Eliciting and integrating expert knowledge for wildlife habitat modelling
Introduction
Wildlife habitat managers require detailed information pertaining to the distribution and abundance of species to help understand their ecology. Such information also may be used to develop spatially-explicit wildlife habitat models. These models can be used to determine habitat preservation priorities, identify potential risks to the species, understand the implications of different land management practices, or identify sites for the reintroduction of an endangered species (Stoms et al., 1992). Wildlife habitat models are commonly developed with the assistance of a geographical information system (GIS). This allows managers to examine spatially the interactions and distributions of a species and its environment.
Habitat may be characterised by a description of the environmental features that are important in determining the distribution and abundance of a species (Burgman and Lindenmayer, 1998). Burgman and Lindenmayer (1998) add that such descriptions are often based on field experience and non-quantifiable human perceptions. One widely used method for these descriptions is habitat suitability index (HSI) modelling. HSIs are based on the habitat evaluation procedure first developed by the United States Fish and Wildlife Service in the early 1980s (USFWS, 1980). The HSI for a given species and area of land represents a conceptual model that relates each measurable variable of the environment to the suitability of a site for the species (USFWS, 1996, Burgman et al., 2001). The indices are scaled from 0 (for unsuitable habitat) to 1 (for optimum conditions). Each environmental variable is represented by a single suitability index (SI), and when combined, these constitute a HSI that expresses the suitability of particular habitat for the species.
The SIs are generally linked either by multiplicative or additive functions. When the value of land as habitat depends on the presence of all variables (i.e. they do not compensate for one another), the geometric mean of their values may represent site suitability best. If the environmental variables are compensatory (one component may substitute for another), an arithmetic mean may be more appropriate (Burgman et al., 2001). Furthermore, factors may be assigned weights reflecting the relative importance of different components of the habitat (Burgman and Lindenmayer, 1998). The construction of HSIs is essentially a process for making a descriptive synthesis of information of the biology and life history of a species. This is based on a combination of the available data together with expert opinion on the species’ biology (Burgman and Lindenmayer, 1998).
Expert knowledge is an important resource that may improve the reliability of modelling (Dzeroski et al., 1997, Venterink and Wassen, 1997, Hackett and Vamnclay, 1998, Horst et al., 1998, Moltgen et al., 1999). It is particularly valuable where no systematic field investigations have been conducted. However, there remains uncertainty regarding its reliability (Fraser and Hodgson, 1995, Horst et al., 1998, Maddock and Samways, 2000). Radeloff et al. (1999) comment that the incorporation of location-specific knowledge of field biologists is a key step to improving GIS wildlife models, and thus improving wildlife management. A GIS may be used to achieve this and help obtain spatially-explicit habitat information from experts (Alonso and Norman, 1998, Haslett et al., 1990). This can often be achieved in real-time, which minimises the amount of data entry that is required with a large cohort of experts. In addition, the GIS provides experts with a spatial context when providing data through the inclusion of other data layers such as digital elevation models, road networks or vegetation distributions. The GIS effectively provides a virtual environment for experts that are already very familiar with what exists in the realm of the study area. The advantages of using interactive computer-based techniques such as GIS for acquiring, archiving and analysing expert knowledge are discussed in detail in Wightmann (1995) and Zhu (1999).
Sambar deer (Cervis unicolor), commonly referred to as sambar, were introduced to Australia in the mid 19th century. The distribution of the species in the state of Victoria is poorly understood. Boyle (1998) suggested that at least 40,000 individuals occur in Victoria, and that the range of sambar is still expanding despite an increase in hunting pressure. The impacts of sambar on indigenous flora and fauna are largely unknown, and information pertaining to the basic biological, ecological and behavioural aspects of sambar in Australia is lacking. Furthermore, management decisions pertaining to sambar are complicated by several factors including the uncertainty about the current abundance of the species, and to the difficulty of balancing hunting objectives with the broader community needs such as plant and animal conservation. Because park managers need to develop deer management policies, there remains an urgent need to collect ecological information on this species.
This paper introduces a methodology that utilises GIS and qualitative interviews to elicit information on the distribution and abundance of sambar from a group of experts with detailed knowledge of the study site. The aim is to use this information to build an accurate wildlife habitat model to assist with the management of the species. First, the paper presents a methodology for structured elicitation of knowledge that combines both quantitative (GIS) and qualitative interview techniques. The paper then examines the agreement between experts regarding the distribution and abundance of the species. This information is synthesised and combined with other data layers in the GIS to develop the habitat model. The research was conducted under conditions that are common to many management scenarios where there is little documented information on the biology of the species or its habitat.
Section snippets
Study site and species management
Lake Eildon National Park (LENP) is situated in the northern foothills of Victoria’s Central Highlands, 90 km northeast of Melbourne (Fig. 1). The Park has a total area of 27,750 ha, consisting of open country, woodland and rugged forested ridges (Parks Victoria, 1997) with an extensive creek system running through the Park. The Park supports a range of rare and endangered fauna, and is managed primarily for ecosystem conservation and appropriate recreation (Parks Victoria, 1997). LENP is valued
Spatial database development
Because spatial context is critical for building habitat models, GIS data layers for LENP were obtained or developed. Multi-scale data layers were used for the study including 1:100,000 scale hydrology (rivers, creeks and lake boundaries) and climate surfaces from the
Species–environment relationships
There was substantial variation in the number of observations provided by each expert, ranging from 13 to 323. This variation may be a function of differential familiarity with the study site or with GIS technology.
For the variables examined, the box-plots did not identify any particular environmental niche for sambar within the Park. Representative results are shown in Fig. 3. For several variables, including elevation and distance to lakes and major rivers, there were some inconsistencies
The habitat suitability model
Sambar is a species that can utilise habitats where some environmental conditions are sub-optional (Moore, 1994). Therefore, the arithmetic mean of suitability indices of the three variables is appropriate for development of the final HSI. Column 4 in Table 1 shows the SI functions, which transform each environmental variable to an index ranging from 0 to 1. The importance of gullies was emphasised by all the experts in addition to the supporting literature, so a weight was assigned to this
Knowledge elicitation from experts-observations from GIS interaction
The sambar experts participated in the knowledge elicitation process with a great deal of interest and enthusiasm. This interest and enthusiasm was particularly evident when they provided responses to the qualitative questions. Some limitations hindered the information elicitation process in the quantitative phase of the study. Experts were commonly unfamiliar with computers or GIS software, and hence there was some reluctance when inputting deer observations. For instance, a number of experts
Conclusions and recommendations
This research has presented and evaluated a quantitative GIS-based and a qualitative interview-based technique for eliciting knowledge from wildlife experts. The aim was to identify suitable sambar habitat to build a habitat map to assist with the management of the species in the Park. The research also sought to answer questions of knowledge reliability as expert knowledge is being increasingly used for natural resource management. This was achieved by analysing the degree of inter-expert
Acknowledgements
We are greatly indebted to Sally Troy and John Wright of Parks Victoria for providing the resources that made the research possible. Special thanks goes to Brian Boyle from Parks Victoria and the anonymous interview participants for sharing their time, knowledge and enthusiasm. We would also like to thank Mark Burgman, Prema Lucas and Kimberly Mueller for their support of this research.
References (47)
- et al.
Apparency of contingencies in single panel and pull-down menus
Int. J. Hum. Comput. Stud.
(1998) - et al.
Modelling habitat suitability for black-tailed deer (Odocoileus hemionus columbianus) in heterogeneous landscapes
For. Ecol. Manage.
(1996) - et al.
Using machine learning techniques in the construction of models. II. Data analysis with rule induction
Ecol. Model.
(1997) - et al.
Mobilizing expert knowledge of tree growth with the PLANTGRO and INFER systems
Ecol. Model.
(1998) - et al.
Introduction of contagious animal diseases into The Netherlands: elicitation of expert opinions
Livestock Prod. Sci.
(1998) - et al.
Habitat and population modelling of roe deer using an interactive geographic information system
Ecol. Model.
(1999) - et al.
A comparison of six models predicting vegetation response to hydrological habitat change
Ecol. Model.
(1997) - Animal Health Australia, 2001. The National Animal Health Information System (NAHIS): bovine malignant catarrh. In:...
- Bentley, A., 1998. An introduction to the deer of Australia, 3rd ed. The Australian Deer Research Foundation Ltd.,...
- Bovill, L., 2000. Bunyip Deer Study: the presence, movement and environmental impact of deer within Bunyip State Park....
Setting reliability bounds on habitat suitability indices
Ecol. Appl.
Lay foibles and expert fables in judgements about risk
The American Statistician
Outline of an environmental information system
Environmental monitoring and assessment
Population dynamics of Sambar Cervus unicolor, in Periyar Tiger Reserve
Indian For.
Cited by (91)
A fuzzy knowledge-based model for assessing risk of pesticides into the air in cropping systems
2022, Science of the Total EnvironmentInterweaving local, expert, and Indigenous knowledge into quantitative wildlife analyses: A systematic review
2022, Biological ConservationPredicting wildlife distribution patterns in New England USA with expert elicitation techniques
2020, Global Ecology and ConservationRoad Permeability Index: Evaluating the heterogeneous permeability of roads for wildlife crossing
2019, Ecological IndicatorsCitation Excerpt :RPI is a replicable, transparent and highly adaptable method that allows users, interest groups, decision makers and land managers to adapt the method for any region or focal fauna taxa worldwide. The RPI can help both identify road kill hotspots (Teixeira et al., 2013b) where no data are available and evaluate the chance of each taxon crossing any road position, based on qualitative expert opinion (Yamada et al., 2003). It also furthers our understanding of the individual and combined contributions of road infrastructure and landscape features to road permeability.