Supplementary source data for "Forest fire occurrence prediction for real-world early-warning tasks: difficult negative sample construction and spatiotemporal adaptive boundary learning"

This dataset provides supplementary source data supporting the manuscript "Forest fire occurrence prediction for real-world early-warning tasks: difficult negative sample construction and spatiotemporal adaptive boundary learning" submitted to Forest Ecology and Management. The dataset includes source data for Fig. 2, Fig. 3, Fig. 4, and Fig. 5, together with data supporting Table 2, Table 3, Table 4, and Table 5. It contains aggregated performance metrics, ablation results, bootstrap confidence intervals, bootstrap replicate gains, sample-size summaries, and main hyperparameter settings. Full intermediate sample-level feature tables, trained model artifacts, gridded intermediate products, and workflow logs are not included because of file size, third-party data redistribution restrictions, and workflow-specific storage constraints. The original public datasets used in the study are available from the data providers cited in the manuscript.