Yield Prediction Using Remote Sensing for Stabilization of Grain Protein Content by Variable Rate Fertilization in the Bread Wheat Cultivar ’Setokirara’

Abstract

We developed a yield prediction model of bread wheat cultivar ’Setokirara’ and verified whether the grain protein content could be controlled by variable rate fertilization according to the predicted yield. In the first and second years, six plots with different fertilization rates were established. The vegetation indices normalized difference vegetation index (NDVI) and green NDVI (GNDVI), panicle numbers and SPAD values were obtained at the full heading stage, and yield was determined at maturity. Using two years of data, we obtained a regression line of yield for each vegetation index. In the third year, wheat was sown in late November (standard sowing) and mid-December (late sowing). The amount of nitrogen top-dressing was calculated from the predicted yield using an existing model and applied at the flowering stage. Regarding the yield prediction model, GNDVI had a higher coefficient of determination than NDVI and a significant correlation with panicle number and SPAD value. The regression line of yield by GNDVI was adopted as a yield prediction model because of its high prediction accuracy. In the variable fertilizer application test, the grain protein content was almost the same as the target value for the standard sowing, but greatly exceeded the target value for late sowing. The reason for this error was considered to be the higher grain protein content due to high temperatures after the heading stage. In conclusion, variable rate fertilization is effective in the timely sown ‘Setokirara’, but the model needs to be improved for late sowing.