Representing the Dimensions of an Ecological Niche

Representing the Dimensions of an Ecological Niche Deana Pennington1 1 Long Term Ecological Research Network Office, University of New Mexico, MSC03 2020, Albuquerque, NM 87131-0001, (505) 277-2595, [email protected] Abstract. Niche theory has driven a century of ecological investigations and has been the starting point for the few studies in existence of ecological ontology. This study investigates how scientists conceptualize niche theory, and uses that information to develop a high level ontological framework. That framework includes process, theory, entities, experiments, space, and time. These components are linked in different ways by different disciplinary perspectives, and the context of space and time differs depending on that linkage. Examples are provided to illustrate the representational issues involved. 1. Introduction Smith [1] constructed one of the first ontologies of ecological information, noting the inherently spatial nature of niche theory. Fonseca [2, 3] further investigated ecoontologies using the niche concept, observing that ecological ontologies differ from geo-ontologies in that they are temporal and recursive. Most recently, Keet [4] investigated the use of ecological models to generate ontologies, and discussed the need to represent ‘flow’. Niche theory has been a central driving concept in ecology and has a long and volatile history [5]. The term niche was first coined by Grinnell in [6] as the collection of environmental conditions describing where a species lives. Elton [7] suggested that an ecological niche is the role that a species plays within an ecosystem. Hutchinson [8] articulated the niche as a hyperdimensional space composed of the set of environmental conditions where a species can exist (functional niche) reduced by spaces where interactions with other species won’t allow existence (realized niche). From this theory, different lines of ecological investigation have evolved, such as habitat modeling and studies of environmental variability in space. Niche theory has been hotly debated within ecology for nearly a century. Formal representation of the niche concept, if achieved, is likely to provide a robust framework for much of ecology. The niche concept has also been useful in thinking about processes outside of ecology (such as market niches). Because of the cross-disciplinary usefulness of niche theory, a robust framework for formal representation of ecological niche theory has implications for frameworks in other domains. Hence, a well thought out approach to the semantics of niche theory in ecology is highly desirable. In this paper, I use cognitive mapping to expose the scientists’ view of niche theory to high level analysis of representational approaches. First I describe an investigation designed to expose different disciplinary views of niche theory, including spatial and non-spatial perspectives. Then I construct a framework that integrates those perspectives into high level ontological categories. Lastly I explore some of the representational issues exposed by the investigation, with special attention to spatial context. 2. Methods Cognition is the process by which an individual acquires, codes, stores, recalls, and decodes information about phenomena in their environment [9]. Concept mapping is a method by which an individual’s cognitive (mental) map is made explicit through diagramming associations between terms relevant to a central topic. The goal of this study was to develop a process by which a group of scientists could collaboratively construct an agreed upon community map, from which an ontology could be engineered. Eleven ecologists were asked to draw concept maps about niche theory. Six identified themselves as biogeographers or as having interests in geospatial approaches. The other five were from non-spatial disciplines. They were instructed to include three terms in their diagram: 1) niche, 2) fundamental niche, and 3) realized niche. They could make it as simple or complex as they desired, but were told to stop when they felt they had included the critical terms for understanding niche theory. The concept maps were redrawn in CMap Tools (http://cmap.ihmc.us/) to standardize appearance, given generic names, and all of the maps were distributed to each participant. They were asked to review the maps, then revise their own in whatever way they wanted. The final goal was to align the maps to generate a community map. A list of terms used on the concept maps was compiled. Minor variations between terms (such as tense) were resolved. Semantic overlap between two participants was calculated as the number of terms used by both compared with the combined number of unique terms. The terms were categorized into high level groups, then subdivided into a second tier. The ecologists were then asked to review the categorization of their terms, and revise it however they felt appropriate. They could move terms to different categories, add new categories, or completely revise the grouping. Lastly, they were asked to respond to some questions about the concept mapping process, both from an individual perspective and as a mechanism for group interaction. 3. Results: Scientists’ Cognitive View How do ecologists conceptualize a niche? The variation in concept maps was truly remarkable (Figure 1). Not only were the maps structured differently, there was less than 20% semantic overlap in the terms used between any two maps (mean = 7.5%, range = 0% to 19.5%). Overlap was slightly higher between maps of traditional ecologists (mean = 8.7%; maximum = 19.5%) than biogeographers (mean = 7.9%; maximum = 18.2%), both of which were higher than overlap between the two (mean = 4.6%; maximum = 13.9%). Only three chose to make revisions after viewing others’ maps, and those revisions were simply adding terms to the edges. None made any structural revisions, precluding development of a community map. A ST ST P S P P B TH P P P P TH S TH P P E TH E TH P TH E E P = Process S = Space S M T E ST S P ST TH = Theory T = Time ST S M = Model E = Entity ST = State Fig. 1. Comparison of concept maps by scientists with differing disciplinary perspective, showing different semantic network structure and usage of different categories of terms. A) biogeographer. B) ecologist. Figure 1 illustrates two concept maps, one from a biogeographer and one from a traditional ecologist. In places the terms are connected via subsumption hierarchy, but there are diverse cognitive structures represented. Most often linkages are between the major elements of scientific method, linking a model term with a process term or an entity term with a theoretical term, reflecting the way science is conducted rather than a hierarchy of concepts. Some differences reflect the uncertainty in understanding inherent in science. For instance, the exercise generated discussion among the scientists about how biotic interactions could potentially expand the range of suitable habitat for a species, a point most had not previously considered. Other differences reflect disciplinary perspectives. One map included no spatial terms whatsoever and another map was composed of terms that were virtually all spatial. Many of the issues currently confronting science require integration between differing perspectives. There is a need to represent these diverse scientific perspectives for purposes of conceptual integration, regardless of whether they enable technical integration. While perspectives of the relationships between major elements of scientific method may be diverse, there was little disagreement about the proposed categorization within elements. Only three scientists made any changes to the hierarchy, less than 5 terms were changed per scientist, and in most cases changes were made at the second tier rather than at the highest level. For example, there might be disagreement about the mechanisms of predator/prey interaction and how that relates to niche theory, but agreement that predation is one of several interspecies processes that limit species’ distributions. Hence a hierarchy of terms could be established that would ostensibly be amenable to community agreement – there can be ontological commitment in spite of process disagreement. This study suggested that the major areas of the scientific method (theory, experiment, process, observation in space and/or time) provide high-level semantic categories, however there are some terms that do not fit that structure cleanly. Many participants used terms that might be called a “state”, a snapshot of an entity in space and time, without decomposing those elements. For example, the term “habitat” was used frequently, referring to the collective state of many environmental variables across space at a point in time. Some participants included scientist’s names in their concept maps, linked along an issue time line. The majority of terms were placed in either the process (33.0%) or state (38.5%) categories. Remaining terms were divided between spatial (9.5%), theoretical (8.0%), models (4.5%), temporal (2.8%), and researcher name (2.8%) categories. 3. Niches, Scientific World View, and Ontological Categories The scientific method provides a structure for high level semantic categories (Figure 2). Science is an approach for understanding processes, therefore much of the language of science can be categorized as process terminology. Other high level categories reflect the approach for observing those processes. One observes entities (entity category) in space and time (space and time categories), then makes theoretical inferences about the relevant processes that govern the entity (theory category). Observations of change may be opportunistic, or they may be controlled through experimental manipulation (experimentation category). These categories are combined during observation (observation category) which places any one category into the context of the others [10]. Hence, any artifact of science (dataset, tool, publication) must be placed in a context that includes all of these dimensions. The language of science also includes the scientific tools for observation (experimentation), and the approach to reasoning about those observations (theory), each of which has spatial and temporal characteristics that are independent from the observed entity space and time characteristics. A theory is developed in space and time and has a state, separate and apart from the spatial and temporal characteristics of the entities it refers to. An understanding of the spatial and temporal characteristics of theory development is important for placing any artifact into the correct theoretical context. Likewise, experimental designs have a space and time context. For instance, field experiments are embedded in nested geographic experimental units. The spatial and temporal context of the experiment is related to, but different from, the spatial and temporal context of the entities being observed. Experimental terms such as replicates, blocks, and treatments, which have no spatial context in lab experiments, are inherently spatial in field experiments (Figure 3). A Processes Observation Theory How? Space Conceptual context Time Who, what, where, when Space/time context Relational context Entities & properties Test of How Does it fit? Implications? Methodological context Experimentation B Processes Observation Niche Theory Abiotic interaction Biotic interaction Species Occurrence Ecosystems Biogeography Evolutionary ecology Time Space Scaling Taxonomy Entities & properties Predict occurrence Change analysis Experimentation Field studies Modeling Simulation Fig. 2. Major components of scientific method linked with space and time, relationship with process, and combined through observation. Each component corresponds to a high level category of ontological conceptualization. A) Generic components. B) Niche theory example. Depending on how the experiment is designed, the same terms may have different spatial meanings even though their experimental meaning is the same. A plot is an experimental unit that occurs at a particular place, therefore the plot has a spatial context that may or may not be georeferenced. The entity under observation occurs within the plot and its spatial context is constrained by the plot, but is not the same as the plot. The same entity could be observed in multiple experiments using different experimental designs, and overlapping plots. In that case, although the entity has not moved, the spatial context of the plots and other experimental units within which it occurs has changed. Therefore one must distinguish between the spatial semantics of the entity under observation and the spatial semantics of the experimental context within which it occurs, yet identify the topological linkage between the two. 1m Sites 1-3 C N C N C N C N C N C N C N C N 0.5 m 1m Trap Plot Replicates 1-8 Treatments N, C 1 sample per plot, 1 week capture time, 2 times per year For each sample count number of aphids and number of ladybugs Quadrats 1-8 Plots 1-8 Sites 1-2 L S L S CT N P N+P CT 5’ x 20’ S L S L Quadrat N+P Blocks 1-4 Trap 1m LF SF Treatments: CT, N, P, N+P, L, S Fences NF, LF, SF 1m 1 sample per quadrat, 48 hour capture time, 1 time per year For each sample count number of aphids and number of ladybugs Fig. 3. Examples of experimental design in ecology, showing inconsistent use of terminology and spatial implications of field experiments. Each observation (column) within a dataset can stand alone, or can provide context for other observations; an ontology must provide mechanisms to describe these linkages [10]. Although all of these entities and their nested relationships can be captured with ontological annotations and used to reconstruct the spatial configuration and experimental design for automated integration, a map is still the most effective representation of this information to human interpreters. To summarize, high level ontological categories in science reflect the major components of the scientific method: process, theory, entities, space, time, experimentation, and observation. Theory, entities, space, time, and experimentation can be visualized as forming a polyhedron, where each point can be considered independently, or in conjunction with adjacent categories (Table 1). Space and time take on different meanings depending on the dimensions under consideration. Observation links all five categories. All of these, collectively, are used to understand process. Any scientific ontology must accommodate these different high level categories. Table 1. Different dimensions of the scientific method polyhedron shown in Figure 2, and examples of perspectives that make use of that facet. Theory, experimentation, entities Space, entities, point in time Time, entities, point in space Entities, theory, time Entities, theory, space Theory, space, time Experiment, space, time Experimentation, theory, entities Ecology Cartography History Evolutionary ecology Spatial analysis Issue timeline Experimental design Statistics 4. Representing Process Fonseca et al. [2, 3] and Keet [4] considered ontological methods for representing process, which Fonseca referred to as ‘recursion’ and Keet referred to as ‘flow.’ The question of interest is, how do we best represent the flow of recursive processes? Generations of scientists have converged on process representation through mathematical and computational models (Figure 4). These modes of representation are explicit, formal, and highly effective at capturing flow and recursion. They are reproducible, and easily modified to accommodate different perspectives on the mechanics of process, which entities to represent, and how to incorporate space and time. They do not require community agreement – there can exist many different perspectives simultaneously. They are not, however, a particularly effective way of communicating the meaning of what they represent without studying the details of the model. Conversely, ontologies are not conceived of, nor designed, for capturing processes, but are designed for expressing precise semantic meaning, exactly what is lacking in models. Combining the best of both, processes can be represented by mathematical or computational models, then annotated with terms from an ontology to precisely describe the semantic meaning that the model represents [11, 12, 13, 14]. For example, predation interactions have been modeled computationally (Figure 4A) and mathematically (Figure 4B). These are isomorphic representations of the recursive flow of processes that control the number of predators and number of prey. The Lotka-Volterra model of predator/prey interactions results in a cyclical pattern of predator and prey numbers because of recursion (Figure 4C). The Lotka-Volterra model has been applied in many ecological studies and has been represented in diverse ways. Scientists with different perspectives studying different parts of the ecosystem differ in how they apply the model. But the semantic meaning of the model is constant (Figure 4D). It is an ecological model for investigating the process of predation which describes interactions between predators and prey using an experimental approach that is computational and/or mathematical. Each of these components has a spatial and temporal context that depends on the specific implementation. Prey Births A Prey Deaths Prey Population Prey death proportionality constant Prey birth fraction Predator birth fraction Predator death proportionality constant Predator Population Predator Births Predator Deaths B C Population size ∫ r*n1 – a*n1*n2 ∫ – d*n2 + b*n1*n2 Prey Predator Time n1 = # of prey a = prey death rate n2 = # of predators b = predator efficiency r = prey growth rate d = predator death rate D hasExperiments EcologicalTheories EcologicalExperiments hasProcess EcologicalProcesses NicheTheory hasProcess EcologicalEntities Predator isa has hasEntity hasModel Predation hasModel EcologicalModels PredationModels hasEntity Prey Lotka-VolterraModels Fig. 4. Alternative representations of the Lotka-Volterra predator/prey model and output. A) Stella visual modeling representation. B) mathematical representation. C) output time series, D) ontological representation. This explicit description can be captured by annotating the models and components (inputs, outputs, parameters) with an ontology, which enables automated discovery and (potentially) automated transformation of resources to act as inputs (datasets, output from other models, etc.). This approach, being developed by the NSF-funded Science Environment for Ecological Knowledge (SEEK) project (http://seek.ecoinformatics.org), allows there to be many different predation models, each representing a diverse scientific view, yet annotated with a community ontology. 7. Conclusions There is a tension between representing the full diversity of scientific perspectives and development of formal ontologies based on community ontological commitment. Science is about alternative explanations and no scientist would ever agree to the notion that diversity should be abandoned. However, as demonstrated by this study, scientific argumentation is often not related to categorical groupings. There are some notable exceptions to this. Vegetation and soil classification are notoriously difficult to achieve consensus on, and species taxonomies undergo radical change constantly as the result of scientific argumentation. The point is that scientists mostly argue about non-hierarchical relationships best captured within an analysis or workflow rather than by an ontology. Agreement on a hierarchy of concepts and a limited number of relationships does not in any way jeopardize one’s freedom to agree or disagree on other kinds of linkages that are not specified by the formal ontology. This, in and of itself, is a reason to exclude process-based relationships from formal ontologies, even if a technical mechanism for incorporating process is identified. Processes, linkages between processes, and the recursive nature of processes are easily represented with mathematics and programs, but not by ontologies. Maps, diagrams, procedures, timelines, visual argumentation; all are very effective at representing certain kinds of information for which ontologies are not very effective. The best approach seems to be to determine what needs to be represented, represent it with the most effective means possible, then annotate it using an ontology for automated discovery. Once discovered, mechanisms for rapid exploration of all of the relevant pieces of information are desirable. The relevant pieces include all of the specific resources of interest plus all of the contextual information needed to make sense of those resources. For science, this includes not only metadata, but also information regarding the spatial and temporal context of theory under which the resource was developed, spatial and temporal context of the experimental design, as well as the spatial and temporal context of the entity. Multiple space and time contexts from these different dimensions of scientific investigation may or may not be linked conceptually, but if linkages exist they must be exposed. Capturing all of the necessary information from scientists, representing it in diverse ways, yet coordinating representation in such a way that the resources can be discovered and manipulated using ontologies requires both multi-disciplinary and interdiscplinary efforts between scientists, developers, knowledge engineers, and semantic experts. No single approach will solve all of the issues. This problem is truly a ‘wicked problem’ (sensu [15]), and will require an investment in resources to manage and facilitate complex problem solving across a diverse group of specialists, projects, and institutions. Acknowledgments This work was funded by the National Science Foundation (NSF) award number 0225665 for the Science Environment for Ecological Knowledge (SEEK) project. The author thanks the entire SEEK KR/SMS team for many months of grueling but thought-provoking discussion. References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. View publication stats Smith, B.: The Niche. Nous 33 (1999) 198-222. Fonseca, F., Martin, J., and Rodriguez, M.A.: From Geo- to Eco-ontologies. In: Egenhofer, M.J. and Mark, D.M. (Eds): GIScience 2002, Lecture Notes in Computer Science Vol. 2478. Springer-Verlag, Berlin Heidelberg (2002) 93-107. Fonseca, F. and Martin, J.: Space and Time in Eco-Ontologies. AI Communications 17 (2004) 259-269. Keet, C.M.: Factors Affecting Ontology Development in Ecology. Proceedings of Data Integration in Life Sciences (DILS2005), 20-22 July 2005 San Diego, Lecture Notes in Computer Science Vol. 3615. Springer-Verlag, Berlin Heidelberg New York (2005) 46-62. Chase, J.M. and Leibold, M.A.: Ecological Niches: Linking Classical and Contemporary Approaches. University of Chicago Press: Chicago and London (2003). Grinnell, J.: The Niche-Relationships of the California Thrasher. Auk 34 (1917) 427-433. Elton, C.: Animal Ecology. London: Sidgwick and Jackson (1927). Hutchinson, G.E.: Concluding remarks. Cold Springs Harbor Symp. Quant. Biol. 22 (1957) 415-427. Downs, R.M. and Stea, D.: Cognitive Maps and Spatial Behavior: Process and Products. In: Downs, R.M. and Stea, D. (Eds.), Image and Environment. Chicago: Aldine (1973) 8-26. Madin, J. et al.: OBOE: An Observation Ontology, Manuscript in preparation. Bowers, S and Ludäscher, B.: Towards a generic framework for semantic registration of scientific data. In Proceedings of the Workshop on Semantic Web Technologies for Searching and Retrieving Scientific Data, CEUR Workshop Proceedings, volume 83, (2003). Bowers S and Ludäscher, B.: An ontology-driven framework for data transformation in scientific workflows. In Proceedings of the International Workshop on Data Integration in the Life Sciences (DILS), Springer, LNCS Vol. 2994 (2004). Bowers S and Ludäscher, B.: Towards Automatic Generation of Semantic Types in Scientific Workflows. In Proceedings of the International Workshop on Scalable Semantic Web Knowledge Base Systems (SSWS), WISE 2005 Workshop Proceedings, LNCS, volume 3807 (2005) 207-216. Ludäscher B., Lin, K., Bowers, S., Jaeger-Frank, E., Brodaric, B., and Baru, C.: Managing Scientific Data: From Data Integration to Scientific Workflows In: GSA Today, Special Issue on Geoinformatics (2006). Rittel, H., and Webber, M.: Dilemmas in a General Theory of Planning. Policy Sciences, Vol. 4. Elsevier Scientific Publishing Company, In., Amsterdam (1973) 155-169.