Gauging the sources of uncertainty in soybean yield simulations using the MONICA model

Abstract

Crop models are an important tool to evaluate crop management strategies and simulate yield for present and future scenarios, however, much uncertainty is present within model parameters, approaches and input variables. Therefore, it is important to quantify the uncertainties in simulated yields as a function of input details from field management decisions and their effect on simulated regional soybean yield. To investigate this, the Model for Nitrogen and Carbon in Agroecosystems (MONICA), calibrated with experimental data, was used in this study. Four sources of uncertainty relevant to field management were considered in simulating soybean yields: technological level, soil type, sowing date and cultivar maturity group. The uncertainties in yield simulation were investigated for 14 sites in Southern Brazil, comparing results to governmental statistics for the crop seasons from 1989/1990 to 2013/2014. The MONICA model was able to simulated soybean grain yield efficiently after calibration of crop phases, growth and root-development parameters. The technological level (TL) was the yield factor with the highest coefficient of variation (CV) among the fourteen sites, with an average of 31.4%, while cultivar maturity group and sowing date both had a CV of 10%, and soil 4.4%. However, uncertainties varied with climate conditions in each crop season, with sowing date and cultivar maturity group both showing higher CV than technological level in some crop seasons. In some cases, for regional yield level, the model demonstrated varied performance by location. The most accurate performance was in simulating yields in the municipality of Passo Fundo (r = 0.87), while in Bagé, the lowest accuracy was achieved (r = 0.07), across all factors. However, in Bagé, when the lower yields simulated by MONICA, for all source of uncertainties were considered, simulated yields were close to those observed for most of the crop seasons. Based on these results, it is important to consider these different sources of uncertainty that stem from farmer decision-making in order to simulate regional soybean yield efficiently.