Spectral assessments of N-related maize traits: Evaluating and defining agronomic relevant detection limits

a Summary: The spectral sensing of nitrogen (N)-related traits and the grain yield of maize (Zea mays L.) is widely used in agricultural practice because it is rapid, non-destructive, and cost-efficient. However, there exists a lack of agronomically supported spectral detection limits. The agronomic aspects have not yet been fully considered based on commonly used statistical measures such as the coefficient of determination (R-2), the root-mean-square error (RMSE), and the mean absolute error (MAE) and should therefore be extended. In the present study, we evaluated regression models of spectral indices derived from an unmanned aerial vehicle (UAV) by capturing Nrelated traits such as above-ground and grain N uptake, the nitrogen nutrition index (NNI), and grain yield covering data sets of two years, sites, and four developmental stages of maize. The results suggest that an agronomic evaluation exclusively adopting widely used statistical measures is not fully adequate. The R-2 is essentially influenced by differentiation of the trait, which in turn depends on year effects and growth stages. Further statistics such as RMSE and MAE average the error and lead to an under- and overestimation for most observations. In this investigation, we defined an appropriate agronomical error interval for above-ground and grain N uptake, NNI, and grain yield of +/- 40 and +/- 25 kg N ha(-1), +/- 0.2, and +/- 1.4 t ha(-1), with a probability of at least 80%. These interval limits are consistent across years and growth stages. The consistency occurs because most spectral indices are dominated by biomass. Across all of them, the best-performing spectral indices combine GREEN, REDEDGE, and NIR bands. For spectral indices using the RED band, the range of the agronomic error interval performed equally with a slightly worse probability of data points inside the error limits. Agronomically based error limits should be included in addition to common statistical measures in the spectral assessment of Nrelated traits of maize and grain yield to optimize the ex-ante and/or ex-post analysis of N-fertilization.

Automated summary

The study evaluates regression models of spectral indices derived from an unmanned aerial vehicle (UAV) for nitrogen-related traits and grain yield of maize. It suggests that agronomic error limits should be included in addition to common statistical measures in the spectral assessment to optimize N-fertilization analysis.