On estimating probability of presence from use-availability or presence-background data

Area of Science:

• Ecology
• Ecological Modeling
• Statistical Ecology

Background:

• Estimating species presence probability conditional on environmental variables is a core ecological modeling task.
• Presence-background data, consisting of observed species locations and random environmental samples, is commonly used.
• Disagreement exists in recent literature regarding the identifiability of presence probability from presence-background data alone.

Purpose of the Study:

• To resolve the disagreement on whether species presence probability is identifiable from presence-background data.
• To demonstrate that additional information beyond presence-background data is required for reliable estimation.
• To evaluate the performance of different logistic model-fitting methods under varying assumptions.

Main Methods:

• Defined seven simulated species with diverse response shapes to environmental variables.
• Applied five logistic model-fitting methods to 1000 presence and 10,000 background samples.
• Repeated simulations 100 times to assess model robustness and variability.

Main Results:

• Methods relying on a strong assumption of parametric form matching showed highly variable predictions and larger RMS error.
• Methods incorporating population prevalence as an additional parameter performed significantly better.
• Six of seven simulated species were grossly under- or overestimated by methods lacking prevalence information.

Conclusions:

• Population prevalence is necessary, not just helpful, for identifying species' probability of presence, except in special cases.
• Methods relying on strong parametric assumptions are fragile and unlikely to be true in practice; their use is advised against.
• Presence-background data must be augmented with additional data, such as prevalence, for reliable absolute probability estimation.