Short-range prediction of fog occurrence at Christchurch aerodrome

Abstract

An objective guidance system has been developed for use in aviation forecasting. It is primarily concerned with the prediction of ground fog occurrence at Christchurch aerodrome over forecast periods of one to twelve hours. However, forecasts of many other weather elements are also calculated. The main stochastic model used is the first order Markov chain, defined for transitions through a three-state space defined by present weather codes and ranges of visibility. Transition frequencies are modelled initially as functions of time, represented as the hour of the day relative to sunrise. This model is shown to perform favourably against two simple reference forecasting systems (persistence and climatology). To extend the basic Markov model, transition probabilities are re-modelled as functions of time of day and of related weather element variables (wind speed, temperature and dewpoint). Use of observed weather elements in this model is shown to improve accuracy slightly in terms of mean squared error and quite noticeably in terms of skill in categorical forecasting. A regression system is developed to predict hourly values of wind, temperature and moisture parameters for input to this enhanced model. It is shown that the slight mean square improvement in model performance through use of observed weather is almost completely lost when forecast weather is substituted. However, categorical forecasts retain some increase in skill over the basic model. The final Markov model selected is shown to perform favourably against predictions issued by the aviation forecaster over the three winter months of 1988. A description is given of this model in a real-time environment. It is concluded that the Markov model of fog states developed here (and its concurrent regression model of related weather elements) is of sufficient accuracy to be of use to the aviation forecaster.