In the Brazilian Amazon region, the State of Para stands out as a major soybean producer, with yields averaging around 2.800 kg ha-1
. Soybean production in this region occurs mostly under rainfed conditions, therefore subjected to climate variability. The use of crop models to simulate the growth and development of plants becomes an important alternative for optimization of natural resources, and for adapting management strategies aiming to increase productivity rather than incorporating new areas. The objective of this study was to use the DSSAT (Decision Support System for Agrotechnology Transfer) to simulate sowing dates, for a climatological series of 30 years (1980-2010) to assess the risk of yield losses. The CROPGRO-Soybean model was calibrated with experimental data obtained in 2006, 2007 and 2008 in Paragominas, northeastern State of Para, Brazil (2o
59' 08'' S, 47o
19' 57'' W), and the cultivar used was BRS Tracajá (conventional cultivar, maturity group 9.2). The DSSAT was programmed to simulate the yield using 12 sowing dates (10 days of interval), beginning on January 1 (doy 1) to April 30 (doy 120). The average yield over the historical series were 985 (± 837) to 3,714 (± 389) kg ha-1
depending on the sowing dates, the simulations with yield that showed more than 3,500 kg ha-1
(± 455) were obtained in 76.7% of cases on days 10, 20, 30, and 50 (doy). The sowing dates early to the doy 50 resulted in losses of approximately 2.3 kg ha-1
whereas for the subsequent sowing dates the losses increased linearly at a rate of 44.2 kg ha-1
for each day of delay.