Groups and neural networks based streamflow data infilling procedures

Hydrologic data sets are often of short duration and also suffer from missing data values. For estimation and/or extrapolation, the presence of missing data not only affects the choice of a particular method of analysis but also the resulting decision making process. Existing methods are based on the single-valued data approach and thus do not involve the effect of seasonal grouping (or segmentation) in hydrologic data prediction. Based on concepts and properties of groups and artificial neural networks, this paper develops a segment estimation model for infilling of missing hydrologic records. Efficacy of the proposed model is demonstrated through applications to a number of natural watersheds. The group-based neural network models are shown to retain relevant properties of the historical streamflows both at the auto- and cross-variate series levels. Further, the group-based neural network models are found to closely infill the missing peak flows and also the moderate flows. The results suggest that infilling of data gaps of streamflows based on the concept of neural networks and group-valued data approach is a reasonable alternative, and warrants further investigations.