Automatic methane plume detection in TROPOMI data
TROPOMI51 on board the European satellite Sentinel-5 Precursor was launched in 2017. It observes backscattered sunlight in the near- and shortwave infrared around the 0.76 µm O 2 and 2.3 µm methane bands, at approximately 1:30 pm local time. Total columns (vertically integrated concentrations) of methane with near vertically uniform sensitivity down to the surface are retrieved from these observations using a full-physics approach that accounts for the interfering impact of surface reflectance, aerosol and other geophysical variables on the shortwave infrared signal (v.2.6.0)34. TROPOMI is a methane flux mapper that offers daily global coverage with a 7 × 5.5 km2 spatial resolution at nadir. In addition to being used in long-term inverse analyses, its imaging capabilities enable the detection of anthropogenic methane emission plumes that arise from the world’s largest emitters29. We employ a two-step machine learning approach to explore TROPOMI data for methane emission plumes automatically4. We analyse and manually verify all plumes detected in 2021 and 2022 with estimated sources within 50 km from any of the landfills targeted by GHGSat. We apply the Integrated Mass Enhancement (IME) method20, calibrated specifically for TROPOMI using atmospheric transport simulations, to quantify the methane emission rate and its uncertainty for each TROPOMI-detected plume4.
Given TROPOMI’s spatial resolution (7 × 5.5 km2) compared with GHGSat’s (25 × 25 m2), we cluster the 151 landfills observed by GHGSat into 130 TROPOMI-relevant urban areas. For each urban area, we first apply a 2σ filter to remove outlier estimates that can be hampered by an unrepresentative plume mask due to variable meteorology or surface effects (for example, plume masks truncated by clouds for lower estimates). Then, relying on the remaining TROPOMI plume detections, we report their mean detected urban-scale methane emissions and their standard deviation. These averages only cover emissions detected as strong plumes and are not representative of mean urban emissions but do provide an indication of urban mitigation potential. Not detecting a plume does not imply that there are no emissions: it means that concentrated emissions are lower than the ~8 t h–1 TROPOMI plume detection threshold, or that the observational or geographical conditions did not allow for TROPOMI detection4. The discrepancy with mean emissions is verified for four different (above-average emitting and often detected) cities (Buenos Aires, Delhi, Mumbai and Lahore) where IME-based rates show a 7–47% overestimation (while agreeing within uncertainties) compared with urban-level methane emission estimates based on atmospheric inversions and TROPOMI data5 (Supplementary Note 1).
GHGSat observations and emission quantification
GHGSat-C1 to -C5 instruments were launched between 2020 and 2022. Satellites C1-2-3 perform measurements in the morning around 10 am (local time), while satellites C4-5 perform measurements in the afternoon around 2 pm (local time). These instruments estimate the total column (vertically integrated content) of methane at ~25 × 25 m2 resolution over targeted 12 ×15 km2 domains52 from backscattered sunlight measurements in the shortwave infrared near 1.65 µm, that provide near-surface sensitivity. The GHGSat instruments have an empirically measured methane column precision range of 1.4–2.9%53, which allows them to observe emission plumes from point (for example, a gas pipeline leak) or very localized sources (for example, active faces of landfills) emitting more than ~100 kg h–1 (this detection threshold increases with wind speed)28. Pixels exhibiting local spatially correlated methane column enhancements above background are clustered together and considered to belong to a plume26. We apply the IME method20 to estimate an emission rate Q based on a delineated plume and the local wind speed sampled from a meteorological model. We have:
$$Q=\frac{{U}_{\mathrm{eff}}}{L}\sum _{i}{\rm{\Delta }}{{X}_{\mathrm{CH}}}_{4,i}{a}_{i}$$
where U eff is the effective wind speed, calibrated against the 10-m wind speed based on a set of large Eddy simulations (LES)5; \(L=\sqrt{\sum _{i}{a}_{i}}\) is the plume length computed as the square-root of the plume total area, where a i is the area of the ith pixel included in the plume; and \({\rm{\Delta }}{{X}_{\mathrm{CH}}}_{4,i}\) is the local enhancement above the background of the methane total column for this ith pixel. Here we use an effective wind speed calibration specific to landfills, based on LES of area sources: \({U}_{{\rm{e}}{\rm{f}}{\rm{f}}}=0.34\times {U}_{10\,{\rm{m}}}+0.66\) (ref. 5), where \({U}_{10\,{\rm{m}}}\) is the 10-m wind speed sampled from the GEOS-FP meteorological reanalysis54. The emission rate uncertainty calculation includes contributions from (1) wind speed error; (2) methane column retrieval error; and (3) IME calibration error26.
The calibration of this mass-balance approach against LES of known synthetic emission rates ensures that the estimated rates correctly account for the different advective transport conditions explored within the set of LES. Beyond this calibration on simulations, numerous real-life validation efforts have been organized, including controlled-release experiments, which are the validation gold standard. Notably, GHGSat participated and showed excellent agreement with metered emission rates in internal controlled releases, as well as in two single-blind controlled-release campaigns, where the true emission rates (and wind speeds) are not known to the satellite data providers and the comparisons are done by a third party (in this case a research group from Stanford University)25,55. Beyond controlled releases, landfill emission rates obtained through aerial methane imagery with an instrument that can detect plumes down to 10 kg h–1 have been validated against traditional aerial mass-balance results18,39. Besides, an in-depth study of two landfills near Madrid that included both similar airborne observations and GHGSat satellite observations showed that GHGSat satellite-based estimates match the total of airborne-detected plumes for same day observations within uncertainties56. Combined, these results show that GHGSat satellite-based observations can provide accurate estimates of methane emissions from waste disposal sites.
Estimating site-level GHGSat averages
Three outcomes are possible for any individual waste disposal site observation during a single overpass by GHGSat: (1) no plume is detected; (2) only one plume is detected; and (3) several plumes (arising from the same site) are detected. In the first case, we conservatively consider the emission rate to be equal to zero, with no uncertainty. In the second and third cases, we apply the IME method to each plume separately to quantify its emission rate and uncertainty. In the third case, we sum together all of the detected plume emission rates (and sum their respective uncertainties quadratically) to obtain an emission rate for the whole site.
... continue reading