CROPGRO-soybean model - Validation and application for the southern Amazon, Brazil

Agricultural models are useful tools for predicting phenology, biomass, and crop yield under different environmental conditions during the growth period. Notwithstanding their use in the Southern Cone of Rondonia (southern Amazon, Brazil) is still limited. Thus, the objective in this work was to calibrate and validate the CROPGRO-Soybean model with soybean data collected in harvests of 2017 to 2019, being the data from the first growing period devoted to model calibration, and crop data from 2018/2019 to model validation. Six soybean cultivars were planted at the municipality of Vilhena, State of Rondonia, Brazil. The model was parameterized using climatic, hydric and physical soil properties, and growth data of the evaluated soybean cultivars. To evaluate the accuracy of model prediction we used the coefficient of determination (r2), percentage difference (Pd), root mean square error (RMSE), and Willmott's index of agreement (d-value). The model accurately simulates the vegetative and reproductive stages, and the yields of the soybean cultivars under the climatic conditions of the Southern Cone of Rondonia. However, the CROPGRO-Soybean model tended to overestimate soil moisture content especially in upper soil layer (0.0-0.1 m depth). Taking the observed data, as the baseline, the model predicted the time to flowering, first pod, seeding, and physiological maturity, with a discrepancy of just one to seven days. In most of the soybean cultivars (five of the six cultivar evaluated), the differences between observed and simulated yields (growing cycle of 2018/2019 - validation harvest) ranged from -6 % to (underestimation) to 27 % (overestimation).