Leaf transmittance measurements can improve predictions of the nitrogen status for winter wheat crop

Abstract

Dynamic crop models can be used to predict the occurrence of nitrogen deficiency during crop growth and optimize nitrogen fertilisation. However, prediction errors can be large and may lead to wrong recommendations. The objective of our work is to study the value of correcting the dynamic model Azodyn using transmittance measurements made with the N-Tester ® (Yara) to predict the nitrogen status of a winter wheat crop. Our approach is to use a Bayesian method called the “interacting particle filter” to fit the model's state variables to measurements obtained over the course of the season. This approach was assessed on 44 experimental plots. Predictions of crop biomass, nitrogen uptake and nitrogen nutrition index were first performed for each plot by using the model without any correction. A second series of predictions was then performed for the same variables by correcting the model with N-Tester measurements at GS 7 on Feekes’ scale. The results showed that the second series of predictions were more accurate. Depending on the prediction dates, model corrections reduced the root mean squared error by 18.1–53.2% for nitrogen nutrition index, by 9.1–10.1% for biomass, and by 17.1–45.0% for nitrogen uptake. The predictions were improved up to 52 days after the measurement but the degree of improvement was higher when the prediction date was close to the measurement date. The results also showed that, when corrected, model predictions were very sensitive to values of N-Tester measurements. It is therefore necessary to use N-Tester measurements which are as precise as possible.