One of the main indicators of frost events is the minimum air temperature, and in Brazil there is not yet an objective method of forecasting this variable. In this research experiences were performed to develop a model of forecasting minimum air temperature during winter, applicable to studies of frost prediction in the coffee growing areas of southern Brazil. This model is kown as the "Model Output Statistics" (MOS). Numerical forecasts from climate models for near surface show systematic errors, and these errors are different for different geographical locations, and seasons of the year. Besides, several climate models do not have minimum air temperatures as output.
Experiments were made for the assessment of the possible use of MOS in the forecast of minimum temperatures in southern Brazil. For this purpose, the output of the CPTEC/COLA climate model and the observed minimum temperature of the previous day were used. The forecast was made by using linear regression equations with an stepwise technique.
The prognostic equations were implemented with the use of 28 variables from the model CPTEC/COLA model output, and minimum air temperature for 95 stations in the staes of Paraná and São Paulo, from April to the first half of September of 1997. The output variables were interpolated for each station. The experiments show that the most important variables from the model output are absolute temperature at the 850 and 1000 hPa, equivalent potential temperature at 850 hPa, specific humidity at 1000 hPa and minimum air temperature of the day before, for a 24 hour forecast. Assesments were made of the mean error, the mean square error, and the mean absolute error. The validation shows that the MOS forecast exhibits the same or even smaller errors than the operational forecasts. Currently, MOS was developed and implemented on operational basis at SIMEPAR and CPTEC.