Outlier detection and missing data filling methods for coastal water temperature data

Abstract

Outlier detection and missing data filling (imputation) processes are essential first step in the statistical analysis of coastal monitoring data. Here, we suggest methods for completing these key processes. An outlier detection method that uses approximate and detailed components is suggested. The decomposition of the time-series data is performed by harmonic analysis. Next, the modified z-score method is applied to the residuals (detailed component) to detect outliers. After removing the outliers in the residuals, the filling process for the missing and removed outlier data is conducted by summing the random and the approximate components. Among the environmental monitoring data, this method is applied to the coastal water temperature data. We used hourly interval coastal water temperature data provided by the NFRDI (National Fisheries Research & Development Institute). In these datasets, the dataset of the Yeong-Deok Geomuyeok (36.58 degrees N, 129.40 degrees E) station, Korea, is only used for this method application. This dataset contains some outliers and missing data. To test the model performance, this method is applied to a daily interval modeling dataset from the HYCOM (Hybrid Coordinate Ocean Model). This method provides reasonable results for outlier detection and for filling in missing data in coastal water temperature datasets.