Model-Based Control of Observer Bias for the Analysis of Presence-Only Data in Ecology

Abstract

Presence-only data, where information is available concerning species presence but not species absence, are subject to bias due to observers being more likely to visit and record sightings at some locations than others (hereafter “observer bias”). In this paper, we describe and evaluate a model-based approach to accounting for observer bias directly – by modelling presence locations as a function of known observer bias variables (such as accessibility variables) in addition to environmental variables, then conditioning on a common level of bias to make predictions of species occurrence free of such observer bias. We implement this idea using point process models with a LASSO penalty, a new presence-only method related to maximum entropy modelling, that implicitly addresses the “pseudo-absence problem” of where to locate pseudo-absences (and how many). The proposed method of bias-correction is evaluated using systematically collected presence/absence data for 62 plant species endemic to the Blue Mountains near Sydney, Australia. It is shown that modelling and controlling for observer bias significantly improves the accuracy of predictions made using presence-only data, and usually improves predictions as compared to pseudo-absence or “inventory” methods of bias correction based on absences from non-target species. Future research will consider the potential for improving the proposed bias-correction approach by estimating the observer bias simultaneously across multiple species.

The recent past and promising future for data integration methods to estimate species’ distributions

“…Recent studies demonstrate it is possible to use the toolbox developed for modelling spatial point processes to estimating distributions for many data types (Dorazio, ; Fithian, Elith, Hastie, & Keith, ; Hefley & Hooten, , ; Renner & Warton, ; Renner et al., ; Warton, Renner, & Ramp, ). The result is that if it is possible to link multiple data types to a common point process, it is possible to combine the methods in an integrated estimator.…”

“…This approach was predicated on the development of methods for analysing presence‐only data using a point‐process framework. Point process models have been adapted to estimate distributions and model observation error from presence‐only data (Warton & Shepherd, ; Warton et al., ) as an alternative to MaxEnt (Phillips, Dudík, & Schapire, ; Phillips et al., ). Other recent papers build on these methods (e.g.…”

“…The relationship between the observed count C , true abundance N , intensity of observed signals per unit effort S , and effort E is as follows:In the case where individuals can be identified and each individual has an equal and constant probability of being detected per unit of effort, N is identifiable when effort varies the observed count will take the form of the following:where p is the probability of detecting an individual per unit effort and E is effort. These represent versions of thinned point processes and can be modelled as such (Warton et al., ).…”

Habitat suitability modelling of rare species using Bayesian networks: Model evaluation under limited data

“…The data were collected under an adaptive survey design, whereby sites were sampled nonrandomly, concentrating efforts near or at sites where the species had previously been observed. This non-random data will bias the model to predicting suitability rather than unsuitability; this is a limitation common to many rare species' models (Warton et al, 2013).…”

“…Therefore biased data, including presence-only data (e.g. museum records, incidental sightings), are often the best form available for rare species (Warton et al, 2013;Wisz et al, 2008;Tantipisanuh et al, 2014).…”