Skip to content

Open a TROPOMI CH₄ granule and screen it

A Sentinel-5P TROPOMI methane L2 granule (S5P_*_L2__CH4____*.nc) spreads one retrieval over four NetCDF groups. LocalL2Source flattens them into a single CF Dataset, renames the variables to the registry names, and applies the quality, bounding-box and time screening in one call. The file itself has to be on disk already (Earthdata acquisition is a separate step), and reading it needs a NetCDF backend: pip install 'xrtoolz-products[local]'.

One granule, screened

from xrproducts import BBox, LocalL2Source, TimeRange

src = LocalL2Source()
ds = src.open(
    "tropomi.ch4",
    path="S5P_RPRO_L2__CH4____20190601T104423_20190601T122553_08532_03_020400.nc",
    bbox=BBox(lon_min=4.0, lon_max=6.0, lat_min=51.0, lat_max=53.0),
    time=TimeRange.parse("2019-06-01T11:00", "2019-06-01T12:00"),
    qa_min=0.5,   # PUM recommendation; None keeps the opener default (0.5), 0.0 keeps all
)
ds

<xarray.Dataset>
Dimensions:                  (time: 1, scanline: 212, ground_pixel: 31, layer: 12, ...)
Coordinates:
    time                     (time) datetime64[ns] 2019-06-01
    scanline_time            (time, scanline) datetime64[ns] ...
    lon                      (time, scanline, ground_pixel) float32 ...
    lat                      (time, scanline, ground_pixel) float32 ...
Data variables:
    xch4                     (time, scanline, ground_pixel) float32 ...
    xch4_bias_corrected      (time, scanline, ground_pixel) float32 ...
    xch4_precision           (time, scanline, ground_pixel) float32 ...
    qa_value                 (time, scanline, ground_pixel) float32 ...
    column_averaging_kernel  (time, scanline, ground_pixel, layer) float32 ...
    ch4_profile_apriori      (time, scanline, ground_pixel, layer) float32 ...
    dry_air_subcolumns       (time, scanline, ground_pixel, layer) float32 ...
    sp                       (time, scanline, ground_pixel) float32 ...
    pressure_interval        (time, scanline, ground_pixel) float32 ...
    ...

What happened:

  • PRODUCT, DETAILED_RESULTS, INPUT_DATA and GEOLOCATIONS were merged on the shared (time, scanline, ground_pixel) swath; the averaging kernel and a-priori profile keep their layer axis.
  • latitude / longitude became the 2-D lat / lon coordinates, so bbox was applied with subset_bbox (where(..., drop=True) — scanlines and ground pixels entirely outside the box are dropped).
  • time is the granule reference time; the per-scanline delta_time became scanline_time, which the time= window masks on.
  • Pixels with qa_value < 0.5 are NaN in every data variable (shape kept).

Only the opener

Skip the screening and keep the whole swath — or the raw product names — with the opener directly:

from xrproducts import open_tropomi_ch4_l2

raw = open_tropomi_ch4_l2(path, qa_min=None, keep_groups=True)   # product names
ds = open_tropomi_ch4_l2(path)                                    # canonical, qa >= 0.5

Several granules

paths=[...] concatenates granules along time in the order given (they must share the swath shape):

ds = src.open("tropomi.ch4", paths=[granule_a, granule_b], bbox=box)

Pressure per layer

The averaging kernel is applied on the product's pressure grid, which is not stored per layer. Rebuild it from sp and pressure_interval:

p_layer = ds["sp"] - ds["layer"] * ds["pressure_interval"]   # Pa, (time, scanline, ground_pixel, layer)

The same call for EMIT and GHGSat scenes

emit = src.open("emit.ch4", path="EMIT_L2B_CH4ENH_...nc", glt_path="EMIT_..._GLT.nc", bbox=box)
scenes = src.open("emit.ch4", paths=[scene_a, scene_b], glt_paths=[glt_a, glt_b])
plume = src.open("ghgsat.ch4", path="GHGSat_plume.nc")

Both come back on (time, y, x) with 2-D lon / lat, so subset_bbox and the xrops.geom operators run on them unchanged. See Local L2 products for the full API.