Structuring the unstructured: estimating species-specific absence from multi-species presence data to inform pseudo-absence selection in species distribution models

摘要: Species distribution models (SDMs) are an increasingly popular tool in ecology which, together with a vast wealth of data from citizen science projects, have the potential to dramatically improve our understanding species behaviour for applications such as conservation and wildlife management. However, many best performing require information regarding survey effort, specifically absence, which is typically lacking opportunistic datasets. To facilitate use models, pseudo-absences locations without recorded presence must be assumed. Several studies suggested that hence likely could estimated presence-only by considering records across "target groups" defined according taxonomy. We performed probabilistic analysis, computing conditional probability recording given particular set also recorded, test validity defining target groups taxonomic order explore other groupings. Based on this quantification associations we outline new method inform pseudo-absence selection comparing predictive performance, measured area under curve (AUC) statistic, against standard series SDMs. Our findings show some support grouping classification based taxonomy but indicate alternative using may more appropriate informing effort consequently absence. Across 49 terrestrial mammal species, proposed outperformed showing improvement performance presence-absence 17 out 22 sufficient elicit significant difference. observed substantial compared (MaxEnt) higher AUC all difference between approaches. conclude produces sensible robust either compliment patterns known presences or, where conflicts occur, explainable terms ecological variables potentially improving behaviour. Furthermore, suggest these provide viable MaxEnt when modelling data.