Distribution modelling of vegetation types in the boreal–alpine ecotone

Abstract

AbstractQuestionsVegetation mapping based on field surveys is time‐consuming and expensive. Distribution modelling might be used to overcome these challenges. What is the performance of distribution modelling of vegetation compared to traditional vegetation mapping when projected locally? Does the modelling performance vary among ecosystems? Does vegetation type distribution and abundance influence the modelling performance?LocationGravfjellet, Øystre Slidre commune, southern Norway.MethodsTwo comparable neighbouring areas, each of 4 km2, were mapped for species‐defined vegetation types. One area was used for model training, the other for model projection. Maximum entropy models were run for six vegetation types, two from each of the ecosystems present in the area: forest, wetland and mountain heath‐ and shrublands. For each ecosystem, one locally abundant and one locally rare vegetation type were tested. AUC, the area under the receiver operating curve, was used as the model selection criterion. Environmental variables (n = 9) were selected through a backwards selection scheme, and model complexity was kept low. The models were evaluated using independent data.ResultsDistribution modelling of vegetation types by local projection gave high AUC values, and the results were supported by the evaluation using independent data. The modelling ability was not affected by ecosystem differences. A negative relationship between the number of points used to train the models and the AUC value before evaluation suggests that models for locally rare vegetation types had better predictive performance than the models for abundant types. This result was not significant after evaluation.ConclusionProvided that relevant explanatory variables are available at an appropriate scale, and that field‐validated training points are available, distribution modelling can be used for local projection of the six tested vegetation types from the boreal–alpine ecotone.