Since the number of samples must be higher than the number of
variables, LDAwas performed using, for each spectral range, the 15
variables with the largest classification weight, selected by means
of the algorithm SELECT (Forina, Lanteri, Casale, & Cerrato Oliveros,
2007; Kowalski & Bender, 1976) implemented in the V-Parvus
package. SELECT is a feature selection method useful both in
classification and in regression problems: at each step, it searches
for the variable with the largest Fisher weight in classification or
with the largest correlation coefficient with the response in
regression.