34174

Assessing biogas potentials of energy crops using near infrared reflectance (NIR) spectroscopy

Susanne Frydendal Nielsen 1 ; Radziah Wahid 2 ; Henrik B. Møller 2 ; Claus Felby 3 ; Uffe Jørgensen 4 ; Rene Gislum 4
Favorite this paper

Biogas potentials are typically measured during 30-100 days batch tests. These batch tests are time consuming to perform with the results difficult to repeat. The aim of this study was to develop a partial least squares (PLS) model that is able to screen various plant materials for their biogas potentials and to identify which variables are important for the model using loading plots and interval PLS (iPLS).
Experimental
The database consists of 53 plant samples of above- and below-ground biomass from Jerusalem artichoke and beets; and aboveground biomass from maize, hemp, Miscanthus x giganteus and Miscanthus sinensis stem and leaves, respectively. Accumulated biogas production was measured in batch test experiments for 90 days and the results used as reference values for the PLS and iPLS models. Near infrared reflectance (NIR) measurements were conducted on ground samples using a cyclone mill (FOSS Cyclotec ™ 1093, Denmark) with a 0.8mm sieve. Each sample was measured in duplicates using a FOSS NIR-spectrometer DS2500. Measurements were performed using a cup (diameter of 7 cm) in the range from 400 to 2500nm. Data analysis was carried out using MATLAB version 7.9, along with the PLS toolbox 7.9.3. Spectra was pre-processed using multiplicative signal correction (MSC) and mean centering. The model was cross validated using venetian blinds containing five samples per split. Principal component analysis (PCA) and PLS was conducted on the entire spectra (400-2500nm) while PLS and iPLS were conducted on the range from 1100-2500nm.
Results and discussion
PCA separated the plant material mainly due to the visual part of the spectra. The PLS model based on the full spectra had an R2 of 0.66 and root mean square error of cross validation (RMSECV) of 62.8. The best model was obtained using iPLS from 1100 to 2500 nm giving an R2 of 0.71 and RMSECV of 53.0. The important variables for the iPLS model were: 1500-1524nm, 1800-1824nm and 2375-2399nm. These variables are related to proteins, cellulose and lipids respectively; these constituents are known to influence the biogas production. We will continue our work with a focus on how to relate the visual and chemical information to biogas potential. The main challenge is that biogas potential is difficult to relate to a few chemical constituents in the NIRS spectra. Furthermore, we have an additional challenge that results from batch testing of biogas potential are difficult to replicate which gives a high predictive error on the PLS and iPLS regression models.
Conclusions
A model for screening purposes is demonstrated; however, some of the variations are unaccounted for due to the complexity of the parameters determining biogas potentials.