An artificial neural network approach to rainfall-runoff modelling

Abstract

This paper provides a discussion of the development and application of Artificial Neural Networks (ANNs) to flow forecasting in two flood-prone UK catchments using real hydrometric data. Given relatively brief calibration data sets it was possible to construct robust models of 15-min flows with six hour lead times for the Rivers Amber and Mole. Comparisons were made between the performance of the ANN and those of conventional flood forecasting systems. The results obtained for validation forecasts were of comparable quality to those obtained from operational systems for the River Amber. The ability of the ANN to cope with missing data and to "learn" from the event currently being forecast in real time makes it an appealing alternative to conventional lumped or semi-distributed flood forecasting models. However, further research is required to determine the optimum ANN training period for a given catchment, season and hydrological contexts.Une approche de la modélisation pluie-deblt par ies réseaux neuronaux artificiels Résumé Ce document traite du développement et de l'application des réseaux neuronaux artificiels (RNA) à la prévision des débits de deux bassins versants du Royaume Uni sujets aux inondations grâce à l'utilisation de données hydrométriques réelles. Partant d'un ensemble restreint de données d'apprentissage, il a été possible de réaliser des modèles pour la prévision des débits au pas de temps de 15 min à échéance de 6 heures pour les rivières Amber et Mole. On a comparé les performances des RNA et des systèmes conventionnels d'annonce de crue. Les résultats obtenus lors de la validation des prévisions des RNA étaient de qualité comparable à ceux obtenus par les systèmes actuellement utilisés opérationnellement sur la Rivière Amber. La capacité des RNA à gérer les données manquantes et à "apprendre" en temps réel à partir de l'événement en cours, fait de ces outils une alternative séduisante aux actuels modèles de prévision agrégés ou semi-distribués. De plus amples recherches sont cependant nécessaires pour déterminer la période d'apprentissage optimale des RNA pour un bassin donné et selon le contexte climatique et hydrologique.

Using support vector machines for long-term discharge prediction

“…Since the 1990s, artificial neural networks (ANN) (ASCE Task Committee, 2000a,b), which are based on the architecture of the brain and nervous system, were gradually used in hydrological prediction (cf. Dawson & Wilby, 1998;See & Openshaw, 2000;Hu et al, 2001Hu et al, , 2005Campolo et al, 2003;Cigizoglu, 2003;Wilby et al, 2003;Giustolisi & Laucelli, 2005;Sy, 2006). In this paper, an attempt is made to examine any possible improvement in forecasting accuracy by employing the support vector machine (SVM) model (Vapnik et al, 1997).…”

Group-based estimation of missing hydrological data: I. Approach and general methodology

“…These include, among others, rainfall forecasting (French et al, 1992), multivariate modelling of water resources time series (Raman & Sunilkumar, 1995), modelling of rainfall-runoff processes (Hsu et al, 1995), flow forecasting (Zealand et al, 1999;Dawson & Wilby, 1998), and river level forecasting (See & Openshaw, 1999). The promising results due to the use of ANNs in water resources make them a feasible technique to be employed in this research.…”

Artificial neural networks and high and low flows in various climate regimes

“…Although ANNs have already been shown to produce river flow predictions well compared to conventional models (Crespo & Mora, 1993;Karunanithi et al, 1994;Hsu et al, 1995;Abrahart & Kneale, 1997;Dawson & Wilby, 1998;Abrahart & See, 2000;Tingsanchali & Gautam, 2000), their ability to capture high and low flows is restricted to the research environment (Minns & Hall, 1996), and they often overestimate or underestimate high and low flows (Dawson & Wilby, 1998;Campolo et al, 1999;Karunanithi et al, 1994).…”

“…For example, Dawson & Wilby (1998), and Campolo et al (1999) suggested that the underestimation of peak flows could be attributed to a lack of information provided to the network, such as the antecedent rainfall. Karunanithi et al (1994) suggested that the problem could be alleviated by including more high-flow patterns in the training data sets, while Hsu et al (1995) proposed log-transformations of flow values to reduce the gap between the high-and low-flow conditions.…”