The random forest classification model was implemented with the “r.learn.ml” add-on in GRASS GIS, which acts as a front-end to the Python scikit-learn machine learning library (Pawley, 2018). A training sample was obtained by combining a set of “pseudo-absence” points and observation point data (Acer macrophyllum pursh, 2018).
Two test models were built using climate normals data from ClimateWNA (Hamann, 2013), which assessed the parameters for their predictive value. The final model was built with 11 bioclimatic and topographic parameter rasters, and a 10-fold cross validation was performed to assess accuracy using a number of metrics including AUC and the kappa statistic. This model was then applied to future climate scenarios to predict changes in suitable habitat.