Fog is a case in which water droplets or supercooled water droplets in the atmosphere while the horizontal visibility is less than 1 km. Unlike clouds, the fog is in contact with the ground at the lower boundary, which affects public transportation, ships, and aviation and requires accurate detection and forecasting. To respond to these needs, the Korea Meteorological Administration (KMA) is currently operating about 290 visibility meters to monitor and forecast the fog. The visibility meters operated by the KMA has a high observation period of one minute, but it has a low spatial representation in observation characteristics. Therefore, in order to analyze phenomena such as fog using this visibility data, quality assessment through quality inspection of visibility data is necessary. In addition, the visibility meter has higher spatial resolution than the naked-eyes, but it cannot represent the visibility of the entire Korean peninsula. High resolution grid data is needed to analyze locally generated fog characteristics. In this study, we developed a visibility data correction method (QC) using temporal and spatial continuity of visibility data and estimated grid data using Kriging method. For qualitative and quantitative verification of the QC technology, visibility data and naked-eyes data provided by the KMA were used. The QC method tries to maintain as much as possible the temporal variability of the initial data, and to eliminate the abnormal values based on the spatio-temporal continuity of visibility. In this study, to improve the spatio-temporal representation of visibility data over the last three years(2016.03 - 2019.02), high-resolution grid visibility data is estimated over South Korea using decision tree type of quality control method and Kriging.
First of all, the physical consistency and missing were checked and if the sum of missing and abnormal values exceeded the 10% of the total visibility data, the station was removed in the following analysis. After that, quality control was conducted based on the temporal continuity of the visibility data. In this process, the thresholds were set dynamically from 2 SD to 6 SD in order to reflect the average and temporal variation characteristics of the visibility data. In addition, the threshold setting method of the Norwegian Meteorological Administration was also used. And spatial continuity of visibility was used to detect abnormal values using the nearest 9 stations data. But it is difficult to set the lower and upper limits duse to the strong spatial variation of visibility data. After removal of abnormal data, the weighted 10-minute average visibility is calculated, more weight to lower visibility. Grid data with 2km resolution were estimated by Kriging using the QC visibility data. The variogram models used for kriging are exponential, Gaussian, and spherical models.
The QC results were visually and statistically evaluated (R, bias, RMSE) with the initial visibility data and the naked-eye observation data. Visual and quantitative evaluation of the QC results confirmed that the QC was satisfactorily performed. Although the difference in the quantitative verification results of the four QC methods is not large, the DWDT_B method is relatively superior.(Total: R: 0.98, RMSE: 1034m, POD: 0.87, FAR: 0.24). Among the variogram models, the exponential model shows relatively better results than the other models (R: 0.51, bias: 238m, RMSE: 3058m).