Reduced visibility is a major safety concern at airports leading to flight delays or diversions. The primary motivation for this study is to enhance under standing of visibility and to build a simple, practical visibility forecasting method across Taiwan in an operational setting by using readily avail able, ground-based observations.
This paper presents for the first time a systematic, quantitative examination of the controls on visibility over the en tire Taiwan region by adopting a statistical approach relating visibility to various physical variables. A multiple linear regression is carried out for the early morning hours during the months of November ~ April, when visibility is especially low. The regression reveals that on the west coast of Taiwan, the concentration of fine particles (PM2.5) and relative humidity (RH) are most related to visibility, and to a lesser ex tent, coarse particles (PM10-2.5) and wind speed. The significantly elevated PM concentrations would, there fore, cause a marked reduction in visibility on Taiwan's west coast - where most of Taiwan's population and anthropogenic PM emissions are found. Visibility on the east coast appears to be con trolled by some what different mechanisms, with rain fall playing a larger role. The probability of occurrence of especially low visibility (<= 1600 m) was revealed by logistic regression to be the most statistically related to RH, and, to a less extent, to PM concentrations.
An un certainty analysis to under stand current limitations in predicting visibility indicated that 24-hour visibility forecasts were dominated by a) errors in fore casting RH and b) inadequacies in the adopted statistical model, followed by c) errors in PM fore casts. Hence to minimize the consider able uncertain ties invisibility fore casts, which could reach standard deviations of several thousand meters, future work needs to adopt a more sophisticated statistical model as well as re duce the consider able errors in predicting RH. Finally, the uncertainties associated with PM can be reduced by improving PM emission estimates through an inverse analysis method.