Applications

Biodiversity Predictions: Integrating Urban Growth Models with Land Cover Data and Species Habitat Information

Christopher B. Cogan

Alfred Wegener Institute, Bremerhaven, Germany

Introduction

Habitat loss and subsequent fragmentation due to urban development are part of a larger suite of anthropogenic impacts on biodiversity, but they now rank among the principal causes of species endangerment in the United States.  Several types of urban growth simulation models have been developed which can supply useful information for biodiversity planning.  In many cases, however, the data required for biodiversity planning may not be compatible with the urban models, leading to analytical inaccuracies and misleading conclusions.  Here, I briefly introduce a case study for biodiversity analysis and examine several lines of logic likely to be employed in such assessments.  I conclude with a discussion of assumptions built into the data and their influence on model outcome.

Techniques for Model Integration

Habitat quality and quantity aspects of biodiversity were examined using three principal inputs: urbanization scenarios, wildlife habitat maps, and species habitat models.  Output from the analyses was reported as loss of habitat area or, in some cases, in terms of impact to the vertebrate species associated with degraded habitats.

A flow chart of the models and analyses provides an overview of the biodiversity sensitivity analysis (Figure 1).
Three different models for predicting patterns of urban expansion were tested.  These included the 500-meter "urban buffer," "Landis" (Landis and Zhang 1998) , and "Clarke" (Clarke and Gaydos 1998) scenarios.  Outputs from the different growth models were then used in conjunction with coarse-grain (100 ha minimum mapping unit) land cover maps from the California Gap Analysis Project (GAP, Davis et al. 1998) .

Figure 1. Flow Chart for biodiversity sensitivity analysis. Three urban growth scenarios and two land cover models combine to evaluate vertebrate and habitat impacts in Santa Cruz County, California.

The Landis and Clarke models were also used with a finer-grain (1 ha) land cover data set.  This map layer was commissioned by the Association of Monterey Bay Area Governments (AMBAG) based on 30-meter Landsat Thematic Mapper (TM) imagery.  Spatial distributions of individual vertebrate species predicted to occur in the study area were made possible by applying wildlife habitat relationship (WHR) models (Airola 1988) to the coarser-grained GAP land cover data.  Potential impacts of urban growth to these species were explored by intersecting scenarios of future urban growth from each of the three models with the WHR-based predicted distributions of the species (e.g., Figure 2).

Figure 2.  Comparison of predicted habitat loss under three growth scenarios in Santa Cruz County, California: 500-meter urban buffer, Landis growth model, and Clarke growth model.  Species and habitat data are from the California Gap Analysis Project (GAP). Habitat classes are rank-ordered based on the results from the Landis model.

Discussion

The species habitat analysis outlined here is a close examination of one major factor in the assessment of biodiversity.  Other biodiversity elements such as ecoregional analysis, restoration potential, special features, and habitat shape are also important, though these were not specifically addressed in this study.  The combination of urban growth models and land cover maps (Figure 1) was used to compare measures of habitat and vertebrate impacts.  Here, habitat impacts were considered to be actual habitat areas converted to urban land use.
For example, if a 1,000 ha forest is reduced to 900 ha after urbanization, the habitat loss is 10%.  If the same forest is reassessed in terms of native vertebrate habitat, it may be more important to consider buffer distances from impacts, non-linear predation effects, and other complex landscape metrics.  These more specific approaches can be valuable in some instances; however, when applied to a regional study with many species, the results can be misleading.  Stated differently, it is challenging to model disturbance effects as realistically as possible while working with a group of dissimilar species over a broad area.

The approach to vertebrate habitat assessment presented here assumed that if a highly intrusive land use such as urbanization entered a habitat patch, then the entire patch was likely to be compromised in terms of habitat quality for vertebrate species.  In some instances, this assumption may have overemphasized the impact of urbanization.  On the other hand, it was also likely that urbanization effects were underemphasized in cases where urban expansion approached (but not actually entered) a habitat area.  An alternate model could employ spatial buffers to model the neighborhood effects of urbanization; however, this approach would introduce additional complexities, such as splitting map polygons, and imposes the need for species-specific analysis.  Both the habitat and species types of impacts are important; however, it is necessary to clarify the conceptual differences between habitat and vertebrate impacts when evaluating or discussing urban growth impacts.  The methods used in this analysis were based upon an underlying logical sequence most simply presented as a flow chart (Figure 3).

A central assumption here was that different urban growth patterns should have measurably different biodiversity impacts.  As with any metamodel, it was also important to ensure that the data and various component models were compatible for integrated analysis.  It is often illuminating to investigate where the logic of a scientific investigation might become unsound, as well as where it is strong.  The logical flowchart outlines key junctions where this type of biodiversity assessment might face impediments and offers explanations and recommendations for each situation.

Figure 3. Logical flow chart for biodiversity analysis with urban growth models.

Given perfectly accurate biodiversity and urban growth models, lack of biodiversity response will still occur if the two models are not spatially or thematically compatible.  An indicator of this type of incompatibility can be seen in the comparison of vertebrate habitat losses following different urbanization scenarios (Figure 2).  One interpretation of this result suggests that vertebrate impacts are much the same following either the Clarke or the Landis models.  Indeed, it seems remarkable that the rank order of species and even habitat impacts is so similar under two independent and seemingly different growth models.  It would seem to require a radically different growth model like the simplistic 500-meter buffer to produce a significantly different outcome.  Another, perhaps more likely, interpretation is also possible. If the GAP data on wildlife habitat relationships are spatially coarser that the growth models, our ability to differentiate between the Landis and Clarke models will be diminished.  In support of this hypothesis, the appearance of the map products and (most importantly) the habitat impacts, indicated substantial differences between each of the three urban models.

The balance of spatial grain and thematic detail is an important consideration when producing and using maps of land cover for use in biodiversity analysis.  Using the AMBAG 30-meter MMU land cover map, the fine map grain results in relatively large areas (up to 49,000 ha) to be mapped as contiguous albeit marginally connected patches.  At slightly coarser map grains, many of the corridors of connecting habitat would merge into other classes, resulting in a very different data set for the habitat modeler.  This example illustrates how fine-grain maps with coarse thematic detail can overemphasize habitat connectivity.  In this case, the assumption that urban disturbance on the edge of a habitat patch impacts the entire patch becomes tenuous when using data with fine spatial grain but coarse thematic grain such as the AMBAG 30-meter land cover map.  As 100-meter or finer-grain urban growth models gain acceptance as a reasonable spatial scale to model the biodiversity land use complex, more research is needed to ascertain the appropriate levels of thematic resolution in land use and land cover mapping.

There are several difficulties associated with measuring regional urban impacts on vertebrate species.  The model presented here used polygons of habitat to represent potential distributions of vertebrate species and assumed that analysis of divided polygons was not a valid application of the data.
Detailed studies of specific divided habitat polygons are possible, given appropriate species-specific data.
However, this local approach will not be effective regionally.  Urban development is sometimes seen as a continuous creeping of small steps, whereby each development project in isolation is difficult to assess for regional biodiversity impact.  The species assessment method presented here used habitat polygons to model impacts, effectively dealing with the "urban creep" issue while maintaining biologically meaningful area units.  The complementary combination of a discrete species metric (e.g., polygon-based) along with a continuous habitat model is a powerful and much needed approach.

As biodiversity models such as those discussed here evolve and build in complexity, our land cover maps and wildlife habitat relationship models will be pressed to deliver more information with higher quality standards.  Some of our data sources have already evolved from simple maps of predicted species location to become temporally dynamic models of predicted species connectivity and spatial pattern.  Unfortunately, most of our current maps are not up to this advanced standard.  Like most modelers, cartographers have long known that the design constraints of producing the best habitat maps will depend on the specific questions being asked of the data.  This fundamental principle is sometimes obscured or overlooked when we allow technological capabilities such as satellite sensor resolution and radiometric spectral response to overly influence our understanding of habitat classification and vertebrate distribution.

These findings were presented to facilitate an improved understanding of habitat and species impact models and to provide direction for future land use and land cover mapping.  The specific models discussed here are important elements of more generalized biodiversity assessments, which are continually improving our understanding of biodiversity and promise to provide additional guidance to minimize the disruptive impacts of urbanization and development.

Literature Cited

Airola, D.A. 1988.  Guide to the California wildlife habitat relationships system.
Jones and Stokes Associates, Sacramento, California.

Clarke, K.C., and L.J. Gaydos. 1998.  Loose-coupling a cellular automaton model and GIS: Long-term urban growth prediction for San Francisco and Washington/Baltimore.  International Journal of Geographical Information Science 12: 699-714.

Cogan, C.B.  2002.  Biodiversity conflict analysis at multiple spatial scales.  Pages in  J.M. Scott, P.J. Heglund, and M.L. Morrison, editors.  Predicting species occurrences: Issues of accuracy and scale.  Island Press, Washington, D.C.

Davis, F.W., D.M. Stoms, A.D. Hollander, K.A. Thomas, P.A. Stine, D. Odion, M.I. Borchert, J.H. Thorne, M.V. Gray, R.E. Walker, K. Warner, and J. Graae.  1998.  The California Gap Analysis Project: Final report. University of California, Santa Barbara, California.

Landis, J., and M. Zhang.  1998.  The second generation of the California urban futures model.  Part 1: Model logic and theory.  Environment and Planning, B-Planning & Design 25: 657-666.