A spatial–temporal projection model for extended-range forecast in the tropics

Abstract

An extended singularity value decomposition based statistical model, namely the spatial–temporal projection model (STPM), was constructed for the extended-range (10–30-day) forecast of tropical outgoing longwave radiation anomalies (OLRA). The special feature of this empirical model is using the spatial and temporal information of predictor–predictand coupled patterns to predict the temporally varying predictand field at all-time leads (i.e., 10–35 days) at once. A 10-year hindcast result shows that, different from previous statistical models, the skill scores of the STPM dropped slowly with forecast lead times. Useful skills can be detected at 30–35 day leads over most tropical regions. The highest temporal correlation coefficient of 0.4–0.5 appears over the Maritime Continent (Indian and western North Pacific monsoon regions) in boreal winter (summer), exceeding a 99 % confidence level. The STPM is also capable in predicting the spatial evolutions of convective anomalies, including the zonal and meridional propagation of OLRA. The forecast of the Real-time Multivariate MJO indices shows that the STPM attains a higher skill than previous statistical models. The STPM also shows comparable skills with the state-of-the-art dynamic models during the Dynamics of the Madden–Julian Oscillation campaign period, especially at 15-day and longer leads.