There have been many studies to improve visibility forecast skills using numerical models, but their performance still remains behind the forecast skills for other meteorological phenomena. This study attempted to improve visibility forecasts using a newly established automatic visibility observation network composed of 291 forward-scattering sensors in South Korea. In the analysis of recent 3-year visibility observations, clear days (visibility above 20 km) were reported for 46% of the days, and fog cases (visibility less than 1 km) accounted for 1.58% of the total observations. The Very short-range Data Assimilation and Prediction System (VDAPS) of the Korea Meteorological Administration (KMA) assimilated the visibility observations based on the Met Office Unified Model with visibility data assimilation of Clark et al. (Q J R Meteorol Soc 134:1801–1816, 2008). Prior to the data assimilation, a precipitation check eliminated visibility data with precipitation (9.4% in total, 23% for visibility less than 1 km), and a consistency check removed visibility observations that were inappropriate to relative humidity, temperature, and pressure. In a case study on two consecutive fog days, visibility forecast skills were improved by applying visibility data assimilation, mostly through modifications of aerosol concentrations. A 3-month model run in the winter of 2016 showed a positive bias in visibility predictions, especially for the low-visibility cases. Visibility data assimilation improved the prediction skills, but the positive effects were limited within 9 forecast hours and were smaller for extremely low-visibility events. Sensitivity experiments were performed using local aerosol observations with a larger number of smaller aerosol particles. Modifications in aerosol properties made better results in frequency bias for the whole forecast ranges and also improved the equitable threat score (ETS) for relatively longer forecast hours (more than 4 h).

中文翻译：

---

使用韩国观测网络的能见度数据同化和预测

已经有许多研究使用数值模型来提高能见度预报技能，但它们的性能仍然落后于其他气象现象的预报技能。本研究试图使用韩国新建立的由 291 个前向散射传感器组成的自动能见度观测网络来改善能见度预测。近3年能见度观测分析显示，晴天（能见度20公里以上）占总观测天数的46%，雾天（能见度小于1公里）占总观测数的1.58%。韩国气象局 (KMA) 的甚短程资料同化和预测系统 (VDAPS) 基于气象局统一模型与克拉克等人的能见度数据同化同化了能见度观测。（QJR Meteorol Soc 134：1801-1816 年，2008 年）。在数据同化之前，降水检查消除了带降水的能见度数据（总共 9.4%，能见度小于 1 公里为 23%），一致性检查消除了不适合相对湿度、温度和压力的能见度观测。在连续两个雾日的案例研究中，通过应用能见度数据同化，主要是通过修改气溶胶浓度，提高了能见度预报技能。2016 年冬季运行的 3 个月模型显示能见度预测存在正偏差，特别是对于低能见度情况。能见度数据同化提高了预测技能，但积极影响在 9 个预测小时内有限，对于极低能见度事件的影响较小。使用具有大量较小气溶胶颗粒的局部气溶胶观测进行灵敏度实验。气溶胶特性的修改在整个预测范围的频率偏差方面取得了更好的结果，并且还提高了相对较长的预测时间（超过 4 小时）的公平威胁评分 (ETS)。