Incorporating spatial non-stationarity of regression coefficients into predictive vegetation models

The results of predictive vegetation models are often presented spatially as GIS-derived surfaces of vegetation attributes across a landscape or region, but spatial information is rarely included in the model itself. Geographically weighted regression (GWR), which extends the traditional regression framework by allowing regression coefficients to vary for individual locations (‘spatial non-stationarity’), is one method of utilizing spatial information to improve the predictive power of such models. In this paper, we compare the ability of GWR, a local model, with that of ordinary least-squares (OLS) regression, a global model, to predict patterns of montane ponderosa pine (Pinus ponderosa) basal area in Saguaro National Park, AZ, USA on the basis of variables related to topography (elevation, slope steepness, aspect) and fire history (fire frequency, time since fire). The localized regression coefficients exhibited significant non-stationarity for four of the five environmental variables, and the GWR model consequently described the vegetation-environment data significantly better, even after accounting for differences in model complexity. GWR also reduced observed spatial autocorrelation of the model residuals. When applied to independent data locations not used in model development, basal areas predicted by GWR had a closer fit to observed values with lower residuals than those from the optimal OLS regression model. GWR also provided insights into fine-scale controls of ponderosa pine pattern that were missed by the global model. For example, the relationship between ponderosa pine basal area and aspect, which was obscured in the OLS regression model due to non-stationarity, was clearly demonstrated by the GWR model. We thus see GWR as a valuable complement to the many other global methods currently in use for predictive vegetation modeling.     

Quantifying patch distribution at multiple spatial scales: Applications to wildlife-habitat models

Multiscale analyses are widely employed for wildlife-habitat studies. In most cases, however, each scale is considered discrete and little emphasis is placed on incorporating or measuring the responses of wildlife to resources across multiple scales. We modeled the responses of three Arctic wildlife species to vegetative resources distributed at two spatial scales: patches and collections of patches aggregated across a regional area. We defined a patch as a single or homogeneous collection of pixels representing 1 of 10 unique vegetation types. We employed a spatial pattern technique, three-term local quadrat variance, to quantify the distribution of patches at a larger regional scale. We used the distance at which the variance for each of 10 vegetation types peaked to define a moving window for calculating the density of patches. When measures of vegetation patch and density were applied to resource selection functions, the most parsimonious models for wolves and grizzly bears included covariates recorded at both scales. Seasonal resource selection by caribou was best described using a model consisting of only regional scale covariates. Our results suggest that for some species and environments simple patch-scale models may not capture the full range of spatial variation in resources to which wildlife may respond. For mobile animals that range across heterogeneous areas we recommend selection models that integrate resources occurring at a number of spatial scales. Patch density is a simple technique for representing such higher-order spatial patterns.     

Quantifying patch distribution at multiple spatial scales: Applications to wildlife-habitat models

Multiscale analyses are widely employed for wildlife-habitat studies. In most cases, however, each scale is considered discrete and little emphasis is placed on incorporating or measuring the responses of wildlife to resources across multiple scales. We modeled the responses of three Arctic wildlife species to vegetative resources distributed at two spatial scales: patches and collections of patches aggregated across a regional area. We defined a patch as a single or homogeneous collection of pixels representing 1 of 10 unique vegetation types. We employed a spatial pattern technique, three-term local quadrat variance, to quantify the distribution of patches at a larger regional scale. We used the distance at which the variance for each of 10 vegetation types peaked to define a moving window for calculating the density of patches. When measures of vegetation patch and density were applied to resource selection functions, the most parsimonious models for wolves and grizzly bears included covariates recorded at both scales. Seasonal resource selection by caribou was best described using a model consisting of only regional scale covariates. Our results suggest that for some species and environments simple patch-scale models may not capture the full range of spatial variation in resources to which wildlife may respond. For mobile animals that range across heterogeneous areas we recommend selection models that integrate resources occurring at a number of spatial scales. Patch density is a simple technique for representing such higher-order spatial patterns.     

Analysis of fine-scale spatial pattern of a grassland from remotely-sensed imagery and field collected data

An important practical problem in the analysis of spatial pattern in ecological systems is that requires spatially-intensive data, with both fine resolution and large extent. Such information is often difficult to obtain from field-measured variables. Digital imagery can offer a valuable, alternative source of information in the analysis of ecological pattern. In the present paper, we use remotely-sensed imagery to provide a link between field-based information and spatially-explicit modeling of ecological processes. We analyzed one digitized color infrared aerial photograph of a serpentine grassland to develop a detailed digital map of land cover categories (31.24 m × 50.04 m of extent and 135 mm of resolution), and an image of vegetation index (proportional to the amount of green biomass cover in the field). We conducted a variogram analysis of the spatial pattern of both field-measured (microtopography, soil depth) and image-derived (land cover map, vegetation index, gopher disturbance) landscape variables, and used a statistical simulation method to produce random realizations of the image of vegetation index based upon our characterization of its spatial structure. The analysis revealed strong relationships in the spatial distribution of the ecological variables (e.g., gopher mounds and perennial grasses are found primarily on deeper soils) and a non-fractal nested spatial pattern in the distribution of green biomass as measured by the vegetation index. The spatial pattern of the vegetation index was composed of three basic components: an exponential trend from 0 m to 4 m, which is related to local ecological processes, a linear trend at broader scales, which is related to a general change in topography across the study site, and a superimposed periodic structure, which is related to the regular spacing of deeper soils within the study site. Simulations of the image of vegetation index confirmed our interpretation of the variograms. The simulations also illustrated the limits of statistical analysis and interpolations based solely on the semivariogram, because they cannot adequately characterize spatial discontinuities.     

A multi-scaled analysis of avian response to habitat amount and fragmentation in the Canadian dry mixed-grass prairie

Previous research has suggested that ducks and songbirds may benefit from prairie landscapes that consist primarily of contiguous grasslands. However, the relative importance of landscape-level vs. local characteristics on mechanisms underlying observed patterns is unclear. We measured effects of grassland amount and fragmentation on upland and wetland songbird and duck density and nest success, and on some nest predators, across 16 landscapes in southern Alberta, Canada. We compared these landscape-level effects with local-scale responses, including distance to various edges and vegetation characteristics. We also evaluated several statistical approaches to comparing effects of habitat characteristics at multiple spatial scales. Few species were influenced by grassland amount or fragmentation. In contrast, distance to edge and local vegetation characteristics had significant effects on densities and nest success of many species. Previous studies that reported effects of landscape characteristics may have detected patterns driven by local mechanisms. As a corollary, results were very sensitive to statistical model structure; landscape level effects were much less apparent when local characteristics were included in the models.     

Multi-scale use of lands providing anthropogenic resources by American Crows in an urbanizing landscape

The conversion of forests and farmlands to human settlements has negative impacts on many native species, but also provides resources that some species are able to exploit. American Crows (Corvus brachyrhynchos), one such exploiter, create concern due to their impact as nest predators, disease hosts, and cultural harbingers of evil. We used various measures of crow abundance and resource use to determine crows’ response to features of anthropogenic landscapes in the Puget Sound region of the United States. We examined land cover and land use composition at three spatial scales: study sites (up to 208 ha), crow home ranges within sites (18.1 ha), and local land cover (400 m²). At the study site and within-site scales crow abundance was strongly correlated with land cover providing anthropogenic resources. In particular, crows were associated with the amount of ‘maintained forest’ cover, and were more likely to use grass and shrub cover than forest or bare soil cover. Although crows did not show a generalized response to an edge variable, they exhibited greater use of patchy habitat created by human settlements than of native forests. Radio-tagged territorial adults used resources within their home ranges relatively evenly, suggesting resource selection had occurred at a larger spatial scale. The land conversion pattern of new suburban and exurban settlements creates the mix of impervious surfaces and maintained vegetation that crows use, and in our study area crow populations are expected to continue to increase. Electronic supplementary material  The online version of this article (doi:) contains supplementary material, which is available to authorized users.     

Introducing spatial grazing impacts into the prediction of moorland vegetation dynamics

Grazing by large herbivores is a major determinant of vegetation dynamics in many semi-natural ecosystems, including the replacement of heather moorland by rough grassland in the British uplands. Herbivore foraging is influenced by vegetation patterns and, in turn, their grazing drives vegetation dynamics. Although vegetation impacts are local, spatially heterogeneous local impacts can have different consequences as would the same impacts distributed uniformly. We constructed a simulation model of the spatial effects of grazing by sheep on the vegetation dynamics of heather moorland, a vegetation community of international conservation importance in the UK. The model comprised three sub-models to predict (1) annual average heather utilisation, (2) spatial variation in heather utilisation (higher near the edge of grass patches) and (3) competition between heather and grass. Here we compare the predicted heather utilisation and vegetation dynamics of the spatial model, relative to those of a non-spatial model. The spatial model resulted in a reduced loss of heather cover for a given sheep stocking rate. The model demonstrates how spatial interactions between large herbivores and their forage drive vegetation dynamics, leading to changes in community structure and composition. Indeed, omitting spatial effects in grazing models may lead to inaccurate predictions. We have shown that ecosystem modelling, based around an iterative dialogue between developers and experienced researchers, has the potential to make a substantial contribution towards the conservation and management of vulnerable landscapes. Combining modelling with experimental studies will facilitate progress towards understanding long-term vegetation/herbivore dynamics.     

Introducing spatial grazing impacts into the prediction of moorland vegetation dynamics

Grazing by large herbivores is a major determinant of vegetation dynamics in many semi-natural ecosystems, including the replacement of heather moorland by rough grassland in the British uplands. Herbivore foraging is influenced by vegetation patterns and, in turn, their grazing drives vegetation dynamics. Although vegetation impacts are local, spatially heterogeneous local impacts can have different consequences as would the same impacts distributed uniformly. We constructed a simulation model of the spatial effects of grazing by sheep on the vegetation dynamics of heather moorland, a vegetation community of international conservation importance in the UK. The model comprised three sub-models to predict (1) annual average heather utilisation, (2) spatial variation in heather utilisation (higher near the edge of grass patches) and (3) competition between heather and grass. Here we compare the predicted heather utilisation and vegetation dynamics of the spatial model, relative to those of a non-spatial model. The spatial model resulted in a reduced loss of heather cover for a given sheep stocking rate. The model demonstrates how spatial interactions between large herbivores and their forage drive vegetation dynamics, leading to changes in community structure and composition. Indeed, omitting spatial effects in grazing models may lead to inaccurate predictions. We have shown that ecosystem modelling, based around an iterative dialogue between developers and experienced researchers, has the potential to make a substantial contribution towards the conservation and management of vulnerable landscapes. Combining modelling with experimental studies will facilitate progress towards understanding long-term vegetation/herbivore dynamics.     

Spatial vegetation patterns as early signs of desertification: a case study of a desert steppe in Inner Mongolia,China

Proper assessment and early detection of land degradation and desertification is extremely important in arid and semi-arid ecosystems. Recent research has proposed to use the characteristics of spatial vegetation patterns, such as parameters derived from power-law modeling of vegetation patches, for detecting the early signs of desertification. However, contradictory results have been reported regarding the suitability of those proposed indicators. We used an experiment with multiple grazing intensities as an analog of a desertification gradient and evaluated the performance of two predictors of desertification: percent plant cover and a transition from a patch-area distribution characterized by a power law to another portrayed by a truncated power law, in a desert steppe in Inner Mongolia, China. We found that spatial metrics, such as the largest patch index and coefficient of variation of mean patch area had negative linear relationships with grazing intensity, suggesting that vegetation patches became more fragmented and homogeneous under higher grazing pressure. Using a binning-based method to analyze our dataset, we found that the patch-area relationship deviated from a power-law to a truncated power-law model with increasing grazing pressure, while the truncated power law was a better fit than the power law for all plots when binning was not used. These results suggest that the selection of methodology is crucial in using power-law models to detect changes in vegetation patterns. Plant cover was significantly correlated with stocking rate and all spatial metrics evaluated; however, the relationship between cover and vegetation spatial pattern still deserves a thorough examination, especially in other types of ecosystems, before using cover as a universal early sign of desertification. Our results highlight a strong connection between the vegetation spatial pattern and the desertification associated with heavy grazing and suggest that future studies should incorporate information about vegetation spatial pattern in monitoring desertification processes.     

Topography-mediated controls on local vegetation phenology estimated from MODIS vegetation index

Forest canopy phenology is an important constraint on annual water and carbon budgets, and responds to regional interannual climate variation. In steep terrain, there are complex spatial variations in phenology due to topographic influences on microclimate, community composition, and available soil moisture. In this study, we investigate spatial patterns of phenology in humid temperate forest as a function of topography. Moderate-resolution imaging spectro-radiometer (MODIS) vegetation indices are used to derive local patterns of topography-mediated vegetation phenology using a simple post-processing analysis and a non-linear model fitting. Elevation has the most explanatory power for all phenological variables with a strong linear relationship with mid-day of greenup period, following temperatures lapse rates. However, all other phenological variables show quadratic associations with elevation, reflecting an interaction between topoclimatic patterns of temperature and water availability. Radiation proxies also have significant explanatory power for all phenological variables. Though hillslope position cannot be adequately resolved at the MODIS spatial resolution (250 m) to discern impacts of local drainage conditions, extended periods of greenup/senescence are found to occur in wet years. These findings are strongly supported by previous field measurements at different topographic positions within the study area. The capability of detecting topography-mediated local phenology offers the potential to detect vegetation responses to climate change in mountainous terrain. In addition, the large, local variability of meteorological and edaphic conditions in steep terrain provides a unique opportunity to develop an understanding of canopy response to the interaction of climate and landscape conditions.     

Cumulative effects of forestry on habitat use by gray wolf (<Emphasis Type="Italic">Canis lupus</Emphasis>) in the boreal forest

Forest harvesting involves the creation of roads and cutblocks, both of which can influence animal habitat use. We evaluated the cumulative effects of forestry on habitat selection by six packs of gray wolf (Canis lupus) widely distributed in Quebec’s boreal forest. Resource selection functions were used to evaluate cumulative effects at two levels. First, we studied how the response of wolves to roads and cutblocks varied within their home range (HR level) as a function of the local abundance of these habitat features. Second, we assessed whether differences in the response to roads and cutblocks observed among packs (inter-HR level) could be explained by variations in their average abundance among individual home ranges. At the HR level, we found that cumulative effects shaped habitat selection of wolves, and the nature of the effects varied during the year. For example, we detected a decrease in the selection of roads following an increase in local road density during the rendez-vous and the nomadic periods, but not during the denning period. At the inter-HR level, we found a functional response to logging activity only during the denning period. Packs with home ranges characterized by a larger proportion of recent cutblocks selected these cutblocks more strongly. We conclude that cumulative effects of logging activities occur at multiple levels, and these effects can have profound effects on habitat use by wolves, thereby influencing spatial predator–prey dynamics. Wildlife conservation and management in boreal ecosystems should thus account for cumulative impacts of anthropogenic features on animal distribution.     

Multi-scale study of bird species distribution and of their response to vegetation change: a Mediterranean example

Land use changes operate at different scales. They trigger a cascade of effects that simultaneously modify the composition or structure of the landscape and of the local vegetation. Mobil animals, and birds in particular, can respond quickly to such multi-scalar changes. We took advantage of a long term study on the response of songbirds to land-use changes on four Mediterranean islands in Corsica and Sardinia to explore the benefits of a multi-scale analysis of the relationships between songbird distribution, vegetation structure and landscape dynamics. Field data and aerial photographs were used to describe the vegetation at three different scales. Birds were censused by point counts. We used statistical variance decomposition to study how bird distribution and vegetation at various scales were linked. We analysed multi-scale vegetation changes (floristic composition, plot vegetation type, and landscape structure) and their consequences on bird distribution with multivariate and non-parametrical tests. The distribution of most species was linked to at least two spatial scales. The weight of a given scale was consistent with life-history traits for species whose biology was well-known. In the examples studied, vegetation composition, vegetation type and landscape changes that resulted from land abandonment negatively affected birds depending on open or heterogeneous areas. Our results emphasize that multi-scale analyses can greatly enhance our understanding of bird distribution and of their changes. Management of these populations should take into account measures at various spatial scales depending on the sensitivity of the species.     

Human appropriated net primary productivity as a metric for land use planning: a case study in the US Great Lakes region

Context

Human appropriation of net primary productivity (HANPP) is employed as a measure of human pressures on biodiversity, though largely at global and national scales rather than landscape to regional scales where many conservation decisions take place. Though gaining in familiarity, HANPP is not widely utilized by conservation professionals.

Objectives

This study, encompassing the US side of the Great Lakes basin, examines how regional distributions of HANPP relate to landscape-based biodiversity proxy metrics used by conservation professionals. Our objectives were (1) to quantify the HANPP of managed lands at the county scale; and (2) to assess spatial patterns of HANPP in comparison to landscape diversity and local habitat connectedness to determine if the metric can provide useful information to conservation professionals.

Methods

We aggregated forest and cropland NPP data between 2005 and 2015 and coupled it with previously published potential vegetation maps to quantify the HANPP of each county in the study region. We mapped the outputs at 500 m resolution to analyze spatial relationships between HANPP and landscape metrics of biodiversity potential.

Results

Area-weighted HANPP across our study region averaged 45% of NPP, down to 4.9% in forest-dominated counties. Greater HANPP correlated with reduced landscape diversity (p?<?0.001, r²?=?0.28) and reduced local habitat connectedness (p?<?0.001, r²?=?0.36).

Conclusion

HANPP could be used as an additional tool for conservation professionals during regional-scale land use planning or conservation decision-making, particularly in mixed-use landscapes that both support important biodiversity and have high levels of primary production harvest.

     

Do bird spatial distribution patterns reflect population trends in changing landscapes?

Strong relations between population trends and spatial distribution have been suggested at the regional scale: declining species should have more fragmented distributions because decline causes range retractions towards optimal habitats, whereas increasing species should have more aggregated distributions, because colonization processes are constrained by distance. Most analyses of the effects of land use changes on animal populations are diachronic studies of population dynamics or synchronic studies of species habitat selection. Few studies take simultaneously into account temporal changes in habitat distribution and changes in species spatial distribution. We applied the above rationale to the landscape scale and analysed how population declines, increases or stability, as diagnosed in a long term study, correlate with population connectivity or fragmentation at that scale. We used data on changes in faunal distribution and information on temporal changes in the vegetation in a Mediterranean area that had been subjected to land abandonment. We found that species declining at the landscape scale had retracting fragmented distributions and that expanding species had expanding continuous distributions. However, for the latter, we suggest that the factors involved are related to landscape structure and not to dispersal mediated meta-population processes, which are of little relevance at this local scale. We also show that even species that are numerically stable can show fragmentation of their distribution and major spatial distribution shifts in response to land use changes, especially in species that have low occurrence levels or that are associated with transitory habitats such as heterogeneous shrublands (e.g. Sylvia melanocephala). Studying the spatial structure of species distribution patterns at the landscape scale may provide information about population declines and increases both at the regional and the landscape scale and can improve our understanding of short-term risks of local extinction.     

Landscape influences on stream biotic integrity assessed at multiple spatial scales

The biological integrity of stream ecosystems depends critically on human activities that affect land use/cover along stream margins and possibly throughout the catchment. We evaluated stream condition using an Index of Biotic Integrity (IBI) and a habitat index (HI), and compared these measures to landscape and riparian conditions assessed at different spatial scales in a largely agricultural Midwestern watershed. Our goal was to determine whether land use/cover was an effective predictor of stream integrity, and if so, at what spatial scale. Twenty-three sites in first-through third-order headwater streams were surveyed by electrofishing and site IBIs were calculated based on ten metrics of the fish collection. Habitat features were characterized through field observation, and site HIs calculated from nine instream and bank metrics. Field surveys, aerial photograph interpretation, and geographic information system (GIS) analyses provided assessments of forested land and other vegetation covers at the local, reach, and regional (catchment) scales. The range of conditions among the 23 sites varied from poor to very good based on IBI and HI scores, and habitat and fish assemblage measures were highly correlated. Stream biotic integrity and habitat quality were negatively correlated with the extent of agriculture and positively correlated with extent of wetlands and forest. Correlations were strongest at the catchment scale (IBI with % area as agriculture, r²=0.50, HI with agriculture, r²=0.76), and tended to become weak and non-significant at local scales. Local riparian vegetation was a weak secondary predictor of stream integrity. In this watershed, regional land use is the primary determinant of stream conditions, able to overwhelm the ability of local site vegetation to support high-quality habitat and biotic communities.     

Modeling vegetation pattern using digital terrain data

Using a geographic information system (GIS), digital maps of environmental variables including geology, topography and calculated clear-sky solar radiation, were weighted and overlaid to predict the distribution of coast live oak (Ouercus agrifolia) forest in a 72 km² region near Lompoc, California. The predicted distribution of oak forest was overlaid on a map of actual oak forest distribution produced from remotely sensed data, and residuals were analyzed to distinguish prediction errors due to alteration of the vegetation cover from those due to defects of the statistical predictive model and due to cartographic errors. Vegetation pattern in the study area was associated most strongly with geologic substrate. Vegetation pattern was also significantly associated with slope, exposure and calculated monthlysolar radiation. The proportion of observed oak forest occurring on predicted oak forest sites was 40% overall, but varied substantially between substrates and also depended strongly on forest patch size, with a much higher rate of success for larger forest patches. Only 21% of predicted oak forest sites supported oak forest, and proportions of observed vegetation on predicted oak forest sites varied significantly between substrates. The non-random patterns of disagreement between maps of predicted and observed forest indicated additional variables that could be included to improve the predictive model, as well as the possible magnitude of forest loss due to disturbances in different parts of the landscape.     

Introducing spatial grazing impacts into the prediction of moorland vegetation dynamics

Grazing by large herbivores is a major determinant of vegetation dynamics in many semi-natural ecosystems, including ghe replacement of heather moorland by rough grassland in the British uplands. Herbivore foraging is influenced by vegetation patterns and, in turn, their grazing drives vegetation dynamics. Although vegetation impacts are local, spatially heterogeneous local impacts can have different conseqences as would the same impact distributed uniformly. We constructed a simulation model of the spatial effects of grazing by sheep on the vegetation dynamics of heather moorland, a vegetation community of international conservation importance in the UK. The model comprised three sub-models to predict (1) annual average heather utilisation, (2) spatial variation in heather utilisation (higher near the edge of grass patches) and (3) competition between heather and grass. Here we compare the predicted heather utilisation and vegetation dynamics of the spatial model, relative to those of a non-spatial model. The spatial model resulted in a reduced loss of heather cover for a given sheep stocking rate. The model demonstrtaes how spatial interactions between large herbivores and their forage drive vegetation dynamics, leading to changes in community structure and composition. Indeed, omitting spatial effects in grazing models may lead to inaccurate predictions. We have shown that ecosystem modelling, based around an iterative dialogue between developers and experienced researchers, has the potential to make a substantial contribution towards the conservation and management of vulnerable landscapes. Combining modelling with experimental studies will facilitate progress towards understanding long-term vegetation/herbivore dynamics.