The air temperature is a climatic element of interest for several agricultural related studies. Hence, daily air temperature from two locations in Brazil (Botucatu and Petrolina, with different climate classification), hourly recorded from January 2004 to December 2005, were analysed to verify if 24 h long air temperature variation can be predicted with 6 h interval records. Three data set were selected, with different first time record of air temperature, and analyses performed using geoestatistics to confirm the temporal dependency of records. Then, ordinary krigging was used to model air temperature variation within 1 h interval, and spherical model presented the best approach, with range of 11.6 h for Botucatu and 12.7 for Petrolina. For both locations, the time of air temperature recording was the most importnat factor on accuracy of estimation of air temperature variation. Records at 2, 8, 14 and 20 h and at 4, 10, 16 and 22 h presented the best approaches, but for records at 9, 15 and 21 h, the approach was worst.
geoestatistics; temporal dependency; climatic element
