A GIS-based multi-scale approach ...
Ecological Modelling 169 (2003) 1���15 A GIS-based multi-scale approach to habitat suitability modeling Ron Store a,���, Jukka Jokim��ki b,1 a Finnish Forest Research Institute, Kannus Research Station, P.O. Box 44, FIN-69101 Kannus, Finland b Finnish Forest Research Institute, Rovaniemi Research Station, P.O. Box 16, FIN-96301 Rovaniemi, Finland Received 18 March 2002 received in revised form 22 April 2003 accepted 12 May 2003 Abstract The aim of this study is to develop a method by means of which it is possible to produce georeferenced ecological information about the habitat requirements of different species. The integrated habitat suitability index approach includes the steps of constructing habitat suitability models, producing data needed in models, evaluating of target areas based on habitat factors, and combining various suitability indices. The method relies on the combined use of empirical evaluation models and models based on expertise in geographical information system (GIS) environment. GIS was used to produce the data needed in the models, and as a platform to execute the models and to present the results of the analysis. Furthermore, multi-criteria evaluation methods (MCEs) provide the technical tools for modeling the expertise and for connecting (standardizing, weighting, and combining) the habitat needs of different species. The main advantages of the method were connected to possibilities to consider the habitat factors on different scales, to combine habitat suitability evaluations for several species and to weight different species in different ways, and to integrate empirical models and expert knowledge. The method is illustrated by a case study in which an integrated habitat suitability map is produced for a group of old-forest species. �� 2003 Elsevier B.V. All rights reserved. Keywords: GIS Habitat suitability evaluation Multi-scale Cartographic modeling 1. Introduction Ecological research currently produces plenty of knowledge about the habitat requirements of the var- ious species. Abundance and habitat suitability mod- els, for instance, have been produced for many species (e.g. ��zesmi and Mitsch, 1997 Elmberg and Edenius, 1999 Radeloff et al., 1999 Whigham, 2000). For this ��� Corresponding author. Tel.: +358-102113423 fax: +358-102113401. E-mail address: email@example.com (R. Store). 1 Present address: Arctic Centre, University of Lapland, P.O. Box 122, FIN-96101 Rovaniemi, Finland. knowledge to be efficiently utilized, one needs meth- ods and channels capable of bringing the knowledge into the field of practical forestry and nature conser- vation. Ecological knowledge is needed especially to evaluate different forest management alternatives according to their consequences to biodiversity con- servation and to evaluate the importance of alternative locations to be set aside from wood production. In Finland, a high proportion of commercial forests is within the sphere of woodlot-specific forest planning, which enables the use of forest planning as a link between ecological knowledge and practical forestry. One way of protecting threatened forest species that are difficult to locate and identify is to protect their 0304-3800/$ ��� see front matter �� 2003 Elsevier B.V. All rights reserved. doi:10.1016/S0304-3800(03)00203-5
2 R. Store, J. Jokim��ki / Ecological Modelling 169 (2003) 1���15 habitats from activities likely to radically alter them. Often this is a question of a small area of relatively little value from the commercial forestry. There is an urgent need to develop methods capable of locating and evaluating suitable sites for threatened species. A special challenge is to develop methods and practices suitable also for non-industrial, private forest holdings. From the viewpoint of forest management planning, it is essential that the decision alternatives be assessed with respect to each of the objectives set for the for- est and its use. An evaluation consisting of only a part of the objectives produces alternatives, which are not necessarily effective in terms of the combination of objectives the forest owner have. Requirements to in- crease efficacy also in the area of nature protection have brought about demands for methods enabling the evaluation of alternative areas in appropriate ways in compliance with a set of nature-related objectives. In the case of ecological objectives, this means that evalu- ation has to be done focusing on all the species and im- pacts we are interested in. The purpose of this course of action is to find the optimal solution, not only for one species but also for a certain group of species. Because the objectives set for the forest and its use vary according to the forest holding and forest owner, it is improbable to have a production model for every possible situation. In the case of nature conservation, this means that empirical evaluation models based on real field data for all species of interest cannot be ex- pected to become available. One way of dealing with this problem is to use expert knowledge to substitute empirical habitat models. This could be the best al- ternative with rare species, for example, when empir- ical evaluation models are not available. Methods and techniques for utilizing expert knowledge in the han- dling of natural resources have recently been devel- oped (e.g. Kangas et al., 1993, 2000 Alho et al., 1996 Alho and Kangas, 1997). Combing the use of empiri- cal data and expert knowledge in ecological modeling can offer possibilities to produce suitability maps for larger sets of species. The importance of a specific habitat structure for habitat selection has been demonstrated in many studies (e.g. Hild��n, 1965 Cody, 1981). As the subdi- vision of natural habitats has increased, many recent studies have pointed out the importance of the land- scape matrix in animal population changes, reserve planning, and management practices (Edenius and Elmberg, 1996 Jokim��ki and Huhta, 1996 Saab, 1999 Howell et al., 2000). For many species, habitat requirements are related both to the structure of the habitat and to the landscape surrounding the habitat (landscape matrix) (e.g. Jokim��ki and Huhta, 1996). For these species, habitat size and shape also are needed to be taken into account along with the spatial pattern of their occurrence in the landscape (Virkkala, 1991 Saab, 1999). Recent studies have also shown that the multi-scale approach is needed also because habitat suitability, e.g. in bird species, is related to different factors on different spatial scales (Wiens, 1989 Jokim��ki and Huhta, 1996). When evaluating habitat suitability, the need to use data from different sources and scales usually makes the task more complicated and leads to increased data volumes. Geographical Information System (GIS) ap- plication have been adopted in ecological modeling as tools for producing the data needed in modeling on different spatial and temporal scales (Barnes and Mallik, 1997 Garcia and Armbruster, 1997 Radeloff et al., 1999 Wu and Smeins, 2000), as platforms on which models are run and data stored (Brown et al., 1994 Ripple et al., 1997 ��zesmi and Mitsch, 1997 Hirzel et al., 2001), and as tools for extrapolating the results from point basis to spatial basis (Littleboy et al., 1996 Osborne et al., 2001). In recent years, also many new techniques, e.g. artificial neural net- works (��zesmi and ��zesmi, 1999), genetic program- ming (Whigham, 2000), and machine learning (Kobler and Adamic, 2000), have been connected with GIS to produce ecological knowledge and models. Spatial modeling, especially cartographic modeling, which is a process of combining maps by linking sev- eral map-algebra statements together, has been applied in locating areas simultaneously fulfilling all the con- ditions set (Bonham-Carter, 1994). Cartographic mod- eling in the association of nature conservation can be used, for example to pinpoint suitable habitats for cer- tain species (Store and Kangas, 2001). Multi-criteria evaluation methods (MCEs) have been used with cartographic modeling techniques to provide a basis for evaluating a number of alternative choices on the grounds of multiple criteria (Nijkamp et al., 1990). While GIS includes the tools for man- aging and producing the georeferenced information in different scales needed (e.g. in the habitat suit- ability evaluation), MCE methods provide technical
R. Store, J. Jokim��ki / Ecological Modelling 169 (2003) 1���15 3 tools for modeling the expertise and connecting the habitat needs of different species. Features typical to the process of evaluation natural resources, such as multi-objectivity, scale dependence, and the need to model expertise, set additional requirements for the tools used in the process. Store and Kangas (2001) presented an approach for integrating GIS and state-of-the-art decision analysis techniques to habitat suitability evaluation. Their work concentrated on using expert knowledge and continu- ous decision variables together with GIS tools to im- prove habitat suitability modeling. The purpose of this study is to extend the presented framework and de- velop a method by means of which it is possible to produce georeferenced ecological information about the habitat requirements of different species, such data being needed in functions such as forest management planning. The main requirements set on the method are as follows: the capability to handle expert knowledge together with empirical evaluation models in the eval- uation process, the possibility to examine habitat fac- tors on appropriate geographic scales and to evaluate the choice alternatives on the basis of several species. The emphasis in this study is on method development and this is why less effort has been directed at pro- ducing and evaluating separate suitability models. The method is illustrated by a case study in which a couple of integrated habitat suitability maps were produced for a group of old-forest species. 2. Multi-scale and multi-species method 2.1. Basic steps in the method In this study, species habitat requirements are described by habitat factors, which cover the most essential habitat characteristics of preferred habitats. Habitat factors are connected either to the local habi- tat its vegetation and soil properties, for instance, or to the area surrounding the habitat the properties of the landscape, e.g. the number of different kinds of habitat types. Habitat suitability is measured by means of a suitability index, which is a unitless variable describ- ing habitat priority with respect to the needs of the species or the group of species under consideration. Habitat suitability modeling method applied in this study composed of the following steps. 1. Constructing habitat suitability models 2. Producing the data needed in models 3. Evaluating a target area based on habitat factors 4. Combining the separate suitability indices The abundances and habitat suitability models of the species used in this study were constructed using both field observatories and expert knowledge. 2.2. Constructing the models Empirical models for the habitat selection patterns are based on the investigation of the relationship be- tween the collected occurrence (or abundance) data and the appropriate background variables. Logistic re- gression analysis is applied in many cases when only presence���absence data are collected, whereas step- wise multiple regression analysis is normally applied when abundance data are available (e.g. Jokim��ki and Huhta, 1996). There are many possible standard methods (e.g. line-transects, point-count surveys, and mapping methods) for collecting abundance data. Line-transects and point counts are single-visit sur- veys, whereas in mapping methods with several visits are made to every study site. The method selected will be dependent on the timetable limits, scale of surveys, budget, etc. The selection of appropriate background variables depends on the species studied, but also the measuring costs of the variables and the purpose of use of the empirical model which have their own im- pact on the choice of the final background variables. If the aim is to use the model in forest management planning, the variables chosen for the model have to be of such kind that they are also measured in the in- ventory associated with forest management planning or such that they can be calculated in the course of the planning process. In the case of species lacking objective models based on empirical field observations, a habitat suit- ability model is constructed on the basis of expert knowledge (e.g. Kangas et al., 1993). In the first phase, an expert of the species in question determines the foremost habitat factors for the species. While all essential habitat factors have to be included, if the number of factors is too high, it leads to a complicated model, which is difficult to use in practice. In the second phase, the relative importance of the factors is evaluated. A great number of methods can