Comparative analysis of methods and models for predicting biochemical methane potential of various organic substrates.

Biochemical methane potential (BMP) corresponds to the maximum methane production at anaerobic digestion infinite time and is a key parameter to evaluate the suitability of substrates to obtain biogas. The main objective of this work is to explore the data available in the literature for ten categories of substrates to compare and develop new methods and mathematical models able to predict BMP. Indeed, experimental procedure is time-consuming, laborious and costly, and the development of methods or models based on properties easily assessed may be very helpful at industrial scale. In this study, three substrates (banana waste, tomato waste and winery wastewater) were tested and compared with >150 results from the literature. The analysis involved four methods (Met_I to Met_IV) and five models developed by multivariate regression (Mod_I to Mod_V). Met_I is related to elemental analysis; Met_II with the organic fraction composition; Met_III is associated with chemical oxygen demand (COD); Met_IV is based on NIR spectra. Regression models are combinations by grouping single variables: C, H, O, N (Mod_I); hemicellulose, lignin (LG), acid detergent fibre (ADF) (Mod_II); volatile solids (VS), COD (Mod_III); proteins (PT), carbohydrates (CRB), lipids (LP) (Mod_IV); and CRB, LP, PT, LG, ADF (Mod_V). The results showed that no significant correlation can be found between BMP and single common properties (e.g. VS or C/N ratio). However, good results may be achieved with models developed by multivariate regression (R from 0.93 to 0.98, and R from 0.91 to 0.96). The prediction of BMP based on Met_IV, which is based on NIR spectroscopy combined with a multivariate regression model, revealed to be a promising method for both data from literature as well as for substrates analyzed in the present work.