Browsing by Author "Farstad, Miia"

Due to the harsh conditions in high latitude alpine and arctic regions, climate or land use changes make them very vulnerable. Thus, it is vital to study the habitats of these regions and increase our understanding of what factors impact species distributions. Species distribution modelling can be used to predict possible habitats for species and further inspect the relationships between different environmental variables and species. Generally, these species distribution models have been created using variables describing the topographical and climatic conditions of the study area. Recently there has been more evidence supporting the inclusion of biotic variables to species distribution models at all scales. Including biotic variables can be difficult, as these relationships can be challenging to quantify. This study uses the Normalized Difference Vegetation Index (NDVI) as a surrogate for plant biomass, thus representing biotic interactions. This study aims to answer what are the relationships between environmental variables and the predicted distributions and will including a biotic variable improve the species distribution models. The study data includes observational data from 683 arctic and alpine plant species from Norway, Sweden, and Finland. The observation data were collected from the three national databanks of Norway, Sweden and Finland and completed with observations from the Global Biodiversity Information Facility and observation data collected by the BioGeoClimate Modelling Lab. The cohesive study area was outlined with the biogeographical regions defined by the European Environment Agency. Overall, six environmental variables are used in this study: annual mean temperature, the maximum temperature of the warmest month, annual precipitation, elevation difference in a cell, bedrock class, and NDVI. The NDVI data was gathered by NASA’s MODIS sensors. The observations and the environmental variables were projected into a grid consisting of 1 x 1 km cells covering the whole study area. This study uses the ensemble modelling technique with four individual modelling methods: generalized linear models (GLM), generalized additive models (GAM), generalized boosted models (GBM) and random forests (RF). The modelling process consisted of two modelling rounds so that the impact of NDVI could be evaluated. The first modelling round included all the environmental variables except NDVI (the topoclimate model) and the second modelling round included all the environmental variables (the full model). The two temperature variables, annual mean temperature and the maximum temperature of the warmest month, had the highest mean variable importance values. With the topoclimate model, annual precipitation ranked third with the rest of the climate variables, but when NDVI was added to the models, it rose above annual precipitation. Overall, among the studied arctic and alpine species, the variable importance values of both the edaphic and topographical variables were low. In general, both the topoclimate models and full models performed very well. The mean AUC- and TSS-values were all higher for the full models, indicating that including a biotic variable improved the models. When the binary predictions of both modelling rounds were compared, it was clear that NDVI refined the projected distributions for most species. The results from this study confirm the discovery that including a biotic variable, such as NDVI, has the potential to increase the predictive power of species distribution models. One of the main problems with including biotic variables in species distribution models has been the difficulty of quantifying biotic interactions. NDVI can thus be a promising tool to overcome these difficulties, as it is one of the most direct variables to describe ecosystem productivity, can be acquired at various scales, and as remotely sensed data, it can also cover areas that are difficult to access.