Comparison of data-driven modelling techniques for river flow forecasting

Abstract

Accurate forecasting of streamflow is essential for the efficient operation of water resources systems. The streamflow process is complex and highly nonlinear. Therefore, researchers try to devise alterative techniques to forecast streamflow with relative ease and reasonable accuracy, although traditional deterministic and conceptual models are available. The present work uses three data-driven techniques, namely artificial neural networks (ANN), genetic programming (GP) and model trees (MT) to forecast river flow one day in advance at two stations in the Narmada catchment of India, and the results are compared. All the models performed reasonably well as far as accuracy of prediction is concerned. It was found that the ANN and MT techniques performed almost equally well, but GP performed better than both these techniques, although only marginally in terms of prediction accuracy in normal and extreme events.Key words streamflow; data-driven modelling; artificial neural networks; genetic programming; M5 model trees Comparaison de techniques de modélisation conditionnée par les données pour la prévision des débits fluviaux Résumé La prévision précise des débits est essentielle pour l'opération efficace des systèmes de ressources en eau. Le processus d'écoulement est complexe et fortement non-linéaire. Par conséquent, les chercheurs essaient de mettre au point des techniques alternatives pour prévoir le débit avec une facilité relative et une précision raisonnable, bien que des modèles traditionnels déterministes et conceptuels soient disponibles. Le présent travail utilise trois techniques conditionnées par les données, en l'occurence des réseaux de neurones artificiels (RNA), la programmation génétique (PG) et les arbres modèles (MT) pour prévoir les débits de rivière un jour à l'avance en deux stations du bassin versant de Narmada (Inde), et les résultats sont comparés. Tous les modèles sont assez performants en termes de précision de prévision. Les techniques RNA et MT ont des performances presque égales, mais PG donne de meilleurs résultats, bien que seulement de façon marginale en termes de précision de prévision pour des événements normaux et extrêmes.

Comparative Study of M5 Model Tree and Artificial Neural Network in Estimating Reference Evapotranspiration Using MODIS Products

“…According to the statistics of RMSE and MAE, both methods performed equally well. Similar results have been reported about close performance of ANN and M5 models in rainfallrunoff modeling [43] and river flow forecasting [44]. The main advantage of M5 model tree is that it provides the results in a simple and comprehensible form of regression equations [1], which can be easily used in the calculation of ET .…”

Genetic Programming: A Novel Computing Approach in Modeling Water Flows

“…Thus Genetic Programming entered in rainfall-runoff modeling. It was also found that GP results were superior to that of M5 Model Trees another data driven modeling technique [12,13]. Apart from these two variables the use of GP for modeling for many hydraulic engineering processes was found necessary for similar reasons.…”

“…However if the technique is to stay here it needs to be explored further for more challenging problems like modeling of infiltration, high flood events, hurricane path, storm surge, tsunami water levels to name a few. [13] …”

Short-Term Forecasting of Water Yield from Forested Catchments after Bushfire: A Case Study from Southeast Australia

“…Previously, short-term estimation and forecast of catchment water yield were based on a range of methods, from purely empirical simple models to highly sophisticated distributed process-based models defined by partial differential equations (e.g., the Systeme Hydrologique Europeen model [11] or the Macaque model [9]). Over the past decades, data-driven models have become increasingly useful in hydrological forecasting on the basis that they avoid having to address the problems of the spatial and temporal variability, and the uncertainty of the inputs and the parameters, as opposed to the physically-based models that require a wide range of catchment and climate information [12][13][14].…”