Machine learning is increasingly used in environmental sciences for spatial predictions, but its effectiveness is challenged when models are applied beyond the areas they were trained on. We propose a Local Training Data Point Density (LPD) approach that considers how well a model's environment is represented by training data. This method provides a valuable tool for evaluating model applicability and uncertainties, crucial for broader scientific and practical applications.

Spatial proxies, such as coordinates and distances, are often used as predictors in random forest models for predictive mapping. In a simulation and two case studies, we investigated the conditions under which their use is appropriate. We found that spatial proxies are not always beneficial and should not be used as a default approach without careful consideration. We also provide insights into the reasons behind their suitability, how to detect them, and potential alternatives.

Estimation of map accuracy based on cross-validation (CV) in spatial modelling is pervasive but controversial. Here, we build upon our previous work and propose a novel, prediction-oriented k-fold CV strategy for map accuracy estimation in which the distribution of geographical distances between prediction and training points is taken into account when constructing the CV folds. Our method produces more reliable estimates than other CV methods and can be used for large datasets.