METALAND: Characterizing Spatial Patterns and Statistical Context of Landscape Metrics

Transcription

1 METALAND: Characterizing Spatial Patterns and Statistical Context of Landscape Metrics Author(s): JEFFREY CARDILLE, MONICA TURNER, MURRAY CLAYTON, SARAH GERGEL, and SETH PRICE Source: BioScience, 55(11): Published By: American Institute of Biological Sciences DOI: / (2005)055[0983:MCSPAS]2.0.CO;2 URL: %5D2.0.CO%3B2 BioOne ( is a a nonprofit, online aggregation of core research in the biological, ecological, and environmental sciences. BioOne provides a sustainable online platform for over 170 journals and books published by nonprofit societies, associations, museums, institutions, and presses. Your use of this PDF, the BioOne Web site, and all posted and associated content indicates your acceptance of BioOne s Terms of Use, available at Usage of BioOne content is strictly limited to personal, educational, and non-commercial use. Commercial inquiries or rights and permissions requests should be directed to the individual publisher as copyright holder. BioOne sees sustainable scholarly publishing as an inherently collaborative enterprise connecting authors, nonprofit publishers, academic institutions, research libraries, and research funders in the common goal of maximizing access to critical research.

2 METALAND: Characterizing Spatial Patterns and Statistical Context of Landscape Metrics JEFFREY CARDILLE, MONICA TURNER, MURRAY CLAYTON, SARAH GERGEL, AND SETH PRICE In ecosystem science, understanding the link between spatial heterogeneity and ecological processes is an active area of current research that requires repeatable, quantifiable methods of comparison. Our research has suggested that interpreting landscape pattern measures across large, contiguous areas can improve our understanding of the statistical and spatial properties of these measures, and can suggest links between patterns and processes. In this paper, we introduce METALAND, a publicly available software tool and attendant database for the research community s use. In two applications, we illustrate how this framework can be employed (a) to establish a statistical regional context for a given landscape and (b) to assist sampling design and hypothesis generation at the regional scale. We offer this toolbox and its large and growing set of intercomparable landscapes to aid ecologists who wish to understand the sources and patterns of spatial variability in ecosystems across large areas. Keywords: landscape ecology, spatial pattern, region, distribution, significance Understanding the causes and consequences of spatial heterogeneity is an important focus of research at all levels of ecological study. In ecosystem science, the interaction between spatial heterogeneity and ecological processes is an active area of current research (Chapin et al. 2002, Reiners and Driese 2004, Lovett et al. 2005, Turner and Cardille forthcoming) that requires methods to describe and quantify spatial patterns. The development of such methods has been rapid since the 1980s (Romme 1982, O Neill et al. 1988, Baker and Cai 1992, McGarigal and Marks 1995, Wickham and Riitters 1995), as spatial data, another key component of such studies, have become widely available. As a result, many ecological studies now routinely incorporate a spatial component in which aspects of landscape composition and configuration are hypothesized to influence ecosystem or population processes (Jones et al. 2001, McGarigal and Cushman 2002, With et al. 2002, Holland et al. 2004). The growing implementation of many broadscale ecosystem management projects also necessitates a quantitative understanding of spatial dynamics (Boutin and Hebert 2002, Liu and Taylor 2002, Wickham et al. 2002). However, despite the widespread application of methods to quantify spatial pattern, and several well-known caveats about their appropriate use (Gustafson 1998, Turner et al. 2001, Cardille and Turner 2002, Fortin et al. 2003, Li and Wu 2004), the application of landscape metrics remains problematic. Several authors have reviewed current methods of spatial analysis and discussed a variety of their pros and cons (Haines-Young and Chopping 1996, Gustafson 1998, Hargis et al. 1998, O Neill et al. 1999). In part because of the inherent limitations of indices, the conceptual flaws of pattern analyses (Li and Wu 2004), and a lack of rigorous statistical and geographic context (Turner et al. 2001, Wu and Hobbs 2002), there are a large number of metrics whose ability to predictably relate observed spatial patterns to real-world ecological processes is rightly suspect (Hess and Bay 1997, Hargis et al. 1998, Tischendorf 2001, Bogaert 2003). To address these difficulties in linking pattern and process, an ambitious direction in recent research has been to focus on the behavior of landscape measures in controlled conditions. In these studies, representative processes are simulated in replicated computer-generated landscapes and then analyzed (Hargis et al. 1998, Fortin et al. 2003, Remmel and Csillag 2003). These experiments continue to yield highly useful results and can be thought of as pattern-analysis laboratory experiments, in which cause and effect can be carefully monitored. Yet as in other scientific endeavors, real-world field studies are also important to the generation of hypotheses and advance of scientific knowledge. Jeffrey Cardille ( cardille@wisc.edu) and Monica Turner work in the Department of Zoology, and Murray Clayton works in the Department of Statistics and the Department of Plant Pathology, at the University of Wisconsin, Madison, WI Sarah Gergel works in the Department of Forest Sciences and the Centre for Applied Conservation Research, University of British Columbia, Vancouver, BC V6T 1Z4, Canada. Seth Price works at the Center for Sustainability and the Global Environment, University of Wisconsin, Madison, WI American Institute of Biological Sciences. November 2005 / Vol. 55 No. 11 BioScience 983

3 What, then, is an ecologist curious about the relationship between landscape pattern and ecological process in the field to do? Despite the caveats surrounding landscape metrics, environmental changes and increased access to satellite-based land-cover and land-use data will continue to encourage ecologists, students, managers, and researchers to ask the basic questions characterizing scientific inquiry of many kinds: What is the pattern I see? How is it different from others I have seen? In what ways is it similar? Has this pattern changed through time? For landscape metrics to be used well by nonexperts, they need to provide answers to these questions from an appropriate context of expected metric values and the locations where those values occur. As noted throughout the literature on landscape metrics (Li and Wu 2004), drawing general inferences from a single observation is fraught with danger, yet relating landscape patterns to ecological processes is a recurrent and ongoing priority for landscape ecology (Wu and Hobbs 2002). We believe that much of the improper or naive use of landscape metrics stems from a typical user s practical inability to establish the appropriate context for inferring relationships between pattern and process. Establishing relevant context for landscapes is difficult for experienced users as well: though forest fragmentation and other properties have been analyzed across the United States using landscape metrics (Riitters et al. 2000, 2002, 2004), it remains difficult to interpret such results to provide substantial context for inference using the frequency distribution of metrics in relevant subsets of landscapes (but see Remmel and Csillag 2003). For more casual users who might wish to explore the relation of landscape pattern to ecological process, the effort required to gain landscape-level context for a study area is likely to be prohibitive, and might involve, for example, the difficult task of locating and analyzing hundreds of landscapes with similar properties, or generating and analyzing landscapes with such properties. The end result, we believe, is that many researchers curious about landscape pattern have no practical option beyond running powerful pattern-measuring software such as FRAGSTATS (McGarigal and Marks 1995) once or twice and reporting the results for a few landscape and metric-calculation settings. Thus an understanding of the frequency distribution and spatial distribution of the large suite of metric values remains elusive for all but the most experienced researchers. To advance the ability of researchers studying real-world landscapes to understand the readily measurable properties of landscapes, we have developed METALAND, a publicly available software tool and attendant database for the research community s use. METALAND adds value to existing metric-computing software by assembling, updating, and providing public access to observed values for a large set of pattern measures in frequent use by ecologists. As a complement to efforts to understand landscape patterns in controlled, simulated landscapes (Gustafson and Parker 1992, Gardner 1999, Fortin et al. 2003, Remmel and Csillag 2003, Neel et al. 2004), METALAND is designed to provide the research and management community with a flexible, expandable interface to real-world landscapes and their spatial patterns. METALAND allows users to explore pattern measures from commonly used land-cover/land-use data sets at any of several grains, extents, and classification schemes. The METALAND interface and growing database have been developed to provide researchers with several capabilities: Additional context for inference from pattern measures in real-world landscapes A sandbox to help generate hypotheses about links between pattern and process across large areas A tool for the rapid and flexible selection of landscapes with particular landscape patterns The ability to test the behavior of peer-reviewed metrics in real landscapes Real-world context for a suite of metric values, using an empirically derived frequency distribution, in data sets commonly used by ecologists and managers A platform for testing the information content of a new metric proposed as an addition to the already large suite of peer-reviewed measures Designed as a living library of landscapes and their calculated landscape pattern measures, the quantitative framework of METALAND is a tool both for those wishing to interpret metric values and for those designing studies to test hypotheses relating spatial patterns to ecological responses. Understanding the realized values of pattern metrics in real-world landscapes will enable researchers to develop a deeper understanding of the relationship between pattern and process. By providing a priori knowledge about spatial and statistical patterns in observed landscapes for commonly used data sources, we hope that this toolbox will assist the ecological community s understanding of the locations, causes, and consequences of changing land-cover patterns across large areas. The METALAND database We used the National Land Cover Data (NLCD) data set for 1992 to illustrate the behavior and analysis of landscape measures across large areas. The NLCD product presents, at 30- meter (m) spatial resolution, a seamless land-cover assessment of the continental United States developed from a combination of satellite data and ancillary data sources (Vogelmann et al. 2001). In a method similar to that employed by Cain and colleagues (1997), we demonstrated the analytical potential of the database by partitioning the data set for the state of Wisconsin into a series of regularly spaced, nonoverlapping, same-sized square landscapes of 216 cells by 216 cells. This regular partitioning created samples with an extent ( kilometers [km]) large enough to minimize unwanted edge effects and suitable for subsequent analyses of the effects of changes in grain size on landscape metric values. Landscapes that did not fully lie within the borders of the state were reserved for later analyses. This resulted in a set of BioScience November 2005 / Vol. 55 No. 11

4 regularly spaced square landscapes across the state, with 30- m pixels and a maximum of 21 categories of land cover and land use (figure 1). Using FRAGSTATS 2.0 (McGarigal and Marks 1995), we computed 82 landscape-level metrics, using the eightneighbor rule, from each of the major metric categories: area/density/edge, shape, core area, isolation/proximity, contrast, contagion/interspersion, connectivity, and diversity. Two class metrics were also calculated for each landscape to illustrate class-level behavior: (1) proportion of deciduous forest and (2) number of patches of deciduous forest. Each square landscape was assigned a unique identifier to map back to a geographic information system (GIS), which makes it possible to consider the spatial distribution of values of landscapelevel and class-level metrics, as was done for ecologically meaningful land units by Heilman and colleagues (2002), for overlapping windows by Riitters and colleagues (2004), and for nonoverlapping tiles by Cain and colleagues (1997). The assembly and analysis of these data derived from the NLCD have been completed for more than 100 scales (as defined by a particular combination of grain, extent, and classification scheme) for the state of Wisconsin. This analysis and addition to the database is ongoing for the entire United States at the native scale of the NLCD (i.e., 30-m pixels, 21 categories), with the extent chosen by us as 216 cells 216 cells. Although the contents of the METADATA database currently exist only for the United States, plans are under way to extend this conceptual basis to other countries with comprehensive landcover/land-use classifications. Automatic updates of the progress of analysis are provided through the METALAND interface, and additional future plans for the United States include analysis, storage, and presentation of values in coincident landscape subsets for the 2001 NLCD. The METALAND interface The interface for METALAND is publicly available online ( where it is possible to select and download landscape patterns and information using a variety of geographic and numeric criteria. The interface allows the real-time selection of landscapes based initially on a variety of derived data sets, by specifying (a) the coordinates limiting the area of interest; (b) rectangular limits based on geographic criteria; (c) any method employed (e.g., majority rule, center pixel), where appropriate, to increase the grain of pixels in the landscape from its original form; (d) any of four extents; and (e) any of three classification schemes, including a binary forest/nonforest interpretation of the NLCD data. After identifying a data set in this manner, users can then select landscapes for further study from that set, using any combination of the landscape metric criteria described above. Users who have identified a set of landscapes in this way have immediate access to an image of the frequency distribution of any of the computed metrics, providing a practical opportunity for statistical context for users trying to determine the relative rarity of a particular landscape metric value (within the subset Figure 1. Centers of landscapes used for study; the region is partitioned into 3118 equal-sized landscapes, each with an extent of 6.48 kilometers (km) 6.48 km. The six numbered locations are cited in the text of the article: (1) Wisconsin River, (2) Driftless Area, (3) Madison, (4) Milwaukee, (5) Chippewa and Taylor Counties, and (6) Lake Winnebago. of landscapes selected by the user) for a landscape under study. The simplest part of the interface to the METALAND database is the ability to view and retrieve landscape metric values for a selected set of target landscapes (figure 2). Users can employ the interface to download landscape-level and class-level metric values and multiple formats of any selected landscapes, including ArcExport, GeoTiff, or PNG (portable network graphics) format. Application 1: Establish basic statistical and spatial context for a target landscape METALAND provides the beginnings of the statistical and spatial context needed for users of real-world land-cover data to make better inferences relating pattern and process. With the large set of values assembled here (n = 3118), the statistical properties of metrics across a large region can be quickly and flexibly considered for any computed landscape metric. The number of patches, for example, in Wisconsin landscapes (of the classification, grain, and extent described earlier) has a mean of 2229 and a standard deviation of 712, with a simple frequency distribution suitable for statistical tests (figure 3b). Employing the regular partitioning of the study area to display metric values in the GIS (figure 3a) showed that variation around the mean number of patches is not distributed arbitrarily in space, but instead reveals prominent features of the biophysical and human landscape, including (a) the patchiness produced by the Wisconsin River within the relatively unbroken land cover of the Driftless Area, (b) the highly dissected landscapes of the major urban areas of Madi- November 2005 / Vol. 55 No. 11 BioScience 985

5 Figure 2. Main elements of the METALAND user interface for displaying results of landscape and metric searches, including orientation map, display of landscape, metric values (only four landscape-level and four class-level metrics are shown here), and real-time download options. son and Milwaukee, and (c) the many tightly interlaced patches of pasture, row crops, wetlands, and forest of Chippewa and Taylor Counties (figures 1, 3). Such state-based sets of values, however, may be far from ecologically relevant. The METALAND interface allows users to flexibly query the database of metrics and landscape using metric values near those for a target landscape say, the boyhood home of John Muir ( degrees [ ] longitude, north latitude). According to the NLCD land-cover classification, the surrounding 6.48-km 6.48-km landscape has 2000 patches, with a Shannon evenness of 0.56, and with deciduous forest covering 37.5 percent of the area. The interface can be used to select subsets of landscapes based on these or other criteria, and provide finer context by orienting the landscape along gradients that may be ecologically relevant to the user s analysis. For example, if a user wishes to consider only landscapes having a proportion of deciduous forest between 30 and 50 percent, with the total edge metric in the upper 60 percent among Wisconsin landscapes, the number of patches in the Muir landscape is located at the far left tail within that that 435-landscape subset, different from its location in the 39th percentile (figure 3b, 3c) along the same gradient among landscapes of the entire state. Users may consider selecting subsets using any value-based criteria, such as Jenks optimization (as used by Heilman and colleagues [2002] to produce ordinal categories from a continuous histogram of metric values displayed in space). The flexibility of the interface permits the viewing of both metric maps and the frequency distributions of any metric for subsets of landscapes chosen in real time. Application 2: Use of statistical and spatial context to aid experimental design and development of hypotheses The statistical and spatial context provided by the set of intercomparable landscapes provides a new tool for ecologists 986 BioScience November 2005 / Vol. 55 No. 11

6 Figure 4. Subsets of landscapes selected using landscape metric values derived across the entire state: location of (a) landscapes having the 80th to 100th percentile of total edge and the 60th to 80th percentile of proportion of deciduous forest, and (b) the intersection of that set with the set having the 40th to 60th percentile of Shannon evenness values in the state. Figure 3. (a) Spatial pattern of number-of-patches metric across Wisconsin, for landscapes 6.48 kilometers (km) 6.48 km (top), and histograms showing the frequency of values for the number of patches metric for (b) the full set of landscapes and (c) the subset of landscapes with deciduous forest covering between 30 and 50 percent of the area and with total edge metric in the upper 60 percent among this compiled set of Wisconsin landscapes. designing experiments over large extents. Consider a hypothesis that has a spatial component for example, the hypothesis that areas with moderate to high measures of total edge near certain forest types are likely to have more frequent outbreaks of Lyme disease (e.g., Guerra et al. 2002, Ostfeld et al. 2005). It is straightforward to use computed percentile information to query the assembled data to locate and select those landscapes having certain characteristics (e.g., the 40th to 60th percentile of the total-edge metric, and 40 to 60 percent coverage by deciduous forest; figure 4a), or to select landscapes with those same criteria intersected with those of a third or fourth criterion (figure 4b). A map of the locations of those landscapes that have the selected properties might then be used in a number of ways when designing experiments: To identify the number and location of potential study sites To rapidly assess whether the locations are appropriately distributed through the study region To contrast with the locations of landscapes satisfying other metric-based selection criteria To estimate driving distances (or other measures of sampling effort) among proposed subsets of sites To generate further hypotheses about the factors influencing the observed spatial distribution of landscapes having the selected attributes Conclusion The vast increase in spatial data across time and space, along with growing interest in spatial pattern and process in ecosystems, makes devising tools for the interpretation of realworld spatial pattern increasingly necessary. As these data accumulate to form part of the record of the interactions among abiotic, biotic, and human processes, ecologists need methods to compare and contrast measures of landscape heterogeneity across multiple spatial and temporal scales. The partition of geographic space into a regular framework of equal-sized landscapes allows ecologists to analyze, both statistically and spatially, variation in measures of landscape pattern. This approach is intended to give ecologists, and users of their results, additional tools for presenting landscape measures in a way that can be more readily understood and rigorously interpreted. Like most researchers, we do not think that landscape metrics, as currently understood, are a panacea for ecological analysis. Instead, the inability of potential users to establish the spatial or statistical context of real-world landscapes appears to be a major factor inhibiting a full exploration of and experimentation with the ever growing library of satellite-derived land-cover data. When users are provided with the ability to establish spatial and numerical context, new hypotheses about pattern and process may emerge, the creation of statistically uninformative new metrics may be inhibited, and managers will be able to quickly, repeatably, and flexibly identify study sites based on real-world pattern criteria. For these reasons, we encourage the free use of and contribution to METALAND, a framework and repository from which as yet unanticipated broadscale context and hypotheses may emerge. November 2005 / Vol. 55 No. 11 BioScience 987

7 Acknowledgments We thank the developers of FRAGSTATS, without whose generous efforts on behalf of the ecological community this project would not be possible. We similarly thank the developers of the NLCD and other land cover data sets, which have contributed so much to scientific understanding. We also appreciate the encouragement and constructive feedback provided by members of the Turner laboratory on the content and approach described here. Three anonymous reviewers provided clear and extremely helpful comments. We also acknowledge funding from the National Science Foundation and the Andrew W. Mellon Foundation in support of this work. References cited Baker WL, Cai YM The r.le programs for multiscale analysis of landscape structure using the GRASS geographical information system. Landscape Ecology 7: Bogaert J Response to Bissonette JA and Storch I. 2002: Fragmentation: Is the message clear? Lack of agreement on fragmentation metrics blurs correspondence between fragmentation experiments and predicted effects. Conservation Ecology 7: r6. Boutin S, Hebert D Landscape ecology and forest management: Developing an effective partnership. Ecological Applications 12: Cain DH, Riitters K, Orvis K A multi-scale analysis of landscape statistics. Landscape Ecology 12: Cardille J, Turner MG Understanding landscape metrics I. Pages in Gergel SE, Turner MG, eds. Learning Landscape Ecology: A Practical Guide to Concepts and Techniques. New York: Springer. Chapin FS, Matson PA, Mooney HA Principles of terrestrial ecosystem ecology. New York: Springer. Fortin MJ, Boots B, Csillag F, Remmel TK On the role of spatial stochastic models in understanding landscape indices in ecology. Oikos 102: Gardner RH RULE: Map generation and a spatial analysis program. Pages in Klopatek JM, Gardner RH, eds. Landscape Ecological Analysis: Issues and Applications. New York: Springer. Guerra M, Walker E, Jones C, Paskewitz S, Cortinas MR, Stancil A, Beck L, Bobo M, Kitron U Predicting the risk of Lyme disease: Habitat suitability for Ixodes scapularis in the north central United States. Emerging Infectious Diseases 8: Gustafson EJ Quantifying landscape spatial pattern: What is the state of the art? Ecosystems 1: Gustafson EJ, Parker GR Relationships between landcover proportion and indexes of landscape spatial pattern. Landscape Ecology 7: Haines-Young R, Chopping M Quantifying landscape structure: A review of landscape indices and their application to forested landscapes. Progress in Physical Geography 20: Hargis CD, Bissonette JA, David JL The behavior of landscape metrics commonly used in the study of habitat fragmentation. Landscape Ecology 13: Heilman GE Jr, Strittholt JR, Slosser NC, DellaSala DA Forest fragmentation of the conterminous United States: Assessing forest intactness through road density and spatial characteristics. BioScience 52: Hess GR, Bay JM Generating confidence intervals for compositionbased landscape indexes. Landscape Ecology 12: Holland JD, Bert DG, Fahrig L Determining the spatial scale of species response to habitat. BioScience 54: Jones KB, Neale AC, Nash MS, Van Remortel RD, Wickham JD, Riitters KH, O Neill RV Predicting nutrient and sediment loadings to streams from landscape metrics: A multiple watershed study from the United States mid-atlantic region. Landscape Ecology 16: Li HB, Wu JG Use and misuse of landscape indices. Landscape Ecology 19: Liu J, Taylor WW Integrating Landscape Ecology into Natural Resource Management. Cambridge (United Kingdom): Cambridge University Press. Lovett GM, Jones CG, Turner MG, Weathers KC, eds Ecosystem Function in Heterogeneous Landscapes. New York: Springer-Verlag. McGarigal K, Cushman SA Comparative evaluation of experimental approaches to the study of habitat fragmentation effects. Ecological Applications 12: McGarigal K, Marks BJ FRAGSTATS: Spatial pattern analysis for quantifying landscape structure. Portland (OR): USDA Forest Service, Pacific Northwest Research Station. Neel MC, McGarigal K, Cushman SA Behavior of class-level landscape metrics across gradients of class aggregation and area. Landscape Ecology 19: O Neill RV, et al Indices of landscape pattern. Landscape Ecology 1: O Neill RV, Riitters KH, Wickham JD, Jones KB Landscape pattern metrics and regional assessment. Ecosystem Health 5: Ostfeld RS, Glass GE, Keesing F Spatial epidemiology: An emerging (or re-emerging) discipline. Trends in Ecology and Evolution 20: Reiners WA, Driese KL Transport Processes in Nature: Propagation of Ecological Influences through Environmental Space. Cambridge (United Kingdom): Cambridge University Press. Remmel TK, Csillag F When are two landscape pattern indices significantly different? Journal of Geographical Systems 5: Riitters KH, Wickham JD,Vogelmann JE, Jones KB National land-cover pattern data. Ecology 81: 604. Riitters KH, Wickham JD, O Neill RV, Jones KB, Smith ER, Coulston JW, Wade TG, Smith JH Fragmentation of continental United States forests. Ecosystems 5: Riitters KH, Wickham JD, Coulston JW A preliminary assessment of Montreal Process indicators of forest fragmentation for the United States. Environmental Monitoring and Assessment 91: Romme WH Fire and landscape diversity in subalpine forests of Yellowstone National Park. Ecological Monographs 52: Tischendorf L Can landscape indices predict ecological processes consistently? Landscape Ecology 16: Turner MG, Cardille J. Spatial heterogeneity and ecosystem processes. In Wu JG, Hobbs R, eds. Key Research Directions in Landscape Ecology. Darwin (Australia): IALE World Congress. Forthcoming. Turner MG, Gardner RH, O Neill RV Landscape Ecology in Theory and Practice: Pattern and Process. New York: Springer. Vogelmann JE, Howard SM, Yang LM, Larson CR, Wylie BK, Van Driel N Completion of the 1990s National Land Cover Data set for the conterminous United States from Landsat Thematic Mapper data and ancillary data sources. Photogrammetric Engineering and Remote Sensing 67: Wickham JD, Riitters KH Sensitivity of landscape metrics to pixel size. International Journal of Remote Sensing 16: Wickham JD, O Neill RV, Riitters KH, Smith ER, Wade TG, Jones KB Geographic targeting of increases in nutrient export due to future urbanization. Ecological Applications 12: With KA, Pavuk DM, Worchuck JL, Oates RK, Fisher JL Threshold effects of landscape structure on biological control in agroecosystems. Ecological Applications 12: Wu JG, Hobbs R Key issues and research priorities in landscape ecology: An idiosyncratic synthesis. Landscape Ecology 17: BioScience November 2005 / Vol. 55 No. 11