Habitat requirements of species were described as map layers within GIS. GIS was used as the platform in managing, combining and displaying the criterion data. Habitat suitability maps were produced for an old-forest polypore, Skeletocutis odora.
Habitat requirements of species were described as map layers within GIS. GIS was used as the platform in managing, combining and displaying the criterion data. Habitat suitability maps were produced for an old-forest polypore, Skeletocutis odora.
Habitat requirements of species were described as map layers within GIS. GIS was used as the platform in managing, combining and displaying the criterion data. Habitat suitability maps were produced for an old-forest polypore, Skeletocutis odora.
Integrating spatial multi-criteria evaluation and expert knowledge
for GIS-based habitat suitability modelling
Ron Store * , Jyrki Kangas Finnish Forest Research Institute, Kannus Research Station, P.O. Box 44, Fin-69101 Kannus, Finland Received 4 April 2000; received in revised form 23 October 2000; accepted 31 January 2001 Abstract GIS data processing and spatial analysis, together with modern decision analysis techniques, were used in this study to improve habitat suitability evaluation over large areas. Both empirical evaluation models and models based on expert knowledge can be applied in this approach. The habitat requirements of species were described as map layers within GIS so that each map layer represented one criterion. GIS was used as the platform in managing, combining and displaying the criterion data and also as a tool for producing new data, especially by utilising spatial analysis functions. Criterion standardisation, weighting and combining were accomplished by means of multi-criteria evaluation (MCE) methods, the theoretical background being based on the multi-attribute utility theory (MAUT). By using continuous priority and sub-priority functions in the evaluation, no classication of continuous attributes was needed and also non-linear relationships between habitat suitability and the attributes could be considered. Sensitivity analysis was applied to consider the temporal factor in the analysis and to nd out the effect of different criteria weights on the spatial pattern of the suitability index. Changing the weights of permanent and time-changeable habitat factors shifted the location of optimal habitats for the species. In the long run, permanent factors such as soil properties dene the habitat potential, which is important to take into consideration; e.g. in forest management planning and species conservation. The method is illustrated by a case study in which habitat suitability maps were produced for an old-forest polypore, Skeletocutis odora. #2001 Elsevier Science B.V. All rights reserved. Keywords: Habitat suitability modelling; Multi-criteria evaluation; Geographical information systems; Expert knowledge; Sensitivity analysis; Skeletocutis odora 1. Introduction Nowadays, the objectives set for forest use are more diverse than in the past. Values such as recreation, landscape beauty and conservation of biodiversity have gained prominence alongside traditional wood- production values. In wildlife management, protection of critical habitats and conservation of endangered species are the foremost tasks from the point of view of biodiversity preservation. When performing these tasks, it is important to know the factors affecting habitat suitability and habitat selection, and also to have methods for determining the suitability of an area for certain species. Habitat suitability modelling is often used to produce probability maps depicting the likelihood of occurrence of certain species and to nd out Landscape and Urban Planning 55 (2001) 7993 * Corresponding author. Tel.: 358-6-8743211; fax: 358-6-8743201. E-mail address: ron.store@metla.fi (R. Store). 0169-2046/01/$20.00 # 2001 Elsevier Science B.V. All rights reserved. PII: S0 1 6 9 - 2 0 4 6 ( 0 1 ) 0 0 1 2 0 - 7 the landscape properties of preferred habitats (e.g. Johnson and Temple, 1986; Pausas et al., 1995). These models are typically made by exploring statis- tically the relationship between existing occurrences of the species and the site properties. However, sta- tistical evaluation models must be based on adequate empirical study material, which is usually expensive and time-consuming to collect, especially in the case of rare species. Furthermore, the number of species we are interested in is continuously increasing. Thus, empirical evaluation models for all the species of interest cannot be expected to be available. One possibility for dealing with this problem is to use expert knowledge when models based on objective information are not available. In this approach, habitat suitability models, for example, are estimated on the basis of expert judgements instead of empirical mea- surement data. In order to enable the use of ecological expert knowledge in habitat suitability modelling, methods for transforming expert knowledge into a numerical form, as well as appropriate tools for hand- ling and producing data about environmental condi- tions, are needed. GIS applications have frequently been used in producing new information by both combining infor- mation from different sources and by spatial analysis of existing data. Spatial modelling, especially carto- graphic modelling, has been applied when looking for areas suitable for a specic forest use or for some species of interest (e.g. Shaw and Atkinson, 1988; Reisinger and Kennedy, 1990; Wadge et al., 1993). Usually the objective in applications involving cartographic modelling is to locate the area or areas where the given criteria apply. In a classication based on Boolean logic, an area is either accepted or rejected based on a given threshold value. The nal outcome of these applications is a map depict- ing areas simultaneously fullling all the conditions set. However, problems have been noted with methods for site selection and resource evaluation that rely on classical Boolean logic (Carver, 1991). In situations where the threshold value is not precise, loss of information or error propagation may occur. Further- more, the method does not offer any analytical pos- sibility for examining which of the areas fullling the criteria are the most appropriate for the purpose in question or which areas are the best beyond the feasible areas. Because of problems with Boolean overlay, multi-criteria evaluation (MCE) methods have been applied instead of Boolean logic (Carver, 1991; Pereira and Duckstein, 1993; Jankowski, 1995). The main aim in using MCE methods is to provide a basis for evaluating a number of alternative choice possibilities on the basis of multiple criteria (Nijkamp et al., 1990). Besides the problems associated with the use of Boolean logic, the accuracy of results depends on the quality of the source data and on the quality of the spatial data analysis used in the process. According to Burrough and McDonnell (1998), the most important factors affecting the quality of spatial data are cur- rency, completeness, consistency, accessibility, accu- racy and precision, and various error sources. Errors in the nal results may originate from any stage of the process, from the collection of the source data to the interpretation of the nal results. Data quality is often described by thematic accuracy, positional accuracy and temporal accuracy. Also error propagation, mean- ing the accumulation of errors from various sources, affects the results of analyses. One approach to MCE in a GIS environment is the additive technique whereby the criterion scores are standardised and the total score for each alternative is calculated by multiplying each criterion score by its weight factor and then adding the results. Weighted linear summation is probably the best-known example in this category (e.g. Berry, 1993). Additive techni- ques are based on the multi-attribute utility theory (MAUT), where the weights are interpreted in the framework of a linear priority (utility) function. Siddiqui et al. (1996) presented an additive approach to a spatial problem based on the analytic hierarchy process (AHP). AHP is one method of producing the criteria weights and also of making criteria of different kinds commensurable (Saaty, 1980). Siddiqui et al. (1996) called their method spatial-AHP and applied it to supporting landll sit- ing. It provides a way to exclude areas which are not suitable for a certain purpose and to rank the remain- ing areas based on area attributes. However, in this method, attribute values have to be classied using a limited number of classes. The problem in doing so is that the classication may lead to loss of information and, on the other hand, dening the threshold values of classes may increase uncertainties in situations where 80 R. Store, J. Kangas / Landscape and Urban Planning 55 (2001) 7993 there is no exact threshold value or the value is not known with certainty. Decision criteria, used in habitat suitability eva- luation, are mainly continuous ones. This is why classication of the data describing the alternatives may lead to increasing uncertainty and loss of infor- mation. This problem can be avoided by utilising continuous priority functions in the evaluation. Then, no classication is needed in the evaluation of habi- tats. In the present study, the evaluation technique utilised in the HERO heuristic optimisation (Pukkala and Kangas, 1993) is applied. It enables the evalua- tion of habitats with respect to continuous attributes by applying so-called sub-priority functions without any classication of attribute values. Also, non-linear relationships between habitat suitability and the attributes can be considered using the sub-priority functions. Suitability indices have been utilised, for example, in various aspects of natural resources planning (e.g. Pukkala et al., 1995). In forest management planning, decision alternatives are considered in a given area on the basis of their consequences within the planning horizon. For evaluating the alternatives, they are assessed with respect to each objective set for the forest and its use. Planning always concerns the future. The objectives, in turn, can be related to the near future or to longer time horizons. When a suitability index is used in planning, it is, therefore, useful to incorporate the temporal factor in the estimation of the index values. In this study, a technique based on sensitivity analysis is presented for taking the tem- poral factor into consideration when producing habitat suitability indices. The aim of the present study is to improve habitat suitability evaluation by using GIS-based data proces- sing and spatial analysis along with state-of-the-art decision-analysis techniques. Important qualities of the habitat-evaluation method aimed at in this study are the abilities to utilise expert knowledge when evaluation models based on empirical studies are not available, and to apply continuous attributes in the evaluation in a reasonable way within a GIS framework. The method is illustrated by a case study in which a habitat-suitability map is produced for an old-forest polypore, Skeletocutis odora. The approach presented is appropriate for use with other species as well. 2. Methods 2.1. The general outline All species have specic habitat requirements, which can be described by habitat factors. These factors are connected to the critical characteristics of the habitat, e.g. to those of vegetation or soil, but also areas surrounding the habitat can inuence the habitat quality (e.g., spatial structure of landscape elements). Habitat factors can also be classied according to the deterministity of the factor: a deter- ministic habitat factor has to be present in a high- quality habitat, but a non-deterministic factor has a trade-off with some other factor. So, a deterministic factor can be taken as a non-compensatory habitat characteristic whereas a decrease in a non-determi- nistic factor can be compensated by an increase in the value of another non-deterministic factor, as expressed in the habitat-evaluation model. Habitat suitability is determined by habitat factors. Habitat suitability can be measured by a habitat suitability index, which is a unitless variable describing the priority of the habitat with respect to the needs of the species (or group of species) under consideration. Typically, it can get values between 0 and 1, and is estimated on grounds of the measur- able habitat characteristics. For producing habitat suitability indices for large areas, methods enabling the management and analysis of large amounts of data are needed as well as the calculation para- meters describing the most essential habitat character- istics. The habitat suitability modelling method applied in this study consists of two basic phases. Firstly, deter- ministic habitat factors are used to nd the potential habitats fullling the absolute prerequisites for the species in concern within the area. The result of this is a feasible area meeting the absolutely necessary con- ditions for the occurrence of the species. Secondly, non-deterministic factors are used to evaluate the potential habitats. The evaluation procedure consists of the following steps: 1. The assessment of a suitability structure: choosing the habitat factors and determining their impor- tance and how they affect the habitat priority. R. Store, J. Kangas / Landscape and Urban Planning 55 (2001) 7993 81 2. Producing map layers: raw data acquisition and transforming to appropriate GIS layers. 3. Cartographic modelling: dening the feasible area and combining the habitat factors. 4. Sensitivity analysis: demonstrating the effect of different criterion weights on the spatial pattern of the suitability index. Step (1) is carried out using an external MCE program which works outside a GIS while steps (2)(4) are executed under a raster GIS. 2.2. Assessment of the suitability structure The rst step in assessing the suitability structure is to determine the habitat factors on the basis of an analysis of existing studies and knowledge. Here, judgements made by experts on ecology can be applied. The next step is to classify the decision factors into deterministic and non-deterministic ones according to their exclusive character. Deterministic habitat factors are Boolean type of variables, which are used like geographical constrains to eliminate areas from further consideration on the basis of certain attribute values. In practice, very few habitat factors can be considered so straightforwardly that a simple Boolean variable is appropriate for describing it. Next, the mutual importance of non-deterministic habitat factors have to be evaluated. This can be done by a number of methods. In this study, the HERO method (Pukkala and Kangas, 1993) was adopted. By its very nature, HERO is a heuristic optimisation method including both the estimation of the priority model and the procedure for maximis- ing the value of the model among the decision alter- natives. It was originally developed for the purposes of tactical forest planning when dealing with a huge number of alternative forest plans, each consisting of a combination of treatment schedules for the forest stands within the area under planning. In this study, only the priority model estimation option of HERO is applied. Instead of forestry objectives, the priority model here is estimated for habitat-evaluation purposes only. Thus, it includes habitat factors as variables. The resulting model can be utilised directly in forestry planning calculations. Combining the habitat factors is based on the multi-attribute utility theory (MAUT) and accomplished by an additive priority function P: P X m i1 a i p i q i (1) where P is the global priority, i.e. the habitat suitability index, m the number of factors, a i the relative impor- tance of factor i ( P a i 1), p i the sub-priority func- tion of factor i, the maximum value for each factor is one, and q i the amount of factor i. The relative importances of the habitat factors are used as coefcients a i in the priority function. The relative priority of the habitats with respect to each factor is estimated on a ratio scale with a sub-priority function. The sub-priority function depicts the change in habitat suitability as a function of the factor. The sub-priority functions scale all factors between 0 and 1, thereby making the objectives commensurable (the best priority value is always 1, other attribute values getting sub-priorities relative to it). Sub-priority func- tions are developed separately for each habitat factor and the total suitability index is the weighted sum of the sub-priorities. There are several alternative methods for estimating the coefcients a i . In this study, the coefcients were solved applying pairwise comparisons carried out by an expert on ecology. The relative importances of the factors are computed using the eigenvalue method of ratio scale estimation (Saaty, 1977). The importances of the factors are compared pairwise using a graphical interface instead of the verbal scale as proposed by Saaty (1980). The relative importances of two factors at a time are dened by adjusting the lengths of the horizontal bars on the computer screen. The method enables the presentation of a habitat factor in a hierarchical manner. A factor may be described by means of a model whose variables are the components describing the factor in more detail and whose coefcients are the components' relative importance. This being the case, the sub-priority functions are dened to depict the impact of the detailed components on the priority obtained through the habitat factors they explain. When estimating the sub-priority function, the maximum and minimum values to be considered in the evaluation are determined rst and displayed by the user interface. In addition, a few intermediate values are selected. The desirability of these values 82 R. Store, J. Kangas / Landscape and Urban Planning 55 (2001) 7993 is then estimated by means of pairwise comparisons and the values are given relative priorities dening the sub-priority function (Fig. 1). The sub-priorities in- between the compared values are solved by linear interpolation. Although a piece-wise linear function is applied, a sub-priority function can show a non-linear relationships between habitat suitability and the envir- onmental variables. The sub-priority function is esti- mated separately for each habitat factor, or for its components. Generally taken, the estimation can be based equally on expertise or subjective value infor- mation. The ratios of the values of objectively mea- surable variables can also be used. 2.3. Producing map layers A GIS application is used for managing, producing, analysing and combining spatial data. Some of the attribute data needed in the suitability-evaluation pro- cess are collected by means of eld inventories and the rest are produced from collected or existing data by using different kinds of spatial functions and analysis. For example, soil moisture is described by the topo- graphic wetness index T i . It is a function of the upstream contributing area and the slope of the land- scape. It is calculated by means of formula (2) (used, e.g. in the TopModel program, Beven et al., 1995). The index can be interpreted as the spatial distribution of soil moisture or as the relative depth of water table. When the catchment area increases and gradient decreases, the topographic index and soil moisture content increase. T i ln a tan B (2) where a is the area draining through a grid square, and tan B the average outow gradient from the square. The data describing the habitat factors are rasterised into 25 m 25 m grid cells and every habitat factor is stored in its own map layer. 2.4. Cartographic modelling Cartographic modelling is applied in producing and combining spatial data describing the habitat factors. Cartographic modelling is used as a means to deter- mine the feasible area, standardisation and transform- ing raw scores for priority measurements, and nally to calculate a suitability index as a combination of single habitat factors. In the rst phase, the feasible area is produced by numerically overlaying a map layer describing the study area and all the map layers considering the deterministic habitat factors. This overlay is carried out as a Boolean overlay, where areas not included in the area determined as being suitable by all the deterministic habitat factors are excluded from further consideration. The remaining areas establish the feasible area. The map layers considering the attribute data used in the MCE process may differ from each other as regards the measurement units. To make the raw scores commensurable, some kind of standardisation is needed. Voogd (1983) presented various kinds of standardisation methods to transform raw scores into scores with ranges of variation 0 to 1. The problem with the most commonly used standardisation meth- ods is that they merely perform a linear transformation between the raw score and the standardised score. The linearity approach does not enable meaningful com- bination of standardised criteria scores in situations where the best value is either not the greatest value or the smallest value, or where the relationship between the raw score and priority is not linear. To avoid those problems, the priority model as applied in HERO is used as a standardisation tool and for transforming the standardised scores into Fig. 1. An example of sub-priority functions indicating a non- linear relationship between habitat suitability and a habitat factor. R. Store, J. Kangas / Landscape and Urban Planning 55 (2001) 7993 83 priorities. In the priority model, the differences in measurement scales and units are not a problem, because the method is based on direct comparisons between signicance or priority of the elements in the model. Non-linearities between raw scores and prior- ity are handled by piece-wisely linear sub-priority functions. In order to technically enable the calculation of sub- priorities and global priority (i.e. the habitat suitability index) within the GIS application, where each habitat factor is its own map layer and the data is rasterised into grids, the following procedure is performed. First, the raw scores in the map layers, describing the non- deterministic habitat factors, are divided into groups according to the ``corner points'' of the sub-priority functions (see Fig. 1). The scores between two ``corner points'', i.e. between the ones compared in the esti- mation of the corresponding sub-priority function, belong to the same group. Thus, the sub-priority values for them can be calculated by the same inter- polation formula. A separate interpolation formula is constructed for each `piece' of a piece-wisely linear function. Then the scores are dealt with group by group to compute the sub-priority produced by a certain value of a habitat factor. The sub-priority value is estimated for each cell this way, resulting in so- called sub-priority grids. Finally, the sub-priority grids are combined by multiplying each sub-priority grid by its weight coefcient a i and by summing the weighted sub-priority grids by the means of arithmetical overlay analysis. The end result is the habitat suitability index for each 25 m 25 m cell. 2.5. Sensitivity analysis Various weighting schemes are applied to the main criteria when using sensitivity analysis. The main purpose here is to examine how sensitive the choices are to the changes in criteria weights. This is useful in situations such as where uncertainties exist in the denition of the importance of different habitat fac- tors. In many cases, it is important also to know how the results will change if the weights are changed. If there are requirements in existence for a certain kind of sensitivity analysis, these can be taken into consideration already at the stage of building the decision hierarchy. For example, criteria based on facts can be classied under one main criterion and those based on changeable preferences under another. Then it is possible to change the weights between the facts and the preference criteria, or one can simply change the weights of different preference criteria. An other example is to classify criteria under permanent factors (e.g. soil characteristics) and time- changeable factors (e.g. vegetation). Permanent factors dene the habitat potential in the long run. Time-changeable factors, together with the permanent ones, dene momentary habitat suitability. The general purpose of sensitivity analysis in this study was to nd out the inuence of different criteria weights on the spatial pattern of the suitability index. In order to investigate the effect of the time horizon in examining the location of the most suitable areas, sensitivity analysis was connected to the temporal factor. By means of this connection, it was possible to investigate how the most suitable area shifts, if suitability is related not only to the situation present at the moment, but more and more to those character- istics, which would not change in the long run. With this purpose in mind in the hierarchy-producing phase, the rst level factors were arranged according to time- permanent and time-changeable factors. 3. Case study 3.1. Study area and general course of case study The case study area, Kivalo forest estate, covers about 6870 ha and is located in Finnish Lapland (66823 0 N, 26823 0 E), about 20 km east of the town of Rovaniemi (Fig. 2). It is state-owned land adminis- tered by the Finnish Forest Research Institute. The difference in elevation within the area is 278 m, the highest point being 360 m above sea level and the lowest point 82 m above sea level. The study area was divided into 1833 compartments composed of relatively homogenous tree stands and having uniform soil characteristics. The dominant tree species is Scots pine (Pinus sylvestris), accounting for 44% of the standing volume. Norway spruce (Picea abies) and Silver Birch (Betula pendula) are the next common tree species, both accounting for 24% of the standing volume. The mean volume of growing stock in the area was only 50 m 3 /ha due to the harsh climate; the maximum volume recorded in the area was 84 R. Store, J. Kangas / Landscape and Urban Planning 55 (2001) 7993 262 m 3 /ha. The age class distribution was 43% <50 years, 18% 51100 years, and 39% >100 years. The soil fertility of the area is mainly medium, with some highly fertile (13%) and poor (15%) parts. The objective of the case study was to provide support for the choice of the most suitable old-forest areas for the polypore species S. odora. The process of assessing the ecological value of an old-forest area is often done by considering the threatened aphyllopor- ales and other old forest indicator species persisting in the area. S. odora is considered to be one of the most important boreal old-growth forest indicator species, e.g. in Finland and Sweden. In the examination of the ecological potential of the area, it was chosen as one of the key species to be considered in landscape ecolo- gical planning. According to a list approved by the Ministry of Environment in Finland, S. odora is classied as a threatened polypore falling into the threat class `rare'. Also in Sweden and Norway it has been found to be vulnerable and classied as a threatened species (Kotiranta and Niemela, 1996). S. odora is an old- growth forest species growing on Norway spruce. Several records have also been made of it growing on Aspen (Populus tremula). It requires a fertile mire- like habitat with a humid micro-climate and downed large-sized boles with the bark still attached (Kotiranta and Niemela, 1996). Scarceness of shady, old-growth forests with a humid micro-climate is the main reason for its rareness. 3.2. Estimation of habitat suitability indices 3.2.1. Data procurement The habitat quality required by S. odora was found to depend on both vegetation and soil characteristics (Fig. 3). Its vegetation requirements are related to the occurrence of a suitable host tree and a humid Fig. 2. Location of the case study area. Fig. 3. Habitat factors of S. odora, as applied in the case study. R. Store, J. Kangas / Landscape and Urban Planning 55 (2001) 7993 85 micro-climate. The following stand-specic mean characteristics were used in the case study: age of spruce, stem volume of spruce, diameter at breast height (dbh) of spruce, and the density of the growing stock. The habitat factors related to the soil's physical properties and topography were: slope direction (aspect), soil moisture and soil fertility. Suitable soil characteristics enable a humid micro-climate and also moist and fertile forest site types. One expert on polypores and another on soil sciences took part in the evaluation process. The expert on polypores chose the most important habi- tat-quality factors for S. odora and also assessed the importance of these factors. Furthermore, in the pro- cess of producing sub-priority functions, the experts' task was to assess the relative priority produced by the different criterion performance scores. The expert on soil science determined the factors inuencing soil moisture and the relative importance of these factors. In addition, the soil expert evaluated the consequences of different soil textures with respect to soil moisture in the study area. Most of the variables needed in the evaluation process were collected by means of ocular compart- ment inventory, and they were then stored into a database of a GIS application. Some of the evaluation criteria, such as the spatial distribution of soil moisture and slope direction (aspect), were elaborated by spa- tial analysis functions using GIS operations. In GIS, the collected and produced data were managed so that the data for every habitat factor were stored in separate map layers. The GIS used in the case study was Arc/ Info workstation (UNIX) version 7.2. Arc/Info is a commercial GIS software product developed by ESRI. The priority functions for habitat suitability were estimated on grounds of the expert judgements. The general priority function for habitat suitability is pre- sented in Table 1. The vegetation factors were con- sidered to be more important than the factors related to soil properties (0.7 versus 0.3). The factors directly related to the properties of the spruces were consid- ered to be clearly more important than general stand properties such as the density of growing stock (0.59 versus 0.11). Regarding soil properties, soil fertility was more important than the factors related to soil moisture (0.17 versus 0.13). The sub-priority functions (Fig. 4) of the habitat factors were estimated using a minimum, maximum and 23 intermediate values in-between them, and then comparing the values pairwise with respect to their sub-priorities. The sub-priority functions for stem volume and mean age of spruce indicated decreasing marginal sub-priority. However, the mean diameter of spruce at breast height and the density of the growing stock were assessed to have nearly linear sub-priority functions. The sub-priority functions for the mean diameter of spruce and the density of growing stock could also show decreasing marginal priority, if the maximum values recorded in the case study area were higher. It is also worth noting that, according to the judgements made by the expert, very high soil moisture values decrease habitat suitability in the case of S. odora. 3.2.2. Overlay analysis Two deterministic habitat factors were used in the combining phase to screen out unsuitable areas. Firstly, the area has to be forest land with the condi- tions sufcient also for tree growth as regards hydrol- ogy. Secondly, the area's site class should be fertile enough to enable a suitable micro-climate and appro- priate host trees with regard to tree species and size. Map layers describing these factors were overlaid numerically to build up a feasible area fullling both deterministic habitat factors simultaneously. The deterministic habitat factors used in this study were not very strict because the purpose was not to rule out any areas where the circumstances are or could, in the future, be suitable for S. odora. Of the 6870 ha of the case study area, 5444 ha were con- sidered to be suitable. Most of the areas deemed to be feasible were suitable according to their soil and Table 1 Habitat suitability criteria and their relative importance Criterion Relative importance Vegetation 0.7 Density of growing stock 0.11 The dbh of spruce 0.25 Stem volume of spruce 0.14 Age of spruce 0.20 Soil characteristics 0.3 Soil fertility 0.17 Slope aspect 0.04 Soil moisture 0.09 86 R. Store, J. Kangas / Landscape and Urban Planning 55 (2001) 7993 vegetation characteristics, but also areas the vegeta- tion of which could in the long run develop suitable were chosen. The value grids, describing the non-deterministic habitat factors, were divided into groups according to the ``corner points'' of the sub-priority functions. For example, the habitat factor `stem volume of spruce' was divided into the following groups: <60 m 3 /ha (I), 60120 m 3 /ha (II), >120 m 3 /ha (III). The correspond- ing linear parts of the piece-wisely linear sub-priority function could then be determined via formulas used in the interpolation, respectively: I y 0:0097x II y 0:0045x 0:31 III y 0:0025x 0:55 where x is the value of habitat factor, and y the sub- priority produced by value x. The habitat suitability index was calculated for each cell by multiplying each sub-priority grid by its weight coefcient and then summing the weighted sub-prio- rities by the means of an arithmetic overlay analysis. The index map resulting in the habitat suitability analysis for S. odora is presented in Fig. 5. In this basic calculation, where weights for vegetation factors and soil characteristics were 0.7 and 0.3, respectively, the maximum suitability index value recorded in study area was 0.83 and the minimum was 0.10 (scale 01). The mean value was 0.44. The areas in the middle part of the study area, located highest above the sea level within the area, seemed to be best suited for S. odora. Especially the north-facing hillsides in these areas got high index values. Also, the biggest continuous high index areas were located in this area while the high index areas in northern and western part of the study area were relatively small. The range of variation in the index value in the study area was high and, furthermore, high and poor index values were clearly clustered. Steep gradients follow the forest stand boundaries, because the coefcients of vegetation factors were high. 3.3. Sensitivity analysis Sensitivity analysis was used to present a technique enabling the temporal factor to be taken into consid- eration in the process of producing the suitability index and, on the other hand, to investigate how changing the weighting of various habitat factors affected the determination of the preferred areas. Special attention was paid to the weighting of time- permanent factors. Giving more weight to them means giving less weight to factors whose values change with time. Correspondingly, emphasising the importance of permanent factors means preferring the future habitat Fig. 4. The sub-priority functions estimated in the case study. R. Store, J. Kangas / Landscape and Urban Planning 55 (2001) 7993 87 potential over the present qualities of habitats. The habitat factors were divided into time-permanent and time-changeable factors for this analysis. The habitat factors connected to soil characteristics, such as fertility and moisture as well as topography, can be considered to be relatively constant. However, habitat factors connected to vegetation mostly change slowly through time, but also rapid and radical changes are possible. Usually the most extensive changes are due to the consequences of human activ- ities, e.g. clear cutting. The weighting scheme, when approaching the problem this way, reects the impor- tance of the present vegetation and the vegetation potential in the future (determined by soil character- istics). In practise, sensitivity analysis was accomplished by applying different weighting schemes for the two main decision criteria. In the basic computation the weight of vegetation (0.7) and the weight of soil characteristics (0.3) demonstrate the importance of the two main factors at the present moment according to opinion of the expert on polyspores. In addition to Fig. 5. The process of combining habitat factors to suitability index. 88 R. Store, J. Kangas / Landscape and Urban Planning 55 (2001) 7993 Fig. 6. Sensitivity analysis, where weights 0.5, 0.7 and 1.0 where applied for soil characteristics. R. Store, J. Kangas / Landscape and Urban Planning 55 (2001) 7993 89 basic calculation, weighting schemes 0.5 and 0.5, 0.3 and 0.7, 0.9 and 0.1, and 0.0 and 1.0 were applied for vegetation and soil characteristics, respectively. The results illustrate how changes in weighting affect the optimal choice of habitats to be conserved. Also, the alternatives where suitability maps are based either only on soil characteristics, or only on vegetation characteristics, were useful in identifying areas where vegetation preservation was less important because soil factors limit the site's suitability. For the purpose of sensitivity analysis, suitability maps for every weighting scheme were created in GIS. In suitability maps the pixels of the index maps were classied into ten categories according to suitability. The class ranges were determined so that the area covered by pixels was equal in each category. Sensi- tivity analysis was done by comparing these suitability maps. In the basic calculation, the most suitable areas were located in the middle parts of the study area. When the importance of soil characteristics was increased, the within-stand variation of the forest compartments also increased. When the importance of soil characteristics increased close to 1, then also the location of the most suitable areas shifted from the middle part's old high-elevation spruce forest to more fertile and moist, but also younger, spruce forest in the northern part of the study area (Fig. 6). This shows that, in the case of S. odora, the temporal factor had a major inuence on the results. 4. Discussion The possibilities of GIS were used in this study in combination with state-of-the-art decision-analysis methods to develop more exible habitat suitability analyses utilising expert knowledge. The theoretical background in the combining phase was provided by the multi-attribute utility theory (MAUT) and conti- nuos attributes and non-linearities were handled by means of sub-priority functions. A GIS was used as a platform enabling the management of the criterion data, production of criterion layers, calculation of attributes by means of spatial analysis, carrying out the combining of decision criteria by means of carto- graphic modelling, and nally conducting sensitivity analyses and production of the maps needed in the evaluation process. The major advantage of the method presented in the present study is in the possibility it offers for produ- cing suitability indices for large areas within a reason- able period of time, and also for species not having statistical suitability models based on empirical data available. This is enabled by using expert knowledge instead of evaluation models based on empirical data and on the other hand by utilising the data processing and production capabilities of GIS. The use of sub- priority functions is one way to handle the non-line- arities between the raw scores and the relative priority produced by these scores. With sub-priority functions it was also possible to deal with continuous variables without classifying them. In the discipline of landscape ecology, discussion has arisen about the nature of boundaries between different features in the landscape. In many applica- tions of landscape ecology, or more generally in the management of natural resources, the boundary is determined by means of crisp borders. However, in real-life landscapes, the boundary between two fea- tures is more often soft or fuzzy rather than crisp. In the case study, the data used were collected for the purpose of enabling forest management planning where the basic spatial entity is a homogenous forest compartment with crisp boundaries. This being the case, spatial patterns of suitability variables based on stand characteristics, such as stem volume and density of the growing stock, follow crisp stand boundaries. However, the data produced by the digital elevation model, such as aspect and wetness index, can be considered as being a continuous surface and they can be discretised into regular grids, for instance. Then each cell can have a different value for the attribute without any unnatural crisp boundaries. Besides the problem of crisp boundaries, the forest compartment approach involves the problem that in reality there is often within-compartment variation as regards stand characteristics. The method presented in this study is based on analysing and overlaying the data in raster format. This, and the technique whereby the evaluation of continuous variables is done by means of partial priority functions, provides a methodical capability for handling both crisp and fuzzy boundaries, and also the within-stand variation. The main factors limiting the full use of the capability were the accuracy and resolution of input data. The nature of the boundaries 90 R. Store, J. Kangas / Landscape and Urban Planning 55 (2001) 7993 and the needed accuracy depends on the requirements of the application. The suitability index calculated for S. odora was not as such the most essential point of this study. The index was produced mainly in order to demonstrate the purposes and qualities of the method, and there- fore, eld checks were not included in the case study. The criteria used in the case study represented just one expert view on the habitat factors of the species. Instead of using only one expert it is possible to use a group of experts and estimate the sub-priority functions and criteria weights according to the opinion of that group (Kangas et al., 1998). The Delphi technique, for example, could be used in searching for the coherence between the views of experts involved in the process. Determining the habitat factors and their weights are two crucial phases in the process. It is possible that even slight changes in weight coefcients have a conclusive effect on the results of suitability analyses. Also, the structure of the decision hierarchy can affect the results. In practice, it is impossible to include all the factors affecting the suitability to an evaluation model. The choice of the model is always a compro- mise between accuracy and costs. Also, the capabilities embodied in the model for handling spatial relationships have bearing on the accuracy of the results in the case of S. odora. For example, variables describing the size of the core area and the distance between habitat patches could increase the accuracy of suitability analysis. Some further research and method development is needed to get more spatial relationships included in the analysis. In a situation where the decision criteria are not independent of each other, the additive assumption as applied in the case study is not valid. In this context, the additive assumption means that a suitability index value can be calculated as the weighted sum of sub- priorities with respect to individual habitat factors. One way to deal with non-additive variables in the model is to estimate the interactions of habitat factors and add the interaction terms to the additive priority function (Keeney and Raiffa, 1976). Another way to handle this independence problem is to use rules of combination method as presented by Hopkins (1977). Using the said method, suitabilities are assigned to sets of combinations of factors and expressed in terms of verbal logic. It is also possible to deal with non- additivities by transforming and combining habitat factors (Pukkala et al., 1997). When HERO-type priority models are applied, also multiplicative parts can be added into the model (Kangas and Kangas, 1998). In sensitivity analysis, the effects of changes in the weights of different habitat factors were investigated. In addition to the uncertainties associated with the weight coefcients, the quality of the nal results was inuenced by the accuracy of the input data and error propagation. The assumption in the approach pre- sented in this study was that the input data are error free. However, the information available in habitat suitability analysis is often uncertain and imprecise because of matters such as measurement errors and errors arising from processing of the data. In the case study, most of the data layers were collected by means of eld inventory and the rest were produced by using spatial analysis. The most signicant factors affecting the quality of the results in the case of S. odora were the uncertainty associated with the data produced by spatial analysis and the uncertainty associated with the modelling of expert knowledge. The data measured in eld inventories, e.g. tree age, stem volume, diameter at breast height, were not error free, but the errors associated to them were not crucial. It is possible to determine the sensitivity of the model to errors and error propagation by means of tools based on the analytical error pro- pagation method and Monte Carlo simulation (Bur- rough and McDonnell, 1998). Vigorous research in the eld of statistical metho- dology of modelling expert knowledge and on analys- ing and reducing uncertainties in expert judgements is going on (Alho et al., 1996, 2001; Alho and Kangas, 1997; Kangas et al., 1998; Leskinen, 2000). In the future, the results of these studies will be integrated with GIS-based habitat suitability modelling. This will, hopefully, alleviate the problems regarding the quality of the analysis results. If the suitability index is used to select a network of areas where selection is not independent of the com- position of previously selected sites, the problem is that the method does not offer any analytical way to take into consideration the effects of earlier selections on the ranking of the remaining areas. In those cases, e.g. nature reserve network selection, optimisation or heuristics algorithms provide a better approach than R. Store, J. Kangas / Landscape and Urban Planning 55 (2001) 7993 91 the method presented in the present study (e.g. Pressey et al., 1997). Sensitivity analyses examined the inuence of the temporal factor by changing the weights given to permanent and time-changeable habitat factors. The spatial pattern of the suitability index changed along with the weighting schema. This means that the optimal values of the time-permanent factors did not follow those of the time-changeable factors. Also, the variety within the forest compartments was increased when more emphasis was given to the time-permanent decision factors. Different weighting of time-permanent and time- changeable factors leading to a different choice among habitats has consequences both for management plan- ning and species conservation. The most effective alternatives in the long run can be found in areas which are not among the best ones at the moment, when a momentary evaluation is made emphasising factors whose values change with time. Especially when dealing with goals related to nature conserva- tion, the time horizon has to be long enough. This is important to take into consideration, e.g. in landscape ecological planning. Sensitivity analysis using the temporal factor is most useful in situations where the objective is related to a long time horizon and when one is not in such a trade-off situation where one would have to give away some areas if one were to want to preserve some others. For example, the temporal factor plays an important role if one is going to increase the size of an existing nature conservation area and at the same time one wants to nd out the optimal solution in the long run. Then it might be better to choose an area with optimal soil characteristics and adequate vegetation characteristics rather than an area with optimal current vegetation characteristics but only average suitability with regard to soil characteristics. On the other hand, species usually require a con- tinuum of suitable habitats in time and place to survive and, therefore, it is important to avoid changes in species habitats that are too sudden and distinct. Certainly it is not appropriate to give away all of today's best habitats because vegetation development takes a long time and involves many uncertainty factors. One needs to nd an appropriate dynamic `habitat slide' if the purpose is to shift habitats in the course of time to the best areas from the viewpoint of time-permanent habitat factors. The modelling approach presented, especially the sensitivity analysis as carried out in the case study, gives valuable support to this kind of dynamic forest planning. However, the method presented is at its best in tasks where the purpose is to rank a set of alternative sites according to the habitat requirements of a certain species or where the aimis to nd sub-areas where it is most probable to nd the species under examination. Developing methods for managing dynamic ``habi- tat slides'' in nature conservation problems, as well in any natural resources planning, is an important topic of future research. Other important topics include the combined use of expert knowledge and existing evaluation models in habitat suitability modelling, including several species' concurrent suitability con- siderations, and taking more spatial factors into suit- ability considerations. Furthermore, dealing with the interdependencies between habitat factors considered in modelling processes is worth studying further. The GIS-based framework presented in this study gives a solid basis for this kind of methodological develop- ment work. References Alho, J., Kangas, J., 1997. Analysing uncertainties in experts' opinions of forest plan performance. For. Sci. 43, 521528. Alho, J., Kangas, J., Kolehmainen, O., 1996. Uncertainty in the expert predictions of the ecological consequences of forest plans. Appl. Stat. 45, 114. Alho, J., Kolehmainen, O., Leskinen, P., 2001. Regression methods for pairwise comparisons data. In: Schmoldt, D., Kangas, J., Mendoza, G.M., Pesonen, M. (Eds.), The Analytic Hierarchy Process in Natural Resources and Environmental Decision Making. Kluwer Academic Publishers, Dordrecht, in press. Berry, J., 1993. Cartographic modelling: the analytical capabilities of GIS. In: Goodchild, M., Parks, B., Steyaert, L. (Eds.), Environmental Modelling with GIS. Oxford University Press, New York, pp. 5874. Beven, K., Lamb, R., Quinn, P., Romanowich, R., Freer, J., 1995. TOPMODEL. In: Singh, V. (Ed.), Computer Models of Watershed Hydrology. Water Resources Publications, pp. 627 668. Burrough, P., McDonnell, A., 1998. Principles of geographical information systems. Oxford University Press, NewYork, 333 pp. Carver, S., 1991. Integrating multi-criteria evaluation with geographical information systems. Int. J. Geogr. Informat. Syst. 5 (3), 321339. Hopkins, L., 1977. Methods for generating land suitability maps: a comparative evaluation. J. Am. Inst. Plann. 43 (4), 386400. 92 R. Store, J. Kangas / Landscape and Urban Planning 55 (2001) 7993 Jankowski, P., 1995. Integrating geographical information systems and multiple criteria decision-making methods. Int. J. Geogr. Informat. Syst. 9 (3), 251273. Johnson, R., Temple, S., 1986. Assessing habitat quality for birds nesting in fragmented tallgrass prairies. In: Verner, J., Morrison, M., Ralph, J. (Eds.), Wildlife 2000. Modelling Habitat Relationships of Terrestrial Vertebrates, pp. 245249. Kangas, J., Kangas, A., 1998. Ekologiset mallit ja ekologisten riskien hallinta metsasuunnittelussa. Metsatieteiden aikaukau- sikirja-Folia Forestalia 2, 207222 (in Finnish). Kangas, J., Alho, J., Kolehmainen, O., Mononen, A., 1998. Analyzing consistency of experts' judgments case of assessing forest biodiversity. For. Sci. 44 (4), 610617. Keeney, R., Raiffa, H., 1976. Decisions with Multiple Objectives: Preferences and Value Trade-offs. Wiley, New York, 569 pp. Kotiranta, H., Niemela, T., 1996. Uhanalaiset kaavat Suomessa. Toinen uudistettu painos. Edita, Helsinki, 184 pp. (in Finnish). Leskinen, P., 2000. Measurement scales and scale independence in the analytic hierarchy process. J. Multi-Criteria Decision Anal. 9, 163174. Nijkamp, P., Rietveld, P., Voogd, H., 1990. Multicriteria Evaluation in Physical Planning. Contributions to Economic Analysis. North-Holland, Amsterdam, 219 pp. Pausas, J., Braithwaite, M., Austin, M., 1995. Modelling habitat quality for arboreal marsupials in the South Coastal forests of New South Wales, Australia. For. Ecol. Manage. 78, 3949. Pereira, J., Duckstein, L., 1993. A multiple criteria decision making approach to GIS-based land suitability evaluation. Int. J. Geogr. Informat. Syst. 7 (5), 407424. Pukkala, T., Kangas, J., 1993. A heuristic optimization method for forest planning and decision making. Scand. J. For. Res. 8, 560570. Pukkala, T., Nuutinen, T., Kangas, J., 1995. Integrating scenic and recreational amenities into numerical forest planning. Land- scape Urban Plann. 32, 185195. Pukkala, T., Kangas, J., Kniivila, M., Tiainen, A.M., 1997. Integrating forest-level and compartment-level indices of species diversity with numerical forest planning. Silva Fennica 31 (4), 417429. Pressey, R., Possingham, H., Day, J., 1997. Effectiveness of alternative heuristic algorithms for identifying indicative minimum requirements for conservation reserves. Biol. Con- serv. 80, 207219. Reisinger, T., Kennedy, D., 1990. A spatial decision support system for opportunity area analysis on the Jefferson national forest. In: Proceedings of the GIS/LIS `90, Anaheim, Canada, pp. 733740. Saaty, T., 1977. A scaling method for priorities in hierarchical structures. J. Math. Psychol. 15, 234281. Saaty, T., 1980. The Analytic Hierarchy Process. McGraw-Hill, New York, 287 pp. Shaw, D., Atkinson, S., 1988. GIS applications for golden-cheeked warbler habitat description. In: Proceedings of the GIS/LIS `88, Falls Church, Virginia, USA pp. 401406. Siddiqui, M., Everett, J., Vieux, B., 1996. Landll siting using geographic information systems: a demonstration. J. Environ. Eng. 122 (6), 515523. Wadge, G., Wislocki, A., Pearson, E., 1993. Spatial analysis in GIS for natural hazard assessment. In: Goodchild, M., Parks, B., Steyaert, L. (Eds.), Environmental Modeling with GIS. Oxford University Press, New York, pp. 332338. Voogd, H., 1983. Multicriteria Evaluation for Urban and Regional Planning. Pion, London, 367 pp. Ron Store is presently a Researcher at the Finnish Forest Research Institute, Kannus Research Station. He received the MSc degree in Forestry in 1995 from the University of Joensuu (Finland). His field of research includes utilising the capabilities of geographical information systems (GIS) in multi-objective forest management planning. His other areas of professional interests are GIS, cartographic modelling, multi-objective forest planning, multi- criteria evaluation. Jyrki Kangas is at present the Director of the Kannus Research Station of the Finnish Forest Research Institute. He has been the Professor of Forest Management Planning (acting) and Associate Professor of Forest Economics (acting), at the University of Joensuu. He received his MSc in 1986, and DSc in 1992 from the University of Joensuu. He has been a Docent at the University of Joensuu since 1994, an Editorial Board Member of Journal of Environmental Management, Managing Board Member of Silva Fennica, as well as a referee of a number of journals, supervisor of many doctoral theses, opponent or peer-reviewer of many theses, and involued in international evaluation tasks. His areas of interest include forest planning, natural resource management, optimisation methods, multicriteria decision support, ecological modelling, expert judgments. Kangas has also received the Scientific Achievement Award conferred upon him by the International Union of Forest Research Organisations (IUFRO). His total number of publications is about 150 of which about 90 are scientific publications and about 45 are peer-reviewed articles or mono- graphs. R. Store, J. Kangas / Landscape and Urban Planning 55 (2001) 7993 93