Predicting soybean development with a simple photothermal dynamic algorithm

SEVERINI, A.D.; ALVAREZ PRADO, S.; MARIA E. OTEGUI; VEGA, C.R.C; ZUIL, S.; KAVANOVA, M.; CERETTA, S.; ACRECHE, M.M.; SCHOLZ DRODOWSKI, R.F.; SERRAGO, R.A.; MIRALLES, D.J.

Rosario

Congreso; Merco Soja: Reformular la soja para impulsar una cadena de conocimiento.; 2019

Predicting the occurrence of the critical period for soybean?s yield determination is important for farmers to decide on variety and sowing date with the objective to expose this period (during which yield is mainly determined) to the best environmental conditions. Simulation models like APSIM (Keating et al., 2003) and DSSAT (Jones et al., 2003) are extremely useful to predict yield under different environments. However, in these models the parameterization of genetic coefficients for simulating phenology and yield is a complex and time-consuming process limiting the number of genotypes available for simulation. Moreover, the modeled genotypes may not be representative of the broad range of genetic material grown by farmers. We created a simple, dynamic model based on photoperiod and temperature to predict flowering initiation (R1), start of grain filling (R5) and physiological maturity (R7, Fehr and Caviness (1977)) in a wide number of commercial soybean varieties ranging from maturity group (MG) II to VI used by farmers in Argentina, Uruguay, and Paraguay. Our aim is to build a model that is simple enough to be calibrated, yet able to predict these stages with reasonable accuracy. The ?CRONOSOJA? model outputs will be freely available in the near future on an interactive website (http://soja.cronos.agro.uba.ar).Materials and methodsSoybean varieties, field experiments and measurementsWe selected 34 soybean commercial varieties from major seed companies of Argentina, Uruguay and Paraguay, covering MG II to VI (Table 2). By previous expert consultation, we ensured that these varieties were representative of the regional soybean seed market and widely used by farmers. To explore a broad range of temperature and photoperiod conditions, we sowed those varieties from October to February, with a gap of around one month between successive sowing dates, during three seasons in the locations listed in Table 1. We determined the phenological stage of each plot every 2-3 days using the scale of Fehr and Caviness (1977).Table 1: Number of different sowing dates tested in each experimental site during three growing seasons. Bold numbers denote datasets that were used for validating the model, while those in italics show datasets that were not included yet (but they will be included soon), neither for calibration nor for validation of the model. The rest of the datasets were used for the initial calibration of the model.SeasonCountrySiteLatitude (°S)Longitude (°W)2016/172017/182018/19ArgentinaCABA34.658.5110ArgentinaChascomús35.658.0300ArgentinaManfredi31.863.7043ArgentinaPergamino33.960.6433ArgentinaReconquista29.259.9333ArgentinaSalta24.965.5002ParaguayCapitán Miranda27.255.8001UruguayLa Estanzuela34.357.7024Model developmentFor model fitting, we transformed soybean developmental stages into a continuous numerical scale: emergence (EM) = 0, R1 = 1, R5 = 2 and R7 = 3. In its current state, the model only simulates the latter three stages but R3 (beginning of pod development) and R6 (full seed size) will be included soon as temperature-corrected calendar-day deviations from R1 and R5, respectively. The equation describing the developmental stage (R) at a specified time (t, in days after EM) isR_t={■(R_(t-1)+1/D_(EM-R1) f(T_t)f(P_t),&if R_(t-1)