Estimating Grassland Biomass Based On Optimized Spectral Indices

Recently, development of hyperspectral remote sensing provides the feasibility in timely and nondestructive deriving biomass of forage grass while the traditional remote sensed indices have limited in estimating biomass in high dense vegetation and complicated biomass canopy structure. The objective of this study was to test the robustness of published and optimized spectral indices in estimating biomass of forage grass in natural and artificial pasture in Inner Mongolia from 2013 to 2014. Chlorophyll index (CI) was calculated based on the concept model while normalized difference spectral index (NDSI) and simple spectral index (RSI) were calculated form all possible two bands combinations from 350 nm to 1150 nm. Linear regression analysis and nonlinear regression analysis were used as variable selection and modeling techniques to predict biomass of grassland grass. Results indicated that:(1) The optimized indices produced higher correlations with biomass as compared to the published indices and there is no saturate problem happened to the published index NDVI1 used in high biomass canopy structure.(2) The best performing spectral index varied in different species of forage grass with different treatments (R2=0.00-0.77) and the influence in canopy structure and biomass. The canopy reflectance has a great difference in different type of forage grass.(3) The best selected NDSI were computed from bands located at 696 and 652 nm and it produced better predictive accuracies (R2=0.77, RMSE (root mean square error of prediction)=756, RE (mean relative error)=37.1%) as compared to CI(R2=0.76, RMSE-954,RE=52.0%) and RSI(R2=0.80, RMSE=927, RE=35.8%). In conclusion, optimizing wavebands combination was a promising algorithm for improving prediction abilities of biomass for complicated dataset of grassland biomass and NDSI in this paper showed higher relationship correlation as compared to others.