Is it possible to forecast the grain quality and yield of different varieties of winter wheat from Minolta SPAD meter measurements?

摘要

To optimize wheat segregation for the various markets, it is necessary to add to genotype segregation, a prediction before harvest of the values of yield and grain protein concentration (GPC) for the different fields of the collecting area. Different tools allowing a prediction of crop production exist. Among them, the evaluation of nitrogen concentration by a chlorophyll meter (Soil-Plant Analysis Development (SPAD) readings), classically used to adapt the nitrogen fertilizer application, has been used in few works to foresee grain yield and grain protein concentration. But the relationships between N crop status and SPAD measurements varies among varieties and this genotypic effect has rarely been incorporated in models of forecasting grain quality. This paper compares several models to forecast yield, nitrogen uptake in grain (NUG) and grain protein concentration from trials carried out in 2001 and 2002 at the INRA experiment station of Grignon (West of Paris). Trials crossed nine varieties by four (2002) or five (2001) nitrogen rates. Input variables of those models are mainly chlorophyll meter measurements (SPAD) on the penultimate leaf at GS65 and on the flag leaf at GS71 Zadoks growth stages and ear number per square meter (NE). A square root model of yield based on NE x SPAD gave the best fit (RMSE=0.6 t ha-1 for both stages) if considering three different groups of genotypes. Based on the same variable, NE x SPAD, a quadratic model for NUG without significant effect of genotypes gave the best fit (RMSE, between 21 and 30 kg ha-1 depending of the growth stage). And, for GPC, considering the same three groups of genotypes, the slope of the linear model with the ratio of predicted grain nitrogen concentration to predicted yield, is the same at both stages and very close to the standard value used to calculate protein concentration from nitrogen concentration (5.7), but the predictive quality of the model is more than 10% higher at GS71 (R2 of 0.77) than at flowering (R2 of 0.64). Finally, the sensibility of the models to delay in the stage of measurement is discussed..