methaneset-s2-finetune/README.md

# methaneset-s2-finetune


Landsat 8/9 plume scenes for supervised methane detection, each with the original target scene
(MARS-S2L) and four clean backgrounds of the same site.

## Samples

Every sample is one plume id, a FOLDER with eight leaves:

| leaf | what it is |
|---|---|
| `target` | the Landsat scene with the plume (from MARS-S2L) |
| `bg0` | the background that MARS itself selected for that target (the original `reference`) |
| `bg1`, `bg2`, `bg3` | three extra clean backgrounds, curated here |
| `ch4` | the MBMP enhancement of the target (from MARS) |
| `plume` | binary plume mask of the target |
| `dem` | Copernicus GLO-30 elevation |

## Backgrounds

`bg0` is the single background that comes with the MARS-S2L pair. `bg1` to `bg3` are three extra
backgrounds that we curate for every target, with the same MARS method: query Landsat 8 and 9
together within ±120 days (before or after), filter by cloud and constellation, discard the target's
own pass, and keep the three most similar by the MARS metric on methane-insensitive bands (B2, B3,
B4, B6). When ±120 days does not yield 40 candidates, the pool is filled with the same day-of-year
±45 days up to two years back. All four backgrounds are clean of methane.

## Columns

All columns describe the target unless prefixed `bg:`, which describe the four backgrounds
(`bg:sensor0` to `bg:sensor3`, and so on).

Some columns are constant in this dataset and are kept for compatibility: `detection:offshore` is
always False (onshore plumes only) and `quality:observability` is `clear` for all but three samples.
`detection:isplume` was always True and is not shipped as a column.

## Dataset Information

**Version**: 1.0.0

**License**: CC-BY-4.0

**Keywords**: methane, landsat, background, plume, taco

**Tasks**: detection

## Dataset Overview

**Partitions**: 20 files
**Spatial coverage**: [-180.00, -90.00, 180.00, 90.00] (WGS84)

## Dataset Structure (Root-Sibling Uniform Tree)

**Root**: FOLDER (3,603 samples)

**Hierarchy**:

- Level 1: FILE → FILE → FILE → FILE → FILE → FILE → FILE → FILE (28,824 samples)

## Metadata Fields

### LEVEL0

| Field | Type | Description |
|-------|------|-------------|
| `id` | `string` | Unique sample identifier within parent scope. Must be unique among siblings. |
| `type` | `string` | Sample type discriminator (FILE or FOLDER). |
| `detection:case_study` | `string` | Case study region (e.g. Turkmenistan, Algeria) |
| `detection:ch4_fluxrate` | `double` | Emission rate of the plume estimated by MARS [kg/h] |
| `detection:ch4_fluxrate_std` | `double` | Uncertainty of detection:ch4_fluxrate [kg/h] |
| `detection:isplume` | `bool` | detection:isplume |
| `detection:offshore` | `bool` | Whether the source is offshore. Always False in this dataset |
| `detection:sector` | `string` | Economic sector of the source (Oil and Gas, Coal) |
| `detection:wind_source` | `string` | Reanalysis used for the wind: ERA5-Land onshore, GEOS-FP offshore |
| `geoenrich:admin_countries` | `string` | Administrative country (may repeat site:country) |
| `geoenrich:admin_districts` | `string` | Administrative district |
| `geoenrich:admin_states` | `string` | Administrative state or province |
| `geoenrich:elevation` | `double` | Elevation of the emission point, Copernicus GLO-30 [m] |
| `geoenrich:population` | `double` | Mean population density around the emission point (HRSL) |
| `geoenrich:temperature` | `double` | Mean climate temperature at the site [K] |
| `majortom:code` | `string` | Internal identifier of the MARS scene |
| `meteo:wind_u` | `double` | Eastward wind component at the emission point [m/s] |
| `meteo:wind_v` | `double` | Northward wind component at the emission point [m/s] |
| `plume:geometry` | `string` | Plume polygon of the target in WKT/WKB, EPSG:4326 |
| `quality:last_update` | `string` | Last update of the MARS record |
| `quality:notified` | `bool` | Whether the emitter was notified upstream |
| `quality:observability` | `string` | MARS observability tag: clear or bad_retrieval |
| `quality:percentage_clear` | `double` | MARS cloud-clear fraction of the target scene [%] |
| `target:sensor` | `string` | Satellite of the target scene: LC08 (Landsat 8) or LC09 (Landsat 9) |
| `target:sza` | `double` | Solar zenith angle of the target [deg] |
| `target:tile` | `string` | Landsat granule id of the target scene |
| `target:vza` | `double` | Viewing zenith angle of the target [deg] |
| `site:country` | `string` | Country of the emission source |
| `site:location_name` | `string` | Site name of the emission source |
| `split` | `string` | train / validation / test split inherited from MARS-S2L, grouped by emitter |
| `stac:crs` | `string` | CRS of the chip (local UTM) |
| `stac:geotransform` | `string` | GeoTransform of the chip: 10 m pixels, 200 x 200 |
| `stac:tensor_shape` | `string` | Shape of the chip: 11 bands, 200 rows, 200 columns |
| `stac:time_start` | `string` | Acquisition time of the target scene |
| `bg:date0` | `string` | Acquisition date of bg0 |
| `bg:sensor0` | `string` | Satellite of bg0 (the MARS reference): LC08 or LC09 |
| `bg:tile0` | `string` | Landsat granule id of bg0 (the MARS reference) |
| `bg:sensor1` | `string` | Satellite of bg1: LC08 or LC09 |
| `bg:date1` | `string` | Acquisition date of bg1 |
| `bg:cloud1` | `double` | Cloud and shadow fraction of bg1 (OmniCloudMask) |
| `bg:difference1` | `double` | MARS similarity of bg1 to the target (lower is more similar) |
| `bg:methane1` | `double` | Fraction of bg1 pixels with a methane enhancement of its own |
| `bg:sensor2` | `string` | Satellite of bg2: LC08 or LC09 |
| `bg:date2` | `string` | Acquisition date of bg2 |
| `bg:cloud2` | `double` | Cloud and shadow fraction of bg2 (OmniCloudMask) |
| `bg:difference2` | `double` | MARS similarity of bg2 to the target (lower is more similar) |
| `bg:methane2` | `double` | Fraction of bg2 pixels with a methane enhancement of its own |
| `bg:sensor3` | `string` | Satellite of bg3: LC08 or LC09 |
| `bg:date3` | `string` | Acquisition date of bg3 |
| `bg:cloud3` | `double` | Cloud and shadow fraction of bg3 (OmniCloudMask) |
| `bg:difference3` | `double` | MARS similarity of bg3 to the target (lower is more similar) |
| `bg:methane3` | `double` | Fraction of bg3 pixels with a methane enhancement of its own |
| `emission:lat` | `double` | Latitude of the emission point [deg] |
| `emission:lon` | `double` | Longitude of the emission point [deg] |
| `target:cloud` | `double` | Cloud and shadow fraction of the target chip (OmniCloudMask) |
| `bg:cloud0` | `double` | Cloud and shadow fraction of bg0 (OmniCloudMask) |
| `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

| Field | Type | Description |
|-------|------|-------------|
| `id` | `string` | Unique sample identifier within parent scope. Must be unique among siblings. |
| `type` | `string` | Sample type discriminator (FILE or FOLDER). |
| `geotiff:stats` | `list<item: list<item: float>>` | Per-band statistics (List[List[Float32]]): categorical mode returns class probabilities, continuous mode returns [min, max, mean, std, valid%, p25, p50, p75, p95] |
| `taco:header` | `binary` | Binary TACOTIFF header (35 bytes + tile counts) for fast reading without IFD parsing |
| `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). |


## Usage

### Python

```python
# pip install tacoreader
import tacoreader

ds = tacoreader.load("/data/databases/METHANESET_TACOS/methaneset-s2-finetune/.tacocat")
print(f"ID: {ds.id}")
print(f"Version: {ds.version}")
print(f"Samples: {len(ds.data)}")
```

### R

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

ds <- load_taco("/data/databases/METHANESET_TACOS/methaneset-s2-finetune/.tacocat")
cat(sprintf("ID: %s\n", ds$id))
cat(sprintf("Version: %s\n", ds$version))
cat(sprintf("Samples: %d\n", nrow(ds$data)))
```

### Julia

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

ds = load_taco("/data/databases/METHANESET_TACOS/methaneset-s2-finetune/.tacocat")
println("ID: ", ds.id)
println("Version: ", ds.version)
println("Samples: ", size(ds.data, 1))
```

## Data Providers

**Image and Signal Processing Group (ISP-UV)** — *producer*

**UNEP IMEO, MARS (background pairs)** — *licensor*




---

Generated with ❤️ using [TacoToolbox](https://github.com/tacotoolbox/tacotoolbox) v0.26.9