Taxonomic identification errors generate misleading ecological niche model predictions of an invasive hawkweed

Abstract

Ecological niche models (ENMs) have been proposed and applied as tools for predicting the extent of exotic species invasion risk and for identifying areas at risk of invasion. Despite the acknowledged concern of relying on occurrence records of variable and (or) unknown quality, the effect of taxonomically uncertain occurrence records on ENMs has not been investigated. We first present a schematic model describing how taxonomic uncertainty could yield varying predictions of invasion potential depending on the spatial characteristics of all versus “reliable” occurrence records. We then explore the issue in more detail by way of a case study on the morphologically and taxonomically difficult yellowdevil hawkweed (Pilosella glomerata (Froel.) Fr.), which is invasive in North America. We compared the climate niche properties and ENM predictions of invasion risk by P. glomerata in North America among models based on (i) all available occurrence records and (ii) records that are taxonomically “reliable”. “Total” records yielded niche properties that were significantly more heterogeneous than reliable records, and consequently, the potential invasion range of P. glomerata based on total records was predicted to be substantially larger. Our results provide rare empirical evidence that vetting occurrence records for taxonomic reliability is of critical importance for niche modeling.

Impacts of Species Misidentification on Species Distribution Modeling with Presence-Only Data

“…A rare example of a study that addressed the impacts of species misidentification on research using SDMs is that of Ensing, et al [31] in a study that predicted the potential distribution of an invasive species in North America. Ensing, et al [31] found that the species distribution modeled using all of the presence records available (possibly including misidentifications) was substantially larger than that based only on records regarded as taxonomically "reliable".…”

“…Ensing, et al [31] found that the species distribution modeled using all of the presence records available (possibly including misidentifications) was substantially larger than that based only on records regarded as taxonomically "reliable". A similar conclusion is drawn by Molinari-Jobin, et al [32] using reliable and non-reliable data to predict the distribution of the Eurasian lynx in the Alps with site-occupancy modeling.…”

“…Arbitrary misidentification refers here to errors that lack a clear pattern, specifically, when the source of error vary. For example, the presumed presence of a species may involve confusion with several species, recorded by different surveyors with different expertise, and following inconsistent methodologies [31,32]. Systematic misidentification refers to the systematically confusion of one species with another.…”

“…It is expected that the effects of contaminated data on modeling are related to the distribution of the species confused. For example, if the contaminating species has a broader distribution than the species of interest, its erroneous inclusion in the data set will expand the predicted distribution of the species of interest, as Ensing, et al [31] and Molinari-Jobin, et al [32] concluded. However, if the contaminating species has a narrower distribution than the species of interest, it will possibly have the opposite effect.…”

“…The latter issue has not been studied in detail. For example, the results of Ensing, et al [31] and Molinari-Jobin, et al [32] were based on data with unknown rates of species misidentification. This paper focuses on the impacts of systematic species misidentification of variable rate on SDM predictions of species presence using presence-only data.…”