Importance of spatial predictor variable selection in machine learning applications – Moving from data reproduction to spatial prediction

•Spatial cross-validation is crucial for reliable error estimates of predictions.•Geolocation as predictor variable can lead to overfitting and unreliable predictions.•Spatial variable selection is required to detect problematic predictor variables.•Models considerably improve when spatial variable selection is applied.•If spatial dependencies are ignored, models can reproduce data but not predict them.