Landscape Ecology 15: 713–730, 2000. © 2000 Kluwer Academic Publishers. Printed in the Netherlands. 713 The effects of future urban development on habitat fragmentation in the Santa Monica Mountains Jennifer J. Swenson1 & Janet Franklin Department of Geography, San Diego State University, San Diego, California 92182-4493, U.S.A. 1 (Corresponding author: Current address: Department of Forest Science, Oregon State University, Corvallis, OR 97331-7501, USA, (E-mail: [email protected]) Received 27 May 1999; Revised 23 February 2000; Accepted 1 April 2000 Key words: coastal sage scrub, habitat fragmentation, landscape pattern indices, land use change, simulation modeling, Santa Monica Mountains Abstract A site suitability model of urban development was created for the Santa Monica Mountains in southern California, USA, to project to what degree future development might fragment the natural habitat. The purpose was to help prioritize land acquisition for the Santa Monica Mountains National Recreation Area and examine to what extent projected urban development would affect distinct vegetation classes. The model included both environmental constraints (slope angle), and spatial factors related to urban planning (proximity to roads and existing development, proposed development, and areas zoned for development). It implemented a stochastic component; areas projected to have high development potential in the suitability model were randomly selected for development. Ownership tracts were used as the spatial unit of development in order to give the model spatial realism and not arbitrarily ‘develop’ grid cells. Using different assumptions and parameters, the model projected the pattern of development from ∼ 5 to ∼ 25 years hence (based on recent development rates in the area). While < 25% of the remaining natural landscape is removed under these scenarios, up to 30% of core (interior) habitat area is lost and edge length between natural vegetation and development increases as much as 45%. Measures of landscape shape complexity increased with area developed and number of patches of natural habitat increased four- to nine-fold, depending upon model parameters. This increase in fragmentation occurs because of the existing patterns of land ownership, where private (‘developable’) land is interspersed with preserved park lands. Introduction The coastal Santa Monica Mountains in California (Figure 1), host numerous habitats from valley oak savanna to coastal sage scrub, many of which are becoming increasingly rare in the face of rapid development in southern California (Westman 1981; Schoenherr 1990; O’Leary et al. 1992). The mountain range is located within the Southwest Ecoregion of California that contains the highest number of endangered plant species in the U.S. except Hawaii (Dobson it et al. 1997). The National Park Service’s Santa Monica Mountains National Recreation Area (SMMNRA) exists as more of an administrative unit than a typical protected area. Its boundary encompasses various parklands, public beaches, cities, and substantial amounts of private residential and commercial land with varying levels of development. Urban and residential development continues in the semi-rural area as increasing amounts of Los Angeles residents seek refuge from the surrounding city. Although the area is ‘patchily preserved’ through efforts by federal and local agencies and private conservation groups, the majority of the land is mostly undeveloped and remains in private ownership (Figure 2a). The National Park Service (NPS) is attempting to acquire land within this patchwork of ownership and jurisdictions guided by its Land Protection Plan. Prioritization of land acquisition according to the biological or cultural significance of parcels is crucial due to the 714 Figure 1. The Santa Monica Mountains in Southern California. extremely high land prices, and the difficulty in acquiring land held fast by private owners. Generally, conservation in urban areas proves difficult, as land costs are prohibitive due to high demand, and local political entities usually favor development over preservation (Murphy 1988). By predicting where future development is likely to occur, conservation agencies and land managers may assess what areas most urgently need acquisition and subsequent preservation in order to prevent large-scale loss and fragmentation of the remaining natural habitat. In this study a simulation model projecting future urban development in the Santa Monica Mountains was implemented in a Geographic Information System (GIS) to examine the effects that future development patterns could have on different vegetation communities and the spatial pattern and connectivity of the natural habitat. Background Habitat fragmentation results from increasing the number of landscape pieces, decreasing interior habitat area, increasing the extent of forest-opening edges [or other habitat types], or increasing isolation of residual forest patches’ (Li et al. 1993; p. 67). Habitat fragmentation has been recognized as the leading factor in species loss, on both a local and global level (Wilcove et al. 1986; Wilcox and Murphy 1985). Animals with relatively large ranges (umbrella species sensu Wilcox 1994), for example, birds and large mammals, are often the first to be affected by habitat fragmentation due to its effects on population viability (Beier 1993). If reserves are designed to meet the needs of these species, other species with smaller area requirements may be protected under this ‘umbrella’ (Murphy 1988; Noss 1990). Conversely, reptiles and small mammals with limited mobility may be separated into distinct populations by narrow geographic barriers (Quinn 1990) such as non-primary roads or urban structures easily crossed by more mobile creatures. These more mobile species can often be sustained on a network of smaller patches, as long as they are, in some way, linked (Soulé 1991b). Urban development creates a high contrast edge around natural vegetation, and may introduce exotic ornamental plant species (Alberts et al. 1993), domestic cats and dogs, air- and water-borne toxins, increased run-off, lowered water tables, and increased general disturbance by human activities (Murphy 1988). Where native vegetation has been cleared for firebreaks around residences and along ridge tops in the Santa Monica Mountains, exotic grass species dominate. Naturally sparse Mediterranean scrub habitats, such as those in the Santa Monica Mountains, are especially vulnerable to edge effects and exotic invasions. This vegetation is physically more susceptible to disturbance such as trampling, and slower to recover, than that of a forest community, for example (Soulé et al. 1992; O’Leary 1995; Zink et al. 1995). Various preservation agencies, principally the National Park Service, are working to acquire more land to maintain the existing natural areas within the National Recreation Area through the Land Protection Plan, (NPS 1984; 1987; 1991; 1994). The National Recreation Area can be considered to be in a state of reserve planning or design as over one third of the Santa Monica Mountains consists of privatelyowned undeveloped land (NPS 1994), which could potentially be purchased by the NPS. The overall goal of designing a biological reserve is to conserve a diversity of ecosystems and thereby conserve the highest diversity of species possible (Franklin 1993). A number of approaches have been suggested for accomplishing this goal where land resources are limited. These include maintaining requisite habitat area for viable populations of animal species, maximizing habitat connectivity, and prioritizing the conservation 715 value of landscape patches based on multiple criteria (Margules and Usher 1981; Church et al. 1996; Stoms et al. 1998). Landscape linkages can enhance biotic movement between patches, provide extra foraging area, serve as refuges during disturbances such as fire, and often significantly add to the preserved areas of native vegetation (Saunders et al. 1991). If the area to be preserved has already been subjected to habitat fragmentation, Soulé (1991a) recommends promoting habitat contiguity and connectedness through landscape linkages (Hudson 1991). In the case of the Santa Monica Mountains, by selectively purchasing land, areas that are currently conserved may be enlarged and the linkage between habitat patches could be maintained. Habitat pattern and connectivity can be quantified using various indices of landscape structure (O’Neill et al. 1988; Turner 1989; Ripple et al. 1991; Turner and Gardner 1991; Li et al. 1993; Haines-Young and Chopping 1996). Quantification of landscape pattern at large spatial scales is crucial in landscape-scale reserve design (Noss 1983), and in management of processes that are dependent on a particular landscape structure (Mladenoff et al. 1993). The process of landscape fragmentation is generally not random (Usher 1987), as in the case of urban development. An advantage of landscape simulation models is that they allow analysis of hypothetical disturbances that may be generated without altering the actual landscape (Turner 1989). The strength of these models lies in their ability to project future landscape scenarios given current circumstances and to test certain practices or dynamics of disturbance. The models discussed in the following paragraphs contributed to the formulation of the simulation model used in this study, which incorporates elements from landscape ecology and urban development modeling. Urban simulation modeling is a multidisciplinary field that originated in the late 1960s and has since expanded to incorporate simulations such as residence location models, urban employment locations, and transportation flow models (Wilson 1974; Putnam 1983; Kain 1986; Landis 1994). Urban development models for land-use planning often incorporate economic and socio-political factors as well as spatial features such as transportation networks and existing land use. Related to these models are site suitability models, which evaluate areas in terms of their suitability for urban development (Hopkins 1977). While site suitability models strive to find places that are ideal for development, the model developed in this study projects where development is most likely to occur. Urban models differ from landscape ecology models (Baker 1989; Sklar and Costanza 1991; Turner and Dale 1991) by depending heavily on socioeconomic data, and incorporating theories of urban development and transportation modeling. The Santa Monica Mountains are impacted strongly by urbanization, yet it does not appear to be easily predicted by the typical growth models used in urban planning. In our study area there are few central places of business, and travel time to and from work is not an essential factor as many of the residences are second homes or have been built for retirement. Because of the complexity of the socioeconomic conditions present in the area, we do not attempt to model these factors, but incorporate into our model spatial factors that appear to influence the urban development of the Santa Monica Mountains. Therefore, the model does not predict future socioeconomic factors that would accompany development, but assumes current conditions will continue. Recently, models of landscape change have striven to simulate urban development or land use change using concepts from both urban modeling and landscape ecology. These models incorporate socioeconomic and transportation factors such as land tenure or land ownership, population, immigration patterns and distance to roads, rivers and cities with spatial and ecological factors (e.g., lot size, vegetation type, soil type, elevation, and slope), when modeling land use change from regional to global scales (Wilkie and Finn 1988; LaGro and DeGloria 1992; Dale et al. 1993; Spies et al. 1994; Turner et al. 1996; Theobald and Hobbs 1998). The objective of many of these studies is to project and identify the effects of land use change on biodiversity (Steinitz et al. 1998) and to guide ecosystem and resource management (Wear et al. 1996; Spies and Johnson 1999). Materials and methods In this study, the current degree of landscape fragmentation in the SMMNRA was examined using indices of fragmentation. Next, a spatial simulation model was developed to predict the pattern of future urban growth on the landscape. The model contains a stochastic component and was replicated while changing the input variables and assumptions, and as a result produces a range of future development scenarios. The spatial pattern of habitat in the simulated landscapes 716 was summarized using the same habitat fragmentation indices as used on the current landscape. The effects of different land development scenarios on landscape fragmentation were measured with these indices and effects of development on different vegetation classes were examined for disproportionate impacts. Study area The Santa Monica Mountains are an east-west trending mountain range adjacent to the Los Angeles Metropolitan Area. The SMMNRA administrative boundary encompasses 60,000 ha and is bordered on the south by the Pacific Ocean, to the north by suburban communities, to the east by the Los Angeles metropolitan area, and to the west by agricultural lands. The study was conducted within the area defined by a one km buffer outside the park boundary, though numerical data are reported only for the area within the recreation area boundary. Within this boundary, approximately half of the land is publicly owned and protected park land, while the remaining half is private land of which an estimated 25% is developed (NPS 1994). The Santa Monica Mountains have been subjected to many waves of development and are currently the object of further urban growth encouraged by the booming economy, less expensive real estate, and the increasing population of nearby Los Angeles (Markman 1994). Existing urban development within the National Recreation Area boundary ranges from sparsely distributed ranch-style homes to relatively dense subdivisions. The mountains are topographically complex, with steep slopes that rise from the Pacific Ocean to 600 m. Because of this dissected topography, there are areas within the park boundary that are relatively isolated from infrastructure and the surrounding urban development. The vegetation of the Santa Monica Mountains reflects the Mediterranean climate conditions of cool moist winters, warm dry summers, and frequent fires. SMMNRA’s ecological communities of coastal sage scrub, chaparral, coast live oak and walnut woodlands, valley oak savanna, grasslands, wetlands, riparian and coastal areas (Keeley and Keeley 1988; Mooney 1977; Sawyer and Keeler-Wolf 1995; Figure 2b) host 50 species of mammals, 384 bird species, and 36 herptiles (NPS 1994). There are multiple species of small mammals occurring in the recreation area that are endemic to chaparral habitats (Quinn 1990) and various animal species that are found on state or federal rare, threatened, or endan- gered species lists (NPS 1994). The mountain lion (Felix concolor), resides in the area to date, yet its long-term persistence is regarded as precarious due to habitat fragmentation by urbanization (NPS 1993). Coastal sage scrub comprises 25% of the Santa Monica Mountains and is of special conservation interest due to its relatively high native biodiversity and increasing scarcity (O’Leary et al. 1992; O’Leary 1995). Once a widespread vegetation type in southern California, it was estimated to occupy only 10–15% of its original range in ca. 1980 (Westman 1981), and continues to decline due to urbanization. Almost 100 species of plants and animals associated with coastal sage scrub are listed as rare, sensitive, threatened, or endangered by federal and state agencies (California Department of Fish and Game 1994). Valley oak (Quercus lobata) is in serious decline in California due to its low recruitment rates and increasing destruction for development and farming as it occupies valley bottoms and low foothills (Pavlik et al. 1992). Grasslands (now occupying less than 5% of the study area), are prime hunting habitat for raptors, such as the golden eagle (Aguila chrysaetos), and have decreased by thousands of hectares due to large-scale urbanization (Plantrich 1990). Digital map data The primary digital map data used in this study were provided by the NPS and include a map of existing vegetation/land cover, land ownership, the County General Plan, elevation, hydrology, and roads (Table 1). The vegetation/land cover map is of special importance to this study, and was produced by the authors for the NPS (Franklin et al. 1997). The map was based on satellite multispectral imagery from the Landsat Thematic Mapper sensor. Vegetation stands were delineated in two ha minimum mapping units and labeled using image segmentation and unsupervised spectral classification (according to the methods of Woodcock et al. 1994; Franklin and Woodcock 1997), and extensively edited based on air photo interpretation and field reconnaissance. The area-weighted thematic accuracy of the vegetation-land cover map was 93%, based on an accuracy assessment method using fuzzy sets (the RIGHT operator; see Gopal and Woodcock 1994). Vegetation class accuracy ranged from 62–100%. The accuracy of a map depicting land cover can affect the measurement of landscape pattern (Wickham et al. 1997). From this digital map, a layer of existing development was extracted (based on areas 717 Figure 2. a) Land ownership in the Santa Monica Mountains (NPS 1984, see Table 1). b) Map of existing vegetation/land cover ca. 1995 (SDSU/NPS, see Table 1). c) Development likelihood map; based on additive overlay of five criteria and widest buffer widths for roads and existing development. ‘Low’, ‘low to medium’, ‘medium’, and ‘high’ classes correspond to the intersection of 1, 2, 3, 4–5 input variables, respectively. 718 Table 1. Metadata for spatial data (digital maps) used in the analysis. Variable Origins Vegetation Satellite image/aerial photo processing Department of Geography San Diego State University (SDSU) NPS Land ownership, tract numbers, owners proposed developments County General Plan Year of last update Permission of use granted by Original data format 1995 SDSU/NPS Raster, 30 × 30 m 1984 1995 NPS Vector 1995 SCAG Vector Elevation Southern California Association of Governments (SCAG) United States Geological Survey variable NPS Hydrology United States Geological Survey variable NPS Roads Thomas Brothers 1994 NPS Raster, 30 × 30 m, 1:24,000 Vector, digital line graph, 1:24,000 Vector mapped as ‘urban’), as it was the most recent mapped information on land cover for the SMMNRA. The 1984 land ownership map delineates ‘ownership tracts’ or contiguous parcels owned by the same person or entity. This digital map also includes information on proposed developments dated as late as May 1995, which range from large urban subdivisions to small structures such as room additions. Spatial modeling was carried out using the Grid module of the Arc-Info GIS software (Arc-Info 6.0, ESRI 1991). Digital geographic layers obtained from the NPS in Arc-Info vector format were converted to raster format with a cell resolution of 30 by 30 m (0.09 ha). The model results were exported from ArcInfo to the FRAGSTATS software (version 2.0; McGarigal and Marks 1994) for the analysis of landscape pattern. Landscape simulation model The landscape simulation model developed in this study estimates a location’s likelihood of future development using geographic planning data, land ownership data, development constraints, and a stochastic selection process. This model can be classified as a whole mosaic cartographic model in which all variables, as Boolean data layers, are combined or summed (Tomlin 1991). The input variables are equally weighted, each consists of layers that are ‘twostate’ or binary, and they represent variables in a multivariate analysis (Hunsaker et al. 1993). The areas in which all or most of the binary variables intersect geographically are considered more likely to be developed than areas where few or no variables intersected. The five input variables are each represented as binary map layers of the presence or absence of likely future development. The layers are overlain geographically and summed (Figure 3a). Several methods (or simulation variations) of selecting areas for development were tested and are discussed later in this section. The number of points used in the selection of ownership tracts was varied to simulate different amounts of development. The values of two parameters, the buffer widths around roads and buffer widths around existing development, were varied to examine the effect on the simulation results. The assumptions underlying this model are that development is most likely to occur (1) relatively level slopes, (2) in areas of proposed developments, (3) in privately-owned areas that have been zoned for development, (4) on lands in close proximity to roads and (5) in close proximity to existing development. Thus, input variables of the urban development model consist of maps of buffered areas of varying widths around roads and around existing development, areas zoned for development in the County General Plan, areas of moderate slope, and housing tracts in which developments are currently proposed. Our rationale in selecting these five variables is as follows: 719 Figure 3. a) Illustration of how the simulation model produces the likelihood of development layer. The five model input variables are combined spatially and areas of extreme slopes (over 60%) and areas already developed are spatially removed. This results in a layer consisting of low, medium, and high likelihood of development. b) Illustration of how developed area is chosen from likelihood of development layer for the three different modeling scenarios. 1. We limited potential development to slopes of less than 60%, and allowed development on slopes from 0 to 25% after consulting the literature (Whitley et al. 1993; Moser 1991; Landis 1994), Park Service personnel, and the geographic data of the study area (examining the distribution of current development with respect to slope). This was accomplished by assigning slopes of less than 25% the value 1 as a model input variable, while using slopes greater that 60% as a mask or filter after the five variables were combined. The limit of 60% may appear quite high, but because of the strong demand for new housing in southern California and the desire for panoramic views, the building industry often develops remote areas as well as hillsides when flatter lands have already 720 been developed (Moser 1991 and National Park Service personnel). Building on steeper slopes is quite common in the Santa Monica Mountains, where a third of the area has slopes of over forty percent. 2. All housing tracts with proposed developments were given equal weight regardless of type or extent of development, due to the difficulty of classifying the 1100 proposed developments. 3. The County General Plan demarcates areas to be developed, according to density or type, and areas to be preserved. Areas that were zoned for higher density development were selected for likely development. The ‘zoned for development’ input variable was not weighted more heavily because the General Plan could be amended in the future. For example, many of the proposed developments (the second input variable) occur on lands that are not zoned for development. 4. Existing roads and highways indicate where development has occurred in the Santa Monica Mountains (Los Angeles County 1981). It was assumed that future development would more likely occur along main roads or highways that enable access out of the mountains into surrounding urbanized areas. In order to estimate the buffer width to use around roads and highways, the current landscape was examined in terms of the percentage of development that existed within a distance of a given road. This was done to identify a road buffer width that included a high percentage of development, but did not contain large amounts of non-developed areas. For this project we wished to have a range of buffer widths to simulate different urbanization strategies and to examine this effect on the natural landscape. After analyzing the current patterns of development, three sets of road buffer distances were chosen because they represented an adequate range of scenarios for the simulation model. The road buffer widths each had two distances, one for primary roads and highways, and the other for secondary roads, and consisted of: 50/70, 70/100, and 100/140 m. 5. According to the Interim Area Plan for the Malibu/Santa Monica Mountains area, new developments in the Santa Monica Mountains will be located next to existing developments, because of the limited availability of infrastructure and public services throughout the mountains (Los Angeles County 1981). Based on literature review (Iverson 1988; Landis 1994; Theobald and Hobbs 1998) and consultation with park personnel, buffer zones of 100, 200, 500, and 1000 m surrounding existing development were chosen to delineate this model variable. The first model variation, that includes the stochastic component, is the ‘Standard’ simulation, which is an equally-weighted five variable site suitability model with random selection of ownership tracts. The five input variables are summed to form an output layer containing values ranging from 0 to 5, signifying areas of low to high development potential (Figure 3a). Figure 2c shows a map of the resulting values based on a model that used the larger values for buffer widths around roads and existing development. From this output map, as with all model variations, areas of extreme slopes and existing development are removed, so as not to be selected. This results in a site suitability map of privately owned, non-developed land on slopes of less than 60% with various levels of development potential. To simulate development, points are then randomly selected from the areas of high and medium development likelihood (classes 3, 4 and 5), and the ownership tracts into which the points fall are considered ‘developed’ (Figure 3b, Standard Simulation). Each Standard simulation thus results in a map of predicted development, which is subsequently overlain on the existing vegetation map. Test runs of the model were performed to determine the number of spatially random points that could be located in the areas of high development potential without exhausting the possibility of selecting a new ownership tract. The number of points was varied in order to simulate different amounts of development. When the number of random points is held constant, the areal extent of development varies with each replication, as the tracts of private land vary from 0.1 to 5462 ha and, with each model run, different tracts are selected. In order to determine the variability of area projected to be developed as a result of the model’s stochastic component, three replicate runs of one Standard model were performed with 50 and then with 120 points while using the same buffer width parameters. The other simulation types do not require replication because they lack the stochastic component of the Standard simulation. The next type of simulation, the ‘Overlay’, produces a map of development as a direct result of overlaying the binary input variables. The Overlay simulation has no stochastic component and does not use ownership tracts as the unit of development. Instead, 721 all grid cells classified as medium to high likelihood of development (classes 3–5) in the site suitability map are converted to development (Figure 3b). Therefore, the edges or spatial units of development are the direct result of the spatial coincidence of the input variables. To simulate an even greater extent of urban development, a simulation termed ‘Maximum Development’, that does not include a stochastic component, was implemented. This site-suitability model simulates development by ownership tracts. All ownership tracts that partially or completely coincide with areas of high or medium development likelihood (classes 3– 5) are converted to development (Figure 3b). Finally, to simulate a hypothetical maximum urban development, all private lands having natural vegetation are converted to development in the ‘Total Build-out’ scenario. Most main and primary roads were not mapped in the vegetation/land cover map (due to the 2 ha minimum mapping unit that was utilized), and therefore are not represented as edges on the vegetation map (Figure 2b). Because these roads can act as barriers to smaller vertebrates we performed an additional analysis to explore how main and primary roads may affect the fragmentation of the contiguous landscape. In the GIS, the main and primary roads of the Santa Monica Mountains were buffered slightly in order to displace a single line of 30 m pixels. These buffered areas were then overlain on the natural vegetation map that resulted from the Maximum Development and Standard model simulations. The fragmentation of the natural vegetation resulting from all modeling simulations described above was then quantified. Two different versions of each map were analyzed; one contained 14 vegetation classes (generalized classes in Figure 2b), while the other clumped all vegetation classes as a single land cover class (natural vegetation). The fragmentation analysis of vegetation classes considered separately provides insight to the effects that predicted urban development could have upon those different vegetation classes, while using a single natural vegetation class emphasizes the high contrast urban and natural interface across the landscape. Landscape pattern analysis The degree of habitat fragmentation that could be caused by potential future development was analyzed by comparing indices of landscape pattern for the map of existing vegetation to their values for the maps derived from each of the urban growth models. The following research questions were posed: 1. In fragmented landscapes, core (interior) area represents habitat unaffected by edges better than the measure of total area (McGarigal and Marks 1994), therefore, while using an edge zone or buffer distance of 100 m to delineate the interior of patches (Franklin and Forman 1987; Spies et al. 1994), will the amount of core area habitat decrease at a greater rate than the decrease in total area? 2. As urban development increases in the Santa Monica Mountains, will the fragmentation of the landscape increase as shown by (a) an non-linear increase in the amount of edge between natural vegetation and developed land when compared with the increased urban area, (b) an increase in the Landscape Shape Index (LSI), a measure of shape complexity and (c) an increase in the number of patches of habitat? The LSI metric represents the complexity of the shapes over the landscape. It is calculated using the total length of edge of the landscape, divided by the square root of the total area, which is adjusted by a constant for a standard simple shape (in this case a square). LSI is equal to one for a perfect square (or circle in the case of vector data) and increases without limit as the shapes become more irregular across the landscape. 3. Because the coastal sage scrub and valley oak woodland vegetation classes often exist along edges of current development and on lower elevation slopes (thereby appearing to be more susceptible to development), will habitat fragmentation affect coastal sage scrub and valley oak woodland disproportionately when compared to other vegetation classes? 4. With a constant search radius of 300 m (Gustafson and Parker 1992), will the Mean Proximity Index (MPI) for each vegetation class decrease as future development scenarios convert more natural vegetation to urban land use? Will the patches of a particular vegetation type become more dispersed? The MPI uses the nearest neighbor distance and patch area to calculate the average proximity between patches of like vegetation classes, and is a unitless measure representing isolation and fragmentation. The MPI decreases as patches in the landscape become more fragmented and isolated from similar patches. 722 Table 2. Results of the replicate runs of the Standard model (using middle buffer values). Measurement Area converted, ha 50 points 120 points Core area remaining, ha 50 points 120 points Edge, km 50 points 120 points Rep 1 Rep 2 Rep 3 2519 3900 2644 3527 2451 5455 48906 47318 48795 47801 48910 45652 1461 1546 1454 1499 1485 1569 Results and discussion Varying the buffer widths around existing development (four values) as well as buffer widths around roads (three values) produces twelve different results (runs) of each model. The model results shown in Figures 4b and 4c were achieved using the middle values for buffer widths (70/100 m road buffers and 200 m development buffers) and all vegetation as one category. For the Standard model, the area of land converted to urban development ranges from ∼ 2900 ha (5% of the existing natural vegetation of the study area) for the most restrictive model, to ∼ 5000ha (8% of the natural vegetation) for the model with the widest buffers. The majority of the Standard model runs produces values ranging from 4000 to 4900 ha. The influence of the buffer widths around development (100–1000 m) plays a more significant role in the resulting fragmentation of the landscape than the road buffer widths (50–140 m), because of its larger range of buffer distances (details not show here; see Swenson 1995). Each model replication results in different amounts of area developed due to its stochastic component and due to wide variability of ownership tract sizes (Table 2). The map of existing natural vegetation (Figure 4a) can be compared with the results of one run of the Standard model (Figure 4b). The Overlay models result in far less area developed than other models, as it disregards boundaries of ownership, as long as the land was privately owned. Areas developed range from 995 to 3574 ha depending on the buffer widths. The Overlay simulation results are not shown cartographically owing to the small amount of area changed. Areas converted to development in the Maximum Development simulations range from 10,701 to 13,295 ha (18 to 22% of the natural landscape) depending on the buffer widths used (Figure 4c). In this simulation, the model with the largest buffers predictably develops the most area, since all ownership tracts that include areas of high or medium development likelihood are developed (and not by a number of randomly selected points like in the Standard Model). Changing model parameters (from smallest to largest, road and development buffer widths) in the Maximum Development Model results in a 10 percent increase in edge length, a 24% increase in number of patches, and a 8% decrease in core area. The Total Build-out scenario develops 57% (34,000 ha) of the natural existing landscape, the remainder of which is currently protected. The development of all remaining private lands is unlikely to occur, although assuming current trends continue, it could hypothetically take place in 71 years (Figure 4d). The area developed by the models can be compared to recent rates of urbanization, which were estimated in this study by analyzing the differences between the area mapped as Developed in the vegetation maps of 1995 and 1984. The development rate was estimated to be about 500 ha per year for that period or 0.8% of the National Recreation Area per year. Another study of this area estimated the development rate at 570 ha per year (Kamrant 1995). Throughout the Standard, Overlay and Maximum Development simulations, fragmentation increases when compared to the existing landscape in all cases. It is difficult to visually distinguish differences between the current landscape and the results of the Standard model (Figures 4a vs. 4b), and therefore illustrates the utility of the fragmentation indices for enhancing visual interpretations. As fragmentation increases with the modeling simulations (Figures 4a–c), most of the remaining vegetation continues to exist as one large interconnected patch. This large area becomes increasingly perforated by development but the majority of the vegetation remains connected and is not cleanly broken up until Total Build-out (Figure 4d). This phenomenon is partly due to current location of existing development (one of the model’s input variables) but also due to data resolution; if the main roads had been mapped during the vegetation mapping, this main patch would have been initially split up (refer to Figure 2a for main roads). The changes in area, core area, edge, number of patches, and landscape complexity as shown by the 723 Figure 4. a) Natural vegetation (habitat) patches of the current landscape. b) Habitat patches remaining after Standard model with medium buffers (70/100 m for roads, 200 m for development). c) Habitat patches remaining after Maximum Development model with medium buffers. d) Habitat patches remaining after Total Build-out scenario. 724 Figure 5. a) Percent of landscape occupied by natural vegetation vs. percent of landscape occupied by core (interior) habitat area (natural vegetation patches less a 100 m buffer). b) The total length of edges between natural vegetation and development vs. vegetation removed by models, ha. c) Number of habitat patches in landscape vs. percent of natural landscape removed by models. d) Landscape Shape Index vs. percent of natural landscape removed by models (ha). ‘No model’= current landscape. Landscape Shape Index, for the models just discussed are shown in Figure 5a–d. The proportion of core area is, by definition, less than that of the remaining total area, and declines (diverges from the 1:1 line) throughout the modeling scenarios. The Maximum Development models remove a higher proportion of core area relative to total area than do the Standard models (Figure 5a). The amount of edge generally increases with the removal of area by the model with a pattern that appears slightly non-linear (increasing rate) when the current landscape, Standard, and Maximum Development models are compared (Figure 5b). One exception is the Total Build-out simulation that results in little remaining natural vegetation area and con- sists of simple shapes delineated by land ownership boundaries, and therefore has little edge. The Overlay model results in a high amount of edge because it converts all areas of high development likelihood, and ignores ownership tracts that generally consist of simple shapes. Actual development, however, may only occur on portions of ownership tracts and therefore the pattern depicted by the Overlay model may not be entirely unrealistic. As more land is converted to development, more individual habitat patches are formed and the number of patches doubles from the Standard model simulations to those of the Maximum Development model (Figure 5c). In Figure 5d, an increase in habitat patch complexity is also shown by an increase 725 Table 3. Fragmentation change when main roads are barriers for different simulations (middle value buffers). Model type Ha. lost with road removal Total core area, ha # Core areas Mean core area, ha Edge, km Standard Standard with roads as barriers Max. Dev. Max. Dev. with roads as barriers N/A 224 N/A 287 47,024 45,947 38,333 37,626 112 141 166 201 130 119 41 39 1519 1699 1935 2063 Table 4. Mean proximity index (MPI) for different simulations using middle value buffers. Model type Chaparral Coastal sage scrub Valley oak Coast live oak No model (current landscape) Overlay Standard Maximum Development Total Build-out 60,585.4 700.6 24.1 25.7 59,761.4 33,310.0 11,910.7 5,211.7 not calculated 603.5 417.7 157.0 25.2 16.9 11.5 1.2 23.5 16.3 19.1 17.4 Table 5. Losses by vegetation class for the Standard and Maximum Development simulations (+/− = more and less than expected by chance, based on χ 2 -square test). Principal vegetation classes Mixed chaparral Coastal sage scrub Grassland Coast live oak woodland Riparian area Valley oak woodland Walnut woodland Total area loss (converted to development) Mixed chaparral Coastal sage scrub Grassland Coast live oak Riparian area Valley oak woodland Walnut woodland Total area loss (converted to development) Loss (ha) % of original landscape Loss per class/ total loss(%) Loss per class/ area of class (%) Standard Model M, S 2164 59.1 50.8− 1232 25.6 29.0+ 265 4.7 6.2+ 229 3.0 5.4+ 58 1.8 1.3 38 0.6 0.9 13 0.2 0.3 4255 6.1 8.1 9.4 12.5 5.4 11.1 10.2 Maximum Development, M, S 6313 59.1 54.4 3506 25.6 30.2 592 4.7 5.1 381 3.0 3.2 177 1.8 1.5 71 0.6 0.6 61 0.2 0.5 11,597 17.8 22.8 21.0 20.8 16.4 20.7 47.6 726 Figure 6. (a) Patches in current landscape with roads as barriers, ranked by size. (b) Patches after Standard model with medium buffers (70/100 m for roads, 200 m for development) and roads as barriers. (c) Patches after Maximum Development model with medium buffers and roads as barriers. 727 in the LSI as development increases throughout the modeling scenarios. The geographical distribution of patch sizes with main roads considered as barriers to dispersal for the current landscape is shown in Figure 6. The roads comprise only 224 to 287 ha when they are buffered and removed, yet considering them to be barriers increases the fragmentation of these simulated landscapes as measured by core area, number of core areas, mean core area size, and edge length (Table 3). change is supported in all cases concerning coastal sage scrub, and partially supported concerning valley oak depending on the method of evaluation. The impact of development, as a function of the loss of each class area (fourth column, Table 5), shows the small woodland classes of valley oak, coast live oak, and walnut consistently have high percentage losses. Individual vegetation classes The fragmentation indices have been calculated for inventory purposes, representation of biological and environmental processes, and for identification of areas for the maintenance of interior habitat. These indices provide quantitative descriptions about the landscape pattern that may not be visually apparent in a map. Although this type of analysis is not a substitute for detailed biological research such as population viability analysis, it would be advantageous to use studies of landscape fragmentation in conjunction with such research. While we examined a number of indices of landscape fragmentation in our exploratory analyses of model results, we found direct measures such as core (interior habitat) area and edge length, and other simple indices based on direct measurements, to be most useful for comparing model outcomes. The indices in general were useful in comparing the same landscape undergoing different simulations, as a method to quantify change. Particular vegetation classes such as valley oak, coast live oak, and coastal sage scrub are impacted more than other classes by the modeled development projections in this study. The preservation of these vegetation types requires special attention because of their increasing rarity in the region. The differential impact on vegetation classes does not appear to be random and most likely results from the modeling input variables. In particular, the modeling assumption that more homes would be built on gentler slopes, which are often covered in coastal sage or valley oak, than on steep topography, commonly covered in chaparral, has integrated biogeographical factors of vegetation type into the model. Where data on species habitat use and minimum area requirements are available, it appears that the best species preservation strategy is the identification of landscape linkages for wide-ranging species such as the bobcat, while keeping in mind the habitat needs and dispersal capabilities of more sedentary small-area species like the avifauna of coastal sage scrub (see The MPI is shown for four vegetation classes as they occur on the current landscape, and in all modeling scenarios using the mid-sized buffers (Table 4). In their current coverage, chaparral and coastal sage scrub constitute over 85% of the area and are shown with data for two small oak woodland classes of special ecological interest. The index decreases when similar patches become more isolated from one another. As expected, the index decreases with increased conversion of area to development, with the exception of some minor discrepancies in the oak classes, possibly due to their small areal extent. Classes of vegetation were affected to different degrees by the development models. Note the percent of the area of each class that was lost and the change in the percent area for each vegetation class (Table 5). For the Standard and the Maximum Development models, the area lost to development for each class is shown in the first column (Table 5), the percent of the study area the class currently occupies is in the second column, the amount of area lost by the class divided by the total area removed by development is shown in the third column, and the percent loss of the class itself in shown in the fourth column. If development randomly occurred across the landscape, it would be expected that the percent of the original landscape occupied by a given vegetation class would be proportional to the percent lost by the class after development (in this case the second and third columns of Table 5 would agree). These models impact coastal sage scrub more than chaparral, and in most cases, impact valley oak to a slightly higher degree than would be expected of random development. Coastal sage scrub, grassland and coast live oak are impacted significantly more by modeled development than expected by random patterns of development, while mixed chaparral is impacted significantly less by modeled development (Table 5). Thus, the research question regarding vegetation class Conclusions 728 Soulé et al. 1988). In general, maintenance of cohesive habitat at many different scales is necessary, yet the projection models of potential patterns of urban development in this study show that without future land acquisitions, this landscape will be come increasingly fragmented. This model used a simple overlay approach based on a site suitability framework in which all variables are given equal weight and entire ownership tracts are developed. This model could be altered by giving ‘importance weights’ for each variable involved according to expert opinion (Hopkins 1977). Further steps could be taken to simulate development in only portions of ownership tracts and at various levels of density for a more detailed prediction of development, as actual development sometimes occurs across entire ownership tracts and sometimes on only a small portion of the tract, with implications for patterns of fragmentation. This modeling framework could also be used in conjunction with predicted future social and economic scenarios to create a simulation model that assumes a changing economy. Soulé (1991a) concludes from his extensive studies of chaparral canyon fragments in an urban matrix that ‘the best way to maintain wildlife and ecosystem values is to minimize habitat fragmentation’ (p. 319). The amount of area converted to development in each of the modeling simulations plays a significant role in the resulting fragmentation of the landscape, but the pattern and placement of development is important as well. If simulated development had taken place in a few isolated clusters rather than scattered across the landscape, the indices of core area, edge, LSI, number of patches, and MPI would have shown far less fragmentation. Under some patterns of development (around the circumference of a preserve, or in one compact area), the fragmentation represented by these indices could stay roughly the same with a decline in overall habitat area. Controlling or decreasing the amount of development for this area is a partial solution, but special attention should be given to the spatial pattern of present and planned future urban and residential development. According to the projections made by this modeling study, the SMMNRA, where possible, should focus its land acquisition efforts on areas in the west-central portion of the NRA to reduce overall habitat fragmentation (Figure 6), and on land which supports the most disproportionately affected habitat types: coastal sage, grassland and oak woodland. Acknowledgements We thank the following people for their contributions: Paul Chamberland, George Cox, Allen Hope, Sherri L. Johnson, Denise Kamrant, Dave McKinsey, Ray Sauvajot, David Shaari, Joseph Shandley, and Ralph Warbington. This study was supported in part by a Rocky Mountain Region National Park Service Student Cooperative Agreement (to J. Swenson). The manuscript was greatly improved by the comments of two anonymous reviewers. References Alberts, A. C., Richman, A. D., Tran, D., Sauvajot, R., McCalvin, C. and Bolger, T. 1993. Effects of habitat fragmentation on native and exotic plants in southern California coastal scrub. In Interface Between Ecology and Land Development in California. pp. 103–110. Edited by J. E. Keeley. Southern California Academy of Sciences: Los Angeles, CA, USA. Baker, W. L. 1989. A review of models of landscape change. Landsc Ecol 2: 111–133. Beier, P. 1993. Determining minimum habitat areas and habitat corridors for cougars. Cons Biol 7: 94–108. California Department of Fish and Game. 1994. Bibliographies on Coastal Sage Scrub and related malacophyllous shrublands of other mediterranean-type climates. Edited by J. F. O’Leary, S. A. DeSimone, D. D Murphy, P. F Brussard, M. S. Gilpin and R. F Noss. California Wildlife Conservation Bulletin No. 10. Department of Fish and Game: Sacramento, CA, USA. Church, R. L., Stoms, D. M. and Davis, F. W. 1996. Reserve selection as a maximal covering location problem. Biol Cons 76: 106–112. Dale, V. H., O’Neill, R. V., Pedlowski, M. and Southworth, F. 1993. Causes and effects of land-use change in central Rondonia, Brazil. Photogram Eng Remote Sensing 59: 997–1005. Dobson, A. P, Rodriguez, J. P., Roberts, W. M. and Wilcove, D.S. 1997. Geographic distribution of endangered species in the United States. Science 275: 550–553. Environmental Systems Research Institute. 1991. Arc-Info version 6.0. Geographic information system software. Franklin, J. F. and Forman, R. T. T. 1987. Creating landscape patterns by forest cutting: Ecological consequences and principles. Landsc Ecol 1: 5–18. Franklin, J. F. 1993. Preserving biodiversity: species, ecosystems or landscapes? Ecol Appl 3: 202–205. Franklin, J., Swenson, J. J. and Shaari, D. 1997. Map of existing vegetation and land cover for the Santa Monica Mountains National Recreation area; summary of map accuracy. Unpublished Technical Report to the SMMNRA, Department of Geography, San Diego State University, CA, USA. Franklin, J. and Woodcock, C. E. 1997. Multiscale vegetation data for the mountains of Southern California: spatial and categorical resolution. In Scale in Remote Sensing and GIS. pp. 141–168. Edited by D. A. Quattrochi and M. F. Goodchild. CRC/Lewis Publishers Inc.: Boca Raton, FL, USA. Gopal, S. and Woodcock, C. 1994. Theory and method for accuracy assessment of thematic maps using fuzzy sets. Photogram Eng Remote Sensing 60: 181–88. 729 Gustafson, E. J. and Parker, G. R. 1992. Relationships between landcover proportion and indices of landscape spatial pattern. Landsc Ecol 7: 101–10. Haines-Young, R. and Chopping, M. 1996. Quantifying landscape structure: a review of landscape indices and their application to forested landscapes. Progress Phys Geog 20: 418–445. Hopkins, L. D. 1977. Methods for generating land suitability maps: A comparative evaluation. J Am Inst Planners 43: 386–400. Hudson, W. 1991. Landscape Linkages and Biodiversity. Island Press: Washington D.C., USA. Hunsaker, C. T., Nisbet, R. A., Lam, D. C. L., Browder, J. A., Baker, W. L., Turner, M. G. and Botkin, D. B. 1993. Spatial models of ecological systems and processes: The role of GIS. In Environmental Modeling with GIS. pp. 248–64. Edited by M. F. Goodchild, B. O. Parks, and L. T. Steyaert. Oxford University Press, New York, USA. Iverson, L. R. 1988. Land-use changes in Illinois, USA: The influence of landscape attributes on current and historic land use. Landsc Ecol 2: 45–61. Kain, J. F. 1986. Computer simulation models of urban location. In Handbook of Regional and Urban Economics, Vol. II. Urban Economics. Edited by E. S. Mills. North Holland, Amsterdam, NL. Kamrant, D. 1995. Evaluating bobcat viability in the Santa Monica Mountains, California. Master’s thesis, California State University, San Bernardino, CA, USA. Keeley, J. E. and Keeley, S. C. 1988. Chaparral. In North American Terrestrial Vegetation, pp. 165–207. Edited by M.G Barbour and W.D. Billings. Cambridge University Press: Cambridge, UK. LaGro, J. A. and DeGloria, S. D. 1992. Land use dynamics within an urbanizing non-metropolitan county in New York State (USA). Landsc Ecol 7: 275–289. Landis, J. D. 1994. The California urban futures model: a new generation of metropolitan simulation models. Environ Planning B: Planning Design 21: 399–420. Li, H., J. F. Franklin, Swanson, F. J. and Spies, T.A. 1993. Developing alternative forest cutting patterns: A simulation approach. Lands Ecol 8: 63–75. Los Angeles County, Department of Regional Planning. 1981. The Malibu/Santa Monica mountains: An interim area plan. Department of Regional Planning, Los Angeles, CA, USA. Margules, C. and Usher, M. B. 1981. Criteria used in assessing wildlife conservation potential: A review. Biol Cons 21: 79–109. Markman, J. 1994. A new push for hillside homes emerges: Conflict over development in Santa Monicas is renewed. Los Angeles Times, 21 November 1994, A1, Los Angeles, CA, USA. McGarigal, K. and Marks, B. J. 1994. FRAGSTATS: Spatial pattern analysis program for quantifying landscape structure. Version 2.0. Mladenoff, D. J., White, M. A., Pator, J. and Crow, T.R. 1993. Comparing spatial pattern in unaltered old-growth and disturbed forest landscapes. Ecol Appl 3: 294–306. Mooney, H. A. 1977. Southern coastal scrub. In Terrestrial Vegetation of California. pp. 471–489. Edited by M. G. Barbour and J. Major. John Wiley & Sons, New York, NY, USA. Moser, W. A. 1991. Design for successful hillside development. J Urban Planning Devel 117: 85–94. Murphy, D. D. 1988. Challenges to biological diversity in urban areas.In Biodiversity, pp. 71–76. Edited by E.O. Wilson. National Academy Press, Washington, DC, USA. National Park Service. 1984. Land protection plan. N. P. S.: Santa Monica Mountains National Recreation Area, CA, USA. National Park Service. 1987. Land protection plan. N. P. S.: Santa Monica Mountains National Recreation Area, CA, USA. National Park Service. 1991. Land protection plan. N. P. S.: Santa Monica Mountains National Recreation Area, CA, USA. National Park Service. 1993. Draft resource management plan. N. P. S.: Santa Monica Mountains National Recreation Area, CA, USA. National Park Service. 1994. Natural resources research prospectus. N. P. S.: Santa Monica Mountains National Recreation Area, CA, USA. Noss, R. F. 1983. A regional landscape approach to maintaining diversity. Bioscience 33: 700–706. Noss, R. F. 1990. Indicators for monitoring biodiversity: A hierarchical approach. Cons Biol 4: 355–364. O’Leary, J. F. 1995. California coastal sage scrub: general characteristics and considerations for biological conservation. In Endangered Plant Communities of Southern California. Proceedings of the 15th Annual Symposium, Southern California Botanists. Special Publication No. 3. pp. 24–41. Edited by A. A. Schoenherr. O’Leary, J. F., Murphy, D. and Brussard, P. 1992. Scientific Review Panel Conservation Guidelines and Documentation. California Department of Fish and Game, California Resources Agency. Special Report No. 2. California Environmental Trust: San Francisco, CA, USA. O’Neill, R. V., Krummel, J. R., Gardner, R. H., Sugihara, G., Jackson, B., DeAngelis, D. L., Milne, B. T., Turner, M. G., Zygmunt, B., Christensen, S. W., Dale, V. H. and Graham, R. L. 1988. Indices of landscape pattern. Landsc Ecol 1: 153–162. Pavlik, B. M., Muick, P. C., Johnson, S. G., and Popper, M. 1992. Oaks of California. Cachuma Press and the California Oak Foundation, Los Olivos, CA, USA. Plantrich, R. F. 1990. Monitoring of golden eagle nests at Santa Monica Mountains National Recreation Area. In Examples of resource inventory and monitoring in national parks of California. pp. 49–55. Edited by C. Van Riper, T. J Stohlgren, S. D. Veirs, and S. Hillyer. U. S. Department of the Interior, Washington, D.C., USA. Putnam, S. H. 1983. Integrated urban models: Policy analysis of transportation and land use. Pion Limited: UK. Quinn, R. D. 1990. Habitat preferences and distribution of mammals in California chaparral. Research paper PSW-202. Pacific Southwest Research Station, Forest Service, U.S. Department of Agriculture, Berkeley, CA, USA. Ripple, W. J., Bradshaw, G. A. and Spies, T. A. 1991. Measuring forest landscape patterns in the Cascade Range of Oregon, USA. Biol Cons 57: 73–88. Saunders, D. A., Hobbs, R. J. and Margules, C. R. 1991. Biological consequences of ecosystem fragmentation: A review. Cons Biol 5: 18–32. Sawyer, J. O. and Keeler-Wolf, T. 1995. A Manual of California Vegetation. California Native Plant Society, Sacramento, CA, USA. Schoenherr, A. A. (ed.). 1990. Endangered Plant Communities of Southern California. Proceedings of the 15th Annual Symposium, Southern California Botanists. Special Publication No. 3. Sklar, F. H. and Costanza, R. 1991. The development of dynamic spatial models for landscape ecology: a review and prognosis. In Quantitative Methods in Landscape Ecology. pp. 239–288. Edited by M. G. Turner and R. H. Gardner. Springer-Verlag, New York, NY, USA. Soulé, M. E. 1991a. Land use planning and wildlife maintenance. J Am Planning Assoc 57: 313–321. Soulé, M. E. 1991b. Theory and Strategy (in chapter, Conservation corridors: countering habitat fragmentation). In Landscape 730 Linkages and biodiversity. pp. 91–104. Edited by W. E. Hudson. Island Press, Covelo, CA, USA. Soulé, M. E., Alberts, A. C. and Bolger, D. T. 1992. The effects of habitat fragmentation on chaparral plants and vertebrates. Oikos 63: 39–74. Soulé, M. E., Bolger, D. T., Alberts, A. C., Wright, J., Sorice, M. and Hill, S. 1988. Reconstructed dynamics of rapid extinctions of chaparral-requiring birds in urban habitat islands. Cons Biol 2: 75–92. Spies, T. A., Ripple, W. J. and Bradshaw, G. A. 1994. Dynamics and pattern of a managed coniferous forest landscape in Oregon. Ecol Appl 4: 555–568. Spies, T. A. and Johnson, K. N. 1999. CLAMS: Coastal landscape analysis and modeling study. http://www.fsl.orst.edu/clams, Oregon State University, Corvallis, OR, USA. Steinitz, C., Binford, M. W., Cote, P., Edwards, Jr., T., Ervin, S., Forman, R. T. T., Johnson, C., Kiester, R., Mouat, D., Olson, D., Shearer, A., Toth, R. and Wills, R. 1998. Alternative Futures for the Region of Camp Pendleton, California. Harvard School of Design, Harvard University, Cambridge, MA, USA. Stoms, D. M., Borchert, M. I, Mortiz, M. A., Davis, F. W. and Church, R. L. 1998. A systematic process for selecting representative Research Natural Areas. Natural Areas J 18: 338–349. Swenson, J. 1995. Examining and predicting habitat fragmentation using a geographic information system in the Santa Monica Mountains. Master’s Thesis, San Diego State University, CA, USA. Theobald, D. M. and Hobbs, N. T. 1998. Forecasting rural land-use change: A comparison of regression- and spatial transition-based models. Geogr Environ Modelling 2: 65–82. Tomlin, C. D. 1991. Geographic information systems and cartographic modelling. Prentice-Hall, Englewood Cliffs, NJ, USA. Turner, M. G. 1989. Landscape ecology: The effect of pattern on process. Annu Rev Ecol Sys 20: 171–197. Turner, M. G. and Gardner, R. (eds.). 1991. Quantitative Methods in Landscape Ecology. Springer-Verlag, New York, NY, USA. Turner, M. G. and Dale, V. H. 1991. Modeling landscape disturbance. In Quantitative Methods in Landscape Ecology. pp. 323– 352. Edited by M. G. Turner and R. H. Gardner. Springer-Verlag, New York, NY, USA. Turner, M. G., Wear, D. N. and Flamm, R. O. 1996. Land ownership and land cover change in the southern Appalachian highlands and the Olympic Peninsula. Ecol Appl 6: 1150–1172. Usher, M. B. 1987. Effects of fragmentation on communities and populations: A review with applications to wildlife conservation. In Nature Conservation: The Role of Remnants of Native Vegetation. pp 103–121. Edited by D. A. Saunders, G. W. Arnold, A. A. Burbidge, and A. J. M. Hopkins. Surrey Beatty and Sons, Chipping Norton, Australia. Wear, D. N., Turner, M. G. and Flamm, R. O. 1996. Ecosystem management with multiple owners: landscape dynamics in a southern Appalachian watershed. Ecol Appl 6: 1173–1188. Westman, W. E. 1981. Factors influencing the distribution of species of Californian coastal sage scrub. Ecology 62: 439–455. Whitley, D. L., Xiang, W. N. and Young, J. J. 1993. Use a GIS ‘melting pot’ to assess land use suitability. GIS World 6: 48–51. Wickham, J. D., O’Neill, R. V., Ritters, K. H., Wade, T. G. and Jones, K. B. 1997. Sensitivity of selected landscape pattern metrics to land-cover misclassification and differences in land-cover composition. Photogram Eng Remote Sensing 63: 397–402. Wilcove, D. S., C. H. McLellan, and Dobson, A. P. 1986. Habitat fragmentation in the temperate zone. In Conservation Biology: The Science of Scarcity and Diversity. pp. 237–256. Edited by M. Soulé. Sinauer Associates, Sunderland, MA, USA. Wilcox, B. A. 1984. In situ conservation of genetic resources: determinants of minimum area requirements. In National Parks, Conservation and Development. pp. 639–647. Edited by J. A. McNeely and K. R. Miller. Smithsonian Institute Press, Washington DC, USA. Wilcox, B. A. and Murphy, D. D. 1985. Conservation strategy: The effects of fragmentation on extinction. Am Naturalist 125: 879– 887. Wilkie, D. S. and Finn, J. T. 1988. A spatial model of land use and forest regeneration in the Ituri forest of northeastern Zaire. Ecol Modelling 41: 307–323. Wilson, A. G. 1974. Urban and regional models in geography and planning. Wiley & Sons, Chichester, U.K. Woodcock, C. D., Collins, J., Gopal, S., Jakabhazy, V. D., Li, X., Macomber, S., Ryherd, S., Harward, V. J., Levitan, J., Wu, Y. and Warbington, R. 1994. Mapping forest vegetation using Landsat TM imagery and a canopy reflectance model. Remote Sensing Environ 50: 240–254. Zink, T. A., Allen, M. F., Heindl-Tenhunen, B. and Allen, E. B. 1995. The effect of a disturbance corridor on an ecological reserve. Restoration Ecol 3: 304–310.
* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project