Space-based observations of tropospheric ethane map emissions from fossil fuel extraction

Ethane (C2H6) is a prevalent volatile organic compound that influences tropospheric ozone production, reactive nitrogen through PAN formation, and the oxidative capacity of the atmosphere. It is co-emitted with methane from fossil fuel sources and thus serves as a tracer for partitioning methane emissions among anthropogenic and natural sources. Despite its importance, large uncertainties persist in the magnitude and spatiotemporal variability of fossil-fuel ethane emissions and in ethane:methane emission ratios, impeding source attribution and impact assessments. Prior satellite ethane measurements, based on solar occultation (ν7) and limb sounding (ν9), have been limited to the upper troposphere and stratosphere, lacking sensitivity to near-surface emissions. This study aims to provide a tropospheric ethane measurement from space using the thermal infrared capabilities of the Cross-track Infrared Sounder (CrIS), convert spectral ethane signals into quantitative column abundances via machine learning, validate against high-quality ground-based FTIR observations, and use the resulting dataset with chemical transport modeling to quantify ethane emissions, with a focus on the major fossil-fuel hotspot in the Permian Basin.

The paper synthesizes prior knowledge that ethane is a key atmospheric VOC affecting ozone and PAN and is co-emitted with methane from fossil fuels. Previous top-down studies have often inferred fossil ethane emissions significantly higher than bottom-up inventories (factors ~1.4–3). Global fossil fuel-related ethane has been estimated at 7–13 Tg yr−1, with additional 2–8 Tg yr−1 from biofuel and biomass burning; geological sources are uncertain (2–6 Tg yr−1). Ethane’s atmospheric lifetime (~2 months) makes it a useful tracer over long distances. Remote sensing of ethane has leveraged the ν7 band near 3000 cm−1 and ν9 near 815 cm−1: ground-based NDACC FTIR retrieved tropospheric columns via ν7; satellite occultation and limb sensors detected ethane aloft but not with near-surface sensitivity. Hyperspectral range index (HRI) methods have been effective for detecting trace gases (e.g., NH3, SO2, isoprene) from IASI and CrIS. The literature also documents methane and ethane emission characterization in major oil and gas regions (e.g., Permian) using TROPOMI and other satellite datasets, and prior ethane estimates from aircraft campaigns and inventories (e.g., FOG).

The study proceeds in three major components: spectral detection, retrieval development, and emission inference.

• Spectral detection via HRI: The hyperspectral range index (HRI) quantifies the ethane spectral signal in CrIS LWIR radiances (800–850 cm−1 encompassing ν9). HRI uses measured spectra, an iteratively derived mean background spectrum, background spectral covariance Sy, and the ethane Jacobian K, normalized so background conditions yield mean 0 and SD 1. Cloud screening is applied using the 900 cm−1 brightness minus surface skin temperature. Validation of ethane detection leveraged a January 2, 2020 Australian fire plume over the South Pacific, comparing CrIS CO columns (optimal estimation), ethane HRI, and an independent brightness-temperature peak-minus-off-peak ethane index from MUSES/TROPESS spectral residuals.
• ROCRv2 retrieval using an artificial neural network (ANN): To convert HRI to ethane column amounts, ROCRv2 implements two key updates relative to ROCRv1: (1) the Line-by-Line Radiative Transfer Model (LBLRTM) replaces ELANOR to compute Jacobians and synthetic radiances, improving accuracy and species flexibility; (2) an explicit vertical sensitivity predictor P90 (pressure below which 90% of the ethane column resides) is added, enabling profile-shape sensitivity and a posteriori adaptation to assumed profiles. • Training data: 2019 GEOS-Chem (v13.3) simulations (12:00–15:00 LT) for VOCs, ozone, ammonia; VOC profiles randomly scaled (μ=1, σ=1). Other species profiles from MOZART climatology; meteorology from MERRA-2. Synthetic radiances with and without ethane are generated in LBLRTM for multiple view angles, with CrIS-like noise added (as per LIB). Synthetic HRIs are computed against zero-ethane backgrounds using spectral covariances from CrIS measurements. The training set spans the range of observed HRI values. • ANN architecture and predictors: Feed-forward ANN with two hidden layers (20 and 10 nodes) and one output node, trained as an ensemble of 10 networks. Predictors include HRI, P90, water vapor column, surface skin temperature, air temperatures (surface, −1 km, −4 km, −10 km), surface pressure, and view angle. Latitude/longitude are excluded. Retrievals are produced on a 0.5° × 0.625° grid for discrete P90 values (150–650 hPa), permitting interpolation to arbitrary vertical profiles.
• Validation and uncertainty: Training performance achieved R2=0.88, slope=0.94, RMSE=4.08×10^15 molecules cm−2; errors increase at low columns, with mean error exceeding 100% below 4×10^15 molecules cm−2 (approximate detection threshold). Bias tendencies: slight overprediction at low columns and underprediction at high columns and lofted cases (P90<300 hPa). Independent evaluation uses NDACC ground-based FTIR ethane columns at 13 sites (2013–2021), with CrIS data averaged over a 3×3 grid around sites and NDACC a posteriori profiles supplying P90 for consistency; altitude corrections applied for high-elevation sites. The CrIS–NDACC comparison yields R2=0.66 with a mean CrIS/NDACC scaling of ~0.61–0.65.
• Emission inference over the Permian Basin: GEOS-Chem v14.1 Classic simulations (MERRA-2, CEDS, US ethane emissions from Tzompa-Sosa et al.) were conducted for 2019 with nested 0.5°×0.625° domain over the Permian. Ethane emissions within 31°–34° N, 101°–105° W were scaled by factors (2,5,6,7,8,10,11). Model columns sampled at CrIS overpass times provided scenario-specific P90 values for retrieval adaptation. The mean CrIS–model column difference over the Permian as a function of emission scale was regressed to determine the scale factor aligning model with observations. The approach was extended annually for 2014–2019 to derive emission trends and compared to EIA oil production, aircraft-based ethane estimates, FOG inventory, and methane-based inferences using assumed ethane:methane ratios.

• Spaceborne detection: CrIS LWIR radiances contain a robust ethane spectral signal detectable via HRI; fire plume case (Jan 2, 2020) showed strong spatial coherence among CrIS CO columns, ethane HRI, and a brightness-temperature ethane index, confirming signal attribution to ethane.
• Spatial patterns: Persistent, strong ethane HRI enhancements align with major fossil fuel production regions. The Permian Basin exhibits the highest global ethane HRI values, with clear sub-basin structure (Delaware and Midland lobes) consistent with production distributions; Delaware produces 54% (oil) and 61% (gas) of Permian output and shows higher HRI.
• Retrieval performance: ANN training achieved R2=0.88, slope=0.94, RMSE=4.08×10^15 molecules cm−2. Detection threshold near 4×10^15 molecules cm−2. Sensitivity dominated by HRI and P90; thermal contrast less influential than for short-lived VOCs.
• Validation: CrIS ethane columns vs NDACC FTIR: R2=0.66; standard major axis fit slope ~0.60, intercept −0.11×10^16 molecules cm−2; mean CrIS/NDACC ratio ≈0.61–0.65, indicating a low bias in CrIS-derived columns relative to ground-based measurements.
• Permian emissions underestimate: GEOS-Chem scaling analysis indicates bottom-up Permian ethane emissions must be multiplied by 7.4 (95% CI: 7.3–7.6) to match CrIS columns, implying ~0.53 Tg yr−1 ethane in 2019 vs ~0.074 Tg yr−1 a priori.
• Global relevance: The Permian alone contributes at least 4–7% of global fossil-fuel ethane emissions.
• Trends (2014–2019): CrIS-derived Permian ethane emissions increase with oil production, with a dip in 2016; 2017 value (0.31 Tg) agrees with aircraft-based estimate (0.28 Tg) but is lower than FOG 2015 estimate and methane-based inferences assuming 17–18% mol/mol ethane:methane. Applying a factor-of-two correction (per NDACC comparison) would bring CrIS-based estimates closer to FOG and between methane-based estimates, but reduce agreement with aircraft results.
• Inventory mapping: Uniform scaling of bottom-up inventories matches overall magnitude but not spatial detail of observed columns, indicating mismapped sub-regional emissions.

The study demonstrates, for the first time, a global tropospheric ethane retrieval from space using CrIS thermal IR radiances, addressing a major observational gap that previously limited ethane source quantification near the surface. By coupling a physically based spectral detection (HRI) with an ANN trained on line-by-line forward modeling and CTM fields, the method captures ethane variability and magnitude sufficiently to constrain emissions. Validation against NDACC confirms significant skill, albeit with a low bias. Application over the Permian Basin shows that inventories substantially underestimate ethane emissions, with a best-fit scale factor of ~7.4, and establishes the Permian as a dominant global ethane hotspot contributing at least 4–7% of fossil-fuel ethane. These findings support the use of ethane as a tracer for methane source apportionment and highlight the need to revise ethane inventories, both in magnitude and spatial distribution. The long-term CrIS-derived ethane record, together with successor instruments, enables monitoring trends, linking emissions to fossil fuel activity, and assessing impacts on ozone and reactive nitrogen budgets.

This work introduces a space-based tropospheric ethane product from CrIS and a retrieval framework (ROCRv2) that translates spectral HRI signals into ethane columns using an ANN with explicit vertical sensitivity (P90). The approach is validated against NDACC and applied with GEOS-Chem to quantify emissions, revealing that the Permian Basin exhibits the strongest global ethane enhancements and that bottom-up inventories underestimate Permian ethane by a factor of ~7.4, implying ~0.53 Tg yr−1 in 2019 and at least 4–7% of global fossil-fuel ethane emissions. The 2014–2019 trend tracks regional oil production. The CrIS record (2012–2023) and continuity from JPSS-1/2 (and future JPSS-3/4) provide a foundation for global, long-term ethane monitoring. Future work should: (1) reduce retrieval bias relative to ground-based columns; (2) incorporate improved, up-to-date spatial proxies of oil/gas activity to refine emission maps; (3) integrate multi-sensor constraints; and (4) extend analyses to other regions and source types (e.g., biomass burning, geologic emissions) to better assess global ethane budgets and co-emitted methane.

• Retrieval bias: CrIS columns are biased low relative to NDACC (mean ratio ~0.61–0.65); results and inferred emissions may be conservative. Training biases include underprediction at high columns and lofted profiles, and overprediction at low columns.
• Detection threshold: Sensitivity decreases at low columns; mean error exceeds 100% below ~4×10^15 molecules cm−2, limiting detection in clean regions.
• Vertical sensitivity and profile dependence: Thermal IR sensitivity depends on the ethane vertical distribution (captured via P90), and background ethane above detection threshold may require post-hoc additions when synthetic training assumed zero background.
• Near-surface sensitivity: LWIR has reduced sensitivity near the surface compared to SWIR; atmospheric absorption and factors like water vapor and ash can interfere, particularly in wildfire plumes.
• Inventory mapping: Emission inversions used uniform scaling within the Permian subdomain; this matches mean magnitude but not spatial patterns, indicating inaccuracies in bottom-up spatial proxies.
• Uncertainty characterization: Confidence intervals for the emission scale factor are statistical (bootstrap) and do not account for potential systematic errors in satellite retrievals or the CTM.
• Data continuity: Suomi-NPP CrIS LWIR ceased in August 2023 (though successor instruments mitigate continuity concerns).