Quantification of environmental and management effects on the yield of late-sown soybean

Abstract Intensification of production is an avenue to sustain farm profitability in non-subsidised agricultural systems. The main objective of this work is to establish risk profiles of late-sown soybean for its application in wheat ( Triticum aestivum L.)/soybean ( Glycine max L. Merr.) double cropping systems in the southern Argentinean Pampas. Field experiments provided data to: (i) identify the key environmental and crop management variables controlling yield of late-sown soybean; (ii) develop and test a parsimonious model of soybean yield for application in risk management. The data set included 125 crops covering a wide range of environmental conditions and management practices, including sowing dates from 20 November to 20 January, inter-row spacings from 0.19 to 0.76 m, seasonal water supply from 230 to 610 mm, average temperature and solar radiation at different stages, soil depth from 0.37 to >1.2 m. Tests using an independent data set indicated empirical models including three to nine variables compared well with CROPGRO, a more mechanistic, CERES-type model in predicting late-sown yields. Empirical models accounted for 79–82% of the variation in yield (vs. CROPGRO: 86%) with root mean square error between 298 and 702 kg ha −1 (vs. CROPGRO: 512 kg ha −1 ). Sowing date was the variable with the largest effect on crop yield ( r 2 =0.59). The relationship between yield and sowing date was non-linear, with large yield reductions for delays in sowing dates after mid-December. This function provided a good description of yield reductions associated with sowing date for crops grown at a wide range of latitudes (from 29° to 42°) and management practices of each region, as inter-row spaces (0.19–1 m) and cultivars (maturity groups from 00 to IX). An empirical model with inputs accessible to growers, i.e. sowing date, soil depth, row distance, plant density, soil water at emergence, rainfall between emergence and R1 and between R1 and R5, was used to determine probability distributions of yield at three locations during a 30-year series. Across sites, yield of crops in deep soils averaged 2.4 t ha −1 for sowings on 25 December, and 1.5 t ha −1 for sowings on 10 January. In shallow soils, yield dropped to 1.8 and 0.9 t ha −1 , respectively. We concluded that: (i) sowing date was the most important variable determining late-sown soybean yield; (ii) empiric models with a low number of variables had high predictive power.