Forecasting Advective Sea Fog with the Use of Classification and Regression Tree Analyses for Kunsan Air Base

摘要

Advective sea fog frequently plagues Kunsan Air Base (AB), Republic of Korea, in the spring and summer seasons. It is responsible for a variety of impacts on military operations, the greatest being to aviation. To date, there are no suitable methods developed for forecasting advective sea fog at Kunsan, primarily due to a lack of understanding of sea fog formation under various synoptic situations over the Yellow Sea. This work explored the feasibility of predicting sea fog development with a 24-hour forecast lead time. Before exploratory data analysis was performed, a geographical introduction to the region was provided along with a discussion of basic elements of fog formation, the physical properties of fog droplets, and its dissipation. Examined in this work were data sets of Kunsan surface observations, upstream upper air data, sea surface temperatures over the Yellow Sea, and modeled analyses of gridded data over the Yellow Sea. A complete ten year period of record was examined for inclusion into data mining models to find predictive patterns. The data were first examined using logistic regression techniques, followed my classification and regression tree analysis (CART) for exploring possible concealed predictors. Regression revealed weak relationships between the target variable (sea fog) and upper air predictors, with stronger relationships between the target variable and sea surface temperatures. CART results determined the importance between the target variable and upstream upper air predictors, and established specific criteria to be used when forecasting target variable events. The results of the regression and CART data mining analyses are summarized as forecasting guidelines to air forecasters in predicting the evolution of sea fog events and advection over the area.