PRIMEWATER

In data-driven modeling, PrimeWater goals are two-fold. First, PrimeWater aims to investigate the potential of a complementary modeling framework to improve the predictions of process-based models by integrating process-based models with add-on data-driven models. The data-driven models are trained to recover relevant information that is probably contained in systematic errors, which are inevitably present, as there will be always mechanisms that a model is unsuitable to describe. Secondly, PrimeWater aims to address the lack of a proper uncertainty quantification in water quality predictions that applies to most machine learning algorithms employed in water quality modeling. Uncertainty is a valuable component in forecast-based decision making and, in order to address the issues of uncertainty in predictions, PrimeWater utilizes probabilistic random forests and Gaussian Process regression models, which have demonstrated their strengths as modelling tools in other domains, regardless..