Analysis of Crop Reflectance for Estimating Biomass in Rice Canopies at Different Phenological Stages

This paper contributes an assessment for estimating rice (Oryza saliva L., irrigated lowland rice) biomass by canopy reflectance in the Sanjiang Plain, China. Hyperspectral data were captured with field spectroradiometers in experimental field plots and farmers' fields and then accompanied by destructive aboveground biomass (AGB) sampling at different phenological growth stages. Best single bands, best two band-combinations, optimised simple ratio (SR), and optimised normalized ratio index (NRI), as well as multiple linear regression (MLR) were calculated from the reflectance for the non-destructive estimation of rice AGB. Experimental field data were used as the calibration dataset and farmers' field data as the validation dataset. Reflectance analyses display several sensitive bands correlated to rice AGB, e.g. 550, 670, 708, 936, 1125, and 1670 nm, which changed depending on the phenological growth stages. These bands were detected by correlograms for SR, NRI, and MLR with an offset of approximately 10 nm. The assessment of the three methods showed clear advantages of MLR over SR and NRI in estimating rice AGB at the tillering and stem elongation stages by fitting and evaluating the models. The optimal band number for MLR was set to four to avoid overfitting. The best validated MLR model (R-2 = 0.82) at the tittering stage was using four bands at 672, 696, 814 and 707 nm. Overall, the optimized SR, NRI, and MLR have a great potential in non-destructive estimation of rice AGB at different phenological stages. The performance against the validation dataset showed R-2 of 0.69 for SR and R-2 of 0.70 for NRI, respectively.