Regression models of ultimate methane yields of fruits and vegetable solid wastes, sorghum and napiergrass on chemical composition

Several fractions of fruits and vegetable solid wastes (FVSW), sorghum and napiergrass were analyzed for total solids (TS), volatile solids (VS), total organic carbon, total kjeldahl nitrogen, total soluble carbohydrate, extractable protein, acid-detergent fiber (ADF), lignin, cellulose and ash contents. Their ultimate methane yields (B(o)) were determined using the biochemical methane potential (BMP) assay. A series of simple and multiple regression models relating the B(o) to the various substrate constituents were generated and evaluated using computer statistical software, Statistical Package for Social Sciences (SPSS). The results of simple regression analyses revealed that, only weak relationship existed between the individual components such as carbohydrate, protein, ADF, lignin and cellulose versus B(o). A regression of B(o) versus combination of two variables as a single independent variable such as carbohydrate/ADF and carbohydrate+protein/ADF also showed that the relationship is not strong. Thus it does not appear possible to relate the B(o) of FVSW, sorghum and napiergrass with single compositional characteristics. The results of multiple regression analyses showed promise and the relationship appeared to be good. When ADF and lignin/ADF were used as independent variables, the percentage of variation accounted for by the model is low for FVSW (r(2)=0.665) and sorghum and napiergrass (r(2)=0.746). Addition of nitrogen, ash and total soluble carbohydrate data to the model had a significantly higher effect on prediction of B(o) of these wastes with the r(2) values ranging from 0.9 to 0.99. More than 90% of variation in B(o) of FVSW could be accounted for by the models when the variables carbohydrate, lignin, lignin/ADF, nitrogen and ash (r(2)=0.904), carbohydrate, ADF, lignin/ADF, nitrogen and ash (r(2)=0.90) and carbohydrate/ADF, lignin/ADF, lignin and ash (r(2)=0.901) were used. All the models have low standard error values, which indicate the amount of spread is less. Thus, considering only the higher r(2) values, six models are proposed for predicting the B(o) based on FVSW data and sorghum and napiergrass data. It would be more convenient if B(o) could be predicted by analyzing the chemical composition of the substrate rather than performing the long-term batch fermentation. To test the validity of the regression models, chemical constituents of FVSW that were not included in the regression analyses were determined and their experimental B(o) were determined by BMP assay. All the six models were used to predict the B(o) from the chemical constituents of these FVSW. It was found that most of the predicted values were within 20% of the experimental B(o) in models 1, 3 and 6. Since models 3 and 6 used the same variables namely, total soluble carbohydrate, ADF, lignin/ADF, nitrogen and ash, B(o) can be predicted from these five chemical constituents which accounts for more than 90% of the variation in B(o) (r(2)>90).