A spatial–temporal projection model for 10–30 day rainfall forecast in South China

Abstract

Extended-range (10–30 days) forecast, lying between well-developed short-range weather and long-range (monthly and seasonal) climate predictions, is one of the most challenging forecast currently faced by operational meteorological centers around the world. In this study, a set of spatial–temporal projection (STP) models was developed to predict low-frequency rainfall events at lead times of 5–30 days. We focused on early monsoon rainy season (mid April–mid July) in South China. To ensure that the model developed can be used for real-time forecast, a non-filtering method was developed to extract the low-frequency atmospheric signals of 10–60 days without using a band-pass filter. The empirical models were built based on 12-year (1996–2007) data, and independent forecast was then conducted for a 5 year (2008–2012) period. The assessment of the 5-year forecast of rainfall over South China indicates that the ensemble prediction of the STP models achieved a useful skill (with a temporal correlation coefficient exceeding 95 % confidence level) at a lead time of 20 days. The amplitude error was generally less than one standard deviation at all lead times of 5–30 days. Furthermore, the STP models provided useful probabilistic forecasts with the ranked probability skill score between 0.3–0.5 up to 30-day forecast in advance. The evaluation demonstrated that the STP models exhibited useful 10–30 days forecast skills for real-time extended-range rainfall prediction in South China.