Data

TROPOMI

We do not know which specific version of the TROPOMI retrieval processor was used in the analysis by Sadavarte, et al., 2021 . As several of the authors are affiliated with KNMI, they may have had access to multiple internal or pre-release data products, including versions not publicly distributed at the time. For example, the first orbit available in the publicly released TROPOMI li${\mathrm{CH}}_4$ dataset is orbit 2818 from 30 April 2018, whereas Figure S2 in the Supporting Information provided by the authors includes data beginning on 27 April 2018. This suggests that at least some of the data used in the original analysis may have originated from internal or early access products. As noted in the Sadavarte, et al., 2021 Supporting Information :

To further test the robustness of our emission estimates, sensitivity tests have been performed varying the number of valid transects, calculating the enhancements by subtracting the median of the study domain instead of the upwind background mean, applying different quality flags and using the operational instead of the scientific TROPOMI XCH4 data product.

The data processing description in Sadavarte, et al., 2021 indicates that multiple filters and pre-processing steps were applied to the TROPOMI observations. These included not only the selection of orbits based on specific quality criteria, but also direct modifications to the underlying data products. The applied adjustments comprised corrections for pixel height, de-striping of imagery. It is important to note that these steps are not intrinsic components of the CSF algorithm itself; rather, they were introduced to mitigate known artefacts and limitations in the TROPOMI retrievals.

Filtering based on correlations between methane mixing ratios and surface albedo or aerosol optical thickness originates from validation studies comparing TROPOMI and GOSAT observations over the 2018-2019 period Lorente, et al., 2021 . These correlation artefacts are understood to arise from retrieval sensitivities rather than true atmospheric variability and are largely addressed in later versions of the TROPOMI processor, notably version 2.4.0. The TROPOMI li${\mathrm{CH}}_4$ product includes two separate variables for both surface albedo and aerosol optical thickness: one associated with the ( Short Wave Infrared Radiation (SWIR) 2305-2385 ($\mathrm{nm}$)) spectral band and another with the ( Near Infrared Radiation (NIR) ; 757-774 ($\mathrm{nm}$)) band. Definitions and details of these variables are provided in Table 1 of L2__CH4___ ATBD Hasekamp, et. all ATBD .

It is unclear whether the correlation filtering is computed using all valid pixels within an entire orbit or only those pixels contained within the analysis domain. As a result, multiple interpretations of the filtering criterion are possible, yielding up to eight distinct correlation coefficients depending on whether the calculation is performed globally or domain-restricted and which albedo or aerosol product is used. This ambiguity can materially affect orbit selection. For example, orbit 03912 fails the correlation threshold when calculated over the analysis domain but passes when evaluated over the full orbit (see Table 7 ), whereas orbit 02990 exhibits the opposite behaviour (see Table 8).

We evaluated the correlation based filters using both domain restricted pixels and full orbit pixel sets. For the remainder of this report, we adopt correlation values calculated over the full orbit and apply only the SWIR based filter, as this configuration provided the closest agreement with the orbit selection reported by Dr Sadavarte.

The initial “contains” filter selected 757 orbits spanning the period from 30 April 2018 to 31 December 2019 (orbits 2818 to 11487) that intersect the analysis domain defined by 146.0°E - 150.0°E and 24.0°S - 20.0°S. This filtering step is not explicitly described in the original study but was introduced here to streamline subsequent processing. Aside from excluding missing orbits and those that do not cover the domain, this filter does not affect the downstream analysis or emission estimates.

The remaining two filters depend on the version of the TROPOMI processor used to generate the data products. For orbits processed with versions 010202, 010300, 010301, and 010302, application of the second filter (qa) reduced the initial selection to 124 orbits. The third filter (r), based on correlation criteria, further narrowed this set to 109 orbits. These 109 orbits were retained for all subsequent analyses. The specific TROPOMI files passing each stage of the filtering process and ultimately used in the calculations presented in this report are listed in the accompanying files:

1. Table_TROPOMIFilter_contains ,
2. Table_TROPOMIFilter_qa and
3. Table_TROPOMIFilter_r

Using orbits processed by processors 020400 the second filter (qa) narrowed the first selection to 217 orbits. The third filter (r) further, narrowed the selection to 204 orbits. Use of processor 020400 effectively doubles number of satellite images available to the algorithm.

ERA5

Once a TROPOMI file is selected, the satellite overpass time for the Hail Creek mine can be determined. Sadavarte, et al., 2021 indicate that ERA5 fields at 04 Z were used in their analysis. In this work, we instead use ERA5 data from the time step closest to the actual satellite overpass. While most overpasses occur closest to 04 Z, a small number are better represented by the 03 Z ERA5 time step.

We do not perform any spatial interpolation of the ERA5 fields. Instead, we use the ERA5 grid point located at 148.25°E, 21.5°S, which is the nearest grid point to the centre of the Hail Creek mine. It is not stated in Sadavarte, et al., 2021 whether any spatial interpolation of ERA5 fields was applied in their analysis.