Grantee Research Project Results
Final Report: Methodologies for Extrapolating from Local to Regional Ecosystem Scales: Scaling Functions and Thresholds in Animal Responses to Landscape Pattern and Land Use
EPA Grant Number: R826764Title: Methodologies for Extrapolating from Local to Regional Ecosystem Scales: Scaling Functions and Thresholds in Animal Responses to Landscape Pattern and Land Use
Investigators: Wiens, John A. , Horne, Beatrice Van
Institution: Colorado State University
EPA Project Officer: Hahn, Intaek
Project Period: October 1, 1998 through September 30, 2001 (Extended to December 31, 2002)
Project Amount: $581,519
RFA: Regional Scale Analysis and Assessment (1998) RFA Text | Recipients Lists
Research Category: Aquatic Ecosystems , Ecological Indicators/Assessment/Restoration
Objective:
In simple terms, the objectives of this research project were to: (1) develop ways of assessing the scaling of landscape patterns and of organism responses to landscape variation; and (2) develop predictive models that can be used to extrapolate landscape-organism linkages among scales. In this context, “scaling” refers to the ways in which ecological patterns or processes change with changes in the spatial scale on which they are viewed. The results of the research can contribute to a methodology for extrapolating among scales of analysis or management and for evaluating the scales at which environmental disturbances or anthropogenic stresses may have the greatest impacts on ecological systems.
In practice, these objectives translate into three questions that guided this research: (1) How do we develop ways to extrapolate among scales, particularly in the face of scaling nonlinearities? (2) Can landscape measures taken over multiple scales be used to predict the scaling properties of communities or ecosystems and thus serve as a basis for deriving extrapolation algorithms? and (3) Do different taxonomic or functional groups of organisms respond differently to broad-scale gradients in environmental or landscape features?
Summary/Accomplishments (Outputs/Outcomes):
Analysis of Existing Data Sets. To explore how landscape patterns and the distributional patterns of organisms relate to each other and how these relationships change with changes in scale, we evaluated three existing data sets, using somewhat different methodologies. One analysis considered the distribution of grasshopper species and communities in relation to land cover, land use, soil, topography, temperature, and precipitation patterns in eastern Wyoming. One thrust of this approach was to contrast the effects of viewing landscape features categorically versus continuously (i.e., as gradients analyzed by ordination procedures). The relationships that emerge using the two approaches differ with respect to geography, and it appears that ordination may provide a more informative framework for evaluating broad-scale environmental responses of organisms to landscape patterns.
A second analysis used information on vegetation coverage and bird distributions from surveys conducted in the Snake River Birds of Prey Conservation Area in Idaho. Here, the focus was on developing causal models of bird community-environment relationships and examining the sensitivity of such models to changes in scale. Vegetation composition was associated with bird species composition at the local scale, and its effect decreased somewhat as scale increased. Fire history had very little influence at a local scale, but its effect increased with increasing scale.
A third analysis focused on a detailed, spatially explicit data set on plant distributions available from the Oosting Experimental Forest in North Carolina. Here, the objective was to use statistical approaches based on spatial covariance to integrate variogram modeling and multiscale ordination to evaluate assembly rules in plant communities. The variogram matrix provides a framework for partitioning spatial covariance and for factoring out specific components. This mathematical approach greatly increases the interpretability of variograms of biotic communities, extends multiscale ordination to nonsystematic spatial samples, and provides a spatial extension and an empirical null model for the variance test of species richness.
Studies Along a Grassland Macrogradient. We used the insights gained in the first phase of this research to structure a series of field studies that incorporated multiscale analyses of landscape patterns and features with surveys of several biological taxa (vascular plants, beetles, butterflies, spiders, and birds), which also could be analyzed at multiple scales. The basic research design involved studies at five sites located on a broad-scale precipitation/land-use gradient from the shortgrass steppe of northeastern Colorado to the tallgrass prairie of eastern Kansas (see Figure 1).
At each site, surveys were conducted on two to four 2-km-long transects that spanned gradients in local topography and vegetation cover, centered on grassland and riparian cover types. During the summers of 2000 and 2001, information was gathered along each transect for environmental variables (soil hardness, soil pH, soil texture, near-surface temperature, and shrub coverage) and biotic variables (bird and butterfly surveys, pitfall trapping for ground-dwelling beetles and spiders, and quadrat sampling of vegetation). The minimum scale of resolution of these data (i.e., sampling grain) was 50 m.
Based on the 2000 community data, zones of interest were designated along two of the transects from each site in 2001. Within each of these zones, two 250-m transects were established that paralleled the main transect. These transects were sampled every 10 m to generate a minimum scale of resolution of 10 m.
Figure 1. Location of the Main Transect With Five Study Sites Along a Precipitation and Productivity Gradient in the Central Plains. Background image shows total Normalized Difference Vegetation Index (NDVI), a measure of seasonal productivity obtained from remote sensing data (dark shades indicate high productivity).
The information obtained from this direct sampling can be combined with indirect assays of landscape pattern and composition obtained from satellite imagery and aerial photography. The data obtained from this remote sensing can be used to assess how the different taxa respond to the surrounding landscape over a range of scales. They also can be used to determine whether such remote sensing, coupled with basic inventory and life-history information, may yield models that can be used to predict species and community responses to landscapes with changing scales and how these relationships vary across broad, geographic gradients. Analyses of the direct field data indicated that the five sites exhibited a reasonably even spacing on the broad-scale environmental gradient.
Scaling Properties of Grasslands in the Central Plains. Cross-site comparisons are often used to understand how ecological patterns and processes vary with environmental conditions. Sites are often contrasted in terms of differences in the mean values of environmental factors, assuming that their spatial structure remains constant. We examined changes in the nature and the scale of landscape patterns along a transect from the Shortgrass Steppe Long-Term Ecological Research (LTER) site in eastern Colorado to the tallgrass prairie at the Konza Prairie LTER site in eastern Kansas. Multiscale ordination of 30-m resolution spectral information (Landsat 7) for a random sample of grasslands along this transect revealed that both the relative importance of PCA factors and the scale of their pattern of variation changed systematically along the transect. Compared to a combination of landscape metrics applied to classified 30-m land-use/land-cover data, multiscale ordination of spectral bands revealed more systematic variation, especially at small scales and within a habitat type, and was less scale dependent. We concluded that cross-site comparisons should consider not only changes in the mean of environmental factors, but also their spatial structure, and that the spatial structure should be described in terms of gradients rather than artificial classifications.
Categorical Versus Gradient Analyses of Scale-Dependent Responses of Birds to Grassland Environments. Changes in community structure have often been associated with changes in landscape structure. For example, declines in grassland birds have been linked to changes in landscape structure as a result of the conversion of grassland for agricultural use. Changes in landscape structure are often evaluated based on classified images such as those produced by the National Land Cover Dataset, in which areas are classified into a predetermined set of categories based on vegetation type or land use. There are several drawbacks to using these classified images. For example, within-category heterogeneity is smoothed over, and all borders are forced to be discrete so there is no way to assess gradual transitions.
To address this problem, we used two sources of landscape structure data, one continuous (Landsat 7 images) and one categorical (Land-Use\Land-Cover images from the National Land Cover Dataset). A moving window analysis was conducted on each image in which a measure of landscape heterogeneity was calculated within each window. For the continuous data, the standard deviation of the gray scale values was computed within each window; for the categorical data, the Shannon Diversity Index was computed within each window. This was done at four window sizes (90 m x 90 m, 330 m x 330 m, 990 m x 990 m, and 3,150 m x 3,150 m) at the five sites described above.
The use of continuous images revealed more heterogeneity in landscape structure than the categorical images. This was especially prevalent in grassland areas. Continuous images also detected transitions that were gradual rather than discreet, as indicated by the categorical images. There was a distinct westward trend toward the expression of heterogeneity at broader scales.
Although the use of continuous images increased the amount of heterogeneity detected, it is important to determine if this additional heterogeneity is relevant in terms of organismal response. We examined the use of continuous and categorical images in explaining the grassland bird response to landscape structure. Moving window images were created from categorical images using the following landscape metrics: Shannon’s Diversity Index, Shannon’s Evenness Index, patch distance, contagion, edge distance, and mean Euclidean nearest neighbor distance. Moving window images were created from continuous images using Landsat bands 2, 5, and 7, NDVI, and Digital Elevation Models. Broad-scale bird point data were obtained from the National Breeding Bird Survey Program (BBS) and finer scale bird surveys were conducted to supplement BBS data. Bird response was measured in terms of species richness, total community abundance, and abundance of ubiquitous species. Around each bird survey point, a circular area was cut from the moving window image. Circles of three sizes were used to examine changes in the landscape structure/bird-response relationship at different scales. Partial linear regression was used to determine variance in bird response explained by continuous landscape structure, variance in bird response explained by categorical landscape structure, and the overlap in variance of bird response explained by both types of landscape structure.
In every case, the use of continuous landscape structure increased the amount of explained variance in bird-community response. The variance in bird response explained exclusively by continuous landscape structure ranged from 5 percent to 29 percent. Although the relationship between landscape structure and bird response varied with scale, there was no pattern to this variation. Scaling patterns may become clearer by dividing birds into functional groups.
These results have important implications for wildlife management as it relates to changes in landscape structure. The strength of the relationship between community structure and landscape structure may be underestimated if measures of landscape structure are based solely on classified images. The use of continuous landscape structure in determining the effects of landscape alteration on organisms can increase our understanding of human effects on wildlife and enable more effective predictions of how organisms may respond to these landscape alterations.
Scale-Dependent Patterns in Different Taxa Across a Regional Gradient. The assessment of scale dependence in the distribution and abundance of different taxa has involved three phases. The first focused on species-area relationships. The species-area curve has long been important in predicting the number of species in locations of differing size and heterogeneity. Our objective was to determine if area and heterogeneity are strong predictors of local species richness along transects across a major geographical gradient. Furthermore, we wanted to elucidate the impacts of measuring heterogeneity in a scale-dependent manner. Specifically, we examined two aspects of the species-area-habitat relationship: Can models of species richness based on area and heterogeneity of one location be used to predict the species accumulation curves of another location?, and How does the scale dependence of habitat heterogeneity influence models of species richness?
To answer these questions, we examined three taxonomic groups: beetles, birds, and plants, as well as the combination of these three groups, which was used as a surrogate for total biodiversity. These taxonomic groups were sampled along 2-km transects at five sites. To measure heterogeneity, we calculated spectral variance of Landsat imagery within several window sizes. The species richness curves of each transect were modeled with area, heterogeneity, and the interaction of these two variables. These models were then used to predict the richness curves of other transects. In general, the models were poor predictors of species richness at other locations. Less than 25 percent of the richness curves of transects was predicted. The results of this analysis, however, showed that when examining species richness at the scale that monitoring and conservation efforts are usually conducted, heterogeneity is a better predictor of species richness than either area alone or interactions of area and heterogeneity.
Second, we asked how aspects of scale and geographic space are related to the structure of biotic communities. Understanding how different taxonomic groups respond to these factors at different locations can elucidate attributes of the communities themselves, as well as the scale at which monitoring should be conducted at a particular site. Much of community ecology has focused on understanding the influences of different factors on a community at a particular site. Our objective in this analysis was to compare the influences of scale and geographic space on three taxonomic groups at the five sites we sampled. Specifically, the objectives of this study were focused on three issues: (1) how the explanatory power of scale and geographic space change across a climatic gradient; (2) the differences in responses to scale and geographic space among three different taxonomic groups; and (3) how to combine hierarchical, multiscale approaches with spatial partitioning.
To elucidate these issues, we measured environmental variables at three scales at each of our sites and used these different scales of variables to explain the community structures of beetles, birds, and plants in these grassland habitats. The results of this analysis indicated that geographic space generally explained more variance in community structure in both plant and beetle communities than it did in bird communities. The research also suggested that beetles responded distinctly to the different scales of environmental variables. In other words, for the bird and plant communities, there was frequent overlap in the variance explained by different scales of environmental variables. This suggests that, to explain bird and plant community structure, it may only be necessary to measure environmental variables at one or two scales, whereas measuring additional scales of environmental variables may explain variance in beetle community composition that is unexplained by other scales.
Third, we used multiscale ordination to determine whether the spatial structure in the beetle, bird, and plant communities is due to the spatially structured environment (spatial dependence) or spatial biotic processes (autocorrelation). Determining how aspects of space and the environment are related to the structure of biotic communities is essential to managing biodiversity. Using direct multiscale ordination with canonical correspondence analysis (MSO), the spatial structure of these communities was examined. MSO indicated that the range of autocorrelation attributed to biotic processes for both plants and beetles changed across the moisture gradient, increasing from the short-grass steppe to the tallgrass prairie. This illustrates that the process-based differentiation between fundamentally different components of spatial structure can provide additional insights relevant to successful management.
Scaling Responses of Ground-Dwelling Spiders. Ground-dwelling arthropod communities respond to fine-scale variations in landscape structure. Lycosids, a family of ground-dwelling spiders, are generalist predators that do not build webs and rarely, if ever, leave the ground. These spiders are particularly sensitive to temperature changes at the soil surface. As a result, small variations in microclimate, created by fine-scale landscape heterogeneity, may play an important role in explaining the distributions of ground-dwelling spiders and the seasonal changes in those distributions. Furthermore, as seasonal climate changes create a more hostile thermal environment, these spiders may actively choose to hunt in cooler microclimates to extend foraging time and increase prey capture rates.
We have sampled the Lycosid community at several locations in the short-grass steppe of Colorado using a series of paired pitfall traps. Each pair of traps has one trap placed under the microcanopy created by short grasses and other vegetation, and the other in unshaded bare ground. Trapping frequency serves as a surrogate for relative activity time. Trapping methods are designed to distinguish diurnal and seasonal differences in activity. The results from field collections are being related to thermal tolerance and preferences for each species of lycosid. These data, in turn, will be related to the patterns of spider distribution and abundance obtained from the transect-based sampling in 2000-2001. These analyses are nearing completion.
Statistical Studies. The main objective of this part of the research project was to develop a methodology for modeling the coupling of the scaling of species’ responses to environmental variation to scale-dependent changes in landscape composition and spatial structure. Specifically, a method should be developed that allows the integrated analysis of patchiness in species distributions, multispecies patterns, and scale-dependent response to environmental factors. Simultaneously, the method should facilitate scaling among ecological levels of organization from populations over functional groups to community composition and species richness.
The main approach was a mathematical integration of formerly unrelated methods that deal with separate aspects of the overall problem: (1) geostatistical analysis reveals autocorrelation in a spatial sample; (2) the variance of species richness was used as an indicator for interspecific interactions due to niche limitation; (3) ordination techniques described multispecies responses to environmental factors; and (4) moving-window analysis can be used to identify the scale of response.
On a conceptual level, the result is the rephrasing of gradient analysis in a spatial paradigm (see Figure 1). Gradient analysis aims to explain the differences in species composition in a biotic community observed at different sampling locations. The structure of biotic communities is inherently spatial for two reasons. First, population dynamics and interspecific interactions operate through individual organisms that exist and interact only within their immediate neighborhood. Such contagious biotic processes create spatial autocorrelation within the community. Second, physical processes create spatial structure in environmental factors, which in turn, causes spatial dependence in the biotic community. Furthermore, the size of the ecological neighborhood, wherein organisms interact with other organisms and with their physical environment, depends on the particular ecological process, the time period, and the organism’s mobility or activity. Thus, the species-environment relationship depends on the scale of response of an organism (scale dependence). The recognition of the spatial nature of biotic communities requires an extension of the research paradigm of gradient analysis (see Figure 2). The internal structure of the biotic community (B) and its response to the environment (E) cannot be fully understood without considering geographic space (S).
Figure 2. Nonspatial (a) and Extended (b) Research Paradigm of Gradient Analysis (see text)
The method uses the fact that any variance-covariance matrix can be partitioned by distance if the spatial coordinates of the observations are known. The resulting set of distance-dependent variance-covariance matrices can be interpreted as an empirical variogram matrix, containing all species variograms and their cross-variograms, and it provides a spatial partitioning of multivariate methods such as regression, ordination, or the variance test of species richness. A diagnostic plot and statistical tests for residual autocorrelation, scale-dependence of the species-environment correlation, and interspecific associations of autocorrelated species variables were developed. The extension to chi-square distance facilitates the use of a unimodal response model. Further development will concentrate on accommodating nonlinear or threshold response.
The first applications of the method focused on published data sets that would serve as an illustration and benchmark for the proposed methods. Application to the Oosting data set of understory vegetation proved the need for a spatial variance test of species richness, as negative associations at short distances were canceled out by positive associations over larger distances. The example of an oribatid mite community of Lake Mead, Ontario, illustrated that the largest part of the spatial structure in a community may be due to spatial dependence. A third example of cattle habitat use in a wooded pasture of Metairie d’Evilard, Switzerland, highlighted the importance of accommodating multiple scales of response to the environment.
The new methodology is ideally suited for untangling the complex spatial structure of biotic communities caused by different processes operating at various, often overlapping scales. A simple spatial analysis of community data from a heterogeneous environment is likely to reveal spatial autocorrelation up to large distances. Assuming that the range of autocorrelation indicates the scale of internal organization in the community, the results may suggest that environmental disturbances and anthropogenic stress affect communities in a large neighborhood. The distinction between true spatial autocorrelation (as a result of biotic interactions within the community) and spatial dependence (as a result of the spatial structure of the environment) will often reveal a much smaller scale of organization in the community. However, management may still need to consider larger scale effects if the organisms respond to landscape structure at larger or multiple scales (scale dependence). Multiscale ordination facilitates the specification of an appropriate model of a locally autocorrelated, scale-dependent response to environmental gradients that can be used to predict animal distributions over a range of scales from geographic information systems and remote sensing data.
Conceptual Developments. A fundamental issue in the analysis of landscape structure, its relationship to biological responses, and the scale-sensitivity of these relationships is related to how landscape patterns are viewed, either as categories or as gradients. Remotely sensed images are increasingly being used to assess landscape and habitat structure, which, in turn, are used to assess biotic response. These images are classified into patches based on a predefined set of categories. There are many drawbacks to classifications; in particular, the smoothing over of within-patch heterogeneity, the creation of discrete boundaries in areas where transitions are more gradual, and the use of classifications that are based on human perception rather than the perspective of the focal organisms. Rather than use these categorical maps, we have developed a method in which the spectral reflectance of the bands of Landsat images can be used to create a gradient map. Figure 3 illustrates the general protocols for comparing categorical classifications with spectral gradient analyses.
Our results show that the gradient approach enhances much of the finer scale variability as well as gradual transition zones. In addition, scale dependence of landscape structure appears to change when using gradient analysis rather than categorical classifications. An east-to-west trend in the scaling of landscape structure is evident along our grassland macrogradient.
As the capstone of this project, we are preparing a paper that synthesizes our findings about scale dependence of landscape patterns and the responses of different taxa to these patterns in the context of the categorical versus gradient views. This perspective then will be integrated with the statistical work, and the overall synthesis related to management issues in these grassland ecosystems and the use of remote sensing in gauging scale dependence of ecological systems.
Figure 3. Use of Landsat Imagery to Conduct a Categorical (left) and a Gradient (right) Analysis of Landscape Structure.
Journal Articles on this Report : 5 Displayed | Download in RIS Format
Other project views: | All 38 publications | 12 publications in selected types | All 5 journal articles |
---|
Type | Citation | ||
---|---|---|---|
|
Bossenbroek JM, Wagner HH, Wiens JA. Taxon-dependent scaling: beetles, birds, and vegetation at four North American grassland sites. Landscape Ecology 2005;20(6):675-688. |
R826764 (Final) |
Exit Exit |
|
Wagner HH. Spatial covariance in plant communities: integrating ordination, geostatistics, and variance testing. Ecology 2003;84(4):1045-1057. |
R826764 (Final) |
Exit Exit |
|
Wagner HH. Direct multi-scale ordination with canonical correspondence analysis. Ecology 2004;85(2):342-351. |
R826764 (Final) |
Exit |
|
Wagner HH, Fortin M-J. Spatial analysis of landscapes: concepts and statistics. Ecology 2005;86(8):1975-1987. |
R826764 (Final) |
Exit |
|
Wiens JA. Riverine landscapes: taking landscape ecology into the water. Freshwater Biology 2002;47(4):501-515. |
R826764 (2001) R826764 (Final) |
Exit Exit Exit |
Supplemental Keywords:
terrestrial ecosystems, animals, indicators, scaling, habitat assessment, grazing, conservation, Environmental Monitoring and Assessment Program, EMAP, Great Plains, CO, NM, ID, land management., RFA, Scientific Discipline, Geographic Area, Ecosystem Protection/Environmental Exposure & Risk, Ecosystem/Assessment/Indicators, Ecosystem Protection, climate change, State, Monitoring/Modeling, Ecological Effects - Environmental Exposure & Risk, Regional/Scaling, Environmental Monitoring, Ecological Risk Assessment, Ecological Indicators, Agricultural Engineering, ecological exposure, EMAP, scaling, landscapes, risk assessment, extrapolation methods, biodiversity, ecosystem assessment, landscape context, Idaho (ID), terrestrial ecosystems, animal responses, spatial scale, New Mexico (NM), conservation, land use change, regional scale impacts, GIS, conservation , landscape patterns, grazing, indicators, land use, land management, Environmental Monitoring & Assessment ProgramProgress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.