Premium
A hidden seasonal switching model for multisite daily rainfall
Author(s) -
CareySmith Trevor,
Sansom John,
Thomson Peter
Publication year - 2014
Publication title -
water resources research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.863
H-Index - 217
eISSN - 1944-7973
pISSN - 0043-1397
DOI - 10.1002/2013wr014325
Subject(s) - environmental science , climatology , meteorology , hydrology (agriculture) , geography , geology , geotechnical engineering
A hidden seasonal switching model for daily rainfall over a region is proposed where season onset times are stochastic and can vary from year to year. The model allows seasons to occur earlier or later than expected and have varying lengths. This stochastic seasonal variation accommodates considerably more of the observed intraannual rainfall variability than can be represented using seasonal models with standard fixed seasons. In essence, the model dynamically classifies daily rainfall time series into seasons whose onsets vary from year to year and within which the model parameters are assumed to be time homogeneous. A variety of nonseasonal models could have been used to describe daily rainfall within seasons. Here a generalization of the Richardson model is adopted which has rainfall states (dry, light rain, and heavy rain) some of which are hidden or unobserved. It is further assumed that the rainfall states generate rainfall that is independent of season (seasonally invariant), so it is only the dynamics of the rainfall states that vary from season to season. A suitable estimation strategy based on maximum likelihood and the EM algorithm is developed for fitting the model across a region. This strategy is validated on simulated data. Various forms of the model are fitted to daily rainfall measurements from 12 sites in southern New Zealand. These results are discussed and compared to those from fitting standard fixed season models.