Benchmark dataset for GenGHG v1.0: generative emulation of greenhouse-gas transport footprints

This dataset provides the training and validation benchmark for GenGHG v1.0 (https://doi.org/10.5281/zenodo.21017062), a generative machine learning model of greenhouse-gas atmospheric transport. It contains more than 40,000 footprint emulation samples for satellite-style column receptors over 60 global urban emission hotspots from 2019 to 2025. The benchmark is split into training, validation and test subsets, and is designed to support the development and evaluation of GenGHG. Each sample includes meteorological input fields from the NOAA Global Forecast System (GFS) and a reference footprint simulated with the Stochastic Time-Inverted Lagrangian Transport (STILT) model. The provided `parsing_pt_to_nc.py` script can be used to convert the PyTorch format (`.pt`) into NetCDF (`.nc`) files containing both model inputs and target footprints. Each footprint represents 24-hour backward transport over a local 6 latitude × 6 longitude emission domain and quantifies how unit surface emissions would influence a downwind atmospheric measurement.