Continental-scale dynamics of avian influenza in U.S. waterfowl are driven by demography, migration, and temperature.

Emerging diseases of wildlife origin are increasingly spilling over into humans and domestic animals. Surveillance and risk assessments for transmission between these populations are informed by a mechanistic understanding of the pathogens in wildlife reservoirs. For avian influenza viruses (AIV), much observational and experimental work in wildlife has been conducted at local scales, yet fully understanding their spread and distribution requires assessing the mechanisms acting at both local, (e.g., intrinsic epidemic dynamics), and continental scales, (e.g., long-distance migration). Here, we combined a large, continental-scale data set on low pathogenic, Type A AIV in the United States with a novel network-based application of bird banding/recovery data to investigate the migration-based drivers of AIV and their relative importance compared to well-characterized local drivers (e.g., demography, environmental persistence). We compared among regression models reflecting hypothesized ecological processes and evaluated their ability to predict AIV in space and time using within and out-of-sample validation. We found that predictors of AIV were associated with multiple mechanisms at local and continental scales. Hypotheses characterizing local epidemic dynamics were strongly supported, with age, the age-specific aggregation of migratory birds in an area and temperature being the best predictors of infection. Hypotheses defining larger, network-based features of the migration processes, such as clustering or between-cluster mixing explained less variation but were also supported. Therefore, our results support a role for local processes in driving the continental distribution of AIV.

Identification of migratory bird flyways in North America using community detection on biological networks.

Migratory behavior of waterfowl populations in North America has traditionally been broadly characterized by four north-south flyways, and these flyways have been central to the management of waterfowl populations for more than 80 yr. However, previous flyway characterizations are not easily updated with current bird movement data and fail to provide assessments of the importance of specific geographical regions to the identification of flyways. Here, we developed a network model of migratory movement for four waterfowl species, Mallard (Anas platyrhnchos), Northern Pintail (A. acuta), American Green-winged Teal (A. carolinensis), and Canada Goose (Branta canadensis), in North America, using bird band and recovery data. We then identified migratory flyways using a community detection algorithm and characterized the importance of smaller geographic regions in identifying flyways using a novel metric, the consolidation factor. We identified four main flyways for Mallards, Northern Pintails, and American Green-winged Teal, with the flyway identification in Canada Geese exhibiting higher complexity. For Mallards, flyways were relatively consistent through time. However, consolidation factors revealed that for Mallards and Green-winged Teal, the presumptive Mississippi flyway was potentially a zone of high mixing between other flyways. Our results demonstrate that the network approach provides a robust method for flyway identification that is widely applicable given the relatively minimal data requirements and is easily updated with future movement data to reflect changes in flyway definitions and management goals.

Patterns of spatial autocorrelation in stream water chemistry.

Geostatistical models are typically based on symmetric straight-line distance, which fails to represent the spatial configuration, connectivity, directionality, and relative position of sites in a stream network. Freshwater ecologists have explored spatial patterns in stream networks using hydrologic distance measures and new geostatistical methodologies have recently been developed that enable directional hydrologic distance measures to be considered. The purpose of this study was to quantify patterns of spatial correlation in stream water chemistry using three distance measures: straight-line distance, symmetric hydrologic distance, and weighted asymmetric hydrologic distance. We used a dataset collected in Maryland, USA to develop both general linear models and geostatistical models (based on the three distance measures) for acid neutralizing capacity, conductivity, pH, nitrate, sulfate, temperature, dissolved oxygen, and dissolved organic carbon. The spatial AICC methodology allowed us to fit the autocorrelation and covariate parameters simultaneously and to select the model with the most support in the data. We used the universal kriging algorithm to generate geostatistical model predictions. We found that spatial correlation exists in stream chemistry data at a relatively coarse scale and that geostatistical models consistently improved the accuracy of model predictions. More than one distance measure performed well for most chemical response variables, but straight-line distance appears to be the most suitable distance measure for regional geostatistical modeling. It may be necessary to develop new survey designs that more fully capture spatial correlation at a variety of scales to improve the use of weighted asymmetric hydrologic distance measures in regional geostatistical models.

Model selection for geostatistical models.

We consider the problem of model selection for geospatial data. Spatial correlation is often ignored in the selection of explanatory variables, and this can influence model selection results. For example, the importance of particular explanatory variables may not be apparent when spatial correlation is ignored. To address this problem, we consider the Akaike Information Criterion (AIC) as applied to a geostatistical model. We offer a heuristic derivation of the AIC in this context and provide simulation results that show that using AIC for a geostatistical model is superior to the often-used traditional approach of ignoring spatial correlation in the selection of explanatory variables. These ideas are further demonstrated via a model for lizard abundance. We also apply the principle of minimum description length (MDL) to variable selection for the geostatistical model. The effect of sampling design on the selection of explanatory covariates is also explored. R software to implement the geostatistical model selection methods described in this paper is available in the Supplement.