The aim of this work was to evaluate a system to forecast soybean crop yield for Brazil, based on regional empirical models to assess crop yield, with data from a national database of soybean cropped area in municipal scale, and from an agrometeorological monitoring system covering all Brazilian States. The models incorporated the conceptual bases proposed by Doorenbos & Kassam, with empirical adjustments for each region of Brazil, considering the diverse potential yield of most used varieties and the regional differences in the soybean production systems. Cultivated area database from each State was structured with data from IBGE and Conab. Soybean yield was estimated for harvests from 2000/2001 to 2005/2006 and compared to Conab surveys. Statistical analysis using Test t point out that there is no significant difference between estimates and official data. Good fittings were gotten for region grouped yield data (R²>0.87), with higher deviations for harvest assessment of Rio Grande do Sul, Santa Catarina, Mato Grosso do Sul, Maranhão, Piauí and Bahia. In national scale, the highest observed deviation was 5.81% for the 2000/2001 harvest, and the smaller one was 0.62% for the 2005/2006 yield.
crop yield; modeling; information technology; agrometeorology
