An exploratory study into synoptic classification techniques for statistical tropical precipitation forecasting

Daily precipitation data for 24 rainfall-reporting stations along the northeast Queensland tropical coast were collated for a period of 295 days during periods of sustained trade-wind flow. Associated grid-point data from the Australian region numerical model analysis were also collected. Parameters that were found to influence precipitation in previous studies were derived from numerical model analysis output and surface observations. Prior to the application of linear regression techniques to develop statistical relationships between daily rainfall and 278 potential predictors, the data were classified according to synoptic characteristics. These synoptically stratified datasets were used to produce probability of precipitation and rainfall quantity equations on a regional and individual station basis. A 90-day comparative trial on independent data tested the resultant rainfall predictive equations for projections of between 24 and 48 hours beyond model analysis on a perfect prognosis basis. The synoptically stratified rainfall occurrence and quantitative precipitation forecast (QPF) equations, derived from regional data, displayed superior skill to the reference climatology and persistence forecasts. The stratified rainfall equations developed from individual station data showed greater skill than objective guidance and persistence for most applications.