Prevalence, thresholds and the performance of presence–absence models

By C. Lawson, J. Hodgson, Robert J. Wilson, S. Richards. 2014

175 citations recorded by the source corpus.

Subfields: Ecological Modeling, Ecology, Nature and Landscape Conservation

View Paper: https://www.semanticscholar.org/paper/9243fd9a235464c354d9ca3c466123c1b79b67a9

How researchers see this paper

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

In Ecological Modeling

In Ecology

In Nature and Landscape Conservation

Citations in the corpus

A comprehensive evaluation of predictive performance of 33 species distribution models at species and community levels (2019, background): “We note that many previous applications of SDMs have evaluated them based on either the predicted speciesspecific marginal occurrence probabilities, or occurrences derived by thresholding the occurrence probabilities (Liu et al. 2005, Jim enez-Valverde and Lobo 2007, Lawson et al. 2014).”
A standard protocol for reporting species distribution models (2020, background): “Whilst there are good arguments for retaining predictions on a continuous scale (Lawson et al. 2014, Guillera-Arroita et al. 2015), some users prefer to threshold them for certain applications.”
Defining and evaluating predictions of joint species distribution models (2020, methodology): “In a single-species context, choice of metric depends largely on factors such as data type (e.g. presence–absence vs. presence-only) and prediction format (binary or probabilistic; Lawson et al., 2014).”
Forecasting the global extent of invasion of the cereal pest Spodoptera frugiperda, the fall armyworm (2018, methodology): “The Area under the Receiver Operating Curve (AUC) and True Skill Statistic (TSS) were used to judge how accurately the models predicted the validation data (Lawson et al. 2014).”
Two alternative evaluation metrics to replace the true skill statistic in the assessment of species distribution models (2019, background): “That is, two different evaluation metrics may represent true model performance better under varying circumstances (Lawson et al. 2014).”
The area under the precision‐recall curve as a performance metric for rare binary events (2019, background): “While AUC- ROC is mathematically independent of prevalence (Ferro & Stephenson, 2011; Jiménez- Valverde, 2014; Lawson et al., 2014), in practice it is related to prevalence in presence–absence datasets (reviewed in Santika, 2011).”
Without quality presence–absence data, discrimination metrics such as TSS can be misleading measures of model performance (2018, background): “These flaws can be strongly problematic for ecologists, and during SDM performance evaluation it is generally preferable to assume that overprediction should be equivalent to underprediction (e.g. Lawson, Hodgson, Wilson, & Richards, 2014).”