When creating a prediction, i wonder where to use the output with the trained test data to generate a predicton of the actual data. So, will there be always created a new model and used on the actual data or is it stored in the background and used when creating a prediction without specifing the model to use? If i want to use the trained model for another region/dataset where has the model to be specified then?
Is this the section you are referring to?
Predicting using a forest-based model
Predicting in a different study area
Forest-Based Classification and Regression (FBCR) tool will train a model from scratch on the training data set and it can optionally apply the trained model on a "test data set". Meaning regardless of running the tool for training a model or predicting, the tool always creates a model in the background. A suggested use pattern for FBCR is first finding a good model in Train Mode (first option of the tool) based on R2 and variable importance distribution and once a satisfactory model is reached, switching over to "Predict to Features" or "Predict to Rasters" to apply the model to a prediction data set. Note that in prediction model FBCR will still train a model first on your original training data and apply it to the prediction data set. If you have another location that contains all of your predictors, you can define it as your prediction data set. As you mentioned, we achieve this by creating a FBCR model internally every time the tool is run.