Gareth Russell: Current Research Projects

Below are some currently active research projects, usually in conjunction with members of my lab or external collaborators.

Dynamics of active dispersers in patchy habitat

Some of my previous work suggested that organisms with the mobility and cognitive ability to make active dispersal choices (e.g., birds) might readily leave sub-optimal habitat patches as long as they are close to larger patches, leaving those patches species-poor even when island biogeography theory says that 'close' islands should be species-rich. We are testing this hypothesis with a large set of data on species richness patterns in patchy habitats, including various taxa in addition to birds. If this phenomenon is reasonably common (and our initial analyses suggest that it is for birds), it has a number of implications for conservation in fragmented landscapes.

"Cascade" metapopulation dynamics of colonial nesters (Funded, PI Russell)

Using a stochastic model of colonial wading bird behavior, we show that the positive site-selection feedback inherent in colonially-nesting species can lead to complex metapopulation dynamics, even in the absence of external triggers. A key feature of the dynamics is the "switching cascade" whereby large numbers of individuals choose the same new nesting site in a given year, leaving previously populous sites almost empty. This phenomenon is readily observed in data on real nesting colonies of herons and egrets.

An automated identification system for coral reef fish (Funded, PI Russell)

We are in the middle of constructing an integrated camera-computer system that will provide an underwater monitoring apparatus. A visual feed of moving reef fish from two video cameras is processed to extract individual fish images, correct for pose, extract shape and pattern features, and use these to identify each fish to species. This will enable a much greater level of population and community monitoring than currently possible, and provide an 'early warning system' for disruptions to the reef community.

A spatially-explicit model of the Cape Sable Seaside Sparrow (Funded, PI Pimm)

We have just completed a large-scale simulation model of the population dynamics of the endangered Cape Sable Seaside Sparrow (Ammodramus maritimus mirabilis) whose entire population is in Everglades National Park. Unlike most such models, this one has a user-friendly interface that allows users to assess the effects on the Sparrow population of manipulating water levels in different regions of the Park. The work was performed for the US Fish and Wildlife Service, who will use the model to help manage the Sparrow under the requirements of the Endangered Species Act.

The implications of range fragmentation for extinction risk

We are using the concept of 'metapopulation capacity' to develop a standardized metric for the impact of range fragmentation (e.g., due to deforestation) on the extinction risk of a species. The new metric takes advantage of the increasing availability of high-resolution imagery that can be used to estimate changes in the extent of a species preferred habitat in great detail. It links the spatial fragmentation of habitat to species persistence via the colonization and extinction functions of a spatially-explicit metapopulation model. We believe that this can provide a more objective, quantitative basis for assigning threatened status on the basis of habitat loss than is currently available.

Movement and habitat selection of bears

In cooperation with the US Forest Service and US Fish and Wildlife Service, a colleague and I are developing a predictive model of grizzly bear habitat preference based on tracking data from numerous bears. Methodologically, this is an extension of our previous work on elephants.

The 'construction' of ecological communities via manipulation of the assembly process

We are analyzing a very large database of information on early old-field succession by fitting Markov models to the 'percent species cover' timeseries, and then applying 'tweaks' to stochastic simulations based on those matrices. The tweaks are the addition of one of the species in the system at fairly high density (i.e., as seeds). The goal is to see if there are consistent 'vulnerable' points in the succession whereby a simple, one-time seeding can send the succession on a different trajectory. This could provide a simple, cost-effective means of improving the quality of many 'novel' habitats for which full restoration or intensive management (essentially gardening) is unrealistic, such as powerline rights-of-way, road verges and medians, etc.

The evolution of ecological communities

This is a project to add evolutionary dynamics to ecological community models and thereby try explain a priori some of the features of real communities. I am approaching this by separating species interactions into an interaction matrix (the usual a_ij terms of the generalized Lotka-Volterra model) which drives the ecological dynamics and an underlying strategy matrix, the terms of which evolve via selection. The interaction matrix is derived form the strategy matrix via a function based on Prisoner's Dilemma-type game theory.

Gareth Russell: Broader Research Interests

These are some general areas of interest that I am looking to develop over the next few years. A number of these will depend upon collaboration with mathematicians, computer scientists and engineers.

Image processing/machine vision for species ID and monitoring

My lab has a number of ventures in this area, but all revolve around the following goal: to teach a computer to identify species from images, and all use the same basic sequence: image extraction and processing, followed by feature detection, and finally classification. Right now Kim Russell is working on static images of various taxonomic groups (spiders, bees) from a microscope, and Gareth with image sequences of tropical reef fish from multiple digital videocameras. Kim's processing steps therefore involve mainly pixel-level operations such as contrast enhancement, removal of lighting artifacts, etc. The fish project also requires image tracking and segmentation and registration of multiple images to get information about position, pose, etc.

For the feature detection, Kim has been using Gabor wavelets because of their ability to pick out 'oriented blobs' (very important for spiders), but we have recently been experimenting with more general filter banks that can be customized to a particular application. For example, for bees, we will be looking at wing venation, so we need a filter that picks out connections in a net. We know these exist, but this is new to us! And we are just starting on methods for the fish.

For classification, Kim has had great success with artificial neural networks, and we expect to continue this. Our feeling is that this step is hardly the limiting factor — given the complexity and noise in the images we are working with, the possible improvements from better image processing or feature detection will hugely outweigh the benefits of a better classification technique. But we are open to ideas!

I'm also looking into taking a 3D model of a fish and using OpenGL to spin the model through almost every possible orientation and render it, using a high-end gaming card to get good frames-per-second. The renderings would not be sent to the screen, but extracted from the texture buffer and compared to a real image of the fish using a simple and quick pixel comparison routine. The goal would be to find the best-matching orientation, and to do so in around one second. The image comparison itself is amenable to parallelization, and so it might be an ideal candidate to also do within the GPU, using a language like nVidia's CUDA, or the proposed OpenCL standard. I am going to be getting the assistance of senior undergraduates from our Computing Science programs in doing this.

Tracking large and small organisms

One of the emerging areas in ecological data collection is the tracking of organisms as they move around a landscape, from large mammals (e.g., elephants, whales) and birds over continents and many years, to insects over short distances and timespans. Until recently such tracking was impossible, or at least difficult and expensive. New technologies such as GPS tracking collars have enabled vast leaps, but they are still costly. There is a great opportunity for lower-cost technologies to provide tracking ability, especially at the smaller scale, which is the scale about which we know the least. One possibility might be RFID, especially the low-cost, no-tag-power-required passive kind. A simple project that has never been done would be to track beetles in a field! We have now idea how they disperse, and so, for example, whether a typical field is a well mixed population, or if the opposite sides of it are essentially isolated from each other. The ability to collect data like these would enormously enhance our understanding of the population dynamics and genetics of the some of the most common and important species on the planet.

Novel ecology

In part through living in the New York/New Jersey metropolitan area, I have become very interested in how we manage the nature that occurs in highly modified environments: 'novel ecosystems'. Human activities such as farming and urbanization have left much of our planet irrevocably altered from its natural state. In these areas we have novel ecosystems, ranging from autogenic examples, such as the collection of weeds in an abandoned lot, to constructed examples, such as an urban park (this category also includes systems in very artificial locales, such as green roofs or enclosed ‘wetland’ water treatment systems). Academic ecologists, who in the past have tended to seek out relatively pristine environments for their research, are coming to realize that we must incorporate novel ecosystems into our planning for a sustainable future. Unfortunately, autogenic processes sometimes lead to novel ecosystems with undesirable properties, such as a preponderance of non-native species, or low diversity. The standard alternative — constructing complete, functioning ecosystems from scratch — is expensive, and often fails. So far, ecologists have not been able to reliably predict the outcomes of real-world community assembly processes, or the consequences of intervention. But new developments in practical assembly modeling, such as the use of Markov set chains, hold the promise of a predictive body of theory. We believe the time is right to explore how these and other advances in complex systems modeling might be applied to the creation and management of ecological communities, especially novel ecosystems, which provide freedom from the obligation to try to recreate a specific 'template' community

I and a few colleagues have come to feel that the way to approach this is via cooperation between academia and industry (in the form of landscape architects and managers, government agencies with responsibility for land management, etc.). Recently we proposed the creation of a research collaboration and data-sharing network that will involve representatives from all these entities. The network will encourage and facilitate the cooperative study of the ecology of unmanaged, managed, and ‘designed’ sites.

Advanced computation for practical conservation

I am committed to the idea of making the latest advances in ecological analysis that have bearing on conservation issues available, in as simple and transparent a form as possible, to the worldwide conservation community (especially in developing countries, which usually have the most pressing issues). My website, www.eco-tools.net, is the prototype for how to do this. I would like to engage in a large-scale expansion of this project, and have begun laying the groundwork for 'version 2' of the site, which will have a number of improvements. One of these will be a simple means of allowing volunteer translators to provide their translations in a timely fashion, allowing the site to expand into many languages.