Biomass characterisation using Visible Near Infrared and Shortwave Infrared Spectral Imaging

The generation of energy from biomass fuels or bioenergy is increasing in popularity. Biomass fuels are heterogeneous in their shape, size and chemical constituents. This causes variability in fuel quality. Slow laboratory analysis inhibits quality control systems that identify poor quality fuels. The aim of this research is to rapidly predict moisture, ash and bark content in biomass using spectral imaging. Scanning biomass within plastic sample bags, this project aims to speed up moisture and ash predictions and allow rapid classification of bark from woodchip. A comparison between the SWIR (978 – 1,678 nm) and Vis-NIR (400 – 1000nm) range was made across the developed models. The Vis-NIR outperformed the SWIR in bark predictions using PLSDA models. The SWIR was better than the Vis-NIR in PLSR models pixel predictions MC% in a small set (n=20) of predetermined woodchip samples (RMSEP 3.09 versus 8.12%). In the larger LMA data set SWIR (n=195) outperformed the Vis-NIR (n=191) in pixel predictions of MC% (RMSEP 3.09 versus 3.92%). Ash % pixel predictions were better in the SWIR than in the Vis-NIR (RMSEP 0.98 versus 1.02%). The models predicted MC% and Ash% in woodchip samples sealed in plastic, which may prove a useful attribute. The moisture models were the most successful of all pixel predictions.