e.g. disturbance, recruitment) influence the pattern.
The Marine Environment Classification was developed at two levels of spatial resolution. First, a broad scale classification was developed of the entire EEZ, covering the area from approximately 25 to 58 degrees South and 158 degrees East to 172 degrees West. The environmental data layers used to define this classification have a nominal spatial resolution of 1 km. Approximately 8.4 million cells are contained within the 1 km grid environmental variable layers describing the EEZ. This resolution enables aesthetically acceptable mapping at scales of 1:4,000,000 and above. While the classification can be mapped at finer scales, the grain of the underlying data will become increasingly prominent as the scale is increased.
Second, a finer scale classification was developed for the Hauraki Gulf region. The purpose of this regional classification was to assess the feasibility of producing higher resolution inshore classifications relevant to the more intensive management issues that frequently occur there. The region is defined by a line drawn eastward from Bream Head (approximately 36 degrees South) to meet a line drawn from south to north and intersecting Cape Barrier on Great Barrier Island (approximately 176 degrees East). This was based on environmental layers with a nominal spatial resolution of 200 m (i.e. consistent with a maximum map scale of 1:250,000). Approximately 220,000 cells are contained within the 200 m grid of environmental variable layers describing the Hauraki Gulf.
3.2.2 Selection of candidate environmental variables
The selection of candidate environmental variables was based on an initial design (see Snelder et al. 2001). The design process used published descriptions of relationships between environmental and biological patterns at extensive spatial scales (i.e. greater than 200 metres for the regional classification and greater than 1 km for the EEZ classification). In conceptual terms, our overall objective was to identify a set of environmental variables that could be used to define classes that maximize the discrimination of variation in the total biological composition, a task that was complicated by the highly diverse range of organisms that occur in marine environments. As a consequence of this diversity, the selection of variables required the careful balancing of generality (i.e. relevance to a broad range of biological groups) and specificity (i.e. relevance to perhaps a narrow set of organisms). Our overall emphasis tended towards the first of these (i.e. we aimed to produce a single classification that would be reasonably relevant to a broad range of ecological components). We anticipated that a general classification might provide discrimination of variation in chlorophyll biomass at the surface, pelagic and demersal fishes and benthic communities.
In practical terms, candidate environmental variables also had to be able to be derived as systematic coverages or layers. By 'systematic' we mean objectively defined data that show the spatial variation in the variable across the area to be classified at a consistent level of resolution.
The aim of the validation work was to confirm that the candidate environmental variables were useful as predictors of biological characteristics and to determine which had the strongest statistical relationships with biological pattern. There was an expectation that this would reduce the initial set of candidate environmental variables to a core set for which quantifiable statistical relationships with biological patterns could be demonstrated. The available biological data was researched (Fenwick 2001) and datasets were assembled and/or groomed for both the EEZ (Image et al. 2003) and the Hauraki Gulf (Fenwick and Flanagan 2002) classifications.
3.2.4 Definition of the classifications
The classifications of the EEZ and Hauraki Gulf were defined using a numerical classification based on clustering. In clustering, classes are defined by iteratively joining individual cells, and then groups of cells (i.e. clusters), based on their similarity according to the combination of environmental variables that are chosen to define the classification (Zonneveld 1994). Each step in the clustering process is shown graphically by a tree structure, or dendrogram, which shows the order in which clusters are joined. The number of classes depends on the 'cut level' in the dendrogram. The classes are then mapped using a Geographic Information System (GIS) to show the mosaic of patches. Patches show the geographic location of cells belonging to the classes. In general, the size of the patches is large at high levels of the classification (i.e. a small number of classes) and patches are smaller at lower levels. The user is able to map the classification at any level so that the number of classes defines a spatial resolution that is suitable for the particular application.
The aims of the testing process were twofold. First, the strength of the classification is dependent on its ability to define classes that are biologically distinctive (i.e. classes should be different from one another in terms of fish assemblages and chlorophyll concentration, for example). In a strong classification, locations belonging to a class should show a high level of similarity to other locations within the class, relative to their similarity to locations in other classes. Thus, the testing aimed to determine whether there are statistically significant differences between classes and to quantify the overall strength of the classification using biological samples. It was expected that the strength of the classification would vary with the level of the classification hierarchy and between different sets of biological data. Therefore, the testing also aimed to establish the levels at which biological distinctiveness was maximised. As a secondary output of the testing process, we aimed to describe the biological characteristics of the classes.