HERBAGE BIOMASS AND QUALITY STATUS ASSESSMENT IN A MIXED SOWN PASTURE FROM AIRBORNE BASED HYPERSPECTRAL IMAGING

摘要

Aerial hyperspectral imaging is one of the potential tools for the site specific agriculture. This study investigates the use of airborne based hyperspectral sensor, the Compact Airborne Spectrographic Imager 3 (CASI-3) data acquired on July 24, 2007, for estimating spatial distributions of green herbage biomass (GBM), nitrogen (N) and neutral detergent fiber (NDF) concentrations comparing the predictive accuracy with in field hyperspectral measurements in a pasture at Hokkaido, Japan. Canopy spectral measurements were made in the field at 42 plots using an ASD FieldSpec FR Pro spectroradiometer, along with concomitant in situ measurements of GBM and concentrations of N and NDF. Three types of partial least squares (PLS) regressions, full-spectrum PLS (FS-PLS), iterative step elimination PLS (ISE-PLS) and genetic algorithm based PLS (GA-PLS), were performed using canopy reflectance data of CASI-3 and ASD, and simulated CASI-3 spectra from ASD (ASD_(CASI)) to predict GBM and concentrations of N and NDF. Among the PLS regressions, GA-PLS showed the best R~2 and lowest RMSECV values in all data set (ASD, CASI-3 and ASD_(CASI)) to predict GBM (R~2 = 0.69-0.86, RMSECV = 48.70-72.68), N (R~2 = 0.61-0.75, RMSECV = 0.25-0.31) and NDF (R~2 = 0.50-0.73, RMSECV = 2.34-3.28). Applying the GA-PLS models on the CASI image, spatial distribution maps of forage GBM and concentrations of N and NDF of herbage were generated.