Supervised versus unsupervised approaches to classification of accelerometry data.

Sophisticated animal-borne sensor systems are increasingly providing novel insight into how animals behave and move. Despite their widespread use in ecology, the diversity and expanding quality and quantity of data they produce have created a need for robust analytical methods for biological interpretation. Machine learning tools are often used to meet this need. However, their relative effectiveness is not well known and, in the case of unsupervised tools, given that they do not use validation data, their accuracy can be difficult to assess. We evaluated the effectiveness of supervised (n = 6), semi-supervised (n = 1), and unsupervised (n = 2) approaches to analyzing accelerometry data collected from critically endangered California condors (Gymnogyps californianus). Unsupervised K-means and EM (expectation-maximization) clustering approaches performed poorly, with adequate classification accuracies of <0.8 but very low values for kappa statistics (range: -0.02 to 0.06). The semi-supervised nearest mean classifier was moderately effective at classification, with an overall classification accuracy of 0.61 but effective classification only of two of the four behavioral classes. Supervised random forest (RF) and k-nearest neighbor (kNN) machine learning models were most effective at classification across all behavior types, with overall accuracies >0.81. Kappa statistics were also highest for RF and kNN, in most cases substantially greater than for other modeling approaches. Unsupervised modeling, which is commonly used for the classification of a priori-defined behaviors in telemetry data, can provide useful information but likely is instead better suited to post hoc definition of generalized behavioral states. This work also shows the potential for substantial variation in classification accuracy among different machine learning approaches and among different metrics of accuracy. As such, when analyzing biotelemetry data, best practices appear to call for the evaluation of several machine learning techniques and several measures of accuracy for each dataset under consideration.

Greater sage-grouse respond positively to intensive post-fire restoration treatments.

Habitat loss is the most prevalent threat to biodiversity in North America. One of the most threatened landscapes in the United States is the sagebrush (Artemisia spp.) ecosystem, much of which has been fragmented or converted to non-native grasslands via the cheatgrass-fire cycle. Like many sagebrush obligates, greater sage-grouse (Centrocercus urophasianus) depend upon sagebrush for food and cover and are affected by changes to this ecosystem. We investigated habitat selection by 28 male greater sage-grouse during each of 3 years after a 113,000-ha wildfire in a sagebrush steppe ecosystem in Idaho and Oregon. During the study period, seeding and herbicide treatments were applied for habitat restoration. We evaluated sage-grouse responses to vegetation and post-fire restoration treatments. Throughout the 3 years post-fire, sage-grouse avoided areas with high exotic annual grass cover but selected strongly for recovering sagebrush and moderately strongly for perennial grasses. By the third year post-fire, they preferred high-density sagebrush, especially in winter when sagebrush is the primary component of the sage-grouse diet. Sage-grouse preferred forb habitat immediately post-fire, especially in summer, but this selection preference was less strong in later years. They also selected areas that were intensively treated with herbicide and seeded with sagebrush, grasses, and forbs, although these responses varied with time since treatment. Wildfire can have severe consequences for sagebrush-obligate species due to loss of large sagebrush plants used for food and for protection from predators and thermal extremes. Our results show that management efforts, including herbicide application and seeding of plants, directed at controlling exotic annual grasses after a wildfire can positively affect habitat selection by sage-grouse.

The intrepid urban coyote: a comparison of bold and exploratory behavior in coyotes from urban and rural environments.

Coyotes (Canis latrans) are highly adaptable, medium-sized carnivores that now inhabit nearly every large city in the United States and Canada. To help understand how coyotes have adapted to living in urban environments, we compared two ecologically and evolutionarily important behavioral traits (i.e., bold-shy and exploration-avoidance behavior) in two contrasting environments (i.e., rural and urban). Boldness is an individual's reaction to a risky situation and exploration is an individual's willingness to explore novel situations. Our results from both tests indicate that urban coyotes are bolder and more exploratory than rural coyotes and that within both populations there are individuals that vary across both spectrums. Bolder behavior in urban coyotes emerged over several decades and we speculate on possible processes (e.g., learning and selection) and site differences that could be playing a role in this behavioral adaptation. We hypothesize that an important factor is how people treat coyotes; in the rural area coyotes were regularly persecuted whereas in the urban area coyotes were rarely persecuted and sometimes positively rewarded to be in close proximity of people. Negative consequences of this behavioral adaptation are coyotes that become bold enough to occasionally prey on pets or attack humans.

Anticoagulant rodenticide exposure and toxicosis in coyotes (Canis latrans) in the Denver Metropolitan Area.

Anticoagulant rodenticides are widely used in urban areas to control rodent pests and are responsible for secondary poisoning in many nontarget wildlife species. We tested the livers of five coyotes (Canis latrans) in the Denver Metropolitan Area, Colorado, US, for anticoagulant rodenticides. All five livers were positive for brodifacoum, with values ranging from 95 ppb to 320 ppb, and one liver was positive for bromadiolone, with a value of 885 ppb. Both of these rodenticides are second-generation anticoagulants, which are more potent and more likely to cause secondary poisoning than first-generation anticoagulants due to their accumulation and persistence in the liver. We concluded that exposure to these rodenticides may have caused the death of at least two of the five coyotes, and urban coyotes in our study area are commonly exposed to rodenticides.

Environmental enrichment affects adrenocortical stress responses in the endangered black-footed ferret.

Potential stressors of wildlife living in captivity, such as artificial living conditions and frequent human contact, may lead to a higher occurrence of disease and reduced reproductive function. One successful method used by wildlife managers to improve general well-being is the provision of environmental enrichment, which is the practice of providing animals under managed care with environmental stimuli. The black-footed ferret (Mustela nigripes) is a highly-endangered carnivore species that was rescued from extinction by removal of the last remaining individuals from the wild to begin an ex situ breeding program. Our goal was to examine the effect of environmental enrichment on adrenocortical activity in ferrets by monitoring fecal glucocorticoid metabolites (FGM). Results demonstrated that enrichment lowered FGM in juvenile male ferrets, while increasing it in adult females; enrichment had no effect on FGM in juvenile females and adult males. These results correspond with our findings that juvenile males interacted more with the enrichment items than did adult females. However, we did not detect an impact of FGM on the incidence of disease or on the ability of ferrets to become reproductive during the following breeding season. We conclude that an environmental enrichment program could benefit captive juvenile male ferrets by reducing adrenocortical activity.