JPL trace gases dataset: Fully Automatic Trace Gas Plume Detection - v1

Description:

This is the official dataset upload for the paper "Fully Automatic Trace Gas Plume Detection". Shortly, it contains the collection of point source events of methane (CH4), ammonia (NH3), nitrogen dioxide (NO2) and carbon monoxide (CO) in data from NASA's EMIT imaging spectroscopy instrument.

Please refer to the official codebase for more instructions and updates: https://github.com/emit-sds/daily-trace-gases/

Folder structure:

Contains detected point source events in zip files with "detections" in name, confounder events with "confounders" and randomly sampled scenes without point source events with "negatives" (these match the number of scenes with events).

General file structure:

cmf.tif - matched filter enhancement product (primary data used by our models)

rgb.tif - RGB bands extracted from the radiance data (useful to check for on the ground structures and clouds, potentially also for future machine learning models)

prediction.geojson - scored predictions made by our model ensemble

Additionally for detections:

label.shp - manually labelled ground truth

Additionally for NH3 point source detections:

ch4-cmf.tif - matched filter enhancement product computed for methane target (to allow checking if the NH3 event overlaps with CH4)

Finally, we provide all detections in csv files (with the names of EMIT scenes and the geometries) for easier access not requiring the download of the larger tif files.

Statistics:

CO: In total 51 distinct detections in 40 scenes (in CO_detections.zip), 40 no event scenes (CO_negatives.zip) and 20 confounder scenes (CO_confounders.zip)NH3: In total 70 distinct detections in 62 scenes (in NH3_detections.zip) + 8 distinct detections in 8 scenes of NH3 overlapping with CH4 (NH3_with_CH4_detections.zip), 70 no event scenes (NH3_negatives.zip) and 20 confounder scenes (NH3_confounders.zip)NO2: In total 219 distinct detections in 153 scenes (in NO2_detections.zip), 153 no event scenes (NO2_negatives.zip) and 20 confounder scenes (NO2_confounders.zip)

For fast inspection of the data, we recommend using tools such as the free software QGIS, or the rasterio library in Python.