In this study,changes in daily weather states were treated as a complex Markov chain process,based on a continuous-time watershed model（soil water assessment tool,SWAT） developed by the Agricultural Research Service at the U.S.Department of Agriculture（USDA-ARS）.A finer classification using total cloud amount for dry states was adopted,and dry days were classified into three states：clear,cloudy,and overcast（rain free）.Multistate transition models for dry-and wet-day series were constructed to comprehensively downscale the simulation of regional daily climatic states.The results show that the finer,improved,downscaled model overcame the oversimplified treatment of a two-weather state model and is free of the shortcomings of a multistate model that neglects finer classification of dry days（i.e.,finer classification was applied only to wet days）.As a result,overall simulation of weather states based on the SWAT greatly improved,and the improvement in simulating daily temperature and radiation was especially significant.
Experiments of forecasting daily bi-variate index of the tropical atmospheric Madden-Julian Oscillation (MJO) are performed in the context of adaptive filtering prediction models by combining the singular spectrum analysis (SSA) with the autoregressive (AR) methods.the MJO index,a pair of empirical orthogonal function (EOF) time series,called RMM1 and RMM2,predicts by the combined statistical SSA and AR models:firstly,according to the index of historic observation decomposed by SSA and then reconstructed by selecting the first several components based on prominent variance contributions;after that,established an AR prediction model from the composite (scheme A) or built the forecast models for each of these selected reconstructed components,separately (Scheme B).Several experimental MJO index forecasts are performed based on the models.The results show that both models have useful skills of the MJO index forecast beyond two weeks.In some cases,the correlation coefficient between the observed and predicted index series stays above 0.5 in 20 leading days.The SSA-AR model,based on the reconstructed composite series,has better performance on MJO forecast than the AR model,especially for the leading time longer than 5 days.Therefore,if we build a real-time forecast system by the SSA-AR model,it might provide an applicable tool for the operational prediction of the MJO index.
