Species distribution modeling can be effectively carried out using open data and data analysis tools with machine learning techniques. Modeling of the distribution of Phyteuma genus in the Carpathian region has been carried out with data from the GBIF database, climatic data from the Worldclim database, and soil properties data from Soilgrids soil information system. Spatial distribution modeling was accomplished with machine learning techniques that have marked advantages over more traditional statistical methods, like the ability to fit complex nonlinear relationships common in ecology.
Four methods have been examined: Maxent, Random Forest, Artificial Neural Networks (ANN), and Boosted Regression Trees. AUC and TSS criteria calculated for testing data with cross-validation have been applied for assessing the performance of the models and to tune their parameters. ANN with a reduced set of predictor variables (6 from initial 21) appeared to fare the best and was applied for predictive modeling. Prospective data based on future climate projections from Worldclim were input to the model to get the prospective distribution of the plant taxon considering expected climate changes under different RCPs