Learning from mistakes - Assessing the performance and uncertainty in process-based models