Freemark, K., C. Hummon, D. White, and D. Hulse 1996. Modelling risks to biodiversity in past, present and future landscapes. Technical Report No. 268, Canadian Wildlife Service, Headquarters, Environment Canada, Ottawa K1AoH3

Please see the original for the full text.

---

Introduction

Maintaining biodiversity is important for supplying vital resources (food and pharmaceuticals), providing economic income and stability, ensuring long-term ecosystem viability, and for ethical considerations such as intrinsic value (e.g. Wilson 1988, Heywood and Watson 1995). Land-use practices are a major cause of the decline in biodiversity in recent decades (Soulé 1991, World Conservation Monitoring Center 1992, p. 235). Conservation efforts have focused on maintaining biological diversity primarily by minimizing exposure to human activities through establishment of networks of protected areas (e.g. Scott et al. 1987, 1993). However, the long-term conservation of biological diversity is dependent not only on establishment of protected areas, but also on maintaining hospitable environments and viable populations within human-dominated landscapes (Noss and Harris 1986, Western 1989, Hansen et al. 1991, Shafer 1994, Freemark et al 1995).

We modeled the impacts of past and possible future landscape changes on the biodiversity of the Western Muddy Creek watershed. This modeling approach, developed by White et al. (in press), requires:

• a GIS habitat map for each landscape scenario (past, present, and alternative futures),
• a list of resident species,
• and the species' associations with the habitat classes on each map.

We modeled biodiversity as breeding amphibian, reptile, bird, and mammal species in the watershed (Tables1 and 2). See Scott et al. (1993) for a justification for using non-fish vertebrates to represent total biodiversity.

Risk to biodiversity was represented by ratios of habitat area in the past and in each future scenario to habitat area in the present. We calculated risks for individual species, and mean risk for all species and for subsets of species. This measure of change in biodiversity is one estimate of change in species populations. More detailed life history requirements of species (e.g. minimum area requirements) were not incorporated in the model because these data were not available for all species. However, in an earlier study, White et al. (in press) found minimal changes to risk results when area requirements were included.

CREATING SPECIES-HABITAT ASSOCIATIONS

We reviewed published and non-published literature and data to determine species' use of habitats during the breeding season, for breeding and/or feeding. We assigned a "1" (species is likely to use the habitat) or a "0" (species is unlikely to use the habitat) for each entry in the matrix of 26 wildlife habitats and 236 species, for a total of 6,136 species-habitat entries (Table 4). Even the most comprehensive sources (attempting to cover all species and all habitats) were not complete, leaving some species-habitat entries blank, due to omission of some species and/or some habitats.

We compiled a list of the data sources that contributed to each of the 26 wildlife habitats. There were between 1 and 10, with an average of 5 data sources for each wildlife habitat. The agricultural habitats had an average of only 3 sources each, with many sources applying to only a subset of the species list. After compiling the data sources, about 15% of the species-habitat entries had no data; most of these information gaps were in the agricultural habitats. For entries that had data, the sources were contradictory in some cases. Local experts (Table 3) were consulted in order to fill in missing species-habitat entries, to resolve entries with contradictory data, and to confirm or modify entries determined from the published sources.

Cross-referencing between mapped land cover classes and wildlife habitat classes

To identify the habitats that each species uses in the breeding season, we first compiled a list of wildlife habitats that occur in the Muddy Creek watershed. To compile this list we created a cross-reference between:

1. land cover classes for the present and five possible future maps;
2. wildlife habitat classes (data from wildlife publications); and
3. Pre-Settlement vegetation classes (Table 5 ).

Creating this cross-reference (Table 6) involved working within the different constraints of these three data sets. Our task was to find a balance between enough detail to capture differences between species' habitat associations, and enough generalization to have a concise set of habitats. The land cover map is constrained by the minimum mapping unit size (pixels of 30 meters x 30 meters), which does not allow the inclusion of some habitat variables, such as small riparian or wet areas and scattered woody vegetation at field edges. In addition, the Pre-Settlement vegetation map was compiled from surveyors' notes from the 1850s based on observations along the grid of section lines that lie one mile apart. These observations were then interpolated by ONHP/TNC to the landscape between the section lines, resulting in a map with lower spatial precision and lower accuracy than the present and future maps. We acknowledge these differences in resolution and land cover definitions between the Pre-Settlement map and the other maps, and caution our interpretations accordingly.

Several refinements were required to improve the way that the land cover classes represented wildlife habitats.

After these habitat refinements, there were 26 different wildlife habitat classes that were cross-referenced to 38 land cover classes, and to 22 Pre-Settlement vegetation classes (Table 6). There were also an equal number of near-water classes that were cross-referenced. In several cases, a wildlife habitat class was assigned to more than one land cover class (e.g. the herbaceous roadside habitat is assigned to 4 different lowland road classes; urban habitat is assigned to commercial areas and to rural structures). Several land cover classes (e.g. trails; intermittent streams of the lowland) have no habitat associated with them, because the surrounding land cover classes would determine the species that are present. The youngest three Douglas-fir forest land cover classes (0-40 yrs, 40-80 yrs, 80-120 yrs) have no equivalent in the Pre-Settlement vegetation classes, according to the surveyors notes, which indicated a wide range of tree ages in each forest patch, and no burned areas.

Modifying land cover maps to represent wildlife habitats

Figure 9

We investigated several methods for splitting upland from lowland areas in the Muddy Creek watershed. Initially we examined topographic maps at 1:24,000, 1:62,500, 1:100,000 and 1:250,000 scales, each having a different contour interval. Based on this examination, we then constructed upland maps for a series of elevation values from 90 to 120 meters by 5 meter increments, and compared these to the spatial distribution of land cover classes. We also created maps of slopes greater than 1% and greater than 2%. After comparing these maps we selected the 110 meter contour as the splitting criterion for upland from lowland. Subsequently, this criterion was partially verified in the field. Using this definition of upland, the percentages of the watershed in upland and lowland are 57% and 43%, respectively (Figure 9). As discussed above, we used the upland/lowland map to divide all four water classes and two road classes (light duty and unimproved roads) into separate upland/lowland classes for the past, present, and each future map.

We created the near-water habitats by buffering a 90 meter expansion zone around all features in the open water or 2+ order stream classes. Most of the 21 "buffer" species (B; those species only using habitats near water; Table 2) use both open water and 2+ order streams. Smaller 1st order streams were not included in the water buffer because these features' near-water habitat would be unsuitable for many of the buffer species. A buffer of 90 meters (3 pixels) was used because this figure seemed to be a reasonable compromise between the 11 species using land up to several hundred meters from water (e.g. beaver, Castor canadensis) and the 10 species using land 10 - 50 meters from water (e.g. tailed frog, Ascaphus truei). This 90 meter buffer was superimposed on each map, including the upland/lowland water and road classes. The resulting maps had a possible 76 land cover classes, 38 inside the water buffer and 38 outside ( Table 6), though no map had more than 70 classes present.

We treated the Pre-Settlement map the same as the present and future maps with one exception. As created by ONHP/TNC , this map had no hydrographic features. In the present and future maps, transportation features were overlaid on top of the hydrologic network by the University of Oregon team (Mike Flaxman, personal communication 1996). This resulted in pixels of road replacing pixels of streams in various places. Thus the upland/lowland splits of streams reflected these sometimes truncated streams. The water buffer, on the other hand, was created from the complete representation of the open water and 2+ order stream classes. For the Pre-Settlement map, we created the upland/lowland split of water classes based on the full representation of these features. Of course, this representation may not accurately reflect the hydrology of the 1850s. Possible errors could have arisen from removal of beaver ponds, addition of farm ponds, channelization of streams in the lowlands, and modification of channels due to indirect effects of change in land cover, including wetlands. We believe that errors in our model introduced by these possible data errors are preferable to errors from not including any water features.

After applying the cross-reference from land cover classes to habitat classes, we tabulated the changing area percentages of habitat classes in the present, possible futures, and past landscapes (Table 7 ). The most obvious differences in the proportions of habitat classes were between the contemporary landscapes (present and possible futures) and the past landscape. In the contemporary landscapes, conifer classes and grass seed dominated, whereas in the 1850s landscape, older age conifer, mixed forest, savanna, and prairie dominated. We reiterate, however, several reasons for being cautious about these differences and the resulting effect on the biodiversity risk results reported later. The spatial mapping resolution of the 19th century land surveys almost certainly resulted in an under-representation of lowland riparian and marsh habitats. The lack of differentiation of forest age classes in the survey notes precluded use of the finer distinctions that we have in the contemporary data, with the result that all conifer forest in the 1850s landscape was assigned to the oldest age class as the most reasonable alternative. Lastly, some of the fine distinctions in floristic composition recorded by the land surveys were lost in our modeling because we did not have habitat association data for these distinctions.

Risk modeling

The objective of our analysis was to measure changes in biodiversity, represented by species' habitat area, between the present and each of the five future scenarios, and between the present and the past. We regarded habitat area as an index of the abundance of breeding units for each species. Habitat area was determined by the sum of the areas of each habitat assigned to a species in the habitat association matrix (Table 4), without regard to spatial configuration. Change in habitat area for a species in each future scenario was calculated as the ratio of future habitat area to present habitat area, using the present as the baseline for comparison. Change in habitat area for each species in the past was calculated in the same manner, as the ratio of past habitat area to present habitat area, also using the present as the baseline for comparison. By using the present as the baseline for both the future and the past, species' habitat ratios and risks are related, allowing a species' future habitat area (or risk) to be directly compared to its past habitat area (or risk). We calculated the risk to a species for a future (or past) landscape as

obtaining a percentage of habitat area at risk in the future (or past) compared to the present.

We calculated summary risk statistics for taxonomic and other groups of species. Because the skewed empirical distributions of the raw habitat ratios appeared approximately lognormal, we transformed these ratios using natural logarithms. We then computed the mean habitat ratio for the set of species using the transformed habitat ratios, for each landscape. Next, we transformed the mean habitat ratio in the logarithm scale back to the geometric mean on the original scale. The geometric mean of each set of ratios was used as the measure of central tendency for a group of species. We obtained a mean percentage of habitat area at risk from

1 - [geometric mean of habitat ratios] x 100.

In this report we express all risks to the nearest whole percent, with a maximum of three significant digits.

Construction of species richness change maps for futures and past

We constructed maps of the Muddy Creek watershed showing changes in species richness (number of species) for each possible future (and the past), compared to the present. For each future or past map, the number of species present in each pixel of habitat for the future (or past) scenario was subtracted from the number of species present in the same pixel in the present. A positive number would indicate a gain in species richness, while a negative number would indicate a loss in species richness, for that pixel of habitat in the future (or past), compared to the present. These species richness change maps aided in the identification of landscape changes that contributed most to changes in species richness in the futures (and past).

RESULTS

Average risk for taxonomic groups of animals

The average risk of habitat loss for the 214 native vertebrate species (excluding extirpated species) is highest in the High-Development future (19% - Table 8 , Figure 11), with lower risk in the Moderate-Development and Plan-Trend futures (5% and 4%, respectively), and improvement in the Moderate- and High-Conservation futures (4% and 6%, respectively) and the Pre-Settlement past (9%).

Figure 12

The number of native species at risk varies for each future scenario (Figure 12): 75% of native species are at risk in the High-Development future, decreasing to 69% in the Moderate-Development future and 56% in the Plan-Trend future, with only 31% at risk in the Moderate-Conservation future, and 40% at risk in the High-Conservation future, and a similar proportion (42%) are "at risk" in the past (less habitat in the past than in the present). In summary, the average risk to all 214 native species (Table 8 , Figure 11) is lowest in the Moderate- and High- Conservation futures and the number of species at risk (Figure 12) is also smallest in the Moderate- and High-Conservation futures. This overall trend of greater risk with development and less risk with conservation is also reflected in trends within individual species: 39% (83/214) of native species show a monotonic trend of increasing or level risk with increased development, while only 14% (29/214) show a monotonic trend of increasing or level risk with increasing conservation.

The trend of higher risk with more development is also present in the averages of risk for the amphibians, the birds, and the mammals (Table 8 , Figure 11). The amphibians are the taxonomic group showing both the highest risk (High-Development, 29%) and the highest improvement (High-Conservation, 19%). The reptiles have almost no change in any of the futures (0% - 3%), but show a dramatic improvement (65% more habitat) in the Pre-Settlement past, compared to the present. The increased reptile habitat in the past is due to the loss of most oak savanna and prairie habitat since the 1850s (Table 7 ). Oak savanna and prairie habitats (habitats 16 and 17, respectively, on Tables 4 and 6) are used by all but one of the reptiles (11/15 species use both habitats, plus 3 species use one habitat; Table 8 ). Within the amphibians, salamanders and frogs both follow the overall trend of the amphibians, with highest risk in the High-Development future, highest improvement in the High-Conservation future, and some improvement in the Pre-Settlement past. The salamanders have greater risk than the frogs in each possible future.

Within the reptiles (Table 8 ), the turtles show the trend of increasing risk with development, but are at risk even in the conservation scenarios. The lizards show the opposite trend of increasing risk with conservation, and the snakes show no trend. All three reptile subsets show improvement in the Pre-Settlement past compared to the present, particularly the lizards and snakes. The differing trends for each reptile subset account for the overall lack of trend within the averaged reptiles.

Within the birds (Table 8 ), 11/21 subsets show a trend of risk with development and improvement with conservation. 8/21 bird subsets have this trend monotonically (herons, hawks, shorebirds, owls, woodpeckers, forest insect-eaters, vireos, tanagers/grosbeaks), and 3/21 bird subsets have this trend more weakly (ducks, hummingbirds, warblers). 8/21 bird subsets have little or no trend (grouse, flycatchers, swallows, crows/jays, wrens, thrushes, blackbirds, finches). 2/21 bird subsets show an opposite trend of risk with conservation and improvement with development (doves/pigeons, sparrows). The bird subsets are highly variable in risk or improvement in the Pre-Settlement past, with some groups showing improvement of at least 75% more habitat in the past than today (owls, hummingbirds, woodpeckers, vireos).

Within the mammals (Table 8 ), 7/11 subsets show a monotonic trend of increasing risk with development (shrews/moles, bats, large rodents, squirrels/gophers, bears/raccoons, weasels, cats), with the bats and the squirrels/gophers showing the highest risk. The other subsets show no consistent trend in the futures (rabbits/hares, voles/mice) or almost no change (coyotes/foxes, deer/elk). Most mammal subsets show only modest changes in the past: 6/11 subsets show some improvement (more habitat) in the Pre-Settlement past, with the bears/raccoons showing the most improvement; 5/11 subsets have risk (less habitat) in the past.

Average risk for other groups of animals

Figure 13: Risk to Habitat Area by Species Status (XXX In Progress - this version not final)	Figure 14: Risk to Habitat Area for Specialist versus Generalist Species.

Grouping the species by their status (e.g. rare, introduced, extirpated) also shows that there is a correlation between some groups' status and their magnitude of risk (Table 9 , Figure 13, Figure 14 ). The introduced species (I - shown in blue) are (somewhat surprisingly) at risk in all the futures, but they also show a trend towards more risk in the conservation futures, which is consistent with their preference for human-dominated (developed) landscapes. The extirpated (E - shown in pink), rare (R - shown in yellow), and vulnerable (V - shown in light green) species show the common trend of high risk with development and improvement with conservation, with improvement (more habitat) in the Pre-Settlement past compared to today (Table 9, Figure 13 ).

Native habitat specialists (defined here as species using < 10% of the 26 wildlife habitat classes; 14 species) also showed the same trend of risk with development and improvement with conservation (Table 9 , Figure 14). Habitat specialists (shown in blue) also show risk (31% less habitat) in the Pre-Settlement past, due in part to the 6 specialists that use only marsh or deciduous riparian habitats, which are both probably under-represented on the Pre-Settlement map due to the coarse resolution of the original survey. Native habitat generalists (shown in pink, defined here as species using > 70% of the 26 wildlife habitat classes; also 14 species) showed very little change in the futures compared to the present, and a small improvement in the Pre-Settlement past.

Individual species at risk of losing > 50% of their habitat

There are 41 high risk species (18% of the 222 native species, including those extirpated) that are at risk of losing > 50% of their habitat in the future (Table 10 ). 29% (12/41) of these high risk species are vulnerable species (Oregon or Federal conservation status); by comparison, only 18% (39/222) of the entire native species list are vulnerable species. 7% (3/41) of the high risk species are "rare" (R); by comparison, only 4.5% (10/222) of the entire native species list are "rare". These figures indicate that the high risk species are more likely to be species of concern (vulnerable; rare) than species that are not high risk. 85% (35/41) of the high risk species have > 50% risk in the High-Development future, while only 15% (6/41) have > 50% risk in the Conservation futures. This indicates that the High-Development scenario is a greater threat to habitat loss and resulting loss of biodiversity than the other futures.

To put these numbers into perspective, there are 27 species (12% of the 222 native species) that have already lost > 50% of their habitat since the Pre-Settlement past (i.e. those species with risk values < 100%; see Table 11 for explanation of risk values for the past). This means that more species (41) could lose > 50% of their habitat in the next 30 years than the number (27) of species that have already lost > 50% of their habitat in the last 150 years. Only 3 species are on the list of species at risk of losing > 50% of habitat in the future AND on the list of species that have already lost > 50% of their habitat (Tables 10 and 11): the Burrowing Owl, Athene cunicularia (a rare "R" species, also an Oregon Sensitive species and Federal Species of Concern), the Marbled Murrelet, Brachyramphus marmoratus (an Oregon Threatened and Federal Threatened species), and the California Condor, Gymnogyps californianus (an extirpated "E" species that is essentially extinct, also a Federal Endangered species). This shifting of risk from one set of species to another suggests that the kinds of habitat changes in the past are somewhat different from those envisioned for the futures, for most species. 30% (8/27) of the species that have already lost > 50% of their habitat are vulnerable, compared to only 18% (39/222) of the entire native species list. 19% (5/27) are rare species, compared to only 4.5% (10/222) of the entire native species list. This indicates that species of concern have already undergone significant decreases in habitat availability.

Although our analyses do not calculate population viability, these quantitative indications of possible species habitat loss could be considered a first step toward a ranking of species of concern. See Mace and Lande (1991) for ranking criteria based on population persistence.

Changes in species richness in the futures and past

Maps showing changes in species richness (number of species) for habitats in the possible futures and the past, compared to the present, are shown in Figures 15, 16, 17, 18, 19 and 20. Overall, there is a trend of larger total area of species loss in the High-Development scenario (Figure 15), decreasing to smaller total area of species loss in the High-Conservation scenario (Figure 19). There is also a corresponding, opposing trend of smaller total area of species gain in the High-Development scenario (Figure 15), increasing to larger total area of species gain in the High-Conservation scenario (Figure 19). In the Moderate-Conservation future, total area of species loss is roughly equal to total area of species gain. In the Pre-Settlement past (Figure 20) areas of changing species richness (gain or loss) dominate the map because of (1) large-scale landscape changes since the 1850s (e.g. increasing human domination and increasing fragmentation of the landscape), and (2) the change in mapping resolution between the Pre-Settlement and present-day maps. In the Pre-Settlement past (Figure 20), the total area of species gain is greater than the total area of species loss.

DISCUSSION

Methods for predicting potential impacts of human activities on biological diversity across a hierarchy of spatial and temporal scales are needed to make land use planning both clearer and better informed (Hansen et al. 1993, Dale et al. 1994, Freemark 1995). We used an approach for estimating potential risk to biodiversity from past and future land cover associated with landscape changes in the west Muddy Creek watershed in the Willamette River Basin in Oregon. Although many of the risks or losses in habitat that we computed by our model are relatively small, it is important to bear in mind that continued change at the same rate has a dramatic compounding effect. For example, a constant rate of loss of habitat of 1% of the remaining stock per year results in a 22% loss from present in 25 years and an 87% loss in 200 years.

Although much conservation biology is concerned with individual species of concern, or with threatened species as a group, the strength of our approach is a consideration of a broader definition of biodiversity, in this case all breeding non-fish vertebrates in the Muddy Creek watershed. Correspondingly, our approach produces less certain results as the focus is changed to smaller groups of species or to individual species because of the simplifying assumptions we made in order to compute risks for all species. For example, the Bald Eagle (Haliaeetus leucocephalus) may hunt along part of the Muddy Creek, but not on the numerous smaller tributaries in the 2+ order streams class. We modeled the Bald Eagle into all 2+ order streams, which leads to an overestimate of its habitat area. In another example, species that only use old growth forest (e.g. Marbled Murrelet, Brachyramphus marmoratus; Vaux's Swift, Chaetura vauxi) were modeled into all forests >120 years in age, without consideration of microhabitat features, which could result in an overestimate of these species' habitat areas.

There were other possible sources of error or uncertainty in our analyses. Each set of input data may have been affected by error. The land cover maps provided by Hulse et al. may have suffered from errors in assigning land cover classes to pixels. The species-habitat association table (Table 4) may have contained errors as well. Both the land cover maps and the species-habitat association table were affected by the classification system that was used. Certain habitats were likely to be better identified than others through the air photo and Thematic Mapper imagery, and certain species were likely to be better represented than others by the classes of habitat that were delineated on the maps. While we did not attempt to model any of these sources of error, some of the error may have been mitigated in the analysis through the calculation of the ratio of habitat area in the future to the same quantity in the present. To the extent that these errors affected the past and future landscapes in a similar way to the present, then error effects may have been partially canceled in the ratio. A further contribution to the robustness of these results was the calculation of averages for change in habitat area across many species, an analysis strategy that may have helped to mitigate errors or weak assumptions for individual species.

Species richness (total number of species) did not change in any of the possible futures, because our definition of species loss was zero pixels of habitat, implying that as long as one pixel of habitat existed, a breeding unit of the species could be supported. Without considering minimum area requirements and intraspecies demographic effects, the loss of a species would require complete elimination of habitat, rather than habitat loss sufficient to reduce populations below sustainable levels. Thus there is a discrepancy between the model results (no loss of species) and reality (8 permanently extirpated "E" species, and 10 rare "R" species). This discrepancy suggests that:

• (1) some species may not have enough habitat to sustain a viable population, and/or
• (2) some of the habitat associations are not accurately reflecting actual habitat use, and/or
• (3) the extirpations are due to factors other than habitat loss or habitat alteration (e.g. extermination of undesirable species, competition with introduced species, sensitivity to disturbance, pollution).

An example of the first explanation is the almost-extinct California Condor (Gymnogyps californianus), which has only 0.5% of the present-day Muddy Creek watershed available as habitat, compared to 114 times more habitat in the Pre-Settlement past (Table 2). If this situation was not sufficient to represent a species loss, the California Condor would surely disappear from the High-Development future, which has only 1/50th of present-day potential Condor habitat.

Model improvements

We want to reiterate some of the simplifying assumptions that we have made in order to analyze a large set of vertebrate species. These include the use of a limited set of habitat classes (Table 6) and a corresponding species-habitat association matrix (Table 4) that only assigns presence or absence in a habitat class. We did not consider area requirements for species, nor the shape or context of a habitat patch, except for proximity to water and upland versus lowland occurrence of water. Each of these assumptions limits the realism of our analyses. For example, while habitat may serve as a useful indicator of vertebrate demography, the relationship is seldom perfect (Block et al. 1994, Wolff 1995). Biotic interactions (e.g. predation and competition), disturbances, chance demographic events, suitability of edge versus interior habitat, differences in habitat quality and configuration, and other factors may all complicate assessments of species-habitat associations (Freemark et al. 1995). Studies we have in progress indicate, in particular, the need for refinements to the initial model to include habitat quality in the species-habitat association matrix (Barczak et al. submitted) and a more restrictive definition of suitable habitat in relation to area sensitivity and interior/edge habitat preferences of some forest bird species (Santelmann et al. in preparation). Our model also assumes 100% occupancy of habitat units. Many species are relatively rare, even in their most preferred habitat (Robbins et al. 1989, Vickery et al. 1994). Rare species are also those most often at risk of extinction (but see Tilman et al. 1994). For these reasons, it is important to validate species-habitat models to determine if the error level is acceptable (Hansen et al. 1993, Block et al. 1994).

CONCLUSIONS

Although further ecological refinement is still required, modeling approaches such as the one presented here can begin to discriminate the effects of potential landscape change on biodiversity and help inform the decision-making process. We see the assessment based on habitat area in this study as a first step toward a more complete assessment of population viability for a set of species. Population viability is strongly related to area of suitable habitat (Laurance 1991) and to population size (Pimm et al. 1988), which is often a function of habitat area. Augmenting our approach with population viability analysis (PVA) would improve the assessment of risk by incorporating the persistence probability of species within landscapes. Because PVA requires additional life history information and the computation of persistence probability for each species (e.g. Lamberson et al. 1992, Armbruster and Lande 1993, Beier 1993), it is not currently feasible to analyze as large a set of species as in this study. In conducting any PVA, it is also critical to consider the regional context of the study area in relation to the range of the species' populations (Freemark et al. 1993, Ruggiero et al. 1994).

Our approach has been useful for developing and engaging local support for land use planning based on biodiversity considerations. It provides a quantitative ranking of landscape alternatives using a methodology that is relatively simple with few parameters (Doak and Mills 1994) and is adaptable to different definitions of biodiversity. Our approach is sufficiently generic that it can be applied to other spatial and temporal scales and to other regions using data of different levels of resolution. As such, it can facilitate a more comprehensive and hierarchical approach to the development of land use plans for the proactive conservation of biological diversity.