Optimizing Methane Detection On Board Satellites: Speed, Accuracy, and Low-Power Solutions for Resource-Constrained Hardware
1 st Jonáš Herec, 2 nd Vít Růžička, 3 rd Rado Pitoňák · 2025
Read Optimizing Methane Detection On Board Satellites: Speed, Accuracy, and Low-Power Solutions for Resource-Constrained Hardware by 1 st Jonáš Herec, 2 nd Vít Růžička, 3 rd Rado Pitoňák (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-01472.txt.
Introduction
Herec, Růžička, and Pitoňák present a study from twenty twenty-five on spotting methane leaks from satellites on board the craft itself. Methane is a potent greenhouse gas, and early leak alerts help climate action. Many missions still rely on slow manual tasking and downlinks. Classic detection math is too heavy for weak onboard hardware. The team speed up the leading algorithm and test two fast detectors new to this task. Learning models are paired with the filters to judge real potential. Two candidates prove accurate enough for strong plumes.
Speed gains run to hundreds of times faster than the classic method. Band choice tricks cut channels without losing accuracy. Code, data, and models are open sourced. The work lays groundwork for timely onboard alerts.
Catalog record
| CREATOR | 1 st Jonáš Herec, 2 nd Vít Růžička, 3 rd Rado Pitoňák |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-01472
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Optimizing Methane Detection On Board Satellites: Speed, Accuracy, and Low-Power Solutions for Resource-Constrained Hardware 1 st Jonáš Herec, 2 nd Vít Růžička, 3 rd Rado Pitoňák Abstract Methane is a potent greenhouse gas, and detecting its leaks early via hyperspectral satellite imagery can help mitigate climate change. Meanwhile, many existing missions operate in manual tasking regimes only, thus missing potential events of interest. To overcome slow downlink rates cost-effectively, onboard detection is a viable solution. However, traditional methane enhancement methods are too computationally demanding for resource-limited onboard hardware. This work accelerates methane detection by focusing on efficient, low-power algorithms. We test fast target detection methods (ACE, CEM) that have not been previously used for methane detection and propose a Mag1c-SAS – a significantly faster variant of the current state-of-the-art algorithm for methane detection: Mag1c. To explore their true detection potential, we integrate them with a machine learning model (U-Net, LinkNet). Our results identify two promising candidates (Mag1c-SAS and CEM), both acceptably accurate for the detection of strong plumes and computationally efficient enough for onboard deployment: one optimized more for accuracy, the other more for speed, achieving up to ~ 100× and ~ 230× faster computation than original Mag1c on resource-limited hardware. Additionally, we propose and evaluate three band selection strategies. One of them can outperform the method traditionally used in the field while using fewer channels, leading to even faster processing without compromising accuracy. This research lays the foundation for future advancements in onboard methane detection with minimal hardware requirements, improving timely data delivery. The produced code, data, and models are open-sourced and can be accessed from https://github.com/zaitra/methane-filters-benchmark . I Introduction Methane is the second most significant greenhouse gas contributing to global warming after \text{CO}_{2} [ 1 ] . While \text{CO}_{2} is more prevalent, methane is more potent, mainly in the short term (84-87 times more than \text{CO}_{2} over a 20-year period [ 2 ] ). A significant share of methane emissions in the oil and gas industry originates from episodic ultra-emission events, often caused by equipment failures and other unpredictable factors [ 3 ] . These events could be detected and mitigated early with the use of satellite monitoring, but it is a challenging task. While methane has a distinct spectral signature with its strongest absorption features in the ~ 2100–2500 nm range, the change in measured radiance is too subtle and cannot be detected by the human eye. However, it can be seen in data from current imaging spectroscopy sensors such as NASA’s AVIRIS-NG or EMIT [ 4 ] and from the sensors of planned missions such as NASA’s SBG [ 5 ] , or the ESA’s CHIME [ 6 ] . These missions carry hyperspectral instruments that nearly continuously capture light across hundreds of spectral channels, allowing to leverage methane’s spectral signature for its detection. The high number of channels is advantageous, but it also hinders the timely delivery of data. Downlinking from satellites is often slow, as data is transmitted via limited-bandwidth radio communication only a few times per day when the satellite passes over a ground station. Increasing transmission speed would significantly raise costs, not only due to the added weight and expense of more powerful satellite antennas, but also because it would require a larger network of ground stations. The approach of many currently flown satellites (for example, PRISMA or EnMAP) is to observe only over manually selected regions, which severely reduces the number of potentially captured methane leak events. To address this, algorithms and machine learning (ML) models can be employed on board, as was explored for disaster events in [ 7 ] , and specifically for methane detection in [ 8 ] . However, since the goal is to reduce the cost and improve the timely delivery of salient data, these models must be designed to fit the constraints of the mission without compromising its objectives. Given the high power requirements of GPUs and their ongoing adaptation for the space environment, the key could be developing efficient algorithms that can run on commonly used solutions [ 9 ] – lower-power CPUs, or FPGAs [ 10 ] . The key part of the pipeline is the computation of the methane enhancement product, which generates a 2D image representing the detection of the methane spectral signature from the 3D hyperspectral cube. This process effectively reduces the number of channels from dozens or hundreds to just one, which can then be used for methane detection by thresholding or by an efficient machine learning model, enabling real-time methane detection. Fig. 1: Visualisation of a strong methane plume event showing all methane enhancement products used in this study. Note that visual scaling was adjusted manually to aid legibility. Typically used product is the Matched filter [ 11 ] , with Matched filter with Albedo correction and reweiGhted L1 sparsity Code (Mag1c) [ 12 ] being a state-of-the-art variant. These products are calculated for each column of the input image; in Mag1c, this is further repeated in several iterations. We argue that these products are prohibitively slow for onboard deployment. Therefore, we explore faster and simpler variants of these methane enhancement products, where, besides other optimizations, we compute the whole filter at once instead of column-wise, to further speed up the process. Since the number of bands is a crucial parameter for these algorithms, we experiment with newly proposed band selection strategies to determine if the runtime can be further reduced by selecting fewer bands without compromising accuracy. We trade in the quality of the products for speed, but this can be mitigated by lightweight machine learning models, which refine the methane product, similarly to [ 13 , 14 ] . Also, we focus on more easily detectable large plumes, which are particularly relevant, as their impact is disproportionately larger than that of smaller leaks [ 15 ] and is thus more critical for onboard detection. Namely, our contributions are: • We propose a new methane product called “Mag1c Sped up with Additional Sparsity” ( Mag1c-SAS ), which is almost 100x faster than the original Mag1c applied on columns, with an acceptable drop in accuracy. • We benchmark three newly proposed band selection strategies and test the influence of the number of channels on accuracy and speed. • We investigate various target detection methods, including the Constrained Energy Minimization (CEM) method [ 16 ] , which is nearly 230× faster than the original Mag1c applied to columns. • Finally, we explore the capabilities of the ML model to refine these products, where we consistently observe improvements in the F1 score by up to ~ 20 points for strong plumes. II Methodology II-A Methane enhancement products Figure 1 shows a selection of methane enhancement products explored in this work. A widely used method for methane detection is the matched filter [ 17 ] , often implemented with various modifications. Its primary goal is to identify and amplify the methane signal while suppressing background noise. This is accomplished by comparing the spectral signature to deviations from the band-wise means at each hyperspectral pixel while considering the background characteristics modeled by the band’s variance-covariance matrix. Here, we use the matched filter as defined for classical target detection tasks in [ 18