Missing Discharge Data Filling with Artificial Neural Network

Volume 9 Issue 2 March - May 2019

Research Paper

Missing Discharge Data Filling with Artificial Neural Network

Janhabi Meher*

Assistant Professor, Department of Civil Engineering, VSSUT, Burla, Odisha, India.

Meher, J. (2019). Missing Discharge Data Filling with Artificial Neural Network, i-manager's Journal on Civil Engineering, 9(2), 24-30. https://doi.org/10.26634/jce.9.2.14657

Abstract

Artificial Neural Network has been widely used for filling in missing nonlinear data and successfully employed in the hydrologic applications. In this study, Artificial Neural Network is applied to fill in missing discharge data of nineteen discharge stations of Mahanadi river basin located at Chhattisgarh and Odisha in India for the five monsoon months (June-October). However, the overall performance of ANN generally depends on selection of input and output datasets, called input data selection method. The input data selection method applied in this study is using data as input having good (r≥0.9) correlation coefficients with target data. It was observed that coefficient of determination (R2) between observed discharge and predicted discharge for the nineteen stations varied from 0.148-0.954, 0.250-0.991, 0.026- 0.972, 0.148-0.972, and 0.213-0.985 for the months of June, July, August, September, and October, respectively. Predicted discharge of a particular station for a particular month having correlation greater than 0.9 with very few numbers of stations or with no station showed very low match with the observed discharge of that station. Therefore, it can be concluded that Artificial Neural Network requires sufficiently large number of inputs to accurately predict the target.