Comparison of Different Modeling Approaches to Predict Soybean Phenology in the Mid-South.

Models that correctly predict soybean phenology with limited data inputs can be a powerful decision-tool for farmers to select a maturity group (MG) for given latitude and planting dates. In this sense, two soybean models (CROPGRO and SOYDEV) were evaluated for their accuracy to predict phenology with input information consisting of MG and weather data. Experiments conducted in Stoneville, MS (33°25’N ) and Fayetteville, AR (36°03’N) from 1979 to 2011, comprising MG from III to VI and with a range of planting dates (March 9^th to July 7^th) (n=316) were used to evaluate the models. Both modeling approaches predict phenology based on the effect of temperature and day length. CROPGRO requires a calibration of 6 phenological cultivar coefficients, whereas SOYDEV has been calibrated for the conditions in Nebraska and only requires as input relative MG. The CROPGRO model was calibrated with data from Stoneville not used in the model evaluation (n=82).

Simulations showed that CROPGRO simulated beginning of flowering (R1) and physiological maturity (R7) with a high accuracy (RMSE ≤ 5 and 9 days, respectively).  Likewise, SOYDEV predicted R1 with accuracy for MG III (RMSE = 5 days), but tended to underpredict R1 date as MG increased (MG VI: Bias = -28 days). SOYDEV predicted R7 well for MG III to V but had a higher deviation from the observed than CROPGRO (RMSE = 8 – 13 days), and underpredicted R7 in MG VI (Bias = - 18 days). The higher deviation of simulated values from the observed with the SOYDEV model in the Mid-South latitudes of this study can be explained by the calibration of the model in more northern latitudes. Results suggest that calibration of both models by MG could allow an accurate prediction of soybean phenological dates (RMSE < 9 days) in the Mid-South latitudes.