Biodiversity and Conservation 11: 2093–2116, 2002. © 2002 Kluwer Academic Publishers. Printed in the Netherlands.
Information pyramids for informed biodiversity conservation JACOB McC. OVERTON1,∗ , R.T. THEO STEPHENS2 , JOHN R. LEATHWICK1 and ANTHONY LEHMANN3 1 Landcare Research, Private Bag 3127, Hamilton, New Zealand; 2 Department of Conservation, P.O. Box 112, Hamilton, New Zealand; 3 Swiss Center for the Cartography of Fauna, Terreaux 14, CH-2000 Neuchâtel, Switzerland; ∗ Author for correspondence (e-mail:
[email protected];
fax: +64-7-858-4964) Received 15 November 2001; accepted in revised form 29 April 2002
Abstract. We discuss a paradigm for informed ecosystem management that provides a quantitative and rigorous foundation for informing conservation decisions and sustainable ecosystem management. Information pyramids incorporate conceptual and technological advances in ecosystem depiction and provide a framework for the integration and generalization of raw data into forms that are spatially extensive and at the appropriate level of generalization for a particular use. The basic tenets of the pyramid are: (1) Higher levels of the pyramid are entirely derived from a foundation of underlying data. (2) The process of generalization and integration upward should be objective and explicit. (3) Pyramids for different purposes often overlap, with common data and common methods for integration. (4) All levels of the pyramid should be developed together, including base data, methods and kinds of integration, and algorithms for using the information for planning and decision-making. Information pyramids are a powerful approach to organizing research science, and provide a mechanism by which research, data collection, storage and generalization can be focused on conservation outcomes. Common data and methods lead to increased efficiency, while also allowing for separate disciplines and programs. A case study of an integrated pyramid from New Zealand is discussed, which illustrates the characteristics of information pyramids. Components of this pyramid are discussed that provide examples of integration and generalization at various levels of the pyramid, from base data, to derived data, to spatial predictions and classifications, to a method of integrating this information into conservation decisions. Key words: Biodiversity indicators, Ecosystem depiction, Ecosystem management, Environmental domains, Generalized regression analysis and spatial prediction, Information pyramids Abbreviations: GRASP – generalized regression analysis and spatial prediction, LCDB – land cover database, LRI – land resource inventory, MCA – measuring conservation achievement.
Introduction The increased focus on ecosystem management has presented a number of challenges to conservation biology. Ecosystem management spans a range of activities at a range of spatial scales. Conservation activities range from site-focused activities to regional and national planning, reporting, and regulation. The resulting information needs are likewise varied in detail and scale. For instance, consider the following list of questions that might be posed by conservation managers:
2094 1. 2. 3. 4. 5.
What threatened species are present at this site? What is the natural distribution of this threatened species? What is the natural vegetation composition of this region? What is the current vegetation condition for this region? What increase in vegetation condition is achieved by potential management actions in this area? 6. What difference in the state of overall natural heritage is achieved by government expenditure on conservation in this country? The questions in this list increase progressively in their generality and spatial scale. The information required to answer them will also need an accompanying increase in its integration and generality. These examples illustrate that biodiversity conservation demands information at a range of spatial scales and a range of levels of generalization. Management activities extend from single species at single sites, to entire communities over regions, to entire natural heritage assets nationally and internationally. Management done at a particular site might take into account local data about the conservation problems at a site, such as a list of weed species present and their abundances, or the abundances of rare animal species and threats to them, such as introduced predators. However, reams of site-specific information will not be useful (or available) for regional or national decision-making. These different types of activities require information at very different levels of detail and generalization. The challenges of supplying information at such a range of scales in a manner that is ecologically meaningful may be one of the reasons that coordinated and informed ecosystem management at regional and national scales is still in its infancy. No conservation agency has the ability to define and quantify conservation outcomes, and attempts to quantify the outputs of conservation and environmental management are few (Metric and Weitzman 1998; Weitzman 1998). The measures proposed generally capture only a narrow portion of total conservation output (Cullen 1995, 1999), and often fail to link the output measure to value creation for society and future generations (Gowdy 1997; Clough et al. 1998). Nevertheless, there has been an increased move to making regional or national conservation decisions on a more systematic and quantitative basis. Regional conservation planning (Margules and Pressey 2000) has focused largely on the issue of reserve selection and design (e.g., Davis et al. 1999; Noss et al. 1999). Stephens (1999) has designed systems to deal with conservation allocation and reporting at national scales. Such conservation efforts are usually supported by coordinated research to provide and integrate a wide range of physical and biological information needs such as that in northeast New South Wales, Australia, summarized by Ferrier et al. (this issue) and Cawsey et al. (this issue). The move to broader ecosystem management has garnered wide support (Christensen et al. 1996), but has also generated controversy. In particular, the ability to adequately define, measure and depict ecosystems has been questioned (Fitzsimmons 1999) and debated (e.g., O’Neill 2001). While much is not known about ecosys-
2095 tems, there is also much that is known, including component parts, processes and emergent properties. Biotic components of ecosystems, such as individual species are reasonably well documented, as are some of the physical components, such as soils or climatic inputs. A fair amount is understood about some of the processes, such as nutrient cycles; and about some of the emergent properties, such as species richness, or net primary productivity. A productive approach is to not get distracted by the unknown, and to work with that which can be measured and described, while continuing to develop ways of dealing with less well known or less tangible properties. Generalizations of various types about ecosystems have been available for a long time. For instance, a map of vegetation types can be considered a generalization about ecosystems. Vegetation is an important component of ecosystems, and a description or classification of the vegetation pattern is thus an ecosystem classification (albeit not a complete one). Furthermore, we can describe some of the emergent ecosystem properties of vegetation, including species diversity, total biomass, and net primary productivity. The prediction across the landscape of particular ecosystem attributes, such as plant distributions, and rigorous methods to distill and generalize information are two essential components for conservation management. Traditional approaches to vegetation mapping have both operational and conceptual difficulties and reflect the technological constraints of a time when maps were made with pens on paper. The traditional approach to vegetation mapping is to decide (usually in advance) on a number of vegetation classes, and then map (predict) the spatial extent of each class. Boundaries of each vegetation type are determined by drawing lines on maps using a combination of field observations and either aerial photo-interpretation or remote sensing image classification techniques. The operational difficulties of such an approach are that it results in a flexible and unscalable classification and that the prediction process is often a subjective interpretation that is neither explicit nor repeatable. Conceptual difficulties with traditional approaches include that they impose a discrete structure onto species associations that is inconsistent with the large amount of independence in species associations, and that the sharp, smoothed boundaries are poor representations of what are often complex, fractal shapes and clinal variation. Advances in computing and spatial analysis allow for more rigorous ecosystem depiction that is consistent with their real properties and allow us to escape existing paradigms for spatial generalization. Biotic and abiotic components of ecosystems can be predicted, such as species distribution (e.g., Franklin 1995; Leathwick 1998), species richness (e.g., Lehmann et al., this issue), vegetation biomass (e.g., Overton et al. 2000), or soil characteristics (McKenzie and Ryan 1999). The resulting spatial predictions of ecosystem characteristics can be integrated further. For instance, species richness and natural community composition can be estimated by summing several species predictions (Leathwick 2001; Lehmann et al. 2002). Alternatively, biotic domains can be defined from the classification of predicted species distribution
2096 (Peters and Thackway 1998; Overton et al. 2000; Leathwick 2001). Environmental domains provide ecosystem classifications derived from environmental variables (Hargrove et al. 2001; Leathwick et al. 2001) with numerous other possible uses of environmental information available (Faith and Walker 1996; Hargrove and Hoffman 2000; Overton and Leathwick 2001). While many of these approaches are computationally (and memory) intensive, increases in computing power and storage allow them to be implemented at both broad scales as well as high spatial resolution. These quantitative approaches to the depiction and generalization of ecosystems are at the heart of information pyramids. Pyramids provide a vision for harnessing conceptual and technological advances for a robust and rigorous approach to developing environmental indicators and informing conservation decisions. In this paper, we first describe and develop the general concept of information pyramids, listing and discussing a number of central tenets of the pyramids. We then detail a case study of an integrated information pyramid developed for making conservation decisions, incorporating a number of quantitative approaches to ecosystem depiction. The top of this example pyramid is the measuring conservation achievement (MCA) process (Stephens 1999; Stephens et al. 2002) developed by the New Zealand Department of Conservation. Various components required by the MCA process are described, and used as examples of information integration and generalization. In particular, example components are chosen that represent recent advances in information integration. The fundamental data required for each step are discussed in the sections in which it is used.
Information pyramids Managers and scientists operate within an information pyramid, in which base data form the bottom of the pyramid, with increasing integration and generalization of these data moving up the pyramid. Figure 1 shows an information pyramid for the management of terrestrial biodiversity, labeled with components relevant to New Zealand. Information pyramids are a metaphor for a powerful paradigm for organizing research science, recognizing common data and common methods, and leading to increased efficiency, while also allowing for separate disciplines and programs. The core of an information pyramid is the integration and generalization of data and knowledge. We use the term integrate to mean combining different information, often of quite different types, such as biological information with environmental information. Generalizing information involves the identification of patterns, or the process of distilling detailed information into reduced forms, such as the classification of the vegetation composition in an area into vegetation types. Our formulation of these ideas was instigated by similar ideas in both the MfE Environmental Performance Indicator Programme discussion documents (Ministry for the Environment 1997; Burley 1998).
2097
Figure 1. An information pyramid. The leftmost visible face of the pyramid shows examples of the biophysical components for terrestrial biodiversity that are the focus of this paper. The right face shows examples of institutional components that are not addressed in this paper. The base of the pyramid is raw data, which is integrated and generalized to middle layers of the pyramid. Middle layers can be used at that level, and can also be integrated and generalized further to provide the information needs of the top of the pyramid.
Tenets and characteristics of the information pyramid Here we list four basic tenets of information pyramids and discuss their characteristics and advantages. 1. Higher levels of the pyramid are entirely derived from a foundation of underlying data. This tenet makes the fairly strict requirement that all generalizations and integrations of data be strictly data defined, ensuring a rigorous and objective pyramid. The example pyramid (Figure 1 and described below) gives a number of different ways in which data are integrated or generalized upwards. While it might well be necessary to use components of the pyramid that are not data defined, this should be viewed as an interim measure, until the data are available. 2. The process of generalization and integration upward should be objective and explicit. The information pyramid should be based on explicit methods of integrating upwards that are as objective as possible. Almost all methods, including quantitative ones, require subjective decisions or assumptions. It is important that these decisions be explicitly recognized and stated. This has a number of advantages. The work can be repeated and the same result should be obtained. This can be a useful check of
2098 procedures and allows the process to be repeated with updated data, or under changed conditions. The process of developing methods of integration and generalization, and the data underlying them, can be fully specified and debated to arrive at the best method. Without these information requirements, conservation decisions are largely based on personal opinion, sometimes of only one person. For example, the recognition of vegetation types is often based on the opinions or judgments of a few people. One could never repeat the process of making a vegetation map and get the same result, under the traditional approach, often even if the same people did it. This also applies to many existing general ecosystem classifications. Using such subjective approaches results in a pyramid that is based on weights of opinion, rather than objective data. 3. Pyramids for different purposes often overlap, with common data and common methods for integrating upward. Unlike Egyptian or Mayan pyramids, information pyramids may be nested and overlapping (Figure 2). Pyramids are nested when the information at the top of one pyramid becomes part of another pyramid. Thus, rather than being composed of components, information pyramids are in fact composed of many nested and overlapping pyramids, where each component pyramid is a generali-
Figure 2. Nested and overlapping information pyramids. Related pyramids share common fundamental data, as well as common methods for integrating upwards. Each pyramid is both composed of a number of smaller pyramids, as well as part of larger pyramids.
2099 zation or integration of information. Information pyramids do not suggest that only one generalization or integration is possible from base data. Instead, the situation is more like that in Figure 2, where multiple generalizations are possible. Pyramids overlap when a number of generalizations use the same underlying information. For instance, climate information may be used to make environmental domains, or predict species distributions or soil properties. Environmental domains, which are classifications of environment, are used as a conservation framework in both the MCA process described here, as well as the Environmental Performance Indicators Programme (EPIP, Ministry for the Environment 1997), and reporting by regional councils. Predictions of pest abundance, such as brushtail possums, can be used in both the MCA process to address issues of reducing biodiversity loss in native forests, and in a national strategy for control of bovine tuberculosis, of which possums are an important reservoir. On a more general level, many of the factors that determine natural community composition and ecosystems properties are identical to those that influence human use of landscape for productive purposes and the risks associated with those uses (e.g., McHarg 1969). For example, information on soils, climate, land cover and remote imagery is useful in assessing issues related to suitability for crops or pollution risks and movement. Thus, pyramids for terrestrial biodiversity will have overlap with pyramids for agricultural productivity or pollution issues. Pyramids also overlap because of common methods for integration and generalization. Database management techniques and metadata development are often quite general techniques. Classification techniques for example can be used to make environmental domains, or biotic domains, or soil classifications. Mechanistic models are often specific to a particular system, but use very general techniques. Spatial modeling techniques, e.g. GRASP, are quite general in approach and can be applied to a wide variety of systems and problems. Spatial prediction techniques, surface fitting, and geographic information systems are further examples of approaches and methods that are common to many types of information pyramids. Distinct but overlapping pyramids provide a powerful solution to the tension between cooperation and competition between and within agencies. The pyramids often correspond to scientific programs or administrative responsibilities. The overlap of the pyramids recognizes the data and methods shared between pyramids, and provides a clear avenue for cooperation between agencies. Of course, one of the inevitable tensions in coordinated research programs is patch protectionism and competition between and within agencies. The pyramid paradigm allows for the recognition of the common data and methods underlying these programs, while also respecting administrative boundaries. 4. All levels of the pyramid should be developed together, including base data, methods and kinds of integration, and algorithms for using the data for planning and decision-making. While information pyramids can be developed from the top, bottom, or middle, the ideal is for these approaches to be pursued together. Working from the top of the pyramid identifies the way in which information will be used,
2100 and gathers data and makes the appropriate generalizations to provide the needed information. A bottom-up approach first identifies the sorts of data currently available, or that can be gathered, such as vegetation plots, or types of remote imagery. The bottom-up approach might also say which kinds of information cannot be measured or are too costly, such as extensive measurements of biomass, or species composition of microbes. The middle parts of the pyramid focus on the ways of integrating and generalizing this information. The methods and approaches to generalizing the information are important constraints on what information can be provided, and advances in ecosystem depiction need to be incorporated into methods of decision-making. Pursuing just one of these approaches leads to short comings. A bottom-up approach can be a very inefficient method of developing an information pyramid. A large amount of data could potentially be gathered, and most of it would never get used. Gathering data without a vision for use, or without providing the data in a form that can be used, is a waste of time. Data gathered in this way might not provide the required information, might lack crucial information (such as spatial locations, or taxonomic information), and might lack the organization or format for others to use. Similar problems exist in the middle of the pyramid. The majority of scientists work in the lower parts of the pyramid. As a result, a common tendency is to make spatial predictions or other models or products, and expect others to recognize their value and develop a way to use them. Similarly, much research is not outcome focused. Equally difficult problems can also occur in trying to create a pyramid solely from the top. A top-down approach could design an elegant system for making conservation decisions that required all sorts of information that could never be measured, would be expensive to collect, or did not take into account underlying knowledge and understanding. The approach of working from the bottom or middle alone also leads to pyramids that are dysfunctional and likely to be abandoned. This is because base data and integrated information have been developed, but little or no mechanism exists for transferring these into a basis for management decisions. This may well be the explanation for why a number of visionary indicators programs have been developed, with excellent procedures for producing rigorous estimates of required parameters, and then abandoned. When pressure developed to cut the funding, there was no one to say, ‘Hey hang on! We use that information all the time to make decisions!’ Building all levels of the information pyramids together provides a targeted approach that recognizes constraints in underlying data while taking advantage of technological and conceptual advances in ecosystem depiction.
A case study from New Zealand Here, we illustrate information pyramids using a case study from New Zealand that provides an example of a coordinated effort to generalize biodiversity information
2101 and use this information to make conservation decisions. The New Zealand Department of Conservation has been criticized both by the public (e.g., Hartley 1997) and by central government control agencies for its inability to report on the difference made by the Crown’s expenditure on conservation to the state of the nation’s heritage. In response, the Department (Stephens 1999) has developed the measuring conservation achievement (MCA) method to measure the value created by conservation management. We use the case study to provide concrete examples of the tenets of information pyramids, and demonstrate the role of recent advances in spatial prediction and quantitative ecosystem depiction. Rather than try to cover all the diverse information needs of the MCA, we have focused on a number of biological and physical components that illustrate the use and application of information pyramids, and provide examples of the integration and generalization of information (Figure 3). These examples demonstrate how different levels of the pyramid are derived from lower, fundamental levels, while contributing to upper, derived levels. This example also illustrates the advantages of building up the entire pyramid simultaneously. During the process of MCA development, new ways of depicting biodiversity information have been utilized and incorporated in the process. Similarly,
Figure 3. Information integration for the MCA process. Here, the pyramid is reduced to two dimensions for simplicity. The MCA draws upon a number of smaller pyramids that integrate and generalize information to forms useable by the MCA and other purposes. While the MCA draws upon a broad range of biophysical and institutional information, this figure shows only the components that are discussed in the text as examples of information integration and generalization.
2102 the types of data collected and spatial predictions and generalizations have been influenced largely by the MCA developmental process. While the case study here relates to a particular pyramid, it is important to keep in mind that all pyramids overlap with others. There are many overlaps between the pyramid described here and other pyramids in New Zealand related to environmental performance and environmental performance indicators. Pieces of this pyramid, such as environmental domains and tree modeling, have been partly or mostly developed under other research programs. Measuring conservation achievement The MCA process is the top of our example pyramid. This process attempts to achieve the largest conservation bang for the limited conservation dollar, and provides a number of tools, including a basis for accountability, the basis for reporting conservation gains, losses and the difference made, and an identification of priority sites for conservation action. To attain these goals, it draws on a broad range of institutional, biological and physical information, such as environmental variables, topographic and cadastral data, land cover, animal pest, weed and other human disturbance data at its base. The MCA process provides an explicit algorithm for integrating these data to estimate representativeness, naturalness, irreplaceability and vulnerability of the areas affected by potential conservation projects. These estimates are then linked to information on design, cost and risk of the potential projects to assess site prioritization, project cost-effectiveness, and achievement reporting. Conservation management aims to defend or improve the condition of natural heritage by controlling exotic pests, weeds and other forms of human-induced disturbance. Conservation achieved is measured by the difference made to the intensity and diversity of human-induced disturbance pressures. Five types of human-induced disturbance pressures are recognized: biota removal by humans, physico-chemical resource alteration, consumption by introduced species, competition by introduced species, and fragmentation. Spatially explicit information is required for each of these types of disturbance. Naturalness is defined as the absence of these disturbances and their effects. The effects of conservation actions are measured by the difference in predicted site value with and without conservation management, and is then discounted for time and risk of nonachievement to provide a measure of conservation output (Figure 4). The output measure enables direct comparison of the value and cost-effectiveness of diverse and seemingly incomparable conservation projects such as possum control, wilding pine control, stock fencing, preventing the drainage of some wetland or creation of a marine reserve. The underlying information pyramid provides the basis for a number of important measurement and reporting tools used to manage conservation business and to measure the performance of the Department’s conservation managers. These include spatially explicit reporting on conservation gains, losses and the difference made across the landscape, identification of priority sites for conservation action, and conservation project cost-effectiveness analysis.
2103
Figure 4. The MCA conceptualization of conservation outcome size and derived measures. The benefit of conservation is the difference in site value with and without management. This outcome size is then combined with estimates of urgency and feasibility to assess project merit. Project merit is combined with net present value cost to assess project cost-effectiveness.
What value does it add? Trial implementations of the MCA process indicate large potential gains in efficiency in conservation achievement. Demonstrations in the Twizel conservancy, on the South Island of New Zealand (Stephens et al. 2002) show that efficacy varies over three orders of magnitude between projects. An optimum set of projects lead to three times as much estimated value, for the same overall budget, as did the current set of projects. As part of this project, the costs and benefits of implementation of the MCA were estimated. The implementation expenses included development, national databases, and operational costs. The benefits were estimated as the increased amount of conservation merit achieved by the optimal rather than the actual conservation measures. The dollar value of the benefits, estimated by the cost of producing those benefits without the MCA process, was 87 times the cost of implementation (Stephens et al. 2002), implying a cost:benefit ratio of 87:1. Climate surfaces Climate surfaces are an example of data generalization and integration that occurs in the lower levels of information pyramids. Climate is an important driver for a wide range of biotic and abiotic patterns. While climate stations are the main source of information for climatic averages, these stations provide only local (point) measurements, which are of themselves of little use for most applications. Point data must be integrated and scaled up into regional or national coverages. When these stations are
2104 used as the basis for extrapolating biodiversity characteristics across the landscape, the value of the investment in climate stations becomes apparent. At the beginning of the work to develop environmental domains and species modeling, climate surfaces for New Zealand were not available. Despite the large expenditure on weather forecasting, and the massive amounts of climate station data available, simple weather spatial predictions of long-term average climate were not available for the country. As a prerequisite for other work that required this component of the information pyramid, climate surfaces were developed (Leathwick and Stephens 1998). Despite spatial patterns in climate being hugely important in determining the productivity and sustainability of many natural resources, it is notable that these surfaces were not produced until required for a particular application. Once available, they are then useful for a wide variety of uses. The climate surfaces for New Zealand include both fundamental surfaces, as well a number of surfaces that are calculated from the fundamental surfaces. The calculation of the secondary surfaces represents the integration of physiological knowledge on the importance of different climatic factors influencing biological processes, mainly in vegetation. These climate surfaces can then be used as important spatial predictors of biotic patterns for GRASP analyses, or are combined with other sorts of environmental information like landform variables, as is done in environmental domains. Remote sensing imagery and derived products Remote sensing imagery provides a powerful source of spatially explicit information about the landscape. Raw satellite imagery provides the base data, which must be corrected and standardized in various ways before they are useful. Usually the image must be orthorectified, and reflectances must be corrected for the angle of the sun and the viewing angle (Shepherd and Dymond 2000). In addition, the slope and aspect of the land surface affect the reflectances, requiring topographic corrections (Shepherd and Dymond, in press). These corrections integrate a number of different types of information. After these corrections have been taken into account, the spectral information is available as a spatial predictor layer for further integration or generalization. A variety of forms of integrated remotely sensed information are also currently being produced, including layers of fraction of photosynthetically active radiation, woody vegetation, and classified imagery (J. Dymond, personal communication), with the aim of producing uniform layers for all of New Zealand. Once corrected or classified, there is still the question of how the patterns of spectral reflectances help predict the patterns of biodiversity or ecosystem characteristics. There are well-developed techniques for carrying out this in remote sensing including both supervised and unsupervised classifications of the imagery, combined with groundtruthing. The land cover database (LCDB) developed by Terralink and a consortium of end users primarily used SPOT3 and SPOT4 satellite imagery, along with auxillary
2105 information, to classify all New Zealand into about a dozen land cover types. Polygons of land cover were hand digitized from imagery, guided by classification of the imagery as well as by auxiliary spatial information. Despite its traditional approaches to spatial integration, the LCDB provides an important national classification of land cover. By combining it with other sorts of information, we can quantify how much information the classifications convey for particular uses. The New Zealand land resource inventory (NZLRI) NZLRI (Newsome 1992) provides classifications of land use capability (LUC) for all New Zealand. The NZLRI used a polygon-based approach developed in the 1980s, which was the only way to get national coverage, given the limited computing power available at the time. Polygons of uniform LUC were defined, and the fundamental base attributes, such as slope, drainage, soil type, etc., were estimated. Heterogeneity within the polygons was depicted by denoting primary and secondary values of the base attributes for each polygon. However, no information was available on the spatial distribution of these different properties within the polygon. The usefulness of this work can be seen by the fact that it is still an important source of spatial information for further levels of classification. Some of the soil and substrate information in the LRI is used as the substrate information in environmental domains and spatial predictions, such as species modeling. Hopefully, new updated layers of soil information will be developed using newer information and techniques. Progress in this direction is being made through work on standardizing a national soils database, and on methods for using this data to provide spatially explicit depictions of soil characteristics. Environmental domains Environmental domains analysis (Mackey et al. 1988; Belbin 1993; Hargrove et al. 2001; Leathwick et al. 2001) is becoming increasingly well known as a means of ecosystem classification. A domains analysis integrates climatic, soil and landform data and generalizes this information by classifying areas into domains of similar environmental characteristics. Thus, domains are defined as areas of similar environment, using environmental variables that have fundamental functional significance in determining physiological performance and/or proven statistical importance in determining species distributions. Domains of similar environment are used as surrogates of areas of similar ecosystem character. Environmental domains classifications provide a method of depicting ecosystems that is consistent with many ecosystem properties. To produce an environmental domains classification, the region of interest is first depicted by a digital elevation model (DEM) comprised of pixels of a given size, each with a location in geographic space. A number of environmental variables are
2106 then predicted for each pixel, thus defining the environmental space for each pixel. Non-hierarchical clustering is used to define a large number (≈1000) of groups of pixels (domains) that are similar in environmental space. Hierarchical clustering is then used on these domains to produce a dendrogram of similarity between the domains. This dendrogram provides the full domains classification for the region, and different number of groups can be produced by cutting the dendrogram at the desired resolution. These groups in environmental space can then be mapped back into geographic space. While pixels within a domain are close in environmental space, they can be scattered and distant in geographic space. Currently, a national domains implementation for New Zealand, Land Environments New Zealand (LENZ, Figure 5), is underway and due for completion by July 2002. LENZ is funded by the Ministry for the Environment (MfE) under the EPIP.
Figure 5. Land environments of New Zealand (LENZ). This is the current national implementation of environmental domains, depicted for Level I (20 group) and Level II (100 group).
2107 Other uses of environmental information provide continuous measures, rather than classifications (Belbin 1993; Faith and Walker 1996; Overton and Leathwick 2001). For instance, environmental distinctiveness provides another method for integrating the same environmental information that underlies environmental domains classifications, providing complementary applications. Environmental distinctiveness does not impose a discrete classification onto the continuous environmental variation, but instead, calculates a continuous measure of the distance of each pixel from some reference set of pixels. Environmental distinctiveness is used directly within the MCA process to assess site distinctiveness. GRASP predictions Generalized regression analysis and spatial prediction (GRASP) has been defined as both a general concept and a specific implementation in Splus (Lehmann et al. 2002). As a concept, GRASP is defined as a combination of regression modeling and spatial prediction that defines relationships in predictor space and uses these relationships to predict in geographic space. Regression modeling is used to establish relationships between a response variable and a set of spatial predictors. These relationships are then used to make spatial predictions of the response. The GRASP process requires point measurements of the response, as well as regional coverages of predictor variables that are statistically (and preferably causally) important in determining the patterns of the response. GRASP is an important method for integrating and generalizing information up the pyramid (Figure 6). GRASP integrates information by combining quite different sorts of information to make spatial predictions. Knowledge is also generalized by identifying the relationships between the different variables. For instance, GRASP could use a survey of the abundance of a species (the response), and existing spatial coverages of environmental (e.g., climate, landform) variables (the predictors) for a region. A multiple regression (here we use GAMs) can be used to establish the statistical relationship between the species abundance and the environmental variables. These regression relationships can then be used to predict species abundance from the environmental surfaces. The generalization of the knowledge comes both from the relationships seen in the regressions, as well as the spatial predictions. A specific implementation in Splus has been developed that facilitates this process (Lehmann et al. 2002). GRASP (the implementation) is an interface and collection of functions in Splus designed to facilitate modern regression analysis and the use of these regressions for making spatial predictions. GRASP standardizes the modeling process and makes it more reproducible and less subjective, while preserving analysis flexibility. The set of functions provides a toolbox that allows quick and easy data checking, model building and evaluation, and calculation of predictions. The current version uses only generalized additive models (GAMs), a modern regression technique.
2108
Figure 6. GRASP. This is an important method for integrating and generalizing information up the pyramid. A multiple regression (here we use GAMs) is used to establish the statistical relationship between the species abundance and the environmental variables. These regression relationships can then be used to predict the species abundance from the environmental surfaces. The GUI for the Splus implementation is also shown.
Ferrier et al. (two papers, this special issue) provide excellent examples of similar work in Australia, modeling up to 2300 species, and using a wide range of different biological information. Natural distributions of canopy tree species Here we describe work that integrates environmental information with existing plotbased surveys of forest canopy composition. Measurements of tree abundances in largely undisturbed forests were combined with climate and landform variables to establish the relationships between tree species abundance and environment in undisturbed forest. These relationships were used to predict what tree composition would be for all New Zealand in the absence of human influence. At the beginning of the work, climate surfaces were unavailable, and data were stored in hard copies in banana boxes in the basement of a research institute. Much work was required to organize the data into a relational database and create climate
2109 surfaces. Much of this work was performed under climate change research, and had strong overlap with environmental domains work in creating and testing the utility of climatic surfaces for the prediction of ecosystem characteristics. Tree species distributions were modeled in relationship to environment (Leathwick 1998, 2001; Leathwick and Austin 2001, Leathwick and Whitehead 2001) using GRASP. Environmental variables consisted of climate surfaces and landform variables from the LRI. Individual tree species abundances were regressed against environmental variables using GAMs. Since the genus Nothofagus is noted for being at non-equilibrium with its environment, the models of Nothofagus included a spatial variable to account for this patchiness. Regressions of many other species included Nothofagus abundance as predictor variables (Leathwick 2001, 2002). These regression models were used to predict individual species abundances for each of the pixels in a 1-km DEM of New Zealand. This results in spatial predictions of forest community composition that incorporates known biogeographic effects and flow-on competition effects. These predictions are useful in themselves, e.g., for understanding species distributions or designing restoration potential. They can also be integrated further in various ways. One example is given in the Biotic domains section below. Vegetation condition, weeds, and pests In two projects specifically designed to address information needs identified by the MCA process, we are predicting vegetation condition, weed distributions, and the distribution of an introduced pest. All these predictions are designed to provide specific information needs for ongoing implementations, as well as provide a general demonstration of the power and insight gained by combining single observations across the landscape into general patterns and spatial predictions. Unfortunately, the data for this were not available, and had to be collected and collated from a wide variety of sources and formats. These projects modeled vegetation or possum data as a function of climatic surfaces, landform variables, or land cover (described above). As such, they integrated information on vegetation, or possum abundance with existing spatial predictions. For the vegetation data, the study area was confined to the central part of the South Island of New Zealand. We used existing information from appropriate plot-based surveys with data stored in the NVS database (Wiser et al. 2001). We identified the distribution of existing information, and collected new information to sample areas and environmental or land cover combinations without data (see Cawsey et al. (this issue) for a similar process). The resulting database of previous and additional vegetation plots was then used to make spatial predictions about vegetation condition and weed distributions. Vegetation condition can be difficult to define. One of the advantages of the explicit, quantitative approach advocated here is that it forces decisions about what is meant by vegetation condition, and makes them explicit. We are predicting a number of indices of vegetation condition,
2110 including: the proportion of species that are native (Figure 7); the proportion of total vegetation cover and vegetation volume that is accounted for by native species; and the richness of native and exotic species. The choice of weed species for prediction was guided by their conservation impact and by species for which
Figure 7. The proportion of species native for the central South Island, predicted using GRASP.
2111 active control is carried out. In addition to the spatial predictions, the relationships of each weed species with environmental variables are used to guide insights into the species ecology and distribution. The brushtail possum, Trichosorus vulpecula, is an Australian endemic that has gone awry in New Zealand. Introduced to establish a fur trade over a century ago, the species now damages native forests, predates nests of native birds, and carries bovine tuberculosis (Clout and Ericksen 2000). Extensive control of possums by poisons and trapping is carried out for conservation outcomes and control of bovine tuberculosis. In 1996, a national trap catch protocol (Warburton 2000) was widely adopted and used as an index of possum abundance before or after control operations. As a result, there now exist numerous trap catch surveys spread around the country. While considerable effort was required to collect and collate even part of this data information, it provides a valuable opportunity for spatial integration and generalization. Integrating scattered trap catch surveys with existing climate and landform variables, using GRASP, provided a rich model of the large scale behavior of the trap catch index and a spatial prediction of it as an index of possum abundance for MCA purposes. This example also illustrates overlapping pyramids; predictions of possum abundance and insights into the effects of season and set type on the trap catch index are of widespread interest for regional and local planning of pest control measures. Biotic domains Individual spatial predictions can be further generalized in various ways. For example, while environmental domains describe areas similar in environmental attributes, biotic domains define areas similar in predicted community composition. This can be done using different sets of species, and at different number of groups, resulting in flexible, scalable classifications of natural vegetation (Figure 8). A biotic domains analysis for tree species (tree domains) is shown in Figure 9. This tree domains analysis is based on the predicted community composition of natural forests, using predictions for 43 individual tree species. It provides a generalization of the patterns of predicted natural forest pattern for New Zealand (Leathwick 2001). Biotic domains follow the ‘predict first, classify later’ approach to vegetation classification, and avoid many of the problems of traditional vegetation maps (Overton et al. 2000). They are based on rigorous spatial extrapolations from objective point measurements, while traditional vegetation classifications are based on non-repeatable spatial predictions of subjective forest classes. Biotic domains are fully scalable. The full realization of the classification is the dendrogram of similarity between groups (extending to the individual pixels). Furthermore, species abundances are allowed to vary independently in this classification. Overall, the properties of this classification are much more consistent with the general properties of ecological community composition.
2112
Figure 8. The biotic domains approach to ecosystem classification. Biotic domains follow the ‘predict first, classify later’ approach to vegetation classification, and avoid many of the problems of traditional vegetation maps. Individual species predictions from GRASP can be classified in different ways and at different numbers of groups, resulting in flexible, scalable classifications.
Synthesis While the vision described here may seem almost common sense, examples of coordinated pyramids of research and indicators informing conservation decisions are rare. The usual situation is one of disorganized and disparate data that cannot be used in a coordinated fashion, and of few methods for integrating upwards and guiding decision-making. When generalizations of ecosystem patterns do exist, they are generally subjective and not based on any underlying data. If indicators exist, there is often little or no mechanism for using them in a coordinated fashion. Information pyramids provide a framework for organizing conservation science and the underlying research and information needs. The advantages of explicitly operating within information pyramids are numerous, including: (1) management decisions focused onto using indicators and other information explicitly derived from monitoring measurements, (2) missing components of the pyramid are more easily identified, resulting in increased efficiency of gathering data and generating indicators, (3) research efforts are focused on understanding important relationships. Information pyramids draw upon advances in quantitative ecosystem depiction to provide the basis for informed ecosystem management.
2113
Figure 9. Tree domains for New Zealand. Individual predictions of natural tree distributions were integrated in a biotic domains classification. Names of the groups were defined post hoc to describe the mean predicted composition in the group.
The example pyramid illustrates a number of properties of the information pyramids. The integration and generalization of data and knowledge upwards are shown in all examples, from base data, such as climate stations or vegetation plots, to climate surfaces or predicted species distributions or environmental domains, to estimation of
2114 distinctiveness or importance of sites, and then to optimal sets of conservation projects. Base data and derived layers are useful on a range of different generalizations. Climate surfaces for instance, can be used for predicting species distributions or soil characteristics, or for inclusion in environmental domains. Environmental domains are useful for the MCA process, the indicators programme, and a broad range of other applications. Quantitative and explicit methods are used throughout for integration and generalization, allowing the methods and results to be verified and repeated with updated information, or for different scenarios. The challenges to the construction of information pyramids are many. Technical and computational barriers remain, and are just recently being overcome through rapid advances in computing power. The science needed to underpin the integration methods is following, with many quantitative and statistical issues remaining. While the cross-disciplinary and integrative nature of the triangles is one of its strengths, it also poses one of its greatest challenges. Many competitive and territorial barriers exist both within and between institutions regarding data and research and management activities. While one of the advantages of the pyramid is that missing pieces of data can be identified, it is often not clear what the best way is to get lower levels of the pyramid responsive to higher levels. Despite the information needs being defined, there may not be a means to actually get that data produced. The lack of the appropriate data remains the largest barrier to using these approaches. Although this paper focuses on biodiversity issues, the information pyramid is very general and can be applied to a wide variety of other situations. Integrated ecosystem management must consider the biodiversity values of the landscape, the potential for human production activities, and also the environmental impacts and risks of these activities. Much of the same biophysical information, such as soils and climate, underlies all these activities, as do the same methods of integration and generalization. Processes such as MCA can be combined with systems such as environmental accounting to assess the societal costs and benefits of various patterns of land use. Overall, information pyramids provide a vision that applies to producing and using the entire range of environmental and social indicators.
Acknowledgements The work described here has had important contribution from many people, including Gary Barker, Kirsty Johnston, Gareth Wilson, Bruce Warburton and Dan Rutledge. We wish to thank Bill Lee for comments on the manuscript and M. Anne Austin for careful editing and Rochelle Holland for revisions. New Zealand’s Foundation of Research, Science, and Technology provided funding under Contract C09642. The Ministry for the Environment funded the development of LENZ under the Environmental Performance Indicators Programme.
2115 References Belbin L. 1993. Environmental representativeness: regional partitioning and reserve selection. Biological Conservation 66: 223–230. Burley J. 1998. Joining the revolution: a strategy for the standardization integration and dissemination of biodiversity information – as a prospective model for the management of other kinds of environmental information. Manaaki Whenua – Landcare Research, Lincoln, New Zealand. Cawsey E.M., Austin M.P. and Baker B.L. 2002. Regional vegetation mapping in Australia: a case study in the practical use of statistical modelling. Biodiversity and Conservation 11: 2239–2274 (this issue). Christensen N.L., Bartuska A.M., Brown J.H., Carpenter S., D’Antonio C, Francis R., Franklin J.F., MacMahon J.A., Noss R.F., Parsons D.J., Peterson C.H., Turner M.G. and Woodmansee R.G. 1996. The report of the Ecological Society of America Committee on the scientific basis for ecosystem management. Ecological Applications 6: 665–691. Clough P., Barnett J. and Gale S. 1998. The marginal value of conservation. Report for The Treasury and Department of Conservation by New Zealand Institute of Economic Research (Inc.), Wellington, New Zealand. Clout C. and Ericksen K. 2000. Anatomy of a disastrous success: the brushtail possum as an invasive species. In: Montague T.L. (ed.), The Bushtail Possum: Biology, Impact and Management of an Introduced Marsupial. Manaaki Whenua Press, Lincoln, New Zealand, pp. 1–9. Cullen R. 1995. Output measurement for evaluation of ecosystem protection programmes. University of Stirling, Ecological Economics Discussion Paper 95/6. Cullen R. 1999. Biodiversity protection: measurement of output. Proceedings of AARES Conference 20– 22 January 1999, Christchurch, New Zealand. Davis F.W., Stoms D.M. and Andelman S. 1999. Systematic reserve selection in the USA: an example from the Columbia Plateau ecoregion. Parks 9: 31–41. Faith D.P. and Walker P.A. 1996. Environmental diversity: on the best possible use of surrogate data for assessing the relative biodiversity of sets of areas. Biodiversity and Conservation 5: 399–415. Ferrier S., Watson G., Pearce J. and Drielsma M. 2002a. Extended statistical approaches to modelling spatial pattern in biodiversity in northeast New South Wales. I. Species-level modelling. Biodiversity and Conservation 11: 2275–2307 (this issue). Ferrier S., Drielsma M., Manion G. and Watson G. 2002b. Extended statistical approaches to modelling spatial pattern in biodiversity in northeast New South Wales. II. Community-level modelling. Biodiversity and Conservation 11: 2309–2338 (this issue). Fitzsimmons A.K. 1999. Defending Illusions: Federal Protection of Ecosystems. Rowman and Littlefield Publishers, Lanham, Maryland. Franklin J. 1995. Predictive vegetation mapping: geographical modelling of biospatial patterns in relation to environmental gradients. Progress in Physical Geography 19: 474–499. Gowdy J.M. 1997. The value of biodiversity: markets, society and ecosystems. Land Economics 73: 25–41. Hargrove W.W. and Hoffman F.M. 2000. An analytical assessment tool for predicting changes in a species distribution, map following changes in environmental conditions. Proceedings of the Fourth International Conference on Integrating GIS and Environmental Modeling (GIS/EM4): Problems, Prospects and Research Needs. Banff, Alberta, Canada. Hargrove W.W., Hoffman F.M. and Sterling T. 2001. The do-it-yourself supercomputer. Scientific American 265: 72–79. Hartley P. 1997. Conservation Strategies for New Zealand. The Business Round Table, Wellington, New Zealand. Leathwick J.R. 1998. Are New Zealand’s Nothofagus species in equilibrium with their environment? Journal of Vegetation Science 9: 719–732. Leathwick J.R. 2001. New Zealand’s potential forest pattern as predicted from current species-environment relationships. New Zealand Journal of Botany 39: 447–464. Leathwick J.R. 2002. Intra-generic competition among Nothofagus in New Zealand’s primary indigenous forests. Biodiversity and Conservation 11: 2177–2187 (this issue).
2116 Leathwick J.R. and Austin M.P. 2001. Competitive interactions between tree species in New Zealand’s old growth indigenous forests. Ecology 82: 2560–2573. Leathwick J.R. and Stephens R.T.T. 1998. Climate surfaces for New Zealand. Contract Report LC9798/126, Landcare Research, Hamilton, New Zealand. Leathwick J.R. and Whitehead D. 2001. Soil and atmospheric water deficits, and the distributions of New Zealand’s indigenous tree species. Functional Ecology 15: 233–242. Leathwick J.R., Overton J.McC., Emmett D., McLeod M. and Lynn I.H. 2001. Environmental domains for the Canterbury and Waikato Regions: a case study in the definition of spatial classifications for biodiversity management in New Zealand. Environmental Performance Indicators, Technical Paper No. 68 Biodiversity. Ministry for the Environment, Wellington, New Zealand. Lehmann A., Leathwick J.R. and Overton J.Mc.C. 2002. Assessing New Zealand fern diversity from spatial predictions of species assemblages. Biodiversity and Conservation 11: 2217–2238 (this issue). Lehmann A., Overton J.McC. and Leathwick J.R. 2002. GRASP: generalized regression analysis and spatial prediction. Ecological Modelling 157: 187–205. Mackey B.G., Nix H.A., Hutchinson M.F., McMahon J.P. and Fleming P.M. 1988. Assessing representativeness of places for conservation reservation and heritage listing. Environmental Management 12: 501–514. Margules C.R. and Pressey R.L. 2000. Systematic conservation planning. Nature 405: 243–253. McHarg I.L. 1969. Design with Nature. Natural History Press, Garden City, New York. McKenzie N.J. and Ryan P.J. 1999. Spatial prediction of soil properties using environmental correlation. Geoderma 89: 67–94. Metric A. and Weitzman M.L. 1998. Conflicts and choices in biodiversity preservation. Journal of Economic Perspectives 12: 21–34. Ministry for the Environment 1997. Environmental Performance Indicators: Proposals for Air, Freshwater and Land. Ministry for the Environment, Wellington, New Zealand. Newsome P.F.J. 1992. New Zealand Resource Inventory. ARC/INFO data manual. DSIR. Noss R.F., Strittholt J.R., Vance-Borland K., Carroll C. and Frost P.A. 1999. A conservation plan for the Klamath–Siskiyou ecoregion. Natural Areas Journal 19: 392–411. O’Neill R.O. 2001. Is it time to bury the ecosystems concept? (with full military honors, of course!). Ecology 82: 3275–3284. Overton J.McC., Leathwick J.R. and Lehmann A. 2000. Predict first, classify later – a new paradigm of spatial classification for environmental management: a revolution in the mapping of vegetation, soil, land cover, and other environmental information. In: 4th International Conference on Integrating GIS and Environmental Modelling (GIS/EM4), Banff, Alberta, Canada. Overton J.McC. and Leathwick J.R. 2001. Measuring environmental distinctiveness. Science for Conservation 174: 20 pp. Peters D. and Thackway R. 1998. A new biogeographic regionalisation for Tasmania. Tasmanian Parks and Wildlife Service GIS section. Shepherd J. and Dymond J. 2000. BRDF Correction of Vegetation in Avhrr Imagery. Remote Sensing of Environment 74: 397–408. Shepherd J. and Dymond J. (in press) Correcting satellite imagery for the variance of reflectance and illumination with Topography. International Journal of Remote Sensing. Stephens R.T.T. 1999. Measuring conservation achievement. In: Blaschke P.M. and Green K. (eds), Biodiversity Now! Joint Societies Conference, Wellington, 1997. Selected papers. Department of Conservation, Wellington, New Zealand, pp. 13–39. Stephens R.T.T., Brown D.J. and Thornley N.J. 2002. Measuring conservation achievement: concepts and their application over the Twizel area. Science for Conservation 200: 114 pp. Warburton B. 2000. Monitoring possum populations. In: Montague T.L. (ed.), The Bushtail Possum: Biology, Impact and Management of an Introduced Marsupial. Manaaki Whenua Press, Lincoln, New Zealand, pp. 132–142. Weitzman M.L. 1998. The Noah’s ark problem. Econometrica 66: 1279–1298. Wiser S.K., Bellingham P.J. and Burrows L.E. 2001. Managing biodiversity information: development of New Zealand’s National vegetation Survey databank. New Zealand Journal of Ecology 25: 1–17.
