Univariate time series methods for forecasting the monthly mean air temperature in Erechim, RS

This paper presents a time series analysis of the minimum and maximum air temperature of Erechim, RS. A comparison between two traditional classes of the forecasting models, namely: ARIMA models and exponential smoothing models is also presented. In the class of ARIMA models using criteria information, SARIMA type models that consider the seasonal characteristics of air temperature were selected, whereas for exponential smoothing models Holt-Winters additive algorithm were used. Smoothing constants are determined to minimize the mean square error between observed and predicted values. This analysis allowed the identification of components such as seasonality and atypical periods. The model predictions were compared for different forecast horizons. The ARIMA class models proved to be more accurate while the adjusted models were adequate for adjusting forecasts of variables of air temperature, being important tools for agricultural climatology.