Moving window analysis links landscape-scale resource utilization to habitat suitability models of feral pigs in northern Australia

Abstract

Habitat suitability (or species distribution) models have been widely used to inform management decisions about species conservation, resource management or invasion control. We developed habitat suitability models of feral pigs across tropical northern Australia to enable more effective management of their negative impacts. However, when investigating wildlife-habitat relationships of highly mobile animals such as feral pigs, home range behaviour must be accounted for as these species can utilize scattered resources and fulfil their living and breeding requirements at different locations within the wider landscape that is accessible to them. We developed a novel approach for incorporating landscape-scale utilization of resources and other habitat requirements into habitat suitability models of feral pigs via moving window analysis. Modelling followed three steps: (1) model pixel values for each of four key habitat requirements (water and food resources as well as protection from heat stress and disturbance) using probabilistic Bayesian networks that combine spatially-explicit explanatory variables, (2) transform them into landscape values via moving window analysis, and (3) model habitat suitability using a Bayesian network that combines all landscape values. Models were implemented in Norsys Netica 5.12 and AgenaRisk 6.1 software and calibrated using expert elicitation. Moving window analyses were implemented in the R 'raster' package and ESRI ArcGIS 10.2 software. Here, we present expert-elicited response functions describing the relationship between landscape-scale patterns of each habitat requirement and its sufficiency to sustain breeding in feral pigs. Two types of response mechanisms were described by each expert: a distance-dependent one where the sufficiency of a requirement diminishes with increasing distance, and a composition-dependent one where the sufficiency of a requirement diminishes with decreasing abundance. These functions were used to parameterize moving window analyses and calculate distance-dependent, composition-dependent and combined distance/composition-dependent requirement landscape values. Experts described three general shapes of response curve: linear decay, where sufficiency diminishes steadily with increasing distance/decreasing abundance; exponential decay, where sufficiency is strongly affected by initial increases in distance or decreases in abundance; and inverted exponential decay, where sufficiency diminishes little initially but strongly at large distances or low levels of abundance. Experts displayed considerable agreement when describing some responses but differed widely with others. Expert assumptions about the characteristic scale of the response, and hence the size of the moving window, also varied considerably. We schematically illustrate that expert assumptions about the characteristic scale and mechanism of resource utilization had a considerable effect on computed sufficiency: Assuming distance-dependence dramatically increased sufficiency of a requirement. This effect became more pronounced at larger characteristic scales of resource utilization. Assuming composition- or combined distance/composition-dependence consolidated computed sufficiency to larger, more continuous resource aggregations and discounted utilization of isolated or scattered resources. These effects were equally more pronounced at larger characteristic scales. Hence, these assumptions require careful consideration when using them as an input into habitat suitability models and their effect on model performance must be validated. Our approach could readily be applied to other mobile species that respond to conditions at the landscape scale.