Applying UAV-based crop height to monitor biomass, N-concentration, and N-uptake in winter wheat

Abstract. Spatial knowledge for supporting precise N fertilization is of key interest in crop management. Therefore, accurate and reliable data on crop dry biomass (DB) and N concentration (Nconc), and N-uptake (Nup) are needed considering spatial heterogeneity. While N uptake in field experiments is computed using in-situ data of DB and Nconc, it also can be directly estimated with remote sensing methods. Usually, these crop traits are derived by using optical remote or proximal sensing approaches. In this contribution, we investigate a paradigm change in providing non-destructive DB, Nconc, and Nup estimates by using non-optical data analyses but structural information extraction. Numerous studies proofed UAV-derived crop height can serve as a robust estimator for biomass. Due to the well-known negative correlation between biomass and N concentration over the growing season crop height might be used as an estimator for Nconc as well. Based on these correlations we investigate three key hypotheses: (i) crop height from UAV images using a Structure from Motion and Multiview Stereopsis (SfM/MVS) workflow serves as a very robust estimator for DB, (ii) Nconc is correlated over the growing season to DB, and (iii) DB is the dominating parameter in determining Nup. Hence, the main research question of this contribution is if UAV-derived crop height (ÚAV-CH) serves as a robust estimator for DB, it also can be used to directly estimate Nconc and Nup, UAV-CH in ultra-high spatial resolution (< 3 cm) is a mixed signal of crop height and density for a given spatial unit, eg. a square meter or a research plot, and therefore provides valuable crop canopy information. We present results from a 3-years effort in UAV- and in-situ data acquisition and analyses which partly support the proposed paradigm change. For each of the three years 2020, 2021, and 2022, DB can be robustly estimated using UAV-CH, having a R² of 0.89, 0.91, and 0.92, respectively. For Nconc and Nup the results are not as promising on a yearly analysis having R² for Nconc of 0.57, 0.21, and 0.41, and for Nup of 0.72, 0.61, and 0.48, respectively, but are comparable to optical approaches. However, for fertilizer recommendation, the performance on growing stage specific level is of more importance. Surprisingly, the proposed approach seems to provide similar or better results compared to optical sensing analysis showing R² for campaign specific days (approx. every 14 days over 3 growing periods) for N uptake between 0.57 and 0.82, and for Nconc between 0.24 and 0.78. We conclude that UAV-CH can be used as a very good and robust estimator for DB, as a moderate for Nconc, and as a moderate but robust estimator for Nup having the advantage of being more flexible in terms of less effected by weather conditions. Finally, the SfM/MVS workflow to derive UAV-CH has the potential to be fully or semifully automated from data acquisition to fertilizer recommendation.