Package landmap and function train.spLearner provide functionality for fitting Ensemble Machine Learning models where both geographical distance and features (gridded layers) are used to produce spatial interpolations. This is an all integrative approach to geostatistical mapping where multiple machine learning techniques are combined with variogram modeling and spatial cross-validation. For geographical distances we use by default oblique geographic coordinates, although these could be further extended. In this block I will demonstrate how to use train.spLearner to train models and produce spatial interpolations (maps) from point data for numeric and factor-type variables. This approach in general produced stable results and is suited for non-linear relationships and training datasets with skewed distributions. Additional steps to improve the model performance and computing speed include: feature selection, model fine-tuning, more efficient spatial CV.