Identification of hyperspectral vegetation indices for Mediterranean pasture characterization

A field experiment was carried out to assess biomass and nitrogen status in Mediterranean pastures by
means of hyperspectral high resolution field radiometric data. Spectral and agronomicmeasurements
were collected at three different pasture growth stages and in grazed–ungrazed plots distributed over
an area of 14 ha. Reflectance-based vegetation indices such as simple ratio indices (SR[i,j]) and
normalized difference vegetation indices (NDVI[i,j]) were calculated using all combinations of two
wavelengths i and j in the spectral range 400–1000 nm. The performances of these indices in predicting
green biomass (GBM, t ha1), leaf area index (LAI,m2 m2), nitrogen content (N, kg ha1) and nitrogen
concentration (NC, %)were evaluated by linear regression analysis using the cross validated coefficient
of determination (R2
CV) and root mean squared error (RMSECV). SR involving bands in near-infrared
(i = 770–930 nm) and in the red edge (j = 720–740 nm) yielded the best performance for GBM
(R2
CV ¼ 0:73, RMSECV = 2.35 t ha1), LAI (R2
CV ¼ 0:73, RMSECV = 0.37m2 m2), and N (R2
CV ¼ 0:73,
RMSECV = 7.36 kg ha1). The best model performances for NC (R2
CV ¼ 0:54, RMSECV = 0.35%) were
obtained using SR involving near-infrared bands (i = 775–820 nm) and longer wavelengths of the red
edge (j = 740–770 nm). The defined indices lead to significant improvements in model predictive
capability compared to the traditional SR [near-infrared, red] and NDVI [near-infrared, red] and to
broad-band indices. The possibility of exploiting these results gathered at field level with high
resolution spectral data (FWHM 3.5 nm) also at landscape level bymeans of hyperspectral airborne or
satellite sensors was explored. Model performances resulted extremely sensitive to band position,
suggesting the importance of using hyperspectral sensors with contiguous spectral bands.