engleski

Analysing consistency of habitat suitability models for invasive plants across different sample sizes and spatial resolutions of data

Chorological data on invasive plants in Croatia are quite heterogeneous with respect to their spatial accuracy. Habitat suitability models are nowadays widely used for assessing the potential spread of invasive species. However, spatial precision and/or resolution can significantly influence the model output. Here, we have analysed consistency of habitat suitability models across different sample sizes and spatial resolutions of species (dependent variables) and environmental (independent variables) data. Data on 15 invasive species, and 5 natives that share similar environmental envelopes as a control were used, as well as 15 environmental variables (climatic, topographic, proxies for disturbance) for habitat suitability models development using MaxEnt. For species with sufficient amount of data within different spatial resolution classes (e.g. Erigeron anuus (L.) Pers., Robinia pseudoacacia L.), models were developed across different spatial resolution, using environmental data on three spatial resolutions (300-meters, 1-km and 5-km), separately. Using a conservative threshold value of 0.5, we reclassified Maxent models output grids into habitat suitability maps, and compared consistencies (i.e. overlapping) of their spatial coverage across combinations of different data sets. In general, models developed with lower spatial resolution of species data, smallest sample size and lower spatial resolution of environmental data tends to predict larger areas as suitable habitats. Habitat suitability models were more consistent with respect to spatial precision of chorological data, then to their sample sizes. Higher spatial resolution of the environmental data (i.e. 300-meters grids) yielded results that are more consistent across different combinations of sample size and spatial precision of species data, than those with 1-km and 5-km environmental grids. Obtained results suggests that we should focus more on spatial precision of our data then sample size in order to decrease uncertainty of our predictive models, and conclusions we yield based on those models.

MaxEnt ; Croatia ; black locust

nije evidentirano