Fig 1.
Land coverage raster created using ArcMap 10.2.2.
The land coverage raster was obtained from the Global Land Cover 2000 Project (GLC 2000, [12]), coordinated and implemented by the Joint Research Centre (European Commission).
Fig 2.
Land slope raster created using ArcMap 10.2.2.
Data used for the creation of this map was obtained from the “Derived soil properties” of the FAO-NESCO Soil Map of the world, which aggregates the GTOPO30 dataset with a spatial resolution of 5*5 arc minutes (approximately 10km grids). Information on data processing and equations is available under the “Global Terrain Slope and Aspect Data” reference on the FAO website provided under the reference list [15].
Fig 3.
Kernel smoothing of sow sites participating in this study (A); and pig density from FAO projection (2005; B), created using ArcMap 10.2.2.
Data for (B) was obtained from the FAO’s GeoNetwork data repository (global livestock densities, modelled data). This raster was predicted for 2005, and adjusted to match FAOSTAT 2005 national totals.
Table 1.
Description of exploratory variables of interest and results from univariable analyzes.
The main outcome of interest was the counts of PRRS outbreaks at the farm level, and an offset was used to account for different periods participating in data collection (2009–2016).
Fig 4.
Causal diagram showing the hypothesized and plausible associations between the outcome of interest and investigated exploratory variables.
Fig 5.
Histogram of distribution of PRRS outbreaks for all swine sites enrolled in the study over the years 2009–2016.
Table 2.
Final multivariable generalized mixed Poisson model; the main outcome modeled herein was the counts of PRRS outbreaks using 706 sow sites as the unit of analysis, and an offset to account for different numbers of years participating in data collection (2009–2016) was used.
The model accounted for clustering of swine sites within production systems using a random effect.
Fig 6.
Normal quantile plots for Ascombe residuals (A) and best linear unbiased predictors (BLUPS; B) for the generalized mixed Poisson model used to model the association between demographic and environmental variables of interest and the occurrence of PRRS outbreaks in sow farms across the United States.