Compiled literature dataset for biochemical methane potential (BMP) prediction

Structured numerical data were extracted from eligible articles (priority tiers A and B) using a large language model (Gemini 2.5, Google) under a fixed 54-variable schema, with every extracted value traceable to its source page and table, and a random subset independently re-verified by hand against the original PDFs. - **Six external, already-compiled datasets**: a lignocellulosic-biomass BMP database (Liu et al., 2014), a fish-waste BMP review, a manure/sludge review (Cabrita & Santos, 2023), an inoculum-comparison study (Hosseini Koupaie et al., 2019), and a global interlaboratory reproducibility study (Hafner et al., 2026).

All records were harmonized to a common unit basis (mL CH4 g-1 volatile solids) and deduplicated across sources using a union-find clustering procedure. The consolidated file comprises **1128 modeling-ready rows** spanning **59 distinct sources (studies)**. Full methodology is described in Section 2 of the accompanying article. | File | Description | |---|---| | `data/dataset_v1_theoretical.csv` | The full consolidated dataset: all 1128 modeling-ready rows, including engineered features (Section 2.2) and the Buswell--Boyle theoretical BMP / biodegradability factor (Section 2.3). | | `data_dictionary.md` | Column-by-column description of every variable used in the article's analyses. | | `LICENSE` | CC-BY 4.0. |