Models for estimating cotton aboveground fresh biomass using hyperspectral data

The hyperspectral reflectance(350 to 2 500 nm) data were recorded at the six key cotton growth stages in a field experiment. Band depth analysis was conducted on Continuum-removed spectra between 550 and 750 nm, and the band depth at the waveband center(Dc) is the maximum band depth; meanwhile, stepwise regression method is applied to analyze the correlation between reflectance and cotton aboveground fresh biomass, and two bands are confirmed being sensitive to cotton fresh biomass at near infrared band 763 nm and red region band 670 nm, combining two bands reflectance into vegetation indices of NDVI(Normalized difference vegetation index) and RVI(Ratio vegetation index). Based on Dc parameter, NDVI and RVI vegetation indices, five single variables of linear and nonlinear function models against cotton aboveground fresh biomass were established, and result shows that the RVI exponential function has a higher correlation coefficient(R=0.7289^**, RMSE=0.8776) between estimating data and testing data of cotton aboveground fresh biomass; it is significant for five types of function models(α=1%), whilst, power function fitting, exponential function fitting and hyperbolic function fitting have a comparatively higher accuracy for estimating cotton aboveground fresh biomass; it is real-time, nondestructive and quantitative for adopting hyperspectral parameter Dc, vegetation indexes NDVI, RVI to obtain cotton aboveground fresh biomass, and it also provides an approach to analysis, simulation, evaluation, and prediction of the dimension of cotton canopy, finally it can offer an evidence to design an optimum cotton canopy and estimate cotton yield by using hyperspectral remote sensing.