Journal of the South African Institution of Civil Engineering
Print version ISSN 1021-2019
ILUNGA, M. Infilling annual rainfall data using feedforward back-propagation Artificial Neural Networks (ANN): application of the standard and generalised back-propagation techniques. J. S. Afr. Inst. Civ. Eng. [online]. 2010, vol.52, n.1, pp. 0-0. ISSN 1021-2019.
Water resource planning and management require long time series of hydrological data (e.g. rainfall, river flow). However, sometimes hydrological time series have missing values or are incomplete. This paper describes feedforward artificial neural network (ANN) techniques used to infill rainfall data, specifically annual total rainfall data. The standard back-propagation (BP) technique and the generalised BP technique were both used and evaluated. The root mean square error of predictions (RMSEp) was used to evaluate the performance of these techniques. A preliminary case study in South Africa was done using the Bleskop rainfall station as the control and the Luckhoff-Pol rainfall station as the target. It was shown that the generalised BP technique generally performed slightly better than the standard BP technique when applied to annual total rainfall data. It was also observed that the RMSEp increased with the proportion of missing values in both techniques. The results were similar when other rainfall stations were used. It is recommended for further study that these techniques be applied to other rainfall data (e.g. annual maximum series, etc) and to rainfall data from other climatic regions.
Keywords : rainfall data infilling; artificial neural network; back-propagation.