FireDataForge: example and reproducibility bundle

Companion data for FireDataForge (https://github.com/xiazeyu/FireDataForge), an open-source Python framework that automates the retrieval, harmonization, and integration of wildfire-related geospatial data from 11 sources. This bundle is the small subset of inputs and reference outputs needed to reproduce the repository's examples, benchmark, and validation without downloading the full FEDS-MTBS archive.

Contents

- examples.zip — eight example FEDS-MTBS fires (2013–2025; 5 California, 3 Colorado; ~9.7k–189k acres), packaged as a self-contained working directory that unzips at the repository root. Includes per-fire GeoPackages (perimeter/fireline), per-year VIIRS firepix CSVs for the six pre-2025 fires, an offline fire list, and events.txt. Fetched automatically by python main.py --fetch-examples.

- benchmark_results.zip — runtime/throughput benchmark (RUNTIME.md, system_info.md, results_summary.json, raw results.jsonl, and per-run logs).

- validation_metrics.csv — per-event validation metrics: categorical-layer accuracy, continuous-field MAE/RMSE, FRP conservation, sub-pixel registration shift, and reprojection round-trip RMSE.

- ZENODO_README.md — full description of every file, the validation-metric column definitions, provenance, and licensing.

Provenance — Six of the eight fires are reproduced from the FEDS-MTBS dataset (Chen et al., https://doi.org/10.5281/zenodo.20187962), an MTBS-constrained extension of the FEDS algorithm (Chen et al. 2022, Scientific Data, https://doi.org/10.1038/s41597-022-01343-0). The two 2025 fires (Palisades, Eaton) were generated with the same algorithm by the same authors. Bundled FEDS-MTBS data is © Yang Chen et al., redistributed here under CC-BY 4.0 with the authors' explicit permission.

If you use this bundle, please cite both this record and the FEDS-MTBS dataset (https://doi.org/10.5281/zenodo.20187962).