4th International Conference on Integrating GIS and Environmental Modeling (GIS/EM4):
Problems, Prospects, and Research Needs. Banff, Alberta, Canada, September 2 - 8, 2000.

Conducting large-scale conservation evaluation and conservation area selection using a knowledge-based system

GIS/EM4 No. 87

Patrick S. Bourgeron
Hope C. Humphries
Keith M. Reynolds

Abstract

Development of a representative regional system of conservation areas requires (1) delineating a set of land units from which candidate areas will be selected, (2) determining the suitability of these land units for conservation based on their ecological conditions, and (3) selecting land units for inclusion in a conservation network. Land units for selection were delineated as 17,227 biophysical land polygons. The conservation suitability of land polygons was determined using a knowledge-based system (KBS), implemented in a GIS application framework, in the 56,000 km² interior Columbia River basin (ICRB). The KBS consists of hierarchically arranged fuzzy-logic networks, which characterize logical relationships among land polygon attributes contributing to conservation suitability. KBS models produced (1) suitability ratings for land polygons based on their overall ecological conditions and (2) suitability of vegetation cover types (35 in the ICRB) as targets of conservation. Land polygons were selected to minimize the 'cost' of conservation networks while including as many of the cover types as possible. Costs in network selection were derived from suitability ratings; results were compared to other methods of determining cost. Ten percent of the land polygons in the study area were found to be suitable, occurring primarily in higher elevation locations. Cover type suitability ranged from 0 to 47%. Selecting land polygons using suitability as cost resulted in a network with a small number of relatively large land polygons; using area as cost produced a network of many small, but largely unsuitable, land polygons. The strengths of the KBS are its ability to integrate current knowledge and available data concerning a conservation target of interest in an explicit and flexible manner, as well as its ability to provide visualization of the spatial implications of decisions as a platform for discussion and negotiation with land managers and planners.

Keywords

Knowledge-based system (KBS), conservation network, suitability, fuzzy logic, site selection, decision support, conservation target, GIS application framework.

Introduction

A primary conservation objective is maintenance of all ecosystem components over time and over large areas to ensure the protection of all aspects of biodiversity (Noss and Cooperrider 1994). Conservation planning includes identification of conservation targets (e.g., species, communities, vegetation types), and selection, design, and management of conservation areas (Bourgeron et al. in press). Increasingly, conservation planning activities are focused on the problem of selecting networks of conservation areas that are representative of conservation targets in large regions (Margules and Pressey 2000).

Problem statement

Three distinct steps are involved in selecting land for inclusion in a representative regional system of conservation areas: (1) delineating a set of land units from which candidate conservation areas will be selected, (2) determining the suitability of these land units for conservation based on their ecological conditions, and (3) selecting land units for inclusion in a conservation network. Most efforts have been directed towards defining methods for the third step (Church et al. 1996, Csuti et al. 1997, Flather et al. 1997, Kiester et al. 1996, Pressey et al. 1993, 1996, Scott et al. 1993). In contrast, less work has been conducted on the delineation of land units and their suitability for conservation based on ecological conditions, but these steps are central to conservation (Austin and Margules 1986). Furthermore, methods are required that are explicit, repeatable, flexible, and easily visualized to facilitate communication of the consequences of conservation decisions to decision-makers.

Approach

Our objectives were to determine the suitability of land areas for conservation using a knowledge-based system (KBS), and to incorporate suitability into the process of selecting candidate areas for inclusion in conservation networks. The KBS operates within a GIS-integrated application framework, which links the KBS with spatial databases, produces output suitability maps, and provides data for network selection procedures to select land units based on their suitability.

Methods

Study area

Our study area is the 56,000 km² interior Columbia River basin (ICRB) in the US Pacific Northwest (Figure 1), a region encompassing diverse ecosystems ranging from desert grasslands and shrublands to cold, wet forests and alpine vegetation (Hann et al. 1997, Reid et al. 1995), as well as a variety of land use and ecological conditions (Quigley and Arbelbide 1997). A high proportion (53%) of the ICRB is public land administered by the US Forest Service or Bureau of Land Management (Quigley and Arbelbide 1997). A broad-scale assessment recently completed in the ICRB produced a large number of databases and GIS themes (Quigley and Arbelbide 1997), providing data on current, historical, and potential vegetation; current and historical disturbance regimes; road density; and land use condition.

Delineation of land polygons

ICRB land units for selection were delineated as biophysical polygons, using the results of spatial and statistical analyses, by overlaying maps of regional potential vegetation and ecological land units derived from classification of climate, hydrological, and biogeochemical variables. The resulting land polygons are intended to represent functional landscapes (Poiani et al. 2000), i.e., land areas whose boundaries enclose ecological patterns and processes expected to respond similarly to management or other perturbations. A total of 17,227 land polygons were delineated, representing 686 biophysical classes. Land polygon sizes ranged from 3 to 6943 km², averaging 32 km² in size (Figure 2).

Figure 1. Interior Columbia River basin study area.

Figure 2. Land polygons delineated as land units for selection in conservation area networks.

KBS and application framework

KBS models were developed using NetWeaver software and were implemented in the Ecosystem Management Decision Support (EMDS) application framework, which links databases containing land polygon attributes with the KBS in a GIS environment (ArcView) that enables data management and display of maps (Reynolds et al. 1997). EMDS provides a platform for conducting analyses in which the user can vary the spatial extent of the study and the attributes included in the analysis.

The KBS model consists of hierarchically arranged fuzzy-logic networks, which characterize logical relationships among land polygon attributes contributing to conservation suitability (Figure 3). Nodes at the terminus of networks contain fuzzy-logic functions that determine the degree to which an attribute value meets a condition. The result of applying a fuzzy-logic function is a continuous value ranging from -1 (completely false) to 1 (completely true). At any location in the network hierarchy, the evaluation of a condition depends on the condition of its components. The structure of the model, as an explicit expression of our understanding of ecological relationships, is easily communicated to decision-makers.

We constructed two types of models. The first model produced suitability ratings for land polygons based on their overall ecological conditions. Suitability was a function of land polygon size and conditions, as well as conditions in a neighborhood surrounding the land polygon (Figure 3a). If a land polygon was very large, the suitability rating was considered to depend only on the land polygon itself, without consideration of its neighborhood. Land polygon suitability was judged to be a function of fuzzy-logic networks we termed defensibility (i.e., compatibility of land use with conservation activities), and viability (i.e., the likelihood that ecological conditions allow persistence of conservation targets). Defensibility was derived from a fuzzy-logic function applied to land use condition. Viability resulted from fuzzy-logic functions applied to road density, and the degree of departure of current vegetation and disturbance regimes from historical conditions.

The second type of model assessed the suitability of 35 naturally vegetated cover types as targets of conservation within land polygons (Figure 3b). Land polygon suitability was a component of cover type suitability, combined with evaluation of cover type viability and adequate size in a land polygon, as measures of the sustainability of a cover type.

Figure 3. General structure of two knowledge-based models for evaluating (a) land polygon and (b) cover type suitability.

Site selection

We selected land polygons as candidate members in a conservation network using Sites 1.0 (Andelman et al. 1999), a site selection toolbox implemented in ArcView that uses a simulated annealing algorithm as a heuristic method for efficiently selecting sets of areas (Csuti et al. 1997, Pressey et al. 1996). A good solution to the problem of selecting a conservation network is considered to be one in which the cost of the network is minimized, but as many of the targets of conservation as possible are included. For our analysis, land polygon suitability ratings were converted to 'unsuitability' ratings by reversing their sign. These ratings were then rescaled to a range similar to the range of land polygon areas. Unsuitability ratings were compared with two other methods of representing the costs of acquiring land polygons for the network, equal land polygon costs, and land polygon area as cost. In addition, we conducted site selection in which only suitable land polygons were included as candidate sites. Our goal was to represent a percentage of the total area occupied by each cover type in the network; this percentage ranged from 5% for widespread cover types to 20% for infrequent cover types. The method of selection was adaptive simulated annealing with iterative improvement.

Findings

A relatively small number of land polygons (576), occupying 10% of the ICRB land area, had positive suitability ratings. These land polygons were located primarily at higher elevations in the study area, including locations in southeast Oregon, southwest and central Idaho, the Rocky Mountains, and the Cascade range (Figure 4a). Cover type areas with positive suitability ranged from no suitable area (eleven cover types) to 47% (salt desert shrub). Most cover types lacking suitable land polygons were rare in the ICRB. Suitability ratings are shown for three of the 35 cover types (Figure 4b), a widespread type (interior Douglas-fir, occupying 8% of the ICRB), an intermediate type (whitebark pine, 1%), and an infrequent type (alpine tundra, 0.1 %). Cover type areas that were suitable comprised 8% for interior Douglas-fir, 29% for whitebark pine, and 34% for alpine tundra.

Figure 4. Suitability ratings for (a) all land polygons, and (b) three vegetation cover types.

The KBS comprises a formal representation of attributes that contribute important information concerning suitability. However, not all land polygon attributes may be available for a particular application. We varied land polygon attribute availability to determine the sensitivity of suitability to the absence of data. The largest differences in suitability were observed when land use condition or road density were unavailable; smaller effects on suitability occurred when vegetation or disturbance regime departure were unavailable.

We compared four cost scenarios for selecting regional conservation networks with existing conservation areas in the ICRB (Table 1). All cover types were represented in the scenarios except scenario 4, in which only suitable polygons were considered for selection. The number of land polygons selected varied widely among scenarios. Using equal costs or unsuitability as cost resulted in a small number of relatively large land polygons; using area as cost produced a network containing many small polygons. The equal costs and area as cost scenarios incorporated the smallest percentages of suitable area. A relatively high proportion of the suitable area in the ICRB (49%) is in existing conservation areas. However, 59% of these areas are rated as unsuitable for conservation.

Discussion and conclusions

The concentration of suitable land polygons in higher elevation areas was not unexpected, because such mountainous areas are less likely than lower elevation areas to have experienced intensive human impacts, are more likely to be in public ownership, and are therefore more likely to have attributes that are compatible with conservation. When area was not a cost consideration in the site selection process, networks contained large, diverse polygons whose actual monetary value may be prohibitively high. The total conservation area required was reduced by nearly half when area was used as cost, but many unsuitable polygons were included in the network. A next step is development of a cost measure that combines area and unsuitability to address tradeoffs between the two. Implementation of a method such as the KBS for evaluating conservation suitability can focus attention on those land areas that are more likely to sustain the targets of conservation over time. Areas can also be identified in which restoration or stewardship activities may be required to improve conservation opportunities.

This is the first application of such a KBS model for conservation planning. The strengths of the KBS are its ability to integrate current knowledge and available data concerning a conservation target of interest in an explicit and flexible manner, as well as its ability to provide visualization of the spatial implications of decisions as a platform for discussion and negotiation with land managers and planners. Great flexibility in constructing and modifying model structure is conferred by the object-based representation of networks in NetWeaver. Output can be examined for all levels in the KBS hierarchy. Changes in a regional conservation network can be implemented to accommodate comments and reviews by specifying the inclusion or exclusion of particular land polygons in the selection process. In addition, the analyses presented can be conducted at multiple scales. For example, a regional-scale analysis can be conducted with regional coarse-grained data to prioritize sites, which can then be further examined with fine-grained data not available regionally.

Acknowledgements

This work was supported by a grant from the US Environmental Protection Agency ("Multi-scaled Assessment Methods: Prototype Development within the Interior Columbia Basin").

References used

Andelman S, Ball I, Davis F, Stoms D. 1999. Sites v 1.0: an analytical toolbox for designing ecoregional conservation portfolios. Santa Barbara, CA: University of California. Manual prepared for The Nature Conservancy.

Austin MP, Margules CR. 1986. Assessing representativeness. In: Usher MB, editor. Wildlife conservation evaluation. London: Chapman and Hall Ltd. p 45-67.

Bourgeron PB, Humphries HC, Jensen ME. Representativeness assessments. In: Bourgeron PS, Jensen ME, editors. An integrated ecological assessment protocols guidebook. New York: Springer-Verlag. In Press.

Church RL, Stoms DM, Davis FW. 1996. Reserve selection as a maximal covering location problem. Biological Conservation 76:105-12.

Csuti BS, Polansky PH, Williams RL, Pressey JD, Camm M, Kershaw AR, Kiester B, Downs R, Hamilton M, Huso M, Sahr K. 1997. A comparison of reserve selection algorithms using data on terrestrial vertebrates in Oregon. Biological Conservation 80:83-97.

Flather CH, Wilson KR, Dean DJ, McComb WC. 1997. Identifying gaps in conservation networks: of indicators and uncertainty in geographic-based analyses. Ecological Applications 7:531-42.

Hann WJ, Jones JL, Karl MG, Hessburg PF, Keane RE, Long DG, Menakis JP, McNicoll CH, Leonard SG, Gravenmier RA, Smith BG. 1997. Landscape dynamics of the basin. In: Quigley TM, Arbelbide SJ, editors. An assessment of ecosystem components in the interior Columbia basin and portions of the Klamath and Great Basins: volume II. Portland, OR: US Department of Agriculture, Forest Service. General Technical Report PNW-GTR-405. p 337-1055.

Kiester AR, Scott JM, Csuti B, Noss RF, Butterfield B, Sahr K, White D. 1996 Conservation prioritization using GAP data. Conservation Biology 10:1332-42.

Margules CR, Pressey RL. 2000. Systematic conservation planning. Nature 405:243-53.

Noss RF, Cooperrider A. 1994. Saving nature's legacy: protecting and restoring biodiversity. Washington, DC: Defenders of Wildlife and Island Press.

Poiani KA, Richter BD, Anderson MG, Richter HE. 2000. Biodiversity conservation at multiple scales: functional sites, landscapes, and networks. BioScience 50:133-46.

Pressey RL, Humphries CJ, Margules CR, Vane-Wright RI, Williams PH. 1993. Beyond opportunism: key principles for systematic reserve selection. Trends in Ecology and Evolution 8:124-28.

Pressey RL, Possingham HP, Margules, CR. 1996. Optimality in reserve selection algorithms: when does it matter and how much? Biological Conservation 76:259-67.

Quigley TM, Arbelbide SJ, editors. 1997. An assessment of ecosystem components in the interior Columbia basin and portions of the Klamath and Great Basins: volume I. Portland, OR: US Department of Agriculture, Forest Service. General Technical Report PNW-GTR-405.

Reid MS, Bourgeron PS, Humphries HC, Jensen ME, editors. 1995. Documentation of the modeling of potential vegetation at three spatial scales using biophysical settings in the Columbia River basin assessment area. Boulder, CO: The Nature Conservancy, Western Heritage Task Force. Unpublished report prepared for the US Department of Agriculture, Forest Service, Contract 53-04H1-6890.

Reynolds K, Slade J, Saunders M, Miller B. 1997. NetWeaver for EMDS version 1.0 user guide: A knowledge base development system. Portland, OR: US Department of Agriculture, Forest Service, Pacific Northwest Research Station.

Scott JM, Davis F, Csuti B, Noss R, Butterfield B, Groves C, Anderson H, Caicco S, D'Erchia F, Edwards TC, Ulliman J, Wright RG. 1993. Gap analysis: a geographic approach to protection of biological diversity. Wildlife Monographs 123:1-41.

Authors

Patrick S. Bourgeron and Hope C. Humphries, Research Associates,
Institute of Arctic and Alpine Research
University of Colorado, 1560 30th St., Boulder, Colorado, USA 80309.
Email: patrick.bourgeron@colorado.edu, Tel.: 303-492-2841, Fax: 303-492-6388
Email: hope.humphries@colorado.edu, Tel.: 303-492-2594, Fax: 303-492-6388

Keith M. Reynolds, Research Forester, Forestry Sciences Laboratory
Pacific Northwest Research Station, 3200 SW Jefferson Way, Corvallis, Oregon, USA 97331.
Email: reynoldsk@fsl.orst.edu, Tel.: 541-750-7434, Fax: 541-750-7329