Modelling establishment probabilities of an exotic plant, Rhododendron ponticum, invading a heterogeneous, woodland landscape using logistic regression with spatial autocorrelation

Catriona M Harris, Monique Lea MacKenzie, Colin Edwards, Justin MJ Travis

Research output: Contribution to journalArticlepeer-review

40 Citations (Scopus)

Abstract

Rhododendron ponticum has become a well-established invasive species throughout the British Isles and is now considered a problematic invasive weed species. The habitat requirements for establishment, however, have only previously been described qualitatively. The aim of this study was to quantify the influence of topographical and environmental characteristics on the probability of establishment of this invasive shrub in a woodland environment. A binomial generalized linear model (GLM), incorporating spatial autocorrelation, was used to model the presence/absence of R. ponticum seedlings (<7-year-old plants)
given substrate type and depth, canopy type and percent cover, altitude, slope, aspect and the distance to the nearest seed source as covariates. Depth and type of substrate along with distance from the closest seed source were found to be the most important predictors of seedling establishment. Specifically, fallen logs or tree stumps, newly colonised by moss, were identified as the most favourable habitat type for R. ponticum invasion. We show that the inclusion of spatial autocorrelation can affect model conclusions and, as a consequence, may influence management recommendations. Knowing how to reduce establishment can assist in managing invasive populations. Similarly, understanding how establishment probabilities vary according to habitat may aid the conservation of threatened populations.
Original languageEnglish
Pages (from-to)747-758
JournalEcological Modelling
Volume193
Publication statusPublished - 2006

Fingerprint

Dive into the research topics of 'Modelling establishment probabilities of an exotic plant, Rhododendron ponticum, invading a heterogeneous, woodland landscape using logistic regression with spatial autocorrelation'. Together they form a unique fingerprint.

Cite this