@MastersThesis{Quaeghebeur-2002,
title = {Analyse et pr{\'e}diction de d{\'e}bit de rivi{\`e}res par des m{\'e}thodes non lin{\'e}aires},
author = {Erik Quaeghebeur},
school = {Universit{\'e} catholique de Louvain},
year = {2002},
abstract = {This text treats of the problem of predicting the flow of a river using past flow measurements, rainfall measurements and rainfall predictions. The objective is to generalize a forecasting method already used to predict daily consumption of electricity. The developed method will be compared to Hydromax, an existing riverflow forecasting model.
The developed method simultaneously generates a series of consecutive flow predictions called a flow curve. This is done by combining separate forecasts for the mean, standard deviation and the normalized profile of the flow curve. Therefore, three separate forecasting models will be used, one for each of the aforementioned flow curve components.
To understand the difficulties involved in riverflow forecasting we first take a look at the relevant hydrological concepts and the data used in constructing and testing the forecasting method. We will principally be interested in the prediction of floods, as these phenomena can have dire socioeconomic consequences if they take place without a timely warning.
Some of the forecasting models will be based on mathematical techniques derived from the field of artificial neural networks. As an introduction we will shortly elaborate on this field before presenting a more profound study of the two derived techniques we will use. These are the so-called `self-organizing maps' and `radial basis function networks'.
An essential part of the forecasting method are linear and nonlinear regression models. We shall take a look at these parameterized prediction models, the associated prediction errors and the parameter estimation involved in constructing them.
Thus being well prepared, we will have at this stage a close look at the Hydromax model and explain the forecasting method we have developed. They will be compared in terms of the data needed and
the hydrological knowledge involved.
Finally, we will show how the developed method is used in practice. The obtained results will be compared with those obtained by Hydromax, and remarks will be made about possible improvements.
We will conclude with showing that the developed method holds promise, but is not yet suitable for practical applications. Suggestions for further research are also included. }
}