A novel method for model uncertainty estimation in rainfall-runoff modelling using machine learning techniques has been developed. This method is referred to as UNEEC (UNcertainty Estimation based on Local Errors and Clustering). In this method, model error is seen as the indicator of the total (overall) model uncertainty. First, the probability distribution of the model error is estimated separately for different hydrological situations, and, second, the parameters characterizing these distributions are aggregated and used as output target values for building the training sets for the machine learning model. This latter model, being trained, encapsulates the information about the model error localized for different hydrological conditions in the past, and is used to estimate the probability distribution of the model error for new hydrological model runs. Average mutual information and correlation analysis are used to determine the relevant parameters characterizing hydrological situations and the input variables for the learning models. Different data-driven modelling techniques are uses used as learning engine.
Novelty of the approach is in the following:
(a) no assumptions are made about the pdf of residuals;
(b) building the uncertainty model specialized for a particular area of the state space (hydrometeorological condition) which is identified by fuzzy clustering; and
(c) building the uncertainty model using machine learning techniques.
The UNEEC method consists of three main parts:
(b) Estimation of the error probability distribution for clusters; and
(c) Building the overall model of the error probability distribution.
Click here to show the detailed picture of the UNEEC method.