Statistical forecasting of daily temperatures using short records of previous temperatures

From previous analysis of the daily minimum, meam and maximum temperatures in Modena, Italy, over more than 100 years, it has already been demonstres that each time series may be well represented by the sum of a seasonal deterministic function and a stationary gaussian stochastic signal. The latter was found to be autocorrelated and well represented by an autoregressive, moving-average process of order 2 and 1 respectively, AR(2)-MA(1). In this paper, starting from subrecords of daily temperatures, taken from the whole record available in Modena, long-term and short-term predictions have been compared with the actually measured values. The positive results obtained by using the previous model suggest that such statistical forecasting can be performed also in other locations of similar climatic behavior, even if only short records of daily temperatures are available.