Detecting drought status and LAI of two Quercus species canopies using derivative spectra

Abstract

Derivative spectrum analysis has the advantage of reducing additive constants and minimizing soil background effects and is a potential method for exploiting hyperspectral data in vegetated areas. The shift of red-edge position (REP) of vegetation reflectance spectra, which is a focus point of derivative spectrum analysis, has been studied by many researchers as an indicator of environmental stresses and LAI. However, it is less satisfactory at canopy level than at leaf level. A field experiment was conducted, using 3-year-old potted Quercus glauca and Q. serrata, to examine the utility of derivative spectrum analysis for detecting drought status and LAI at canopy level and to find the optimal bands that can independently detect those variables. Five levels of drought status (including a control) and three levels of LAI were set. Measurements were made of canopy reflectance spectra at approximately 3 nm intervals, xylem water potentials, leaf water contents and LAI. Two measures for representing high or low drought status were chosen. One was water-cessation duration in hours (WD) and the other was leaf water content (LWC). They corresponded to a gradual change and an abrupt change, respectively, during drought development. The results showed that REP and REP-relevant indices were not very successful for independent detection of WD, LWC or LAI at canopy level. The best single bands for detecting WD, LWC and LAI were 611.4 nm in the first derivative ( r=0.807), 519.6 nm in the first derivative ( r=0.916) and 676.0 nm in the second derivative ( r=0.828), respectively. The initial part of the red-edge peak was a better indicator than the top (i.e. REP or the first derivative at REP) for the independent detection of LAI. A simulation to test lower spectral resolution proved that the wavelengths at 10 nm intervals that approximated the desirable bands at 3 nm intervals retained similar correlation coefficients for the three variables.