IWA Publishing
 IWA Publishing Journals   Subscriptions   Authors   Users   Librarians   FAQs 

Hydrology Research In Press, Uncorrected Proof © IWA Publishing 2014  |  doi:10.2166/nh.2014.168

Application of self-organising maps and multi-layer perceptron-artificial neural networks for streamflow and water level forecasting in data-poor catchments: the case of the Lower Shire floodplain, Malawi

F. D. Mwale, A. J. Adeloye and R. Rustum

School of Built Environment, Heriot Watt University, Riccarton, Edinburgh EH14 4AS, UK E-mail: fdm4@hw.ac.uk
School of Built Environment, Heriot Watt University, Dubai Campus, United Arab Emirates

First received 4 October 2012; accepted in revised form 22 January 2014. Available online 12 February 2014


ABSTRACT

With a paradigm shift from flood protection to flood risk management that emphasises learning to live with the floods, flood forecasting and warning have received more attention in recent times. However, for developing countries, the lack of adequate and good quality data to support traditional hydrological modelling for flood forecasting and warning poses a big challenge. While there has been increasing attention worldwide towards data-driven models, their application in developing countries has been limited. A combination of self-organising maps (SOM) and multi-layer perceptron artificial neural networks (MLP-ANN) is applied to the Lower Shire floodplain of Malawi for flow and water level forecasting. The SOM was used to extract features from the raw data, which then formed the basis of infilling the gap-riddled data to provide more complete and much longer records that enhanced predictions. The MLP-ANN was used for the forecasting, using alternately the SOM features and the infilled raw data. Very satisfactory forecasts were obtained with the latter for up to 2-day lead time, with both the Nash–Sutcliffe index and coefficient of correlation being in excess of 0.9. When SOM features were used, however, the lead time for very satisfactory forecasts increased to 5 days.

Keywords: data-poor catchments; forecasting; multi-layer perceptron artificial neural networks; self-organising maps; streamflow; water level


Full Text PDF