COMPLEMENT METHOD OF MISSING OBSERVATION FLOW DATA BY MEANS OF DEEP LEARNING METHOD

Abstract

 Streamflow data are important for river maintenance, water resources planning, and flood forecasting. However, the flow data may be missing due to various reasons. Therefore, this study proposed a novel method using deep learning to complement missing data of flow discharge time series. This study utilized hourly flow data before and after missing and precipitation data as input. As the deep learning methods, this study selected 1D CNN and MLP. The results showed the high capability of 1D CNN and MLP. Especially, 1D CNN was able to estimate the missing flow rate well. In addition, it the results in this study indicate that it is important to use precipitation data in addition to flow discharge data as input in order to accurately estimate missing flow discharge.