Introduction to Methane Detection: Satellites & Techniques

Methane (CH4) is the second most important greenhouse gas, and its atmospheric concentration is increasing by around 1% annually. With a lifetime of about a decade, it's short-lived compared to carbon dioxide (CO2). However, its global warming potential is a staggering 80 times higher than that of CO2. This fact underscores the importance of detecting and mitigating large methane leaks, especially from the oil and gas industry, to tackle climate change.

In this article, we explore the role of satellite imagery in monitoring methane emissions by examining the available sources of information and how these images permit us to identify methane anomalies.

The physical basis for methane detection

When light traverses gasses in the atmosphere, it is absorbed on specific wavelengths, as shown in the image below. Methane absorption occurs in the range of near-infrared (NIR) and short-wave-near-infrared (SWIR) wavelengths. This can be leveraged to detect methane by using models that account for the effects of its concentration in the atmosphere on the image captured by a satellite capable of sensing these spectral ranges.

Absorption spectra of different gasses in the atmosphere  (from The "Greenhouse Effect," and Global Warming)

When emitted, methane quickly spreads into the atmosphere, affecting it globally. Therefore, any hyperspectral image will show traces of methane. Detecting and quantifying emission sources involves two main steps: identifying the methane contributions in the image data and detecting anomalies that deviate from baseline methane concentration levels. The approach to solving both problems greatly depends on the specific sensor being used.

Constellations Capable of Methane Sensing

Both private companies and space agencies have developed satellites capable of methane detection. 

In the private sector, GHGSat (US-based) and Absolut Sensing (Europe-based) have designed payloads specifically targeted to detect and measure methane accurately. For now, GHGSat has the only operational and commercial fleet of twelve satellites dedicated to measuring CH4. Absolut Sensing plans to launch a novel constellation of satellites (named GESat) next year to monitor greenhouse gas emissions, with methane as a first use case. Other private satellites have methane detection capabilities, and companies offer these services, although in those cases, their instruments were not designed with the primary purpose of methane emission estimation.

Space agencies have developed the following three notable hyperspectral satellites that allow the detection of methane: Sentinel-5P, PRISMA, and EnMAP.

• Sentinel-5P: was developed by the European Space Agency (ESA). It provides two features that make it ideal for methane detection: fine spectral sampling, which allows the isolation of methane absorption spectrum features, and daily Earth coverage, which enables time series analysis.
• PRISMA: was developed by the Italian Space Agency (ISA). It carries both hyperspectral and panchromatic camera modules and allows monitoring of both methane and carbon dioxide concentrations in the atmosphere.
• EnMAP: is a German hyperspectral satellite that aims to monitor and characterize Earth’s environment on a global scale by providing high radiometric resolution and stability in the bands of interest for detecting methane.

Multispectral sensors instead of hyperspectral ones are not perfectly adapted for precise methane detection, but satellites with spectral bands that cover the SWIR range can still be used to detect and quantify significant CH4 emissions. This is the case of the Sentinel-2 and Landsat-8 missions, among others.

The satellites mentioned above (and others) vary in capabilities, precision, revisit time, and swath. Therefore, each is better suited for different use cases, such as global mapping (measuring methane levels of entire basins), area mapping (quantifying regional-scale emissions), or detecting and accurately quantifying emissions from point sources.

Methods for Detecting Emissions

We present the three main methodologies for detecting methane, exemplified by recent research papers (several of which our team members participated in). These methodologies utilize these satellites’ data to detect methane leaks from industrial sources.

Attenuation

These methods detect the attenuation in the SWIR spectral bands due to excess methane. This property is exploited in articles using different satellites:

• Sentinel-2 and Landsat-8: In (Ehret et al., 2022), the authors leverage the SWIR bands, which provide good spatial resolution, a low revisit time, and freely available data sources. They show that methane emissions from oil and gas infrastructure below the detection threshold of Sentinel-5P can be detected and quantified using Sentinel-2 imagery.
• Sentinel-5:
• In (Ouerghi et al., 2022), the authors present a three-step procedure involving pixel reconstruction, angle correction, and plume detection with a time series. They show how the Sentinel-5P L2 methane product can be complemented by their automatic detection of methane plumes that depend only on the Sentinel-5P L1B product.some text
  • In (Ouerghi et al., 2021), the authors propose other main steps: background subtraction, detection of local maxima in the negative logarithmic spectrum of each pixel, and anomaly detection in the background-free image. They show that the proposed method guarantees a very low number of false alarms, which can help experts focus on meaningful detections. However, the presented approach does not work in the presence of clouds or over water where the albedo is very small.

Both papers validated the methods by comparing detected plumes against a manually annotated dataset from the Sentinel-5P L2 methane product for plume detection.

Spectrometers with matched filtering

These methods rely on sensors that capture spectral bands with sufficient resolution to permit methane detection by matched filtering against its known absorption spectral profile. 

In (Ouerghi et al., 2023), the authors automatically detect sources of methane leaks using images from the PRISMA satellite. They use a variation of the Matched Filter (MF) called the Adjusted Spectral Matched Filter (ASMF) to detect methane plumes in satellite images. They compare the plumes' orientation to the wind direction extracted from the standard meteorological reanalysis product ERA5 provided by the Copernicus Climate Change Service (C3S) to remove false positives. Their proposed approach obtains a better detection rate than deep learning methods or the standard matched filter on a dataset with manually annotated plumes on PRISMA images.

Inversion methods

The methods in this category (like 5AI) allow for precise quantification of methane concentration. They rely on complex simulations of a physical model of the atmosphere. Therefore, they are computationally expensive and require extensive input data, which may often not be available.

More specifically, they are based on a direct radiative transfer model that predicts hyperspectral observations from a state vector, including the concentration of various gasses in the atmosphere, albedo, and the presence of aerosols. Based on this physical model, prior information on the state vector, and the hyperspectral observations from the satellite, an optimization algorithm is used to estimate methane concentration in the atmosphere by posterior maximization.

This procedure is so complex and computationally expensive that cloud-based services have been developed to make them more user-friendly. Also, neural networks have been successfully used to drastically reduce the computational cost of the direct radiative transfer model (see Stegmann et al. 2022) or its inversion (see David et al. 2021)

Conclusions

Although methane doesn't receive the same public attention as carbon dioxide, it is of paramount importance that we tackle its emissions to slow global warming and climate change. Initiatives in the public and private sectors are improving the data available to monitor CH4 emissions. It is up to us to develop algorithms, tools, and systems that can leverage this data to provide insights and enforcement capabilities to policymakers and regulatory agencies that can control methane pollution. In Digital Sense, we have vast experience working with remote sensing data, from every stage of satellite image processing pipelines to applying computer vision and analytics on top of these satellite images (you can check some of this out here). We are eager to continue using that knowledge to help fight climate change.

References

• Ehret, Thibaud, et al. "Global tracking and quantification of oil and gas methane emissions from recurrent sentinel-2 imagery." Environmental science & technology 56.14 (2022): 10517-10529.
• Ouerghi, E., Ehret, T., de Franchis, C., Facciolo, G., Lauvaux, T., Meinhardt, E., and Morel, J.-M.: AUTOMATIC METHANE PLUMES DETECTION IN TIME SERIES OF SENTINEL-5P L1B IMAGES, ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., V-3-2022, 147–154, https://doi.org/10.5194/isprs-annals-V-3-2022-147-2022, 2022.
• Ouerghi, E., Ehret, T., de Franchis, C., Facciolo, G., Lauvaux, T., Meinhardt, E., and Morel, J.-M.: DETECTION OF METHANE PLUMES IN HYPERSPECTRAL IMAGES FROM SENTINEL-5P BY COUPLING ANOMALY DETECTION AND PATTERN RECOGNITION, ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., V-3-2021, 81–87, https://doi.org/10.5194/isprs-annals-V-3-2021-81-2021, 2021.
• E. Ouerghi, T. Ehret, Gabriele Facciolo, E. Meinhardt, J.-M. Morel, et al.. METHANE PLUMES DETECTION ON PRISMA L1 IMAGES WITH THE ADJUSTED SPECTRAL MATCHED FILTER AND WIND DATA. IGARSS 2023 - 2023 IEEE International Geoscience and Remote Sensing Symposium, Jul 2023, Pasadena, United States. Pp.7598-7601, 10.1109/IGARSS52108.2023.1028221. Hal-04497732
• Patrick G. Stegmann, Benjamin Johnson, Isaac Moradi, Bryan Karpowicz, Will McCarty, A deep learning approach to fast radiative transfer, Journal of Quantitative Spectroscopy and Radiative Transfer, Volume 280, 2022, 108088, ISSN 0022-4073, https://doi.org/10.1016/j.jqsrt.2022.108088.
• David, L., Bréon, F.-M., and Chevallier, F.: XCO2 estimates from the OCO-2 measurements using a neural network approach, Atmos. Meas. Tech., 14, 117–132, https://doi.org/10.5194/amt-14-117-2021, 2021.

‍