Landscape-scale richness estimation

At scales greater than local surveys we must rely upon statistical approaches to infer richness and there is a long history in ecology developing a myriad of approaches. My contribution to this literature emphasizes the importance of species turnover in these richness estimates (Jobe 2008). Landscapes are heterogeneous, and the vast majority of methods used to estimate species richness assume a homogeneity that is unrealistic and consequently their estimates are always low. My research has evaluated and improved statistical metrics that explicitly include species turnover in their estimates, whether by estimating species entropy or by statistical rarefaction. As an outcome of this work, I was invited to be part of a larger working group evaluating many different approaches for the British countryside plant dataset (Kunin et al. in prep)

Turnover (Beta diversity)/productivity relationships

The relationship between primary production and richness is well-documented and scale-dependent. At local scales the classic hump-backed model is commonly supported where richness is highest at intermediate productivity. Regional productivity-diversity relationships are typically positive linear. These patterns suggest that there is a positive relationship between species turnover (beta-diversity) and regional productivity. There are two competing hypotheses for this pattern: spatial heterogeneity in more productive regions and species packing limitations in more productivity regions. I tested the local, regional, beta diversity productivity relationships using a large dataset of plant occurrences in the eastern United States (Jobe in prep). Local richness-productivity patterns were hump-backed, but regional richness-productivity relationship was neutral rather than positive linear. As a result, the beta-diversity productivity relationship was also neutral. Upon examination of this lack of relationship between beta diversity and productivity, we found support that environmental heterogeneity was driving beta diversity-productivity relationships rather than species packing. While mean productivity itself showed no trend with beta diversity, heterogeneity in productivity at both local and regional scales was strongly related to beta diversity.

In addition to broad patterns of biodiversity I am interested in the ecology of individual species. I am particularly interested in how the dispersal behavior of organisms interacts with landscape structure and its consequence for conservation planning. My key studies in this area have been: a multi-species habitat conservation assessment for endangered species on Ft. Bragg, North Carolina, the spatio-temporal dispersal of bird flu (H5N1) in Vietnam, and the implementation of maximum entropy models for species of the All-Taxa Biodiversity Inventory of the Great Smoky Mountains National Park.

Multi-species habitat connectivity assessment

My current postdoc is focused on multi-species habitat connectivity of four threatened or endangered species on Ft. Bragg, North Carolina with Aaron Moody, Nick Haddad, Jeff Walters, and Bill Morris. This work extends two much-studied topics in ecology: umbrella species and habitat connectivity. The key contribution of my research has been a unifying framework for developing conservation plans that preserve connectivity for multiple species simultaneously regardless of the species data available (Jobe, Haddad et al. in prep). We leverage widely adopted analysis tools (maxent, Circuitscape, Zonation, and agent-based models) that allow researchers and practitioners alike to produced conservation priority raster maps. We developed models of varied complexity and compared their efficacy in selecting appropriate conservation habitat for our species. One of our major results has been to show that red-cockaded woodpecker, the flagship species for our study, is a poor umbrella for preserving connectivity of our other rare species (Jobe and Moody et al. in prep).

The spatio-temporal distribution of H5N1 in Vietnam

I worked with Maggie Carrell and Mike Emsch, disease ecologists at the University of North Carolina, on a study asking whether genetic evolution in bird-flu (H5N1) was associated with geographic distance in Vietnam through time (Carrel et al. 2010). We documented genetic evolution to be strongly correlated within southern and northern Vietnam, but distant between northern and southern Vietnam. This was the first study to document the explicit spatial structure of H5N1 evolution.

Maximum entropy modeling

Maximum entropy modeling of species distributions is a common approach to modeling rare species where limited occurrence data is available. There are few studies that apply maximum entropy modeling across a broad array of species within a single landscape to assess changes through time. I was contracted by Great Smoky Mountains National Park to develop a species distribution modeling approach where the only available data for species are occurrences. I developed the framework for Park researchers to implement this modeling strategy (Jobe 2009).

My research connects field-collected data at local scales to larger environmental patterns that are measured at landscape scales via by remote sensing. There are two statistical challenges to making these connections. The first is that field data are rarely collected as a simple random sample and thus require more sophisticated statistical inference than classic parametric tests. The second is that remote sensing data is treated as a known quantity without error when used by ecologists. Of course remotely-sensed data is a sample just as field data are, and it has its own sources of error that we should account for when trying to understand ecology. But, how should these errors be treated? Both of these challenges are concerned with the relationship between sampling design and statistical inference, and my research addresses them in three ways. The first is through understanding consequences of sampling bias introduced when biologists sample close to roads to the exclusion of remote locations. The second is through my development of sophisticated inferential models to deal with multi-stage clustering in samples for long term ecological monitoring. The third study addresses sampling phenomenon for remotely sensed data, specifically, laser altimetry or lidar. In this study I developed a new Bayesian method for estimating 3-D forest structure that could propagate lidar sampling error through to other ecological models for more robust ecological inference.

Accessibility bias

I am interested in how human-induced bias in data collections changes our understanding of ecological patterns. I developed a model of accessibility that estimates the energetic cost to humans walking through landscapes. This model considers the physiological cost associated with walking up and down slopes, through vegetation and across streams and was parametrized using empirical results from published research on human physiology. Using the model I demonstrated road-side bias in the location vegetation samples in Great Smoky Mountains National Park. This bias results in a difference between observed and actual patterns of vegetation (Jobe 2009). I have also used this model to evaluate remote locations as potential restoration sites given their decreased probability of human disturbance (Jobe in press).

Inferring 3-dimensional forest structure from lidar data

Remotely-sensed data, like field-collected data, have their own set of biases and sampling errors. Yet, the error structure of remotely sensed data is rarely used explicitly in to ecological inference. I developed a model for airborne laser altimeter data , lidar, that accounts for the sampling error of these data when used to infer the 3-dimensional distribution of canopy area in forests. I use a hierarchical Bayesian approach which allows the model to be directly incorporated into species distribution models. This approach is particularly effective at inferring the density of low-canopy structures. Particular application has been made to modeling the distribution of Saint-Francis Satyr on Ft. Bragg, North Carolina.

Long-term monitoring of cobblebars

How much sampling is enough? This is the question answered by power analysis and few ecologists use power analysis to assess long-term monitoring of ecological data because the sampling designs are complex. Such data are collected using multi-level designs, because they allow researchers to make inference at multiple scales and across a larger extent than simple random samples. I am the first ecologist to apply a Bayesian simulation approach to calculating power for long-term ecological monitoring. I developed this approach for a long-term study of vegetation along cobblebars in Big South Fork National River (Tennessee and Kentucky, US)along this monitoring project which consisted of many transects-clustered along multiple cobble bars on this River system (Jobe 2010).

The major facet of my research that I will expand on over the next five year is to examine the contribution of spatial dependence to ecological processes and patterns. There are at least three directions that are currently active. The first is the relationship between beta-diversity and productivity. The research that I have completed revealed the importance of heterogeneity in productivity in producing species turnover, but it is still unclear to what degree this heterogeneity is correlated with geographic distance. I plan to develop a new model that separates the effects of spatial auto-correlation from the pure effects of spatial heterogeneity on species turnover. Second, I want to add the effects of spatial dependence to the Bayesian model for forest canopy cover from lidar data that I developed. This new model would account for the fact that forest canopy structures are spatially auto-correlated. Third, I will expand my current research on species richness estimation to ask not only about richness, but also the relative abundance of species and their range sizes. What environmental patterns correlate with changes in species abundance and range size, and how do these change as we move along latitudinal gradients and from local to landscape scales?