Detecting leaf nitrogen content in wheat with canopy hyperspectrum under different soil backgrounds

• Five types of spectral index were significantly affected by soil backgrounds.
• Novel wavebands were extracted for LNC detecting using three different methods.
• Newly developed CASI performances better as compared with other spectral indices.
• The quantitative model was developed for LNC at wheat canopy scale under soil backgrounds.

Hyperspectral sensing techniques can be effective for rapid, non-destructive detecting of the nitrogen (N) status in crop plants; however, their accuracy is often affected by the soil background. Under different fractions of soil background, the canopy spectra and leaf nitrogen content (LNC) in winter wheat (Triticum aestivum L.) were obtained from field experiments with different N rates and planting densities over 3 growing seasons. Five types of vegetation index (VIs: normalized difference vegetation index (NDVI), ratio vegetation index (RVI), soil adjusted vegetation index (SAVI), optimize soil adjusted vegetation index (OSAVI), and perpendicular vegetation index (PVI)) were constructed based on three types of spectral information: (1) the original and the first derivative (FD) spectrum, (2) the spectrum adjusted with the vegetation coverage (FVcover), and (3) the pure spectrum extracted by a linear mixed model. Comprehensive relationships of above five types of VI with LNC were quantified for LNC detecting under different soil backgrounds.The results indicated that all five types of VI were significantly affected by the soil background, with R2 values of around 0.55 for LNC detecting, with the OSAVI (R514, R469)L=0.04 producing the best performance of all five indices. However, based on the FVcover, the coverage adjusted spectral index (CASI = NDVI(R513, R481)/(1 + FVcover)) produced the higher R2 value of 0.62 and the lower RRMSE of 13%, and was less sensitive to the leaf area index (LAI), leaf dry weight (LDW), FVcover, and leaf nitrogen accumulation (LNA). The results demonstrate that the newly developed CASI could improve the performance of LNC estimation under different soil backgrounds.