Estimating SPAD value, chlorophyll and mineral components using hyperspectral data of maize leaves

Background: Visible-infrared hyperspectral data have been widely used recently in remote sensing for nondestructive crop-quality estimation in the field. The authors applied hyperspectral remote sensing to the field of feed maize to investigate the estimation of feed contents of the whole maize plant (including leaves, stems, and grains) from the hyperspectral data of maize community.

Methods: In this study, as a preliminary step to the estimation of feed contents, we attempted to estimate the SPAD value, chlorophyll (a, b and a+b), and mineral components (T-N, T-P, and T-K) contained in leaves from the hyperspectral data (390-983 nm, 60 bands) of maize leaves.

Results and discussions: Regarding the estimation method, we compared the estimation accuracy of two kinds of partial least squares regression (PLSR) using either all bands (60 bands) or only selected ones as explanatory variables. When all bands were used as explanatory variables, estimation was possible with accuracy that is sufficient for practical use for all parameters except chlorophyll b, phosphorus (T-P) and potassium (T-K) (R2 = 0.82-0.90, El = 19.7-24.5, El Rank= B). When waveband selection was conducted, it was judged that all parameters except phosphorus (T-P) and potassium (T-K) can be estimated with accuracy that is sufficient for practical use (R2 = 0.78-0.91, El = 19.6-21.7, El Rank= B). Based on the relation between measured values and estimated ones in verification, it was judged that actual estimation was possible for three parameters: the SPAD value, chlorophyll a+b and nitrogen (T-N).

Conclusion: The results described above demonstrate that the SPAD value related to the greenness (depth of green color) of the leaf blade, chlorophyll a+b and nitrogen (T-N) can be estimated by applying PLSR, or PLSR with band selection, to hyperspectral data of maize leaves.