• seasonal predictability;
  • Indonesia;
  • hidden Markov model


The seasonal predictability of rainfall over a small rice-growing district of Java, Indonesia is investigated in terms of its daily characteristics during the September–December monsoon-onset season. The seasonal statistics considered include rainfall frequency, mean daily intensity, median length of dry spells, as well as the onset date of the rainy season. General circulation model retrospective seasonal forecasts initialized on August 1 are downscaled to a set of 17 station locations using a nonhomogeneous hidden Markov model. Large ensembles of stochastic daily rainfall sequences are generated at each station, from which the seasonal statistics are calculated and compared against observations using deterministic and probabilistic skill metrics. The retrospective forecasts are shown to exhibit moderate skill in terms of rainfall frequency, seasonal rainfall total, and especially monsoon onset date. Some skill is also found for median dry-spell length, while mean wet-day persistence and daily rainfall intensity are not found to be predictable. Copyright © 2008 Royal Meteorological Society