methaneset-bank/README.md

# MethaneSET bank

A bank of synthetic methane plumes for injection into real hyperspectral scenes. Each sample is a
GeoTIFF with the XCH4 enhancement (ppb) of one plume emitting 3000 kg/h, as seen from above with the
sun at a given position. 279 physical plumes × 855 sun positions = 238 545 samples.

## Source simulations

The plumes come from the WRF-LES 4.7.0 simulations of Gorroño et al. (2026), *Atmos. Meas. Tech.*
19, 1245-1257, with data on Zenodo (doi:10.5281/zenodo.18161182). The inner domain has 120 × 90
cells of 20 m and 49 vertical levels of 20 m, with one output every 30 s. There are two simulation
types, `multi` (point sources p1 to p9) and `area` (area sources a1 to a9, from smallest to
largest), each run at 11 wind speeds from 2 to 10 m/s.

The 5-cell relaxation band is removed on every side, which leaves a domain of 110 × 80 cells
(2200 × 1600 m).

## Snapshot selection

The wind always blows towards +x: the **upwind** edge is where it enters the domain and the
**downwind** edge is where it leaves. A snapshot is included only if its vertical column map meets
all of these conditions:

1. **The plume fades before the downwind edge.** The logarithm of the maximum per column decreases
   over the last 30 columns (600 m), and the value at the downwind edge is below the lowest value in
   the middle of the domain.
2. **The plume closes before it leaves.** Its crosswind extent above 50 ppb, the noise of a typical
   EMIT scene, is at most 60 m at the downwind edge (`edge:downwind_width_m`). One EMIT pixel is
   60 m, so whatever leaves the domain is narrower than a pixel of the sensor and the cut cannot be
   seen once the plume is injected.
3. **The upwind edge is clean.** Its maximum is below 2% of the peak. This excludes tracer entering
   from the periodic outer WRF domain, where the same sources also emit.
4. **The top and bottom edges are clean.** Their maxima are below 2% of the peak, for the same
   reason.
5. **Almost no methane from outside.** The plume itself cannot reach the strip upwind of the source
   (from the upwind edge to 100 m before the source), so any tracer there comes from the outer WRF
   domain. Its mass per column, extended over the 110 columns of the domain, must be below 0.2% of
   the mass of the map (`qa:foreign_mass_frac`). Since the IME method gives Q proportional to the
   IME, this keeps the bias in Q below 0.2%, twice the budget accepted for the zeros. The estimate
   assumes the inflow is even along the domain, which it is not exactly: it comes in gusts.

Up to three snapshots are kept per run: those closest to minutes 37.5, 45 and 52.5, at least
8 minutes apart. No snapshot of the three largest areas (a7 to a9) meets the conditions.

## Geometry

- The frame is local, not geographic: x is the wind direction and y is across it. There is no CRS.
  Images are stored with +y up and the wind towards +x, so the plume runs to the right. Coordinates
  are in metres from the source, which is at (0, 0). The four edges of the domain are named after
  this frame: **upwind** (where the wind enters) and **downwind** (where it leaves) along x, **top**
  (+y) and **bottom** (−y) across it. In the stored image the wind runs to the right, so **upwind**
  is its left side and **downwind** its right side. They only become north, south, east and west
  once the plume is rotated to the wind direction of the scene it is injected into.
- The sensor looks straight down (VZA = 0), so the ground-to-sensor path does not shift the plume.
  Along the sun-to-ground path, each level at height z is shifted by z·tan(SZA), away from the sun.
- The two paths are combined as (A_s·sun + A_v·view) / (A_s + A_v), with A_s = 1/cos(SZA) and
  A_v = 1/cos(8.4°), where 8.4° is the median EMIT viewing zenith angle. Using 0° instead would
  change the values by about 1%.
- The sun positions cover 17 values of SZA, from 0 to 71.0°, chosen so that a layer at 330 m
  shifts by 60 m from one value to the next. At each SZA, the azimuths are spaced so that neighbouring
  positions are also 60 m apart.
- RAA is the azimuth of the **sun** relative to the wind direction, measured counterclockwise from
  +x. The solar-path image of the plume is displaced towards RAA + 180°.
- The emitting source is a rectangle centred on (0, 0), spanning x from −W/2 to +W/2 and y from
  −H/2 to +H/2, with W = `methane:source_width_m` and H = `methane:source_height_m`. Point sources
  occupy a single 20 × 20 m cell (the source pixel). Area sources range from 60 × 80 m (a2) to
  420 × 440 m (a6); they are always one cell taller than wide, so their centre falls on a pixel
  edge in y (`array:source_row` ends in .5).

## Choosing the sun position for a scene

Use the row whose `sun:sza` and `sun:raa` are closest to those of the scene. The error of this
choice grows with the height of the plume. Neighbouring sun positions, in both zenith and azimuth,
shift the plume by 60 m × h / 330 m relative to each other, where h is `methane:mean_height_m`. The
nearest position is therefore off by at most half the diagonal, about 42 m × h / 330 m: about 4 m
for a plume close to the ground (30 m), and about 42 m (2 pixels) for the highest plumes (330 m).

## Choosing the emission rate

Every value in the bank is the enhancement of a plume emitting 3000 kg/h. To use it at another
emission Q, multiply the image by Q / 3000: at 1500 kg/h every value is halved, at 6000 kg/h every
value is doubled.

The plume leaves the LES domain at its downwind edge, so it ends in a straight cut there. How much
methane is left at that cut is `edge:downwind_ppb`, again at 3000 kg/h. The cut is invisible while it
stays below the noise of the scene you inject into. So, for a scene with noise r (ppb), a plume can
be used up to

    Q = 3000 × r / edge:downwind_ppb        [kg/h]

For example, a plume with 40 ppb at its downwind edge can be injected up to 3700 kg/h in a scene with
50 ppb of noise, but only up to 2000 kg/h in a quieter scene with 27 ppb. Select plumes with

```python
r = 50                                    # noise of your scene, ppb
q = 3000                                  # emission you want, kg/h
ok = t[3000 * r / t["edge:downwind_ppb"] >= q]
```

In this bank the median of that limit is about 2700 kg/h for r = 50 ppb.

## Zeros

Values below a threshold are set to 0. The threshold is the highest value that removes at most
0.1% of the plume mass, capped at 1e-4 of the maximum of the image. The cap applies in 84% of the images. Each image is then cropped to the bounding box of its non-zero values.


## Rescaling

Values scale linearly with the emission rate: for a plume emitting Q kg/h, multiply the image and
the `_ppb` columns by Q / 3000. The `_frac` columns do not change.

## Known limitations

- The plume is cut at the downwind edge of the LES domain (`edge:downwind_frac`: median 4%, up to 51% of
  the peak). With a low sun, this edge appears as a straight line inside the image.
  See "Choosing the emission rate" below to pick plumes whose cut stays hidden in the noise.
- The images contain the enhancement only, with no noise and no background.


## Dataset Information

**Version**: 1.0.0

**License**: CC-BY-4.0

**Keywords**: methane, plume, LES, WRF, XCH4, hyperspectral, EMIT, synthetic

**Tasks**: other

## Dataset Overview

**Partitions**: 2 files
**Spatial coverage**: [-180.00, -90.00, 180.00, 90.00] (WGS84)

## Dataset Structure (Root-Sibling Uniform Tree)

**Root**: FILE (238,545 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). |
| `methane:plume_uid` | `string` | Physical plume: emitter, wind speed and snapshot (e.g. a1_w02.0_s078). Shared by the 855 rows (sun positions) of the plume |
| `methane:sim_type` | `string` | LES simulation type: 'area' (area sources) or 'multi' (point sources) |
| `methane:emitter` | `string` | Emitter: a1 to a9 are area sources, from smallest to largest; p1 to p9 are point sources |
| `methane:wind_speed` | `double` | Wind speed of the LES run [m/s] |
| `methane:snapshot_index` | `int64` | LES output index (one output every 30 s) |
| `methane:snapshot_minute` | `double` | Simulation minute of the snapshot (index / 2) |
| `methane:scale_factor` | `double` | Factor that converts the arbitrary WRF tracer to an emission of 3000 kg/h |
| `methane:peak_ppb` | `double` | Maximum of the plume with the sun overhead (SZA = 0) [ppb]. The same for all rows of a plume; the reference for every _frac column |
| `methane:mean_height_m` | `double` | Mass-weighted mean height of the methane [m]. The solar-path image of the plume shifts by about mean_height * tan(SZA); the nearest sun position in the bank is off by at most ~42 m * mean_height / 330 m |
| `methane:width_max_m` | `double` | Widest crosswind extent of the plume above 50 ppb on the vertical column map [m], at 3000 kg/h |
| `methane:source_width_m` | `double` | Size of the emitting rectangle along x, the wind direction [m]; 20 for point sources. The source at (0, 0) is its centre |
| `methane:source_height_m` | `double` | Size of the emitting rectangle along y [m]; 20 for point sources |
| `sun:sza` | `double` | Solar zenith angle [deg] |
| `sun:raa` | `double` | Azimuth of the sun relative to the wind direction [deg], counterclockwise from +x. The solar-path image of the plume is displaced towards raa + 180 |
| `edge:upwind_ppb` | `double` | Maximum along the upwind edge (column 5), where the wind enters [ppb]. Below 2% of the peak in every sample |
| `edge:upwind_frac` | `double` | edge:upwind_ppb / methane:peak_ppb. It does not change when the plume is rescaled to another emission rate |
| `edge:downwind_ppb` | `double` | Maximum along the downwind edge (column 114), where the wind leaves [ppb]. The plume is cut there; below edge:downwind_ref_ppb in every sample |
| `edge:downwind_frac` | `double` | edge:downwind_ppb / methane:peak_ppb. It does not change when the plume is rescaled to another emission rate |
| `edge:downwind_slope` | `double` | Slope of log(maximum per column) over the last 600 m before the downwind edge; a negative value means that the plume fades towards it. Negative in every sample |
| `edge:downwind_ref_ppb` | `double` | Reference for the downwind edge: the lowest value of the maximum per column in the middle of the domain, columns 40 to 84 [ppb]. Every sample has edge:downwind_ppb < edge:downwind_ref_ppb |
| `edge:downwind_ref_frac` | `double` | edge:downwind_ref_ppb / methane:peak_ppb. It does not change when the plume is rescaled to another emission rate |
| `edge:downwind_width_m` | `double` | Crosswind extent of the plume above 50 ppb at the downwind edge [m]. At most 60 m, one EMIT pixel, in every sample: the plume closes before leaving the domain |
| `edge:top_ppb` | `double` | Maximum along the top edge (row 84, columns 5 to 114, the +y side) [ppb]. Below 2% of the peak in every sample |
| `edge:top_frac` | `double` | edge:top_ppb / methane:peak_ppb. It does not change when the plume is rescaled to another emission rate |
| `edge:bottom_ppb` | `double` | Maximum along the bottom edge (row 5, columns 5 to 114, the -y side) [ppb]. Below 2% of the peak in every sample |
| `edge:bottom_frac` | `double` | edge:bottom_ppb / methane:peak_ppb. It does not change when the plume is rescaled to another emission rate |
| `array:width` | `int64` | Image width [pixels of 20 m] |
| `array:height` | `int64` | Image height [pixels of 20 m] |
| `array:source_col` | `double` | Column of the source pixel in the image (0 = upwind side) |
| `array:source_row` | `double` | Row of the source pixel in the image (0 = top, +y) |
| `array:peak_ppb` | `double` | Maximum of this image [ppb]. It changes with the sun and equals methane:peak_ppb when sza = 0 |
| `array:grow_upwind_px` | `int64` | Number of pixels by which the image extends beyond (+) or falls short of (-) the 110 x 80 LES domain on the upwind, where the wind enters side |
| `array:grow_downwind_px` | `int64` | Number of pixels by which the image extends beyond (+) or falls short of (-) the 110 x 80 LES domain on the downwind, where it leaves side |
| `array:grow_top_px` | `int64` | Number of pixels by which the image extends beyond (+) or falls short of (-) the 110 x 80 LES domain on the top (+y) side |
| `array:grow_bottom_px` | `int64` | Number of pixels by which the image extends beyond (+) or falls short of (-) the 110 x 80 LES domain on the bottom (-y) side |
| `qa:threshold_ppb` | `double` | Zero threshold of this image [ppb]: lower values were set to 0. It is the highest threshold that removes at most 0.1% of the mass, capped at 1e-4 of the image maximum |
| `qa:threshold_rel` | `double` | qa:threshold_ppb / array:peak_ppb; equal to 1e-4 when the cap applies |
| `qa:ime_before` | `double` | Sum of the image before zeroing [ppb x pixel] |
| `qa:ime_after` | `double` | Sum of the image after zeroing [ppb x pixel]; ime_after / ime_before is the fraction of mass kept |
| `qa:foreign_mass_frac` | `double` | Estimated fraction of the mass of the vertical column map that comes from the outer WRF domain: the mass in the strip upwind of the source, where the plume itself cannot reach, extended over the 110 columns of the domain. Below 0.002 in every sample |
| `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). |


## Usage

### Python

```python
# pip install tacoreader
import tacoreader

ds = tacoreader.load("/data/databases/METHANESET_TACOS/methaneset-bank/.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-bank/.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-bank/.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*

**LARS-UPV, Gorroño et al. (WRF-LES simulations)** — *licensor*




---

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