engleski

Invading the habitat suitability models of invasive plants

Invasive alien species (IAS) related problems does not seems to decrease. Conse‑quently, neither as our attempts to better understand mechanisms of their success‑ful spreading worldwide. Among various aspects of our survey on IAS is predict‑ing their distribution, whether for purpose of optimising demanding field work (in terms of resources i.e. time and money) or for estimation of their survival/spread driven by possible future scenarios e.g. climate change. Whichever our objective is, our models will be influenced by: model type (e.g. regression, CART, DA, GAM, etc.) ; dependent data set (sample size and data precision) ; independent data set i.e. predictors (their relevancy for modelled IAS). Here, I have compared MaxEnt based habitat suitability models for several IAS plants (e.g. Erigeron anuus, Robinia pseu‑doacacia, Reynoutria × bohemica, Ailanthus altissima) for territory of Croatia. Models were developed based on different sample sizes and spatial accuracy of data on the IAS, and different sets of predictor variables in terms of their spatial resolution and content (e.g. relief, climate and disturbance). Different sets of predictor variables can yield almost incomparable outputs in terms of their spatial overlap and predicted area, especially when comparing climate‑only data with other combinations. When varying different sample size and spatial accuracy of IAS data, spatial overlap was more congruent among different models when comparing spatially more precise data (17% for Robinia and 22% for Erigeron) then with less spatially precise data (just 1% for Robinia and 2.5% for Erigeron). Comparison of different sample sizes, within same class of spatially precision of data, were much more consistent with larger overlap with increase of the spatial precision (e.g. up to 35% for Robinia). Observed inconsistencies emphasize the necessity for critical evaluation of predictive models, in which we should use our knowledge of ecology on species of interest, and not vice‑versa.

Croatia, logistic output, predictive modelling, ecology

nije evidentirano