Machine Learning in Prediction of Residual Uncertainty: UNEEC Method

This chapter presents a novel methodology to analyse, model, and predict the residual uncertainty which, for the purpose of this thesis, is considered to be represented by the model residual errors. The methodology aims to predict the uncertainty of an optimal (calibrated) model output by building machine learning model of the probability distributions of its historical residuals. The novel feature of the method is that the data to train the model is constructed in such a way that it carries the information about the dependencies characteristic to the particular regions of the input space (hydrometeorological and flow conditions). These regions correspond to natural clusters of data in input space. This method is referred to as UN certainty E stimation based on Local E rrors and C lustering (UNEEC). This abbreviation may also stand for “uncertainty prediction method developed at UNE sco-ihe in the framework of a project funded by EC ”.