Abstract
A 1129-day high quality rainfall dataset was collated, from the months of March to November 1993–97, where trade wind flow predominates across the northeast Queensland tropical coast. This dataset was matched with numerical model fields from the Australian regional limited area model. Synoptic precipitation regimes were identified by subjective classification of a major portion of the development data utilizing radar, surface, and upper-air data. With the aid of classification trees, a series of objective sorting criteria were developed to stratify the numerical model dataset into weather regimes prior to the development of a series of predictive equations for rainfall through screening regression. In addition to raw numerical model output, potential predictors with established links to trade wind precipitation were offered to the regression procedure. Rainfall equations were also developed by seasonal stratification of the development data. Further comparison was made with rainfall occurrence output from a synoptically tuned classification tree and other objective and subjective prognostic methods. The relationships established through synoptic and seasonal stratification were tested on a 226-day independent dataset for the verification periods 0–24 and 24–48 h beyond the defining model analysis.
Objective probability of precipitation (PoP) forecasts were produced on a regional and individual city basis. The synoptically stratified PoP forecasts, tuned for measurable rain prediction, showed skill against reference methods and against the subjective Bureau of Meteorology operational forecasts at the 24–48-h projection. At the shorter 0–24-h range, no skill advantage was shown when compared to the Australian Bureau of Meteorology forecasts. For rain events ≥2.5 mm day−1, the synoptically stratified methodology was significantly more skillful than other objective techniques at the 95% confidence interval.
Quantitative precipitation forecasts (QPFs) were developed for application to rainfall prediction across a city rainfall network. The QPFs for objective techniques based on numerical model output displayed skill against climatology. However, only slight skill gains over persistence were obtained for the 0–24-h interval beyond analysis. Higher objective QPF skill gains over persistence were realized for the 24–48-h forecasts.
* Additional affiliation: JCU Physics, James Cook University of North Queensland, Townsville, Queensland, Australia.
Corresponding author address: G. J. Connor, Townsville Meteorological Office, Bureau of Meteorology, RAAF Base Garbutt, Queensland 4814, Australia.
Email: G.Connor@bom.gov.au
