Application of Machine Learning Method to Predict Residual Uncertainty

This chapter presents the application of UNEEC (machine learning method to predict residual uncertainty) method to estimate uncertainty in the model prediction for the four different case studies. In first one, synthetic data set (where the exact solution is known) is used to demonstrate the method’s capability. The second case study is a typical river flow forecasting problem of the Sieve catchment in Italy. In this study the river flow forecasts are made by several data-driven techniques such as artificial neural networks, model trees, and locally weighted regression. The third case study considers a conceptual rainfall-runoff model of the Brue catchment in United Kingdom. The Hydrologiska Byråns Vattenbalansavdelning (HBV) model is used to simulate hourly river flows data. In the fourth case study, we apply the UNEEC method to a rainfall-runoff model of the Bagmati catchment in Nepal. The HBV model is used to simulate daily river flows data. In all these case studies, the results are compared with other existing uncertainty methods including the generalised likelihood uncertainty estimation (GLUE), the meta-Gaussian, and the quantile regression methods.