Using remote sensing to estimate forage biomass and nutrient contents at different growth stages

Lignocellulosic biomass is an important feedstock for the second generation bioenergy production. A sustainable supply of biomass feedstock having consistent composition is critical for a biorefinery. This requires a timely monitoring and estimation of the biomass yield and composition in the field. It is not clear if one can use the near infrared (NIR) vegetation canopy reflectance measured in the field and build a calibrate model to estimate biomass yield and nutrient contents (compositions) for the vegetation from vegetative growth through dormancy stages. In this study, the NIR canopy reflectance of a grass/legume mixture was measured in a field with a spectroradiometer with wavelength ranging from 400-2500 nm. The plants were then clipped, dried, and ground to measure the biomass yield, neutral detergent fiber (NDF), acid detergent fiber (ADF), and crude protein (CP). The measurements were conducted at the boot, peak growth, and at dormancy stages. Partial least significant regression (PLSR) models were built using data from each individual growth stage as well as all stages combined. Except for the yield model at peak growth stage and the NDF and ADF models at dormancy stage, models developed from each of the individual stages generally estimated the yield, CP, NDF, and ADF poorly, with R2CV ranging between -0.31 and 0.42. When data from all three growth stages were included, the accuracy of all models was greatly improved, with R2CV ranging between 0.77 and 0.80. Furthermore, multiple linear regression (MLR) models developed with 7-9 most significant wavelengths selected from 400-2500 nm estimated the yield, ADF, NDF, and CP equally well compared with the PLSR models. The estimates from MLR model showed strong correlations between the measured and estimated values, with R2 of 0.72, 0.67, 0.78, and 0.66 for the yield, ADF, NDF, and CP, respectively. These results indicate that biomass feedstock yield and composition can be estimated by the in-situ NIR canopy reflectance using a multiple linear regression model with 7-9 wavelengths. Further calibration is needed to use the model to other geological locations.