Diagnosis of nitrogen status in winter oilseed rape (Brassica napus L.) using in-situ hyperspectral data and unmanned aerial vehicle (UAV) multispectral images

This study aimed to investigate whether the optimal vegetation indices (VIs) derived from the in situ hyperspectral data to estimate the nitrogen nutrition index (NNI) can also be used at the local scale using unmanned aerial vehicle (UAV) multispectral images, and whether texture metrics derived from UAV images could improve the remote estimation of the NNI in winter oilseed rape. Three field experiments with different N fertilization levels were conducted in two sites in Hubei Province, China. The mechanistic and empirical methods were both employed to estimate NNI. With the in situ hyperspectral data, the empirical method based on structural VIs (R-2 is about 0.70) or the photochemical reflectance index (PRI) (R-2 = 0.73) provided more accurate estimations of NNI than the mechanistic method did (R-2 = 0.62). Although most of the studied VIs were strongly correlated with the NNI, they had different responses to the NNI at the low N fertilization and the optimal to excessive N fertilization rates. For the UAV multispectral images, the mean VI of all pixels within the region of interest (ROI) (referred to VI(-)mixed) outperformed the mean VI of vegetation pixels within the ROI (referred to VI(-)pure). The mean normalized difference vegetation index (NDVI(-)mixed), the modified soil adjusted vegetation index 2 (MSAVI2(-)mixed), and the red edge chlorophyll index (CI(red edge-)mixed) of all pixels within the ROI yielded more accurate NNI estimates than the other Vls. Furthermore, the stepwise multiple linear regression models with VIs and texture metrics of VIs provided more accurate NNI estimations than the models based solely on VIs. Results of this study suggested the great potential of UAV multispectral images in monitoring the crop N status at local scales.