Forest chemistry mapping with hyperspectral data

For forest chemical concentration mapping with hyperspectral imagery, it is a common practice to relate chemical measurements to image spectra by one of several linear regression techniques. To improve the mapping accuracy, we apply arithmetic transformations to the image spectra to reduce the spectra variations due to differences of fractional compositions within pixels. Canopy endmember fractions, derived from a linear spectral unmixing, are used to adjust the chemical measurements to reflect the pixel fractional composition. It is found in this study that the 2^nd derivative of absorbance spectra have the best correlation with foliar nitrogen measurements. Moreover, the adjustments with canopy endmember fractions can improve this correlation. Finally a foliar nitrogen concentration map is created by using a multiple linear regression to relate the canopy-fraction-adjusted nitrogen measurements to the 2^nd derivative absorbance spectra.