TIME-SERIES MODELS FOR BEACH POLLUTION

Beach pollution was assessed visually by beach inspectors on a five-po int ratings scale as 0, 1, 2, 3 or 4 corresponding to pollution levels of None, Low, Trace, Medium or High, respectively The data consisted of 640 days of pollution ratings at 34 beaches in Sydney, Australia, t ogether with concomitant wind, rain, ocean current and temperature inf ormation. Development of statistically significant relationships betwe en the more subjectively measured pollution data and the more objectiv ely quantified physical variables not only served to explain the occur rence of pollution but lent credibility to the ratings scale itself as a useful measure of visual pollution. Methods for analysing qualitati ve data were combined with time series models to account for the influ ence of the physical variables, whose effects were subject to delay an d dissipation over time. The GLIM statistical language is suitable for modelling ratings data. It is not generally used for time series mode lling. Most commonly available time series software is not suitable fo r ratings data. The application here required a combination of the two methodologies. Special purpose models were formulated and then softwa re written in the GLIM language to estimate time series models with th e survey ratings scale data as the dependent variable and the physical data as the independent variables. The work also raises the possibili ty of developing forecasts for pollution-free days and led to the deve lopment of a new estimation scheme for time series models. Copyright ( C) 1996 Elsevier Science Ltd