Skip to main content

Table 2 Discussion of best practices for pseudo-absence selection method

From: Where did they not go? Considerations for generating pseudo-absences for telemetry-based habitat models

Scenario A: Model purpose is to understand broadscale distribution of species habitat often averaged across multiple years [47, 57]. Background sampling has been used to understand where species could have been but were not sighted. These plots are useful for long-term planning and understanding general patterns of habitat use, for example planning military uses in the ocean, shipping lane designation, or off-shore energy sites. Based on Johnson [40] four orders of resource selection, background sampling can be targeted towards a species’ geographic range (1st order) or an area within the geographic range (e.g. a home range; 2nd order). Specific care needs to be taken to ensure that the background sampling extent represents the potential habitat and not beyond because oversampling can lead to inflated model skill. Background sampling often has the greatest environmental separation between presences and absences of the pseudo-absence methods explored.
Scenario B: Model purpose is to describe fine-scale dynamic habitat of species [4, 30, 56, 67]. Correlated random walk sampling is used to create where an individual could have gone in the environment but did not choose to go. This approach is better at capturing fine scale changes in habitat as a function of changes in the environment, for example producing daily maps of predicted habitat to reduce bycatch, or ship-strike risk as a function of the changing environment. Reverse CRWs have also been used to counter the effects of tag-location bias on habitat selection [53]. CRW and reverse CRW both address Johnson [40] third-order of an area within the home range, and can be responsive towards more dynamic selection of habitat. These two approaches had intermediate separation between presences and absences of the pseudo-absence methods explored.
Scenario C: Model purpose is to understand the factors that drive decision-making at each step for tagged individuals. Habitat models with buffer sampling are restricted to each location [19, 22]. Buffer pseudo-absence generation is used to assess individual potential steps rather than the track at a whole. This approach is best suited for understanding the fine-scale factors that influence habitat selection rather than broader habitat preferences, for example which habitat variables and anthropogenic features influence animal movements as they move through the landscape. Buffer sampling for species distribution models address similar aims as resource selection functions (RSF [16];) targeting Johnson [40] 4th order for specific site or resources within broader habitat. This method often results in the least environmental separations between presences and absences of the pseudo-absence methods explored.