Correction note: Poisson point process models solve the “pseudo-absence problem” for presence-only data in ecology

By David I. Warton, L. Shepherd. 2010

334 citations recorded by the source corpus.

Presence-only data, point locations where a species has beenrecorded as being present, are often used in modeling the distribu-tion of a species as a function of a set of explanatory variables—whether to map species occurrence, to understand its associationwith the environment, or to predict its response to environmentalchange. Currently, ecologists most commonly analyze presence-onlydata by adding randomly chosen “pseudo-absences” to the data suchthat it can be analyzed using logistic regression, an approach whichhas weaknesses in model speciﬁcation, in interpretation, and in imple-mentation. To address these issues, we propose Poisson point processmodeling of the intensity of presences. We also derive a link betweenthe proposed approach and logistic regression—speciﬁcally, we showthat as the number of pseudo-absences increases (in a regular or uni-form random arrangement), logistic regression slope parameters andtheir standard errors converge to those of the corresponding Poissonpoint process model. We discuss the practical implications of theseresults. In particular, point process modeling oﬀers a framework forchoice of the number and location of pseudo-absences, both of whichare currently chosen by ad hoc and sometimes ineﬀective methods inecology, a point which we illustrate by example.

Subfields: Applied Mathematics, Artificial Intelligence, Geometry and Topology

View Paper: https://www.semanticscholar.org/paper/fd4427fba43c25d2aed0fe87a3b1e974342946ee

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Applied Mathematics

In Artificial Intelligence

In Geometry and Topology

Citations in the corpus

Opening the black box: an open-source release of Maxent (2017, background): “…an independent and identically distributed (i.i.d.) sample from pl, whose size (number of points) is a Poisson random variable with mean λ( ) .z dz D∫ Warton and Shepherd (2010) suggested modeling the occurrence records for a species as arising from an IPP whose intensity l(z) is a log-linear…”
MaxEnt versus MaxLike: empirical comparisons with ant species distributions (2013, methodology): “In instances when differences in probability thresholds between MaxLike and MaxEnt were significant, MaxLike had a higher probability threshold than MaxEnt, except for bias-corrected models for Monomorium emarginatum.”
Practical guidance on characterizing availability in resource selection functions under a use-availability design. (2013, methodology): “Such assessments could follow the methods present- ed here, and those suggested elsewhere (e.g., Manly et al. 2002, Warton and Shepherd 2010, Aarts et al. 2012) where multiple samples of varying sizes are tested until coefficient estimates converge.”
The effect of sample size on the accuracy of species distribution models: considering both presences and pseudo‐absences or background sites (2019, neutral): “2006, Warton and Shepherd 2010, Chakraborty et al.”
Nonparametric estimation of the dependence of a spatial point process on spatial covariates (2012, background): “This is equivalent to pixel-based logistic regression of the probability of the presence of a gold deposit against distance to nearest fault [1, 9, 67].”
Improving species distribution model predictive accuracy using species abundance: Application with boosted regression trees (2020, background): “A number of studies have focused on data deficiency (Warton and Shepherd 2010; Elith et al. 2011; Fithian and Hastie 2013; Yackulic et al. 2013; Radosavljevic and Anderson 2014; Renner et al. 2015).”
Integrated species distribution models: combining presence‐background data and site‐occupancy data with imperfect detection (2017, background): “…specifically formulated under the assumption that species could be present or absent at locations in the background sample: Maxent models (Phillips, Anderson & Schapire 2006), case-augmented binaryregression models (Lee, Scott & Wild 2006), and Poisson point-process models (Warton & Shepherd 2010).”
The area under the precision‐recall curve as a performance metric for rare binary events (2019, methodology): “It is recommended to use a large number of background locations (Warton & Shepherd, 2010; Northrup, Hooten, Anderson, & Wittemyer, 2013), and as a result, the ratio of ones (presence) to zeros (background, which could be true presences or absences) in data used for model estimation and evaluation…”
On the selection of thresholds for predicting species occurrence with presence‐only data (2015, methodology): “Point process modeling has also been introduced to build models with presence-only data (Warton and Shepherd 2010; Chakraborty et al. 2011; Aarts et al. 2012).”
Hierarchical Species Distribution Models (2016, methodology): “As a result, variants of the Poisson point process model are popular for modeling species’ distributions using presenceonly data [83, 99].”
A comparison of Maxlike and Maxent for modelling species distributions (2014, methodology): “…loglinear model and its inability to estimate an intercept can be understood by its connection to inhomogeneous poisson process (IPP) models (cf. Warton & Shepherd 2010; Chakraborty et al. 2011) although other derivations are possible (Phillips, Anderson & Schapire 2006; Halvorsen 2012; Merow,…”
Comparison of five modelling techniques to predict the spatial distribution and abundance of seabirds (2012, background): “More sophisticated models that simultaneously model the observation and process uncertainty and can thus account for imperfect detection may provide better estimates of spatial abundance patterns in the future (Dail and Madsen, 2011; Hutchinson et al., 2011; Warton and Shepherd, 2010).”
Comparative interpretation of count, presence–absence and point methods for species distribution models (2012, methodology): “Furthermore, in Poisson GLMs, the choice of the spatial scale of the grouping is arbitrary, while for the IPP the number of availability points can simply be increased until an acceptable approximation is achieved (Warton & Shepherd 2010).”
Establishing the link between habitat selection and animal population dynamics (2015, methodology): “Use can be equally well defined infinitesimally on single points in space as the spatially varying intensity function of an inhomogeneous Poisson process (Warton and Shepherd 2010, Aarts et al. 2012).”
Protected areas network is not adequate to protect a critically endangered East Africa Chelonian: Modelling distribution of pancake tortoise, Malacochersus tornieri under current and future climates (2020, background): “1) are a general class models that predict count data (i.e. the number of individuals; Warton and Shepherd 2010, Aarts et al. 2012, Fithian and Hastie 2012, Renner and Warton 2012, Royle et al. 2012).”