Supporting materials

Downloads

Download the full report.

Methodology

Estimating methane emissions from TROPOMI observations

This is a technical description of the method used to estimate coal mine methane emissions from TROPOMI observations. It applies the wind rotation technique to enhance the visibility of the concentrations and uses both the cross-sectional flux (CSF) and integrated mass enhancement (IME) approaches to quantify emissions over six coal mine clusters. 

The method is used to estimate annual methane emissions for six coal mining regions in Australia from January 2019 until December 2023. Emissions are calculated for both the calendar year and the financial year (July 1 to June 30), as Australian reporting standards follow the financial year timeline. The modelling work was carried out by environmental intelligence company Kayrros.

Wind rotation method

The wind-rotation method consists of rotating the Sentinel-5P TROPOMI L2 CH4 V02.05.00 images over an area in the direction of the ERA5 wind, so that the wind blows southwards. In July 2022, TROPOMI released newly reprocessed data for its Sentinel 5P images. However, this study incorporated only the previous release of reprocessed products to ensure consistency. This process forces the methane plume to move in a specific direction (in our case south) from the centre of the image. Rotating all the TROPOMI images over a year allows us to average them, in order to obtain a mean plume for the area of interest (AOI). By leveraging the high temporal resolution of Sentinel-5P, this approach enhances the signal-to-noise ratio of TROPOMI data and thus, helps distinguishing the methane plume from its background more easily. 

The procedure starts with image filtering, retaining only those with a high percentage of valid pixels following the recommendations from the Sentinel-5P user guide. Specifically Kayrros incorporated TROPOMI pixels with qa_value equal to 40 and 100.

Then, TROPOMI images are oversampled, taking advantage of shifts in satellite orbits from day to day resulting in partial overlap between pixels. The next step involves  finding the optimal rotation point to identify  the methane emission source within the AOI. To achieve this, wind-rotation is evaluated on a dense grid of rotation points encompassing the whole area. By finding which rotated average shows the largest downwind enhancement, it is possible to pinpoint the optimal rotation point within a few kilometres.

To reconcile the wind direction with the methane propagation direction, an artificial correction is applied to the hourly ERA5 100m wind speed. This correction involves determining the delta angle that, added to the wind direction, best aligns with a simulated Gaussian plume-oriented southwards. After subtracting the mean background upwind of the source, we have a time-averaged plume of the area for the period. 

Emissions quantification

The plume mask is computed statistically, considering the methane plume as an outlier in the enhancement image. The CSF and the IME methods are then employed to estimate the methane flow rate based on the masked enhancement image and the mean wind over the period. The uncertainty in emissions (error bar) is calculated using the range of emissions calculated by the two quantification methods. 

Modelling assumptions

Several assumptions were made for this modelling work. 

Assumption 1. The dominant methane source in the clusters is from coal mining

Operational oil and gas pipelines were found to cross through clusters 1, 3, 4, and 5, as identified by overlaying data from the oil and gas database onto the cluster regions. The assumption is that any leaks from these pipelines would be short-lived and unlikely to significantly affect the mean-averaged plume. It is assumed that agricultural emissions are generally too diffuse to be detected by TROPOMI. 

Gridded emissions inventories for 2020 estimate that the majority of methane emissions, (87%) originate from coal. This is illustrated below, where emissions within the cluster regions are summed using data from the  Global Fuel Exploitation Inventory (GFEI) v3 and EDGAR 2024 solid waste and landfill data.

While the inventories indicate small methane contributions from gas and landfill overall, satellite detections over the cluster regions show that major emissions events within the clusters are exclusively from coal. Analysis of point-source emissions data from both open and commercial providers confirms that all detected emissions within the cluster boundaries are attributed to coal mining. 

We aggregated emissions data from Kayrros Methane Watch, the Methane Alert and Response System (MARS), and Carbon Mapper. These detections come from a range of satellite sensors, including TROPOMI, PRISMA, EMIT, EnMap, and Landsat. GHGSat data was provided to Ember through the UK Space Agency Catapult programme. In total, 425 methane detections were identified, all of which were attributed to coal mining, with none linked to gas infrastructure or landfill sites.

However, within specific clusters of this study’s focus, gas and waste have the potential to influence our findings within clusters 3 and 5. This study was not able to conclusively rule out or estimate the potential influence of these emissions sources.

Assumption 2. There is no transport of methane into the clusters from sources outside the clusters

To minimize interference from methane emissions coming from neighbouring assets, images with wind directions aligned with nearby sources are filtered out. 

Assumption 3. The wind rotation method, often used for smaller regions, can be applied to these large clusters

Traditionally, CSF and IME methods have been used for individual plume quantifications. This analysis applies these methods to emissions from larger regions. A similar approach was used by Schneising et al (2020)  using the CSF method to study emissions over oil and gas basins. Our analysis assumes the IME method is also appropriate over larger scales.

Cluster selection

The six clusters were selected for study because they represent regions of intense coal mining,  and account for approximately 90% of the country’s metallurgical coal production. 

Four clusters are modelled for Queensland, accounting for 79% of coal production in the state. Two clusters are selected for New South Wales which covers 60 to 64% coal production in the state.

TROPOMI data coverage

TROPOMI data coverage over Australia is generally good in contrast to other regions, as demonstrated by Gao (2023). This is largely due to low cloud cover and the presence of flat reflective surfaces, which create favourable conditions for satellite detection.   A study by Ember also showed that environmental conditions in Australia are favourable for satellite-based methane monitoring. The study estimated that 91% of coal production occurs in regions well-suited for satellite monitoring. 

Despite this, cloud cover did impact the number of images required to estimate emissions at some of the clusters during some years (see table below).  This limited the comparison of satellite estimates with national reported values to the financial year 2020 where all six clusters are modelled. 

In Queensland, a comparison with state reported data was possible for the financial year 2020. In New South Wales,  it was possible to compare satellite emissions to state reported data for the financial years 2020 and 2021.

Uncertainties

Several uncertainties of the wind-rotation methodology are intrinsically linked to the TROPOMI sensor onboard Sentinel-5P. Primarily, measurements are significantly influenced by observational conditions. Factors such as cloud cover and water surfaces, which have low reflectance in the short-wave infrared (SWIR) bands, make observations difficult. Filtering images with excessive cloudy pixels may introduce a seasonal bias, as yearly estimations could predominantly rely on periods of favourable weather. However, the cloud free conditions in Australia are expected to minimise this bias. Additionally, the daily overpass time of Sentinel-5P at approximately 1:30 PM local time may introduce a temporal bias.

The methane detection threshold of the TROPOMI sensor varies between 4.2 tons per hour under optimal conditions and up to 25 tons per hour under less favourable conditions. Consequently, smaller emission sources may remain undetected. Furthermore, the spatial resolution of TROPOMI, which is 7.5 km x 5.5 km, can pose challenges in attributing emissions to specific sectors or assets. To mitigate this, a cluster-based approach is employed to isolate emissions. However, for a cluster that is close to others, images with wind directions aligned with these neighbouring clusters are discarded in order to prevent their contamination in the final methane estimate.

Alternative methods to estimate methane emissions from satellite

Satellites measure methane concentrations, but an extra step is needed to convert these observations into emissions estimates that can be compared with reported figures. This study uses the wind rotation method and the CSF and IME methods to quantify the emissions.  However, alternative methods are also available for estimating emissions from satellite observations.

Integrated Methane Inversion

An alternative approach, as used by Shen shown above, involves the integrated methane inversion (IMI), a tool for estimating methane emissions. It uses methane concentration data from the TROPOMI sensor to derive emission estimates. 

This approach begins with an initial guess of emissions, which is then fed into a model that tracks how methane moves through the atmosphere. The model’s results are compared to satellite measurements, and the emissions estimate is adjusted to reduce any differences. This process helps refine the original estimate, making it more accurate.  One key benefit of the IMI method is its ability to spatially distinguish emissions by source—such as oil, gas, coal, and agriculture—enabling the estimation of emissions by both location and source type.

The OpenMethane project, run by the Superpower Institute, focuses exclusively on Australia, allowing it to incorporate higher-resolution weather data and more detailed initial emissions estimates. This higher-resolution inversion builds on the approach outlined by Shen and can potentially lead to more accurate methane emissions estimates that could also play a valuable role within an emissions verification approach.

Acknowledgements

Contributors

We extend our gratitude to Kayrros for conducting the emissions modelling for this study and to our funder, Boundless Earth, for their generous support.

This work was significantly improved by the contributions of Reynoldo Dizon, Sabina Assan and Rini Sucahyo as well as reviews conducted by Rajasekhar Modadugu and Eleanor Whittle. We also greatly appreciate the external reviewers who contributed to improving this work.

Cover image

View out of the International Space Station which hosts the EMIT methane sensor.