Identification and Removal of Negative Biomass Samples via Scatter Plot Analysis to Improve GWP Predictive Modeling

¹ Department of Agricultural Engineering, School of Engineering, King Mongkut’s Institute of Technology Ladkrabang, Bangkok, 10520, Thailand, 66016181@kmitl.ac.th.
² Department of Mechanical Engineering, School of Engineering, Kathmandu University, Dhulikhel, PO Box 6250, Nepal, 63601254@kmitl.ac.th
³ Department of Agricultural Engineering, Faculty of Engineering, Khon Kaen University, Khon Kaen 40002, Thailand, jetspo@kku.ac.th
⁴ Department of Food Engineering, School of Engineering, King Mongkut’s Institute of Technology Ladkrabang, Bangkok, 10520, Thailand, pimpen.po@kmitl.ac.th
⁵ Department of BioEngineering, University of Washington, Seattle, William H. Foege Building 3720, 15th Ave NE, Seattle, WA 98195-5061, USA, shrestha@ku.edu.np
⁶ Institute of Catalysis Research and Technology (IKFT), Karlsruhe Institute of Technology (KIT), Hermann-von-Helmholtz-Platz 1, 76344 Eggenstein-Leopoldshafen, Germany, axel.funke@kit.edu

^* Corresponding author: jetspo@kku.ac.th and shrestha@ku.edu.np

Abstract

Accurate prediction of Global Warming Potential (GWP) from biomass constituents is essential for evaluating the sustainability of bioenergy sources. However, the inclusion of biomass samples with weak or negative correlation to key elemental components such as Carbon (C). Hydrogen (H). Nitrogen (N). and Oxygen (O)—can reduce model accuracy and lead to misleading conclusions. This study utilizes scatter plot regression analysis to evaluate and remove "negative biomass samples." defined as those with consistently low R² values across constituent-GWP relationships using HHV = 0.2949C + 0.82 50H developed for wood biomass by Yin. Regression models were generated for each biomass species using elemental concentrations as predictors of GWP. Notably, several non-wood species (e.g.. Zea Mays-Shell. Bagasse. Bamboo) exhibited very low R- values (often <0.05) for model between elemental composition and GWP. where all elemental correlations indicated weak predictive relationships. In contrast, wood-based species such as Alnus demonstrated significantly higher R² values, especially with Carbon (R² = 0.69). Hydrogen (R² = 0.57). and Oxygen (R² = 0.68), suggesting a stronger linear influence on GWP. Removing these low-contributing samples improved the clarity and reliability of the predictive model related to HHV and each type of element (C.H.N and O) as evidenced by a sharper regression slope of a graph plotted between predicted GWP and measured GWP of positive species and better fit (increased R²) for the remaining samples. These results highlight the value of preliminary scatter plot analysis in identifying biomass species that obscure rather than support predictive modeling. This filtering step ultimately enhances the robustness and inteipretability of constituent-based GWP prediction frameworks, particularly when applying FT-XIR spectroscopy and chemometric modelling.