Gareth J. Russell

Rates of species loss from Amazonian forest fragments

In the face of worldwide habitat fragmentation, managers need to devise a time frame for action. We ask how fast do understory bird species disappear from experimentally isolated plots in the Biological Dynamics of Forest Fragments Project, central Amazon, Brazil. Our data consist of mist-net records obtained over a period of 13 years in 11 sites of 1, 10, and 100 hectares. The numbers of captures per species per unit time, analyzed under different simplifying assumptions, reveal a set of species-loss curves. From those declining numbers, we derive a scaling rule for the time it takes to lose half the species in a fragment as a function of its area. A 10-fold decrease in the rate of species loss requires a 1,000-fold increase in area. Fragments of 100 hectares lose one half of their species in <15 years, too short a time for implementing conservation measures.

A century of turnover : community dynamics at three timescales

We calculated the observed species turnover of the bird communities on 13 small islands off the coast of the British Isles and the Republic of Ireland for different census intervals. For seven of these islands, the maximum intervals exceeded 80 years. We developed a non-linear, asymptotic model to describe how observed turnover should change with census interval. Our assumptions were traditional ones, based upon the assumption of dynamic equilibrium in familiar island biogeography theory, even though we knew that such equilibrium was rare in these islands. Furthermore, the model considered the average dynamics of the species present and not the dynamics of the individual species. This model showed that neither the observed turnover calculated over different intervals, nor the turnover rate obtained by dividing this observed turnover by the interval, are statistics that permit comparison between islands. Observed turnover increased over time, thus 1-year turnover underestimated the turnovers over a decade or a century. The increase was less than linear, however, so dividing observed turnover by the number of years in its calculation produced a statistic that declined progressively with that number. The model provided a significant overall fit to the data, but underestimated turnover at both the shortest and longest census intervals. We modified the model to reduce the amount of underestimation, by incorporating the long-term changes in the number of species on each island and hence removing the assumption of equilibrium. This non-equilibrium model provided a much improved fit to the data, but it still failed to describe turnover at the very shortest intervals. These, however, are known from other studies to be inflated by individuals-floaters-that nest only once or twice on the islands. The improved, non-equilibrium model made good predictions of the observed turnover over a 4-year interval. These predictions may be used to compare islands, including those for which empirical data on 4-year turnovers are sparse. We divided the non-equilibrium model into an intrinsic and an extrinsic component, representing the influence of within-community and external factors, respectively. Even the intrinsic components of turnover are large, involving differences in species composition of 6-36% between widely separated censuses. How these intrinsic components of turnover vary from island to island is not clear because previous studies have been unable to compare turnover at different time scales. The number of islands in this study was too few for this purpose and we leave a more broadly based comparison for a future paper.

The effect of hydrological patterns and breeding‐season flooding on the numbers and distribution of wading birds in Everglades National Park

Field ecologists in Everglades National Park know that the dynamics of water flow affect the breeding success of wading birds. A number of recent studies have suggested foraging success as the primary causal link. Data on the number and location of foraging birds are available from the Systematic Reconnaissance Flights, monthly aerial surveys of wading birds and surface water condition. A set of regression models were developed that predict the number of foraging birds observed in the Park at the beginning of May, a crucial period in the breeding season of almost all wading birds in this area. Predictor variables were obtained by converting the observations of surface water condition into three indexes that describe (1) the amount of surface water in the Park in January (near the beginning of the ‘dry’ season), (2) the rate at which it dries over the subsequent months, and (3) the amount of disruption to that drying process. An information-theoretic measure, ICOMP(IFIM), was used to choose on the basis of parsimony between the large set of possible models that incorporate these predictors. Most species were best predicted by the same few models, and the fitted model parameters were also similar, indicating that the same pattern of surface water dynamics was optimal for most species. The optimal pattern was: intermediate water levels at the beginning of the dry season, a rapid rate of drying, and no disruption in the drying process. A number of disruptions in drying since 1985 have been the result of releases of water from the flow-control structures at the northern boundary of Everglades National Park. Reducing or eliminating these unnatural hydrological events should help wading bird populations to increase.

Quantitative Analysis of Forest Fragmentation in the Atlantic Forest Reveals More Threatened Bird Species than the Current Red List

Habitat loss and attendant fragmentation threaten the existence of many species. Conserving these species requires a straightforward and objective method that quantifies how these factors affect their survival. Therefore, we compared a variety of metrics that assess habitat fragmentation in bird ranges, using the geographical ranges of 127 forest endemic passerine birds inhabiting the Atlantic Forest of Brazil. A common, non-biological metric — cumulative area of size-ranked fragments within a species range — was misleading, as the least threatened species had the most habitat fragmentation. Instead, we recommend a modified version of metapopulation capacity. The metric links detailed spatial information on fragment sizes and spatial configuration to the birds’ abilities to occupy and disperse across large areas (100,000+ km2). In the Atlantic Forest, metapopulation capacities were largely bimodal, in that most species’ ranges had either low capacity (high risk of extinction) or high capacity (very small risk of extinction). This pattern persisted within taxonomically and ecologically homogenous groups, indicating that it is driven by fragmentation patterns and not differences in species ecology. Worryingly, we found IUCN considers some 28 of 58 species in the low metapopulation capacity cluster to not be threatened. We propose that assessing the effect of fragmentation will separate species more clearly into distinct risk categories than does a simple assessment of remaining habitat.

Data-driven discovery of the spatial scales of habitat choice by elephants

Setting conservation goals and management objectives relies on understanding animal habitat preferences. Models that predict preferences combine location data from tracked animals with environmental information, usually at a spatial resolution determined by the available data. This resolution may be biologically irrelevant for the species in question. Individuals likely integrate environmental characteristics over varying distances when evaluating their surroundings; we call this the scale of selection. Even a single characteristic might be viewed differently at different scales; for example, a preference for sheltering under trees does not necessarily imply a fondness for continuous forest. Multi-scale preference is likely to be particularly evident for animals that occupy coarsely heterogeneous landscapes like savannahs. We designed a method to identify scales at which species respond to resources and used these scales to build preference models. We represented different scales of selection by locally averaging, or smoothing, the environmental data using kernels of increasing radii. First, we examined each environmental variable separately across a spectrum of selection scales and found peaks of fit. These ‘candidate’ scales then determined the environmental data layers entering a multivariable conditional logistic model. We used model selection via AIC to determine the important predictors out of this set. We demonstrate this method using savannah elephants (Loxodonta africana) inhabiting two parks in southern Africa. The multi-scale models were more parsimonious than models using environmental data at only the source resolution. Maps describing habitat preferences also improved when multiple scales were included, as elephants were more often in places predicted to have high neighborhood quality. We conclude that elephants select habitat based on environmental qualities at multiple scales. For them, and likely many other species, biologists should include multiple scales in models of habitat selection. Species environmental preferences and their geospatial projections will be more accurately represented, improving management decisions and conservation planning.

Self-organized criticality in ecology and evolution

We thank Sergey Gavrilets and other members of the ‘Complexity in Biological Systems’ seminar group at UT for stimulating discussions.

Sociability leads to instability Site-switching cascades in a colonial species

We present a general stochastic model showing that colonial breeding can lead to complex multi-colony population dynamics when combined with nothing more than (inevitably) imperfect decision-making by individuals. In particular, frequent “switching cascades”—mass movement of individuals between locations from one breeding season to the next—arise naturally from our model, bringing into question the need to invoke a separate, fitness-based explanation for this commonly observed real-world phenomenon. A key component of the model is the development, at the beginning of each breeding season, of a set of colonies, based on sequential choices by individuals about where to breed. Individuals favor the colony they bred in previously, but are also attracted to colonies that are rapidly establishing, and may switch locations. This provides a positive feedback that leads to switching cascades. We examine the effect on the dynamics of individuals’ access to (and ability to act on) information, as well as the overall size of the colony system and of individual colonies. We compare the model’s dynamics to the observed population dynamics of a set of heron and egret breeding colonies in New York Harbor.

Phytoplankton biomass and nutrient distributions in the Amazon River plume: Environmental correlates

The distribution of phytoplankton biomass in the plume of the Amazon River over the Brazilian continental shelf is analyzed by the use of multiple regression. Previous attempts to assess how different parameters control phytoplankton biomass have used pairwise correlations. A multiple regression approach, however, allows the elucidation of collinearity between these parameters. This approach reveals that phytoplankton biomass may be predicted largely by the following three groups of collinear variables that resemble the “factors” of factor analysis: suspended-sediment concentration and transparency (which generally describe irradiance availability), salinity and temperature (which describe vertical stratification, a measure of water-column stability), and the ambient concentrations of nutrients (phosphate, nitrate, silicic acid, and nitrite). The effects of water clarity and nutrients have been previously described, but the importance of vertical stability has never been separated from the other two. Additional important single variables were oxygen, ammonia, and urea. The strength of the contribution of particular variables to a regression model depends on the season of the cruise and hence on the volume of riverine discharge.

Sampling designs matching species biology produce accurate and affordable abundance indices

Wildlife biologists often use grid-based designs to sample animals and generate abundance estimates. Although sampling in grids is theoretically sound, in application, the method can be logistically difficult and expensive when sampling elusive species inhabiting extensive areas. These factors make it challenging to sample animals and meet the statistical assumption of all individuals having an equal probability of capture. Violating this assumption biases results. Does an alternative exist? Perhaps by sampling only where resources attract animals (i.e., targeted sampling), it would provide accurate abundance estimates more efficiently and affordably. However, biases from this approach would also arise if individuals have an unequal probability of capture, especially if some failed to visit the sampling area. Since most biological programs are resource limited, and acquiring abundance data drives many conservation and management applications, it becomes imperative to identify economical and informative sampling designs. Therefore, we evaluated abundance estimates generated from grid and targeted sampling designs using simulations based on geographic positioning system (GPS) data from 42 Alaskan brown bears (Ursus arctos). Migratory salmon drew brown bears from the wider landscape, concentrating them at anadromous streams. This provided a scenario for testing the targeted approach. Grid and targeted sampling varied by trap amount, location (traps placed randomly, systematically or by expert opinion), and traps stationary or moved between capture sessions. We began by identifying when to sample, and if bears had equal probability of capture. We compared abundance estimates against seven criteria: bias, precision, accuracy, effort, plus encounter rates, and probabilities of capture and recapture. One grid (49 km2 cells) and one targeted configuration provided the most accurate results. Both placed traps by expert opinion and moved traps between capture sessions, which raised capture probabilities. The grid design was least biased (−10.5%), but imprecise (CV 21.2%), and used most effort (16,100 trap-nights). The targeted configuration was more biased (−17.3%), but most precise (CV 12.3%), with least effort (7,000 trap-nights). Targeted sampling generated encounter rates four times higher, and capture and recapture probabilities 11% and 60% higher than grid sampling, in a sampling frame 88% smaller. Bears had unequal probability of capture with both sampling designs, partly because some bears never had traps available to sample them. Hence, grid and targeted sampling generated abundance indices, not estimates. Overall, targeted sampling provided the most accurate and affordable design to index abundance. Targeted sampling may offer an alternative method to index the abundance of other species inhabiting expansive and inaccessible landscapes elsewhere, provided their attraction to resource concentrations.

An automatic identification and monitoring system for coral reef fish

To help gauge the health of coral reef ecosystems, we developed a prototype of an underwater camera module to automatically census reef fish populations. Recognition challenges include pose and lighting variations, complicated backgrounds, within-species color variations and within-family similarities among species. An open frame holds two cameras, LED lights, and two ‘background’ panels in an L-shaped configuration. High-resolution cameras send sequences of 300 synchronized image pairs at 10 fps to an on-shore PC. Approximately 200 sequences containing fish were recorded at the New York Aquarium’s Glover’s Reef exhibit. These contained eight ‘common’ species with 85–672 images, and eight ‘rare’ species with 5–27 images that were grouped into an ‘unknown/rare’ category for classification. Image pre-processing included background modeling and subtraction, and tracking of fish across frames for depth estimation, pose correction, scaling, and disambiguation of overlapping fish. Shape features were obtained from PCA analysis of perimeter points, color features from opponent color histograms, and ‘banding’ features from DCT of vertical projections. Images were classified to species using feedforward neural networks arranged in a three-level hierarchy in which errors remaining after each level are targeted by networks in the level below. Networks were trained and tested on independent image sets. Overall accuracy of species-specific identifications typically exceeded 96% across multiple training runs. A seaworthy version of our system will allow for population censuses with high temporal resolution, and therefore improved statistical power to detect trends. A network of such devices could provide an ‘early warning system’ for coral ecosystem collapse.