An Autoregressive Moving Average Model of Monthly Surface Temperature Field over China

For designing a time series of the successive monthly surface temperature in 1952－1980 over China a stochastic model is formulated by an AEMA (p, q) model. The monthly temperature fields composed of records at 60 stations are expanded by means of EOF (Empirical Orthogonal Function) with the various sample sizes 348, 336 and 300 months so as to examine the stability of EOF expansion. Then the first four principal components, i. e. z1, z2, z3, z4 are taken as the variables of multivariate time series, since their total variance contribution is 99.26%. The major periods in first four principal components series are also revealed by using the periodogram and the maximum entropy method. The model identification of ARMA (p,q) for univariate variables zi (i=1,2,3,4) is made by the Pandit-Wu method, Then, the empirical models are obtained. The extrapolations of s models are used for predicting a monthly temperature field. The score of hit ratio of departure forecasts is 78.3%, which is better than that of the recent operational long-range weather forecast.