Analysis code and derived data for: Artificial Intelligence Across the Wildfire Management Lifecycle — A PRISMA-Guided Survey of Techniques, Validation Practices, and the Global-South Gap

This deposit contains the complete analysis pipeline and the derived datasets behind the review article "Artificial Intelligence Across the Wildfire Management Lifecycle: A PRISMA-Guided Survey of Techniques, Validation Practices, and the Global-South Gap". The review follows a reproducible PRISMA-2020 protocol. A single Scopus query, executed on 29 June 2026 and published verbatim in this deposit, returned 4,463 records, reduced to an eligible corpus of 2,086 journal articles (2015–2025). The article characterises that corpus bibliometrically, formalises how the surveyed methods work, and quantifies validation practice on the corpus itself: fewer than 1% of eligible abstracts mention spatially blocked validation, and only one of the 41 primary studies synthesised reports a spatially blocked protocol. Contents code/ — Python and R scripts that regenerate every reported count, table and figure: PRISMA flow, bibliometric profile, stratified selection pool, supplementary tables, and the keyword co-occurrence network of Figure 3. data/corpus/ — the corpus accession list identifying all 2,086 eligible records (DOI, Scopus EID, year, source, document type, open-access status, citation count, author keywords), plus the exact query string and retrieval date. data/derived/ — annual production, source counts, author-keyword frequencies, the top-50 co-occurrence matrix, Louvain community assignments, PRISMA exclusion logs and the abstract-level term scan results. data/supplementary_tables/ — supplementary tables S1–S3 as CSV. docs/ESM_1.pdf — the article's supplementary material, sections S1–S5. On the raw Scopus export. The raw export is deliberately not included: it carries the Abstract and References fields, which are copyrighted content owned by the individual publishers and licensed to the authors through an institutional Scopus subscription. The accession list identifies the corpus record-for-record, so it can be reconstructed exactly by anyone with Scopus access. Every analysis in the article is reproducible from the files shipped here, with one exception: the abstract-level term scan searches abstract text and therefore requires the licensed export. Its results are included; re-running it requires supplying your own export. See README.md, Section 3. Licences. Code is released under MIT; derived data and documentation under CC BY 4.0. Changes in version 1.1.0. The supplementary material (docs/ESM_1.pdf) is replaced by the version accompanying the manuscript submitted to Environmental Modelling & Software. Section S5 now publishes verbatim the three regular expressions used for the corpus-level term scan, together with the terms they do not match and the direction of that bias. The risk-of-bias wording is aligned with the manuscript, and the reported relation in spread modelling is stated as reported score tracking the breadth of the evaluation rather than its rigour, with the counterexample in the table declared explicitly. The code and the derived data are unchanged from version 1.0.0.