인공신경망을 이용한 해양관측자료 보간 기법 소개

Marine Data Interpolation Using Neural Networks

In the framework of “The Comprehensive Data Analysis for Areas under the National Jurisdiction” Project, the Korea Hydrographic and Oceanographic Agency first checks data for missing or wrong values, then provides the data to the public. For stationary observation points, removing wrong or missing values entails temporal discontinuity which may not be best restored with interpolation. Manual correction is time-consuming and the decision criteria may differ among employees. The less is the quality of the data, the more specialists are to be involved, and the less objective are the models’ outcomes that use the data as an input. The focus of the current study is to overcome this by a temporal and spatial interpolation of marine data using neural networks. The experiments were done in two cases: an interpolation based on the single observation point data, and an interpolation based on the neighborhood observation points’ data. Then, each case was further subdivided into an interpolation based on six and twelve-hours time horizons. This is because, the variation in data is not large for the data within an hour timeframe, so simple nearest interpolation or linear interpolation models are sufficient. On the basis of this, the feasibility of interpolation of missing marine data was verified.