A quantitative precipitation forecast experiment for Puerto Rico

A quantitative precipitation forecast (QPF) experiment is conducted for the island of Puerto Rico. The experiment ranks the utility of six objective rainfall models and an operational forecast issued by the National Weather Service Forecast Office, San Juan. This is believed to be the first experiment to rank the utility of rainfall forecast schemes in the tropics. Using an analysis of variance tool called common factor analysis (CFA), the island of Puerto Rico is divided into convective rainfall regions. These regions are statistically independent and represent the forecast domains for the experiment. All forecasts and realizations are area-averaged over each convective region. The QPF experiment is conducted in real-time over three 6-week periods in 1998. The periods fall in three separate rainfall seasons. All seven forecast schemes are configured to produce an area-averaged 24 h rainfall forecast. Forecasts are realized through a network of 114 data rain gauges, whose 24 h values are also area-averaged within convective region. We conduct this experiment in a Bayesian framework. Users may determine the ex ante value of forecast products through the Bayesian correlation score (BCS). Over each of the three seasons, the climatology forecast held the highest ex ante utility for users. Although objective forecast utility scores for heavy rain events are low, they yield higher BCS values than operational forecasts.