Appropriated probabilistic models can give mathematical precision in historical climatological series studies. The "extreme value distribution" model (Gumbel) was used to estimate the probabilities of monthly (May, June, July, August, September) and yearly frost occurrence for Campinas region in the State of São Paulo, Brazil, based on a 110 years (1891 to 2000) historical series of absolute minimum air temperature. The main objective was to estimate and compare the monthly and yearly frost risk during different studied periods of years (1891-1910, 1911-1940, 1941-1970 e 1971-2000). The "extreme value distribution" model showed good agreement between observed and estimated probabilities for all the monthly and yearly periods. Despite the increase of the average values of the absolute minimum air temperature since 1891 to 2000, the estimated monthly and yearly probabilities showed a great variability during the different periods of years studied. The most recent period, 1971/2000, showed yearly probability (24.2%) similar to the 1911/1940 period (24.5%), and to the entire series analyzed of 1891/2000 (24.4%). The yearly highest probability of low temperature occurrence (34.2%) was during the 1891/1910 period, while the yearly lowest probability (18.4%) was during the 1941/1970.
Frost risk; probabilities; minimum air temperature; Gumbel distribution
