Bayesian forecasting of disease spread with little or no local data

Rapid and targeted management actions are a prerequisite to efficiently mitigate disease outbreaks. Targeted actions, however, require accurate spatial information on disease occurrence and spread. Frequently, targeted management actions are guided by non-statistical approaches that define the affected area by a pre-determined distance surrounding a small number of disease detections. As an alternative, we present a long-recognized but underutilized Bayesian technique that uses limited local data and informative priors to make statistically valid predictions and forecasts about disease occurrence and spread. As a case study, we use limited local data that were available after the detection of chronic wasting disease in Michigan, U.S. along with information rich priors obtained from a previous study in a neighboring state. Using these limited local data and informative priors, we generate statistically valid predictions of disease occurrence and spread for the Michigan study area. This Bayesian technique is conceptually and computationally simple, relies on little to no local data, and is competitive with non-statistical distance-based metrics in all performance evaluations. Bayesian modeling has added benefits because it allows practitioners to generate immediate forecasts of future disease conditions and provides a principled framework to incorporate new data as they accumulate. We contend that the Bayesian technique offers broad-scale benefits and opportunities to make statistical inference across a diversity of data-deficient systems, not limited to disease.

The role of drought as a determinant of hemorrhagic disease in the eastern United States

Bluetongue virus and epizootic hemorrhagic disease (HD) virus are globally distributed, vector-borne viruses that infect and cause disease in domestic and wild ruminant species. The forces driving increases in resulting HD may be linked to weather conditions and increasing severity has been noted in northerly latitudes. We evaluated the role of drought severity in both space and time on changes in HD reports across the eastern United States for a recent 15 year period. The objectives of this study were to: (a) develop a spatiotemporal model to evaluate if drought severity explains changing patterns of HD presence; and (b) determine whether this potential risk factor varies in importance over the present range of HD in the eastern United States. Historic data (2000-2014) from an annual HD presence-absence survey conducted by the Southeastern Cooperative Wildlife Disease Study and from the United States Drought Monitor were used for this analysis. For every county in 23 states and for each of 15 years, data were based on reported drought status for August, wetland cover, the physiographic region, and the status of HD in the previous year. We used a generalized linear mixed model to explain HD presence and evaluated spatiotemporal predictors across the region. We found that drought severity was a significant predictor of HD presence and the significance of this relationship was dependent on latitude. In more northerly latitudes, where immunological naivety is most likely, we demonstrated the increasing strength of drought severity as a determinant of reported HD and established the importance of variation in drought severity as a risk factor over the present range of HD in the eastern United States. Our research provides spatially explicit evidence for the link between climate forces and emerging disease patterns across latitude for a globally distributed disease.

Territorial landscapes: incorporating density-dependence into wolf habitat selection studies

Habitat selection is a process that spans space, time and individual life histories. Ecological analyses of animal distributions and preferences are most accurate when they account for inherent dynamics of the habitat selection process. Strong territoriality can constrain perception of habitat availability by individual animals or groups attempting to colonize or establish new territory. Because habitat selection is a function of habitat availability, broad-scale changes in habitat availability or occupancy can drive density-dependent habitat functional responses. We investigated density-dependent habitat selection over a 19-year period of grey wolf (Canis lupus) recovery in Michigan, USA, using a generalized linear mixed model framework to develop a resource selection probability function (RSPF) with habitat coefficients conditioned on random effects for wolf packs and random year intercepts. In addition, we allowed habitat coefficients to vary as interactions with increasing wolf density over space and time. Results indicated that pack presence was driven by factors representing topography, human development, winter prey availability, forest structure, roads, streams and snow. Importantly, responses to many of these predictors were density-dependent. Spatio-temporal dynamics and population changes can cause considerable variation in wildlife-habitat relationships, possibly confounding interpretation of conventional habitat selection models. By incorporating territoriality into an RSPF analysis, we determined that wolves' habitat use in Michigan shifted over time, for example, exhibiting declining responses to winter prey indices and switching from positive to negative responses with respect to stream densities. We consider this an important example of a habitat functional response in wolves, driven by colonization, density-dependence and changes in occupancy during a time period of range expansion and population increase.

Spatial Distribution of Black Bear Incident Reports in Michigan

Interactions between humans and carnivores have existed for centuries due to competition for food and space. American black bears are increasing in abundance and populations are expanding geographically in many portions of its range, including areas that are also increasing in human density, often resulting in associated increases in human-bear conflict (hereafter, bear incidents). We used public reports of bear incidents in Michigan, USA, from 2003-2011 to assess the relative contributions of ecological and anthropogenic variables in explaining the spatial distribution of bear incidents and estimated the potential risk of bear incidents. We used weighted Normalized Difference Vegetation Index mean as an index of primary productivity, region (i.e., Upper Peninsula or Lower Peninsula), primary and secondary road densities, and percentage land cover type within 6.5-km2 circular buffers around bear incidents and random points. We developed 22 a priori models and used generalized linear models and Akaike's Information Criterion (AIC) to rank models. The global model was the best compromise between model complexity and model fit (w = 0.99), with a ΔAIC 8.99 units from the second best performing model. We found that as deciduous forest cover increased, the probability of bear incident occurrence increased. Among the measured anthropogenic variables, cultivated crops and primary roads were the most important in our AIC-best model and were both positively related to the probability of bear incident occurrence. The spatial distribution of relative bear incident risk varied markedly throughout Michigan. Forest cover fragmented with agriculture and other anthropogenic activities presents an environment that likely facilitates bear incidents. Our map can help wildlife managers identify areas of bear incident occurrence, which in turn can be used to help develop strategies aimed at reducing incidents. Researchers and wildlife managers can use similar mapping techniques to assess locations of specific conflict types or to address human impacts on endangered species.