PS 99-187
A global estimate of the potential distribution of Sus scrofa: Implications for modeling the distributions of other widespread invasive species

Friday, August 14, 2015
Exhibit Hall, Baltimore Convention Center
Christopher L. Burdett, Dept. of Biology, Colorado State University, Fort Collins, CO
Sarah J. Garza, Dept. of Biology, Colorado State University, Fort Collins, CO
Ryan S. MIller, APHIS-VS-CEAH, USDA, Fort Collins, CO

Sus scrofa, the wild-pig species that includes both the wild boar and feral domestic pigs, is a widespread and destructive large mammal that is invasive in much of its global range. Wild pigs are known to spread over 30 zoonotic pathogens and cause over $1 billion in annual damage to natural and agricultural ecosystems in the United States (U.S.). Populations have been increasing throughout the species’ native and non-native ranges in recent decades, often due to introductions in novel locations far from existing populations. Because anthropogenic introductions circumvent the ecological limitations of population growth and dispersal ability, it’s critical to have estimates of the potential distributions of problematic invasive species like the wild pig. We developed a model that forecasted the potential global distribution of S. scrofa. We collected 9,537 occurrence records that were filtered for a 25 km resolution from across the native Eurasian and non-native North American, South American, Australian, and southern African ranges of the wild pig. We used an ensemble of seven distribution-modeling methods to relate the current distribution of the species to various climatic variables.


The distribution of S. scrofa correlated with seasonal extremes of both temperature and precipitation. Parametric methods like generalized linear models (GLM) generally estimated broader potential distributions than the more restricted distributions estimated by non-parametric methods like random forests (RF). Our ensemble model validated well (area under the curve or AUC > 0.90). However, validation scores varied across methods. Methods estimating broader distributions generally had lower AUC scores (e.g., GLM AUC = 0.89) than those estimating restricted distributions (e.g., RF AUC = 0.98), whose extremely high AUC scores suggest over-fitting problems. Parametric distribution-modeling methods may therefore provide broader, more conservative forecasts of the potential distributions of widespread species from which thousands of occurrence records can be collected. Still, a major limitation when modeling the distributions of widespread species is whether a comprehensive occurrence sample can be collected from throughout the global range. There are up to 17 sub-species of S. scrofa, and we obtained fewer records from a large part of the species’ native range in Russia, China, India and southeastern Asia. A more comprehensive sample would likely estimate a larger potential distribution for S. scrofa, suggesting that invasive pig populations are capable of persisting in most introduction locations.