Combined use of in situ hyperspectral vegetation indices for estimating pasture biomass at peak productive period for harvest decision

Since the typical destructive methods for measuring aboveground biomass (AGB) have many limitations, a variety of non-destructive techniques have been developed. In this paper, the potential of ground-based hyperspectral remote-sensed data for non-destructive assessment of semi-arid pasture AGB at the peak productive period was investigated. The reflectance spectrometric and AGB data were sampled at the end of the growing season (almost peak biomass) over two locations at pastures in the southern Horqin sandy land, eastern Inner Mongolia, China. All combinations (two-band and three-band) of narrow bands in the forms of simple ratio vegetation index (SRVI), normalized difference vegetation index (NDVI), soil adjusted vegetation index (SAVI) and enhanced vegetation index (EVI) were used in a linear regression analysis against AGB. The predictive performance of the stepwise multiple linear regression (SMLR) using 4 best VIs as input variables was compared with the performance of multivariate partial least squares regression (PLSR) using all reflectance bands as input variables to estimate AGB. It was observed that the relationship between AGB and single band spectral reflectance was low, while the estimation performance of the best VIs based on all available wavebands was improved considerably. In addition, the best VIs based on all available wavebands had considerably better fitting performance than those based on traditionally used wavebands for estimating AGB. In comparison to PLSR using the full individual reflectance as input variables, SMLR using the best VIs as input variables performed much better, with the maximum decrease in RMSECV of 37% and the relative mean absolute errors always below 12.5%. The study demonstrated the high potential to estimate pasture AGB, which is a proxy for pasture forage yield, at the peak productive period using a hyperspectral technique.