Dataset for Deep Learning-Based Separation and Quantification of Overlapping Methane Plumes from Satellite Observations

<p dir="ltr">This dataset supports training a spatial-attention U-Net 3+ model, evaluating overlapping methane plume separation, and estimating individual source emission rates using integrated mass enhancement (IME).</p><p dir="ltr">The overlapping_plumes(2-8)-V2 directory includes 28,000 training, 2,800 validation and 700 test samples. The overlapping_plumes(2-8)-V2_Add_S10 directory contains the same 700-sample test subset with additional wind metadata.

Each split includes paired combined_plumes and labels NPZ files and images_png previews.</p><p dir="ltr">The archive contains 96,601 files. No trained checkpoint is included. See DATASET_README.md and dataset_inventory.json for data organization and reproduction instructions.