Commentary |
Corresponding author: Darren Kriticos ( darren@cervantesagritech.com ) Academic editor: Ingolf Kühn
© 2014 Darren Kriticos, Louise Morin, Bruce Webber.
This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Citation:
Kriticos D, Morin L, Webber B (2014) Taxonomic uncertainty in pest risks or modelling artefacts? Implications for biosecurity policy and practice. NeoBiota 23: 81-93. https://doi.org/10.3897/neobiota.23.7496
|
Various aspects of uncertainty have become topical in pest risk modelling discussions. A recent contribution to the literature sought to explore the effect of taxonomic uncertainty on modelled pest risk. The case study involved a high profile plant pathogen Puccinia psidii, which causes a major disease of plants within the Myrtaceae family. Consequently, the results and recommendations may attract a wide range of interest in the biosecurity and pest risk modelling communities. We found the study by
Biosecurity, eucalyptus rust, extrapolation, guava rust, MaxEnt, methods, myrtle rust, niche modelling, Puccinia psidii , species distribution modelling, Uredo rangelii
In a recent issue of Australasian Plant Pathology,
For at least one hundred years, several names had been applied to different populations of the rust fungus found on Myrtaceae in South America, now all regarded as P. psidii s.l. About thirty years ago, it was found that two different types of urediniospores are present in voucher specimens of P. psidii (
The modelling system (MaxEnt;
The covariate importance rankings in the MaxEnt models of
MaxEnt requires the modeller to specify pseudo-absence data in the form of a defined background layer that spans the known positives, and to a greater or lesser degree, regions in which there are no presence records (
In framing the modelling treatments,
It may be argued that the taxonomic in silico experimental treatments in
Figure 2 of
MaxEnt requires the modeller to specify how the model should fit covariate response functions beyond the range of conditions experienced by the training dataset (Fig. 8 in
Risk maps of modelled climate suitability and selected fitted response curves from
Developers of MaxEnt have called repeatedly for “extreme care” when extrapolating to novel climates (e.g.
Maps are potent communication devices, and two aspects of the way they have been presented in
Secondly, a reasonable interpretation of these maps appears to be that the extent of the geographical area of concern, at all levels of modelled suitability, is likely to be greater from U. rangelii than P. psidii s.l., despite U. rangelii distribution records having a narrower geographical range. Indeed, this is the interpretation intended by
“Recognition ofP. psidiisensu lato (Puccinia_94) would lead managers to place lower priority on surveillance and containment in Western Australia, and to increase the focus of activities in Australia’s northern and eastern neighbours (e.g. New Caledonia). Recognition ofU. rangelii(e.g. Uredo_27) would lead managers to increase the priority for these activities in New Zealand, Tasmania and Western Australia (Fig. 2).”.
According to both set theory and ecological reasoning this result and conclusion are implausible. In ecological terms, based on the critical assumptions underlying correlative modelling methods and niche theory, the broader the range of environmental tolerances encompassed by an organism in its native range, the broader the range of conditions we might suppose it is at least capable of inhabiting in an introduced range. That is, if Areasmall and Arealarge refer to the environmental space occupied by taxa with a smaller or larger environmental envelope respectively, the corresponding environmental space projected to be suitable (potential niche breadth, Habitatsmall and Habitatlarge) should conform to the same inequality i.e., because Areasmall ⊂ Arealarge then Habitatsmall ⊂ Habitatlarge. The results presented in
From ecological theory, we expect that closely-related species (such as U. rangelii and P. psidii s.s., if they are eventually confirmed as being different species) may competitively exclude each other from otherwise suitable habitat (
A corollary of this theory is that the inferred differences in climatic preferences based on native range sampling relate only to the realised niche (
Based on their modelling,
In order to better gauge the biosecurity implications of taxonomic uncertainty (avoiding the pitfalls in the MaxEnt example discussed here), we may be better off using true presence-only correlative models such as BIOCLIM/ANUCLIM (
Pest taxonomy is a subject that continues to challenge biosecurity agencies (
The specific issues we raise in this paper fall within the context of a relatively immature, rapidly evolving field of science, where methods are being adapted, developed and tested at such a rate that a consensus view of best practices has yet to emerge. As has been so carefully emphasised in the past by the authors of both this paper and
Some recent developments in computing technologies have been focused on making ecological modelling tools accessible to the masses (e.g.,
For end users there is a need to become more familiar with how various modelling choices affect the meaning and utility of the model results. Moreover, modellers have a responsibility to foster an effective understanding of these issues amongst biosecurity risk managers. For example, it is not possible to look at a map of a modelled species distribution, and to know instinctively what it means in terms of pest risks. There are many different ways in which the risk modelling problem can be framed, and the meaning of the model results changes accordingly.
There is clearly still much work to be done in this space, and we will need contributions from both the ecological modelling and biosecurity communities to achieve our goals of advancing best practice for pest risk modelling.
We thank John Walker for contextual information on the taxonomic history of P. psidii s.l. and Jane Elith and Mark Burgman for constructive discussions on their modelling decisions and interpretation, clarifying the background definitions and making available some of the data from their models to allow us to interrogate the output in greater detail and to construct Figure