Reconstruction of missing daily streamflow data using recurrent neural network

Abstract

    Streamflow data is one of the most important quantities in hydrology because of these data closelyrelated to water resource management problems as well as flood forecasting problems. The lack ofthese data can lead to inadequate scientific analysis. Therefore, reconstruction of missing data is animportant step to get more reliable and accurate information. The objective of this paper is to introducean effective approach based on the recurrent neural network model to reconstructing missing dailydischarge data. Lai Chau hydrological station, located upstream of the Da River basin, was selected asa case study. The findings of this study demonstrated that the recurrent neural network model yieldsreliable estimates for the problem of missing data. As a result, the RNN model can be applied to otherhydrological stations upstream where the flow data is missing.