Facility-scale multi-source methane plume simulation dataset for oil and gas emission source localization and contribution-ratio inversion

This dataset contains 5,000 physically parameterized synthetic multi-source methane plume samples designed for facility-scale oil and gas methane emission source localization and contribution-ratio inversion. Each sample is stored as a NumPy NPZ file and includes methane column concentration enhancement fields, valid observation masks, wind components, true source coordinates, sparse Gaussian heatmap supervision, and normalized source-contribution labels. The dataset covers five active-emission scenarios: separated multi-source plumes, close-merge plumes, clustered sources, long-tailed source intensities, and out-of-frame sources. Stopped-emission/residual-plume samples are excluded. The release also provides metadata, deterministic train/validation/test split files, a data dictionary, checksums, and a minimal Python loading example.