Simulated WRF-Chem Plume Output Resampled to TROPOMI Pixel Footprints

This record contains two synthetic sets of methane plume observations, simulated with WRF-Chem and resampled to TROPOMI pixel footprints. These datasets comprise the training and validation datasets developed for the training of ML-SPERE (the Machine Learning-Superemitting Plume Emission Rate Estimate), a convolutional neural network framework for estimating emission rates from TROPOspheric Monitoring Instrument (TROPOMI) methane plumes. These datasets consist 104,988 TROPOMI scenes with methane plumes (94,159 in the training dataset and 10,829 in the validation dataset) simulated using the WRF-Chem atmospheric transport model, using NCEP meteorological data. Plumes were simulated at 7 regional domains globally for 3 months in 2019, with simulation output sampled once daily during that time period. We augmented the total number of scenes using spatial transformations, and simulation output was scaled to emission rates sampled from the distribution of observed methane super-emissions reported in Schuit et. al. (2023). Please see the corresponding associated publication (listed below) for a complete description of processing steps taken in transforming raw WRF-Chem simulation output into the resampled netCDF4 files presented here. To replicate the environment used to run the demo.ipynb notebook, use pip to create a virtual env with the python version specified in python-version.txt, and install the packages specified in requirements.txt.