MethaneSET-EMIT: Hyperspectral Methane Plume Detection from EMIT

Description

methaneset-emit provides analysis-ready EMIT hyperspectral scenes for methane plume detection and quantification. 721 granules (700 with plumes + 21 verified plume-free), selected from the granules where the UNEP-IMEO and CarbonMapper catalogs BOTH report a plume (consensus), with 0% onboard nodata and complete IMEO flux quantification, balanced across flux class (all 812 weak plumes under 1,000 kg/h retained), year, region and sector. Each sample carries 10 sensor-geometry layers with 128x128 internal tiling: the full 285-band L1B radiance cube, per-pixel lat/lon, the GLT orthorectification lookup, Copernicus GLO-30 elevation, an ERA5-Land 10 m wind field (u/v, time-interpolated to the acquisition minute), three matched-filter retrievals (mf, rmf, mag1c), and dual instance-coded plume masks from the IMEO portal and CarbonMapper analysts, where a matched pair shares the same code in both files. A leakage-free train/val/test split grouped by emitter is proposed; all metadata is queryable in level0.parquet without opening a raster.

Dataset Overview

Keywords

methane EMIT hyperspectral plume detection matched filter IMEO CarbonMapper

ML Tasks

segmentation regression

TACO Structure (Root-Sibling Uniform Tree)

Hierarchical structure showing representative samples across levels. The "..." notation indicates additional samples following the same pattern. All samples at the same level share identical structure (RSUT constraint).

Hierarchy Details

Level	Types	Total Samples	Sample IDs (preview)
Level 0	`All FOLDER`	721	Root level samples
Level 1	`FILE + FILE + FILE + FILE + FILE + FILE + FILE + FILE + FILE + FILE`	7,210	`radiance.tif, latlon.tif, glt.tif...`

Metadata Fields by Level

These fields are available for querying with SQL when using TacoReader.

LEVEL0 (65 fields)

Field Name	Type	Description
id	string	Unique sample identifier within parent scope. Must be unique among siblings.
type	string	Sample type discriminator (FILE or FOLDER).
detection:n_imeo	int64	Number of distinct IMEO emitters with a plume in this granule
detection:n_cm	int64	Number of distinct CarbonMapper sources with a plume (upper bound: CM can register one plume under multiple source identities)
detection:coverage_imeo	double	Percent of scene pixels covered by IMEO plume masks
detection:coverage_cm	double	Percent of scene pixels covered by CarbonMapper plume masks
detection:coverage_union	double	Percent of scene pixels covered by the union of both masks
detection:imeo_ids	string	JSON list of IMEO emitter ids (source_name) present in the scene
detection:cm_ids	string	JSON list of CarbonMapper source names present in the scene
detection:imeo_flux	string	JSON dict IMEO emitter -> CH4 flux rate (kg/h); flux completeness enforced at selection
detection:cm_flux	string	JSON dict CM source -> CH4 flux rate (kg/h); null = unquantified (structural in CM's catalog, not size-biased)
detection:sectors	string	JSON dict sector -> {imeo: [...], cm: [...]}; CM IPCC codes mapped to IMEO names (97% cross-agreement on matched pairs)
detection:sector_list	string	JSON list of sectors present in the scene
detection:sector	string	Dominant sector by plume count
detection:n_imeo_weak	int64	IMEO plumes with 0 < flux < 1,000 kg/h (weak, following STARCOP)
detection:n_imeo_strong	int64	IMEO plumes with flux >= 1,000 kg/h (strong)
detection:n_cm_noflux	int64	CarbonMapper sources without quantified emission
detection:n_cm_dup_pairs	int64	CarbonMapper plume pairs from DIFFERENT sources overlapping with IoU >= 0.5 in this scene (duplicate identities). If > 0, treat n_cm and the CM source partition as an upper bound; IMEO is the reference catalog
detection:cm_dup_groups	string	JSON list of GROUPS of CarbonMapper sources that describe the same plume in this scene (connected components of the IoU >= 0.5 overlap graph). A group can hold more than two sources: up to six in this dataset. Use it to check whether the CM source matched to your IMEO plume has twins, in which case its flux and mask area are a lower bound for that plume
detection:imeo_flux_total	double	Sum of IMEO plume fluxes in the scene (kg/h)
detection:imeo_flux_max	double	Largest IMEO plume flux in the scene (kg/h)
detection:cm_flux_total	double	Sum of quantified CarbonMapper fluxes (kg/h)
detection:cm_flux_max	double	Largest quantified CarbonMapper flux (kg/h)
detection:imeo_cod	string	JSON dict IMEO emitter -> instance codes in plume_imeo.tif band 1 (a matched IMEO-CM pair shares the same code in both files)
detection:cm_cod	string	JSON dict CM source -> instance codes in plume_cm.tif band 1
match:pairs	string	JSON list of matched IMEO-CM pairs (Hungarian assignment on geometry alone: overlap + source distance; flux and sector never used, keeping them free for cross-catalog studies)
match:orphans	string	JSON dict of unmatched plumes per catalog
match:n_pairs	int64	Number of accepted IMEO-CM pairs
match:n_perfect	int64	Pairs without same-catalog overlap ambiguity
match:n_orph_imeo	int64	IMEO plumes without a CM counterpart
match:n_orph_cm	int64	CM sources without an IMEO counterpart
sensor:shape_rows	int64	Sensor-grid rows
sensor:shape_cols	int64	Sensor-grid columns
sensor:sza_mean	double	Mean solar zenith angle (deg); intra-scene range is <1 deg
sensor:vza_mean	double	Mean view zenith angle (deg)
sensor:saa_mean	double	Mean solar azimuth angle (deg)
sensor:vaa_mean	double	Mean view azimuth angle (deg)
sensor:amf_mean	double	Two-way air mass factor 1/cos(SZA)+1/cos(VZA)
sensor:phase_mean	double	Mean solar phase angle between sun and view vectors (deg); sunglint indicator
sensor:path_length_mean	double	Mean sensor-to-ground path length (m)
sensor:earth_sun_distance	double	Earth-Sun distance (AU)
meteo:wind_u	double	Scene-mean 10 m eastward wind from the wind.tif layer (ERA5-Land, time-interpolated); per-pixel field in wind.tif
meteo:wind_v	double	Scene-mean 10 m northward wind (m/s), ERA5-Land
meteo:wind_speed	double	Scene-mean 10 m wind speed (m/s), ERA5-Land. NOTE: this is MethaneSET's own reanalysis wind, not the wind either catalog used to invert its fluxes; for that see imeo_wind and cm_wind
meteo:imeo_wind	string	JSON dict IMEO emitter -> {u, v, speed} (m/s): the PER-PLUME 10 m wind published by UNEP-IMEO, i.e. the wind IMEO used to invert its own flux rate. Independent of the ERA5-Land wind.tif layer shipped with the scene
meteo:cm_wind	string	JSON dict CarbonMapper source -> {speed, dir, u, v}: the PER-PLUME wind published by Carbon Mapper (wind_speed_avg_auto and wind_direction_avg_auto), i.e. the wind CM used to invert its own emission. dir follows the meteorological convention (degrees FROM which the wind blows); u and v are derived from speed and dir. Null when CM ships no wind for that plume
emit:flight_line	string	EMIT flight line identifier
emit:time_start	string	Acquisition start (UTC)
emit:time_end	string	Acquisition end (UTC)
radiance:min	double	Minimum radiance over the full 285-band cube (negative values are calibration noise in dark SWIR bands, kept unclipped)
radiance:max	double	Maximum radiance over the full cube
radiance:elev_min_m	double	Minimum elevation (m, Copernicus DEM GLO-30 2024)
radiance:elev_max_m	double	Maximum elevation (m)
site:country	string	Country of the scene (from the IMEO portal)
spatial:bbox_west	double	Bounding box west (deg, EPSG:4326)
spatial:bbox_south	double	Bounding box south (deg)
spatial:bbox_east	double	Bounding box east (deg)
spatial:bbox_north	double	Bounding box north (deg)
spatial:imeo_points	string	WKT MULTIPOINT of IMEO emission source points
spatial:cm_points	string	WKT MULTIPOINT of CarbonMapper source points
selection:split	string	Proposed train/val/test split (70/15/15, grouped by emitter: zero source leakage; users may re-split, everything ships)
selection:is_free	bool	True = verified plume-free granule from IMEO's full-tile test pool
selection:flux_bin	string	has_weak / strong_only / free (weak = any IMEO plume under 1,000 kg/h, following STARCOP)
internal:current_id	int64	Current sample position at this level (0-indexed). Enables O(1) random access and relational JOINs (ZIP, FOLDER, TACOCAT).
internal:parent_id	int64	Foreign key referencing parent sample position in previous level (ZIP, FOLDER, TACOCAT).

LEVEL1 (5 fields)

Field Name	Type	Description
id	string	Unique sample identifier within parent scope. Must be unique among siblings.
type	string	Sample type discriminator (FILE or FOLDER).
internal:current_id	int64	Current sample position at this level (0-indexed). Enables O(1) random access and relational JOINs (ZIP, FOLDER, TACOCAT).
internal:parent_id	int64	Foreign key referencing parent sample position in previous level (ZIP, FOLDER, TACOCAT).
internal:relative_path	string	Relative path from DATA/ directory. Format: {parent_path}/{id} or {id} for level0 (ZIP, FOLDER, TACOCAT).

Loading the Dataset

# pip install tacoreader
import tacoreader

# Load dataset
ds = tacoreader.load("methaneset-emit.tacozip")

# Basic info
print(f"ID: {ds.id}")
print(f"Version: {ds.version}")
print(f"Samples: {len(ds.data)}")

# Coming soon: R support is planned but not yet available
# install.packages("tacoreader")
library(tacoreader)

# Load dataset
ds <- load_taco("methaneset-emit.tacozip")

# Basic info
cat(sprintf("ID: %s\n", ds$id))
cat(sprintf("Version: %s\n", ds$version))
cat(sprintf("Samples: %d\n", nrow(ds$data)))

# Coming soon: Julia support is planned but not yet available
# using Pkg; Pkg.add("TacoReader")
using TacoReader

# Load dataset
ds = load_taco("methaneset-emit.tacozip")

# Basic info
println("ID: ", ds.id)
println("Version: ", ds.version)
println("Samples: ", size(ds.data, 1))

Providers & Curators

Data Providers

NASA JPL producer

UNEP IMEO producer

Carbon Mapper producer

Universitat de Valencia, ISP processor

Dataset Curators

Name	Organization	Email
Julio Contreras	Universitat de Valencia, Image and Signal Processing (ISP)	julio.contreras@uv.es
Cesar Aybar	Universitat de Valencia, Image and Signal Processing (ISP)	cesar.aybar@uv.es

Publications & Citations

How to Cite This Dataset

If you use this dataset in your research, please cite:

10.5067/EMIT/EMITL1BRAD.001

Green, R. O., et al. (2023). EMIT L1B At-Sensor Calibrated Radiance and Geolocation Data 60 m V001. NASA LP DAAC.

10.1038/s41598-023-44918-6

Ruzicka, V., et al. (2023). Semantic segmentation of methane plumes with hyperspectral machine learning models. Scientific Reports.

10.5270/ESA-c5d3d65

European Space Agency (2024). Copernicus DEM GLO-30.

10.24381/cds.e2161bac

Munoz-Sabater, J., et al. (2021). ERA5-Land hourly data. Copernicus Climate Data Store.

BibTeX

@dataset{methaneset-emit1,
  title = {MethaneSET-EMIT: Hyperspectral Methane Plume Detection from EMIT},
  author = {Julio Contreras and Cesar Aybar},
  year = {2022},
  version = {1.0.0},
  publisher = {Universitat de Valencia, Image and Signal Processing (ISP)}
}