Evaluating and defining agronomically relevant detection limits for spectral reflectance-based assessment of N uptake in wheat

Spectral detection of the N uptake of wheat is a widely used method, as it is non-destructive, rapid and cost-efficient. However, to date, agronomically supported spectral detection limits have not been sufficiently described. Concurrent statistical measures such as R2, RMSE, or MAE do not fully satisfactorily address the agronomical relevance, bearing in mind that sensing is frequently carried out in either very low or excessively high nitrogen fertilization applications, which may not be indicative of current farming practice. This study, therefore, evaluates regression models of spectral indices in capturing N uptake using hyperspectral ground-based and multispectral unmanned aerial vehicles (UAV) based on data sets covering several years, sites, vari-eties, and developmental stages of wheat (Triticum aestivum L.). The results suggest that solely adopting commonly used statistical measures is not sufficient for an agronomic evaluation. Whereas the common statis-tical measure R2 is essentially influenced by differentiating N uptake, which primarily occurs at later stages of development, the use of further statistics such as RMSE and MAE averages the error and should be extended by self-set confidence intervals based on agronomical decisions. For these studies, we therefore defined an appro-priate error interval of +/- 15 kg N uptake ha-1 up to BBCH 50, with a probability of at least 80%. Interval limits may be narrowed for earlier developmental stages and wider for later ones. Extreme N levels in field trials can bias models and should be limited to N-fertilization ranges that are indicative of the current practice in a given region, so as not to overemphasize the extremes. Differentiation of biomass was revealed to be more crucial than that of N content in detecting N uptake. Essentially, both terrestrial-and UAV-based sensing were equally well suited, with combinations of the REDEDGE and NIR bands being particularly effective for detecting the N uptake of wheat. Agronomically based detection limits should be considered besides common statistical measures in the spectral assessment of wheat N uptake.

Automated summary

This study evaluates regression models of spectral indices in capturing N uptake using hyperspectral ground-based and multispectral unmanned aerial vehicles (UAV). It suggests that solely using commonly used statistical measures is not sufficient for an agronomic evaluation and agronomically based detection limits should be considered in addition to common statistical measures in the spectral assessment of wheat N uptake.