Cerrado Wildfire Data Package

This dataset supports the article A Retrospective Benchmark of Spatiotemporal Covariates for Daily Active-Fire Detection in Cerrado Conservation Units. The package provides derived modeling tables, metadata, configuration files, evaluation summaries, and figure-support tables used to audit the reported results. The study evaluates remote-sensing and environmental covariates for daily active-fire detection in Brazilian Cerrado Conservation Units and their buffer zones in Minas Gerais, Brazil. The package includes Stage 1 through Stage 4 modeling tables and independent AOI test tables in Parquet format, together with source-data manifests, split metadata, feature dictionaries, label definitions, preprocessing summaries, model-evaluation tables, and reproducibility files. Large raw source files, raw Google Earth Engine extraction outputs, raster caches, model binaries, SHAP arrays, logs, cache files, and temporary artifacts are not included when they are derivable from public data sources and documented workflows. Associated preprint: https://arxiv.org/abs/2606√.04170 Executable workflow and source code: https://github.com/julianopadua/cerrado-wildfire-prediction-reproducibility本数据集支撑论文《巴西塞拉多保护区每日明火探测的时空协变量回顾性基准研究》。 本数据集包提供衍生建模表格、元数据、配置文件、评估汇总结果，以及用于核验已报道成果的配图支撑表格。本研究针对巴西米纳斯吉拉斯州境内的巴西塞拉多保护区及其缓冲区，开展了每日明火探测所用遥感协变量与环境协变量的评估工作。 本数据集包包含第1阶段至第4阶段的建模表格，以及Parquet格式(Parquet)的独立感兴趣区域(AOI, Area of Interest)测试表格，同时附带源数据清单、拆分元数据、特征字典、标签定义、预处理汇总结果、模型评估表格与可复现性文件。 若大型原始源文件、谷歌地球引擎(Google Earth Engine)原始提取输出、栅格缓存、模型二进制文件、SHAP数组、日志文件、缓存文件与临时人工产物可通过公开数据源与已记录的工作流获取，则此类文件不会被包含在本数据集中。 相关预印本：https://arxiv.org/abs/2606.04170 可执行工作流与源代码：https://github.com/julianopadua/cerrado-wildfire-prediction-reproducibility