Methane Emissions from Underground Coal Mines

Abstract

Methane emissions from Kentucky underground coal mines were measured using near-infrared spectrometers deployed on a vehicle, an airplane, and a small uncrewed aerial system (sUAS), offering insights into the effectiveness of these methods under real-world conditions. From 2021–2022, surveys covered 14 active, 4 inactive, and 4 abandoned mines across Kentucky’s coal-producing basins. Vehicle-based surveys at 13 active mines detected methane anomalies at 9 sites with anomaly lengths spanning tens to hundreds of meters and peak emissions of 665 ± 229 kg h^–1 from two Cardinal mine fans. Airborne GHGSat surveys identified anomalies at 3 sites, including a peak of 1062 ± 386 kg h^–1 at Cardinal, consistent with the EPA’s Greenhouse Gas Reporting Program (GHGRP) ranges. sUAS emissions measured at Straight Creek were 65 ± 22 kg h^–1 (below the GHGRP reporting threshold). Aeris vehicle-based estimates more closely matched GHGRP values than GHGSat estimates and exhibited smaller uncertainties. For example, at Cardinal Nebo, Aeris reported 228 ± 142 kg h^–1 versus GHGRP’s 360 kg h^–1, while GHGSat reported 716 ± 355 kg h^–1. These results have important implications, where terrain and road access permit, vehicle-based methods can yield emissions estimates comparable to aircraft- and satellite-based approaches. Additionally, the higher detection rate of vehicle-based surveys suggests superior performance in identifying methane anomalies. This study highlights the spatial and temporal variability of methane emissions from underground coal mines and emphasizes the importance of integrating multiple observational strategies to improve monitoring in underrepresented regions. It also provides a transferable framework for areas where limited data availability has hindered effective methane tracking and mitigation planning.

Introduction

Fossil fuels supplied 81% of the global energy demand in 2024, with natural gas, coal, and oil contributing 23%, 26%, and 32%, respectively. Methane is not only the major component of natural gas, but also a potent greenhouse gas (GHG). Methane contributes 30% of the global GHG emissions to the atmosphere as it escapes to air during fossil fuel exploration, transportation, and storage. − Methane’s global warming potential (GWP) is 82.0 ± 1.0 over 20 years and 28.5 ± 1.5 over 100 years, relative to CO₂. Consequently, abating methane emissions is important for reducing the increase in global warming over the upcoming decades.

Emissions from both active and abandoned underground or surface coal mines account for about 9% of global methane emissions. Active underground mines contribute 60% of the previous figure, mainly through fan ventilation systems technically known as ventilation air methane (VAM). In Kentucky, coal is mined primarily in the Appalachian and Illinois Basins in eastern and western Kentucky, respectively. In 2018 there were 51 active underground mines in eastern Kentucky and 7 in western Kentucky. According to data reported to the U.S. Environmental Protection Agency’s (EPA’s) Greenhouse Gas Reporting Program (GHGRP), active coal mines in Kentucky emitted an average of 3.90 × 10⁵ metric tons of CO₂-equivalent methane per year in the Appalachian Basin and 4.29 × 10⁵ metric tons per year in the Illinois Basin between 2011 and 2018, based on a 100-year GWP time horizon.

The EPA’s emissions estimates are primarily derived from quarterly single-point-in-time measurements of methane concentrations at or near ventilation fans in the mine. The methane concentrations are multiplied by the volumetric flow rate of the fan to estimate emissions. Such ground-based measurements are inherently limited by the frequency and geographic distribution of sampling campaigns. Moreover, the methane emissions estimated for a coal-producing basin per EPA’s guidelines are an underestimate as only sources emitting more than 25,000 tons CO₂ equivalent per year are required to report. One potential solution to this limitation involves utilizing remotely deployed sensors on small uncrewed aerial systems (sUAS), − vehicles, aircraft, and satellites for measurements. Both aircraft and satellites have been used to measure methane emissions from landfills, oil and gas infrastructure, and underground coal mines. , However, the consistency between ground-based, aircraft, and satellite measurements remains uncertain (e.g., Caulton et al.; Fiehn et al.). , Concerns about the reliability of each method are valid. For example, remote sensing integrates methane concentrations through the atmospheric column. However, rapid atmospheric dynamics at low altitudes may hinder the gathering of accurate information, necessitating ground-level measurements.

This work presents an 11-month field study that, for the first time, characterized methane emissions from underground coal mines in Kentucky using a combination of vehicle- and airplane-based, and sUAS measurements. Fence-line vehicle- and sUAS-based measurements used near-infrared spectrometers to analyze the distribution of methane emissions around active and inactive coal mines. Airplane-based methane measurements commissioned to Greenhouse Gas Satellite (GHGSat) were also conducted over mapped potential emissions sources (e.g., ventilation fans, belt portals) over preselected targeted mines. In addition to the previous quantitative objective, the work has a 2-fold aim, to compare the performance of different measurement platforms and their accuracy in representing the magnitude of emissions, and to compare the present field-based measurements to historic EPA and Mine Safety and Health Administration (MSHA) measurements.

Experimental Methods

Screening and Mine Selection

Using emissions and coal production data from the EPA, the United States Energy Information Administration, and MSHA, 14 active, 4 nonproductive active, and 4 abandoned underground mines were selected for study in eastern and western Kentucky (Figure S1, Supporting Information). − The active mines in Table S1 (Supporting Information) represent a range of mine sizes in terms of coal production and presumably methane emissions. The range provides the opportunity to assess the effectiveness of different measurement methods for different levels of emissions. Possible emissions from abandoned and nonproductive active mines were also evaluated as these types of mines will become more numerous in the future.

The positions of individual sources of methane within a given mine, such as, ventilation fans, belt portals, and coal piles were located using the Kentucky Mine maps server, based on accurate latitude and longitude (NAD 83) coordinates. Possible methane sources in the mines were the targets of methane measurements using near-infrared spectrometers deployed from vehicles, a sUAS, and an airplane. Starting in October 2021 and ending June 2022, measurements from a vehicle, a sUAS, and a GHGSat airplane were conducted as detailed below.

Vehicle-Based Measurements

Efforts to gain access to mine properties for measurements were unsuccessful and we therefore assessed opportunities to conduct fence-line surveys along public roads adjacent to the mines. Rapid, efficient, and noninvasive quantification of methane emissions (e.g., from well pads) has been demonstrated before using mobile surveys conducted from public roads without requiring site access. Of the 21 mines visited in person, 15 were assessed as viable for completing duplicate surveys (on different dates) (Table S2, Supporting Information). Two portable spectrometers, the Aeris Pico Mobile LDS and LI-COR 7810, measured methane and water vapor concentrations at 1 s intervals with precision <0.001 ppm, while surveying in real time the selected locations. Each instrument has an optical cavity where the concentration of methane is measured through laser absorption (Figure S2, Supporting Information). The Aeris Pico spectrometer also measures concentration of ethane. The ratio between ethane and methane can distinguish between thermogenic versus biogenic methane and hence its source even from coal mine emission. The results from measurements with the Aeris Pico spectrometer are visualized using QGIS 3.42.

The work performed involves at least two vehicle-based transects per site to estimate methane emissions from single point sources. Previous work in the natural gas and oil field sector for simpler terrains by Caulton et al. showed that uncertainty with two transects can range widely (from 0.05× to 6.5×) for the true emission rate estimation. Although Caulton et al. showed that using 10 or more transects can greatly reduce uncertainty in the presence of multiple point sources, this study deals with simpler single-point sources. The feasibility of mobile surveys for detecting methane emissions under field conditions using two transects has been demonstrated by von Fischer et al., although their estimation approach differs from ours.

To ensure measurement stability under field conditions, both the Aeris Pico and LI-COR 7810 analyzers were operated during periods of moderate wind speed (<4.5 m s^–1), minimizing the potential for ambient air turbulence to influence gas sampling. The cutoff of 4.5 m s^–1 was selected based on established Gaussian plume modeling principles, which indicate that higher wind speeds lead to rapid dispersion and reduced concentration gradients, increasing uncertainty in emission estimates. , Importantly, both instruments rely on active sampling systems in which air is continuously drawn into the measurement chamber via an internal pump. This setup enables consistent sample delivery at a controlled rate, independent of external airflow conditions. Furthermore, the analyzers operate at a high temporal resolution, with sampling frequencies of 1 Hz, allowing for real-time tracking of gas concentrations while maintaining robustness against short-term environmental fluctuations. As a result, wind variability did not introduce detectable artifacts in the recorded data, supporting the reliability of the measurements under the observed meteorological conditions.

Both the Aeris Pico and LI-COR LI-7810 gas analyzers are designed to ensure accurate methane measurements under variable environmental conditions by incorporating advanced mechanisms for temperature and humidity compensation. The Aeris Pico employs a pressure-stabilized sensor core to maintain constant internal pressure, thereby minimizing the influence of external temperature fluctuations on gas density and absorption signals. It operates in the mid-infrared (mid-IR) spectral region, where methane and ethane exhibit strong absorption features, enabling robust quantification of dry mole fractions even in the presence of water vapor. In contrast, the LI-7810 utilizes optical feedbackcavity-enhanced absorption spectroscopy (OF-CEAS), a near-infrared laser-based technique, and reports methane and carbon dioxide concentrations as dry mole fractions to correct for water vapor dilution. It also compensates for spectroscopic interferences and temperature-related effects through a temperature-controlled optical bench maintained near 55 °C. Both instruments implement real-time signal processing algorithms and required a warm-up period (at least 30 min were used in this work) prior to operation to ensure thermal stability and measurement reliability. These features collectively enable high-precision methane detection across a range of humidity and temperature conditions in both laboratory and field environments. Both instruments were calibrated in the laboratory using certified gas standards (AWG, UHP) with a relative uncertainty of ±0.05%. These procedures, following manufacturer-recommended protocols, ensure traceability to the CH₄ mole fraction scale. The resulting calibration curves exhibited excellent agreement, with a coefficient of determination (R ²) of 0.999 between the two instruments when measuring the same standards. This high degree of correlation further confirms that the measurements fall within the calibration specifications provided by each manufacturer. However, field colocation data for the LI-COR instrument was not validated for this study.

Driving surveys usually involved two transects combined to form a round-trip (e.g., T1–T2 represents a round-trip, Table S2, Supporting Information). Vehicle speed was kept at 8 to 16 km h^–1. Sample transect locations and measurements were recorded with a GPS pod interfaced with the Aeris Pico spectrometer. Surveys were conducted exclusively in the absence of precipitation to optimize the detection of methane anomalies and ensure high data quality. Anomalous methane levels (Δ­[CH₄]) are defined as methane molar ratios that consistently exceed the environmental background. To ensure accurate quantification of methane enhancements, baseline methane concentrations were determined for each survey in the area of the target but away from its influence (typically 1 km). This allowed us to accurately characterize the variation in background methane among sites and minimize uncertainty in emission estimates. These background values varied across sites, with statistical analysis yielding a mean (±standard deviation) concentration of 1.977 (±0.016) ppm and a coefficient of variation of 0.82%, indicating very low variability. This low variability aligns with the minor differences in local sources, atmospheric transport, and meteorological influences observed across the sampled mines. Nevertheless, the presence of site-specific variation supports the use of localized background definitions rather than a fixed global value. A Kestrel 4000 pocket weather tracker was used to measure temperature, wind speed, humidity, and barometric pressure. Wind direction was expressed in degrees relative to true north for modeling the emissions, indicating the direction from which the wind originated.

Vehicle-based emission rates (Q _e ) were estimated from fence-line transects sampling methane plumes downwind of selected mine sources surveyed by GHGSat. The position and altitude of the exhaust fans relative to the locations of maximum Δ­[CH₄]_max observed downwind along the wind direction, combined with the position and altitude of the sampling port, were used to estimate Q _e (assumed to be constant) at the point source using a Gaussian plume dispersion model. In this model, Q _e from a point source located at (x ₀, y ₀, z ₀) is estimated based on the measured Δ­[CH₄] concentration (assumed to be in steady state) at a downwind location (x ₁, y ₁, z ₁) using eq

$$Q_e=\frac{2\pi u{\sigma }_y{\sigma }_z{{{\mathrm{\Delta }[\mathrm{CH}}_4]}_{\max }\exp }^{({(y)}^2/2{\sigma }_y^2)}}{[{\exp }^{(-({(z-H)}^2/2{\sigma }_z^2))}+{\exp }^{(-({(z+H)}^2/2{\sigma }_z^2))}]}$$ 1

where Q _e is given in g s^–1, u (m s^–1) is the measured wind speed, σ _y and σ _z (m) are the dispersion coefficients in the crosswind and vertical directions, respectively, and H (m) is the effective height of the fan (physical height plus plume rise). The variables x, y, z represent the downwind distance, crosswind offset, and height above ground of the sampling port, respectively. The dispersion coefficients are functions of atmospheric stability and downwind distance, estimated using empirical relationships based on Pasquill-Gifford stability classes: σ _y = ax ^b and σ _z = cx ^d , where a, b, c, and d are empirical constants specific to each stability class. The empirical relationship is valid for wind speeds from 1.0 to 6.0 m s^–1 (encompassing the selected moderate winds of <4.5 m s^–1) and for downwind distances typically exceeding 100 m. This distance threshold was chosen based on practical considerations and is supported by literature indicating that plume dispersion generally stabilizes beyond 100 m under moderate wind speeds. , All measurements were conducted under moderate wind conditions (<4.5 m s^–1), within the model’s applicable range. The Gaussian plume model assumes steady-state emissions, homogeneous terrain, and a uniform wind field, allowing methane to disperse as a passive tracer with a Gaussian distribution. However, it does not account for temporal variability, complex terrain, or plume development at short distances (<50–100 m), and its accuracy diminishes under very low wind conditions (u < 0.5 m s^–1). Additionally, the model simplifies vertical mixing and is less suitable for multiple or spatially distributed sources. Because eq is differentiable with respect to all inputs, its sensitivity to measurement uncertainty can be evaluated rigorously using partial derivatives and first order Taylor expansion.

The conversion of measured methane enhancement from ppm units by volume in dry air to g m^–3 needed for eq was performed using the ideal gas law as determined by eq

$$\mathrm{\Delta }[{\mathrm{CH}}_4](\mathrm{g}{\mathrm{m}}^{-3})=\mathrm{\Delta }[{\mathrm{CH}}_4](\mathrm{ppm}){·10}^{-6}\frac{M·P}{R·T}$$ 2

where Δ­[CH₄] (ppm) is the enhanced concentration in ppm, M is the molar mass of methane (16.04 g mol^–1), P is atmospheric pressure (Pa), R is the universal gas constant (8.314 J mol^–1·K^–1), and T is the ambient temperature in Kelvin.

For the analysis, emission sources were geolocated using GPS coordinates corresponding to the fan exhausts. Time series of Δ­[CH₄] were mapped and analyzed to identify candidate plumes, defined as two or more contiguous elevated readings. Plumes within 100–900 m of the emission source were selected for initial qualitative analysis using three key metrics: (1) the maximum Δ­[CH₄]_max observed within the plume; (2) the integrated plume Area (A, ppm·m), calculated as A = ∑(Δ­[CH₄]_,i × L _i ), where L _i is the detection length; and (3) the plume Kurtosis index (κ, m), defined as the ratio Δ­[CH₄]_max/A, providing insight into plume shape and peakedness. In addition, to ensure accurate source attribution, plumes were confirmed to align with the wind direction vector and result from overlapping vehicle transects. Only plumes exhibiting coherent spatial alignment and elevated kurtosis were used for emission modeling.

A regression analysis was performed using Microsoft Excel Solver v. Sixteen by minimizing the sum of squared residuals (SSR) between modeled and measured methane concentrations, as defined in eq

$$\mathrm{SSR}=\sum _{i=1}^n{({\mathrm{\Delta }[{\mathrm{CH}}_4]}_{\mathrm{measured},i}-{\mathrm{\Delta }[{\mathrm{CH}}_4]}_{\mathrm{model},i})}^2$$ 3

The model was evaluated using more than one measurement locations positioned at approximately equidistant crosswind distances from the plume centerline, within a downwind range (e.g., 100–900 m from the fan). The optimization procedure involved adjusting the crosswind offset of the measurement locations from an initial estimate, under a crosswind distance constraint applied within the framework of the Gaussian plume dispersion model. The lateral concentration profile in the plume model is described by the exponential term (e^{(y 2/2σ_y )}), which characterizes the decay of concentration with increasing crosswind distance from the plume centerline. Significant concentrations are typically confined within ±2σ _y to ±3σ _y of the centerline. Therefore, the regression constrained the distance to the region where the concentration remains above 5% of the peak value, which is represented by values of y = 2.5σ _y . Beyond this range, the contribution to the measured concentration is considered negligible and was excluded from the regression analysis.

Following the SSR minimization, an analysis was conducted by exploring a range of emission rates above and below the initial estimate to assess the robustness of the optimization and build confidence in the resulting emission rate. This approach enabled the determination of the most probable methane emission rate consistent with the observed enhancements, given the known wind speed and spatial configuration of the sampling transects.

Nonlinear Uncertainty Propagation

The propagated uncertainty in the emission rate Q _e is propagated using the standard first-order Taylor expansion for differentiable nonlinear functions. In this framework, the total variance is a form that remains valid for all variables (r _i ) even when the governing expression contains exponential terms. − A complete derivation for the Gaussian plume formulation (eq ) is provided in the Supporting Information. Because Q _e is differentiable with respect to u, σ _y , σ _z , y, z, and H, derivative-based propagation is mathematically consistent with standard atmospheric dispersion analysis. Uncertainties in wind direction (θ) and plume travel time (t̅), which do not appear explicitly in eq , influence the plume geometry and transport through their effects on y and u, respectively; these contributions are incorporated into Δy and Δu using the chain rule following established uncertainty-propagation practice. Numerical evaluation of eq yields propagated uncertainties ΔQ _e /Q _e of 41.1% for Cardinal Wolf Hollow, 62.4% for Cardinal Nebo, 50.2% for E4–1, and 41.8% for the ANR gas pipeline.

$${({\mathrm{\Delta }Q}_e)}^2=\sum {\left(\frac{{\partial Q}_e}{{\partial r}_i}\times {\mathrm{\Delta }r}_i\right)}^2$$ 4

GHGSat Airplane-Based Measurements

GHGSat employs advanced aircraft-mounted near-infrared spectrometers to measure atmospheric methane concentrations, supporting the documentation of anthropogenic emissions. Under the grant supporting this work, GHGSat provided high-resolution measurements with spatial detail finer than 1 m² per pixel and detection thresholds ranging from 10–35 kg h^–1, optimized for cloud-free, low-wind conditions.

Precise location data for ventilation fans, belt portals, and coal piles were provided for 22 selected mines (Table S1, Supporting Information). The Straight Creek mine was excluded due to persistent cloud cover during four survey attempts in November 2021. The remaining 21 mines were surveyed at least twice, with 10 surveyed three times and 2 surveyed four times. Flights were conducted at ∼3000 m altitude and 120 knots airspeed, yielding an across-track swath width of ∼750 m (Figure S3, Supporting Information).

The aircraft was equipped with a compact fixed-cavity Fabry–Pérot imaging spectrometer, which detects methane by analyzing solar backscattered radiation in the shortwave infrared (SWIR) range of 1630–1675 nm. During each overflight, overlapping two-dimensional images were captured, with methane absorption appearing as spectral rings sensitive to column density. Approximately 200,000 spectra were used to calculate methane column densities for ground cells of ∼0.75 m².

These data were geolocated using SWIR surface reflectance imagery to generate Δ­[CH₄] maps, which were compared to known emission sources such as ventilation fans. The determined Δ­[CH₄] in the column, expressed in mol m^–2, was calculated relative to a nominal background of 0.67 mol m^–2 (∼1.90 ppm).

Emission rates were estimated using plume inversion modeling based on the integrated mass enhancement method, which combines observed plume characteristics with wind data − Wind speed and direction were obtained from NASA’s GEOS-FP meteorological data set, consistent with prior applications. Wind speed at 10 m height was sourced from the ECMWF ERA5 reanalysis product, which provides hourly data at a spatial resolution of approximately 0.25° × 0.25°. Reported uncertainty estimates for this method range from 44% to 73% in this study, which are comparable to previous related work. −

sUAS-Based Measurements

Vehicle-based scouting enabled safe deployment of a modified DJI S1000 octocopter, equipped with an Aeris Pico spectrometer (Figure S4, Supporting Information), to conduct methane measurements at ∼21 m altitude (above ground level) near mine infrastructure. Each ∼15 min flight, operated semiautonomously via a Pixhawk autopilot, completed horizontal transects at a constant speed of ∼5 m s^–1, which enables the use of time-integrated concentration values as a proxy for spatial integration. Meteorological parameters, including pressure (10 ms, ±1.5 hPa), temperature (1 s, ±0.3 °C), and relative humidity (0.6 s, ±0.5 %RH), were recorded at 1 Hz using an iMet-XQ sensor. Wind speed and direction were registered via four Calypso ULP ultrasonic anemometers mounted equidistantly in the rotor plane to minimize rotor wash influence, then fused into a motion-corrected 3D wind vector using a custom algorithm. Gas samples were drawn into the spectrometer through 1/4 in. PTFE tubing extending 60 cm beyond the rotor tips, with the spectrometer aligned and calibrated prior to launch. Additional details are provided in the Supporting Information.

Measurements at the Straight Creek mine were unsuitable for the standard Gaussian plume dispersion model due to several critical limitations. First, sampling occurred within 50–71 m of the emission source, a range where Gaussian models are known to be unreliable. Second, the site’s complex terrain, dense vegetation, and misalignment between the sampling path, emission source, and prevailing wind direction further compromised Gaussian plume model assumptions. Third, emissions were below the EPA GHGRP reporting threshold, suggesting lower-intensity plumes. The limitations of the Gaussian plume model at the Straight Creek site, specifically the proximity to the source (<100 m) and the influence of complex topography, necessitated a transition to a mass-balance flux-plane approach. This method treats the sUAS flight path as a ‘sampling plane,’ integrating the total mass passing through it and avoiding reliance on empirical dispersion coefficients (σ _y , σ _z ), which are highly uncertain in complex terrains and at short distances. While the flux-plane approach shares the steady-state emission assumption (∂[CH₄]/∂t = 0) with the Gaussian model, this assumption was supported by consistent time-integrated concentrations from multiple plume detections, providing confidence that the source operated in a quasi-steady state during the observation window. The flux-plane approach adopted for the Straight Creek analysis is physically more robust in near-field environments because it does not require the plume to conform to a Gaussian distribution, which is often distorted by local obstacles or low-altitude turbulence.

The most significant plumes provide consistent measurements and were used to estimate the methane emission rate from the fan by using the integrated peak areas per unit time (ϕ_CH4 ), ppm s, which were converted into g s^–1 units using the same principles than in eq for measured P = 1.00 bar, T = 305.05 K (1 ppm of CH₄  6.32 × 10^–4 g m^–3). The flight and environmental parameters were: sUAS flight path, 97 m in length flown in a straight line; altitude: 21 m above ground level and constant; wind direction, from southeast to northwest hitting the rough terrain and bouncing to the west transport of the methane plume to intersect mainly the last 32 m of the sUAS’s path. The cross-sectional area was A _CS = 775.356 m², which considered the part of the transect intercepting the plume and the distances to the point source. The rationale for using a cross-sectional area stems from observing emissions that are dispersed across an area connecting the vertical emitting point source at 6 m above ground level and the downwind linear path segment of the sUAS. The methane mass flux (Q _e,flux) of each of the four plumes was calculated by integrating the measured enhancement across a virtual cross-sectional plane (A _CS) downwind of the source during a time interval Δt = 1 s. The relationship to calculate Q _e,flux is defined by eq

$$Q_{e,\mathrm{flux}}=\frac{{\mathrm{\Phi }}_{{\mathrm{CH}}_4}·{10}^{-6}·M·P·u·A_{\mathrm{CS}}}{\mathrm{\Delta }t·R·T}$$ 5

where Φ_CH4 = ∫₀ Δ_CH4 dt is the time integrated methane plume enhancement (area under the peak), u = 0.897 m s^–1, P = 1.00 bar, T = 305.05 K, M = 16.04 g mol^–1 for CH₄, and R = 8.314 J mol^–1·K^–1.

The propagated error for the mass flux estimated with the sUAS for the selected plumes at Straight Creek used eq

$$\left(\frac{{\mathrm{\Delta }Q}_{e,\mathrm{flux}}}{Q_{e,\mathrm{flux}}}\right)\approx \sqrt{{\left(\frac{\mathrm{\Delta }{\mathrm{\Phi }}_{{\mathrm{CH}}_4}}{{\mathrm{\Phi }}_{{\mathrm{CH}}_4}}\right)}^2+{\left(\frac{\mathrm{\Delta }u}{u}\right)}^2+{\left(\frac{\mathrm{\Delta }{A}_{\mathrm{CS}}}{A_{\mathrm{CS}}}\right)}^2+{\left(\frac{\mathrm{\Delta }\theta }{\theta }\right)}^2+{\left(\frac{\mathrm{\Delta }t}{\mathrm{t̅}}\right)}^2}$$ 6

for a breakdown of relative error for the four selected plumes of ΔΦ_CH4 /Φ_CH4 , Δu/u, ΔA _CS/A _CS, Δθ/θ, and Δt/t̅ of 3%, 10%; 5%, 12.5% and 31.2%; respectively, corresponding to a total uncertainty of 35.1%.

Bias Analysis

To evaluate the consistency of emission rate estimates across observational platforms, a bias analysis was conducted using the Aeris Pico data as a reference. This approach systematically quantifies the average deviation relative to vehicle- and sUAS-measurements versus aircraft-based methods and the GHGRP values. The mean bias (MB) is calculated as the mean difference between paired observations based on eq

$$\mathrm{MB}=\frac{1}{n}\sum _{i=1}^n(Q_i-R_i)$$ 7

where Q _i represents the emission rate measured by a given platform, R _i is the corresponding Aeris Pico reference value, and n is the number of matched observations. A positive bias indicates systematic overestimation, while a negative bias suggests underestimation relative to the reference. This analysis enables the identification of platform-specific tendencies and supports the interpretation of cross-platform discrepancies in emission quantification. The normalized mean bias (NMB) expresses the MB deviation as a percentage of the reference according to eq

$$\mathrm{NMB}=\frac{\frac{1}{n}\sum _{i=1}^n(Q_i-R_i)}{\sum _{i=1}^n{R}_i}\times 100\%$$ 8

To capture both systematic and random errors, the root-mean-square error (RMSE) was also calculated with eq

$$\mathrm{R}\mathrm{MSE}=\sqrt{\frac{1}{n}\sum _{i=1}^n{(Q_i-R_i)}^2}$$ 9

Additionally, linear regression analysis of measurements by each method versus the Aeris Pico data is used to evaluate if there is a direct correlation to assess proportional and additive biases, providing insight into the consistency and reliability of each measurement platform relative to the Aeris Pico reference. Detailed calculations for bias analysis are provided in the Supporting Information.

Results and Discussion

EPA, MSHA, and EIA Database Analysis

Data from 2011 shows that 24 underground mines in Kentucky reported methane emissions to the EPA GHGRP. However, by 2020, only 4 mines submitted emissions reports. Such a decline in reporting may result in undercounted emissions in this region as emissions from small, nonreporting emitters (those emitting <25,000 t CO₂-e per year ≡ 100 kg CH₄ h^–1) constitute a large share (>70%) of active mines contributing to total emissions. Nevertheless, overall emissions from underground mines in Kentucky and many other mining states will likely continue to decrease due to the drop in the number of active mines especially those with sufficiently large enough methane emissions to be required to report to EPA (>25,000 t CO₂-e per year).

This reporting gap highlights the need for independent, high-resolution monitoring strategies capable of detecting emissions from both large and small sources. Our multiplatform approach that combining sUAS-, vehicle-, and airplane-based spectroscopy enables the determination of methane emissions across a wide range of spatial and temporal scales. This is particularly valuable for identifying emissions from small nonreporting mines that fall below the EPA’s 25,000 t CO₂-e per year threshold but may still contribute significantly to regional totals. By capturing intermittent or spatially diffuse sources that are often missed by conventional inventories, this approach helps close critical data gaps and supports more accurate methane accounting in coal-producing regions like Kentucky.

For mines reporting to the EPA in this study, methane emissions’ rates (kg h^–1) can vary up to 62× between quarters for Paradise #9 mine (Figure ). When analyzing the spatial distribution of emission sources in a mine, the reported emission locations may represent below-ground rather than surface locations. Additional information (e.g., from the operator or Kentucky mine maps) is needed to corroborate surface emission locations. Emission magnitudes can also vary up to 20× for different sources in a mine.

1.

Annual coal production (short tons ≡ 907.18 kg) versus methane emissions (tonnes ≡ 1000 kg in the left axis) for study mines surveyed by aircraft and/or vehicle. Emissions data correspond to the period 2011–2019, during which the Dotiki, Paradise #9, Genesis, and E4–2 mines became inactive. Open and filled symbols represent mines in western and eastern Kentucky, respectively. The short-dashed black line shows the linear fit across all mines (see text), with long-dashed black lines indicating the 95% confidence interval. The dotted brown line marks the EPA’s methane emissions reporting threshold.

Of the 14 active mines studied, 13 reported coal production throughout the fourth quarter of 2021. Methane emissions are generally positively correlated with coal production, but the amount of methane emitted can vary significantly for similar coal production levels (Figure ). This variation is especially pronounced in eastern Kentucky mines. Western Kentucky mines tend to produce more coal and methane emissions compared to eastern Kentucky mines. The dash line in Figure corresponds to a linear fitting of the collective annual data presented. The line is described by eqs and

$$\mathrm{annual}\mathrm{methane}\mathrm{emissions}({\mathrm{CO}}_2‐\mathrm{e},\mathrm{tonnes})=8.18\times {10}^3+5.69\times {10}^{-2}\mathrm{CP}(\mathrm{tons})$$ 10

$$\mathrm{methane}\mathrm{emissions}(\mathrm{kg}{\mathrm{h}}^{-1})=3.343\times {10}^{-3}+2.3\times {10}^{-4}\mathrm{CP}(\mathrm{tons})$$ 11

which relate coal production (CP) with emissions both with a coefficient of determination R ² = 0.674, reflecting a moderately strong linear relationship. The long-dashed lines show the 95% confidence interval for the fit, providing a statistical estimate of the uncertainty around the regression line. This interval represents the estimated range within which the true linear relationship between annual coal production and methane emissions is expected to lie, with 95% confidence, based on the variability observed in the data.

As noted, EPA methane emission inventory rates varied significantly across quarters, with the Paradise #9 mine exhibiting the largest variation (62×) between quarters. To better understand this variability, quarterly data was analyzed for Paradise #9, revealing that emission variability (CV%) ranged from 122% to 176%, with a mean of 136.5% and a standard deviation (SD) of 26.4%. This high variability was not consistently mirrored in flow rate (CV% range: 78–96; mean: 89.5%; SD: 7.9%) or CH₄ molar ratio (CV% range: 73–130; mean: 102.3%; SD: 28.0%), suggesting complex emission dynamics. Correlation analysis showed that emissions were strongly associated with flow rate in Q2 (r = 0.88), but only weakly in Q3 (r = 0.35), where CH₄ molar ratio instead showed a strong correlation (r = 0.94). These findings indicate that both operational and atmospheric factors may influence emission patterns. Additionally, when analyzing the spatial distribution of emission sources, reported locations may reflect below-ground infrastructure rather than surface emission points. Further information (e.g., from mine operators or Kentucky mine maps) is needed to verify surface-level emission sources. Within individual mines, emission magnitudes can also vary by up to 20-times between different sources.

To estimate emissions below the EPA GHGRP reporting threshold, we consider the seven active mines shown in Figure that reported values under this limit. Combined annual emissions from these mines are approximately 4064 tonnes of CH₄ (115,824 tonnes CO₂-e). Although this represents an upper-bound estimate, it highlights the substantial contribution of small mines to regional methane budgets and reinforces the need for independent, high-resolution monitoring strategies.

Only 10 mines in this study reported methane emissions to the EPA over the past decade (Table S1, Supporting Information). This includes 7 mines that have reported quarterly methane emissions for at least 3 years from 2016 to 2022. Five mines (Genesis, Paradise #9, #9A, Dotiki, and Mine #4) that previously reported emissions are no longer active. Examples from two mines are discussed next to illustrate the spatial and temporal aspects of the emission’s reports.

The No. 77 mine is a long-lived mine in Perry County with over 39 years of coal production reported to MSHA. EPA data shows emissions (Figure ) that are from four subsurface sources (with large magnitude differences) but channeled into a central ventilation fan oriented horizontally on the surface. Therefore, it is expected that any generated methane plume would coincide with this single surface ventilation fan. This is particularly true since the mine is situated in a narrow, north–south trending valley. Ignoring the zero value for the first quarter in 2019 (Q1 2019) in the data in Figure , there is a 16× variation in the magnitude of estimated hourly emissions by quarter. An even larger temporal variation in methane emissions (62×) is found at the aforementioned Paradise #9 mine in Muhlenberg County. Though no longer active, estimated hourly emissions at Paradise #9 ranged from 31 kg h^–1 in the second quarter of 2017 to 1928 kg h^–1 in the first quarter of 2019.

2.

Estimated No. 77 mine (Perry County) hourly methane emissions by quarter (Q1–Q4) for 2018–2020. Bars represent emissions from individual measurement locations of flow rates and methane molar ratios. The blue column to the right of each quarter indicates the sum of emissions from all locations. For comparison to the EPA data only, the dashed green line (10 kg h^–1) and pink line (35 kg h^–1) represent the detection thresholds reported by GHGSat.

In contrast, the Cardinal mine in Hopkins County reports emissions from two sources (Figure ). Based on geographic names for surface features, the sources, Nebo and Wolf Hollow fans, are separated by 7.5 km. Emissions data over the period from 2019 to 2022 show that the Nebo fan had higher emissions compared to the Wolf Hollow fan in all but one-quarter. The rate of hourly emissions by quarter varied 3× as they ranged from 537 to 1662 kg h^–1.

3.

Estimated Cardinal mine (Hopkins County) hourly methane emissions by quarter (Q1–Q4) for 2018–2020. Bars represent emissions from individual measurement locations of flow rates and methane molar ratios. The blue column to the right of each quarter indicates the sum of emissions from both locations. For comparison to the EPA data only, the dashed green line (Wolf Hollow ∼ 346 kg h^–1, Table ) and pink line (Nebo ∼ 716 kg h^–1, Table ) represent estimated Cardinal’s methane emissions based on the aircraft measurements.

Vehicle-Based Methane Analysis

Of the 15 mines identified for fence-line surveys, two (Cavalier #4 and Blackberry #1) were excluded due to limited road access. Methane anomalies were detected at 9 of the 13 accessible sites, corresponding to a 69% detection rate in this study, consistent with previously reported mobile lab performance (80  ±  11%). In contrast, the airplane-based survey achieved only a 21% detection rate, well below the 55–83% range reported for aircraft platforms. These findings suggest that vehicle-based surveys offer a robust and effective strategy for methane emissions measurement.

Anomalies varied widely, from short segments (∼20 m) to continuous plumes over 2.5 km, with maximum Δ­[CH₄] ranging from 0.29 to 90.3 ppm (Table S2). When plotted on Google Earth, these anomalies were consistently linked to ventilation fans (e.g., Cardinal Nebo, Cardinal Wolf Hollow, Straight Creek, E4–1, D11 Panther, and Riverview). Given that fans are recognized as the primary emission pathway for underground coal mines, the following analysis focuses on quantifying emissions from this infrastructure.

Figure A for the Cardinal mine shows significant plumes detected to the north of the fan with ratios C2/C1 > 0.01 (nonbiogenic source), while Figure B displays in red circles the position of peaks 1 and 2 with Δ­[CH₄]₁ = 30.771 ppm and Δ­[CH₄]₂ = 28.583 ppm. These peaks are located at x ₁ = 196.1 m and x ₂ = 196.3 m to the north northeast of the Wolf Hollow fan (Figure ), the only mine infrastructure methane source at this location. Given the wind speed of u = 2.2 m s^–1, originating from the south–southwest, it is evident that methane emitted from the horizontally oriented fan was transported toward sampling points 1 and 2, as indicated by the cyan arrow in Figure at the time of the survey. The figure also shows the crosswind distances from the wind vector to these points, with y ₁ = 26.9 m and y ₂ = 27.8 m, respectively. All the previous information is included in Table , which also indicates a consistent regression result by the sum of squared residuals, SSR = 4.3 × 10^–23. The Gaussian plume model indicates Q _e = 438.0 ± 180.0 kg h^–1. The previous error rate represents a 41.1% of Q _e and arises from eq and the analysis based on the Supporting Information with uncertainties for Δ­[CH₄]_max, 1%; u, 10%; σ _y , 5%; σ _z , 5%; y, 6.5%; z, 5%; θ, 11.6%; and t̅, 22.0%.

4.

Emissions adjacent to Cardinal Wolf Hollow (Hopkins County) were evaluated based on (A) maximum methane enhancements (Δ­[CH₄]_max, left axes and red trace) and ethane-to-methane ratio (C2/C1 > 0.01, right axes and dashed green trace) of (*) significant peaks registered on 06/30/2022. (B) Pink and blue dashed lines represent the downwind (x _i ) and crosswind (y _i ) distances from the fan emission source (located to the south) to red circles indicating localized methane plumes 1 and 2 with Δ­[CH₄]_max. The downwind vector (cyan arrow) transporting the plume from the fan originates at 194° (relative to the true north). Values for x _i , y _i , and Δ­[CH₄]_max are listed in Table and used to estimate the emission rate as described in the text. Meteorological conditions during measurements: Moderately unstable atmosphere (class B) due to partly sunny skies, temperature of 32.7 °C, relative humidity of 56%, atmospheric pressure of 1.01 bar, with wind from the southwest at 2.2 m s^–1.

2. Parameters for CH₄ Emission Determination for Cardinal Mine at the Wolf Hollow Fan.

peak ID	Δ[CH4]max (ppm)	C2/C1	Δ[CH4]max (g m–3)	x i (m)	y i (m)	σ y (m)	σ z (m)	Q e (kg h–1)
1	30.771	0.027	1.966 × 10–2	196.1	26.9	18.51	23.54	438.0
2	26.583	0.027	1.826 × 10–2	196.3	27.8	18.53	23.56	438.0
3	26.856	0.028	1.716 × 10–2	200.7	28.6	18.90	24.08	438.0
4	29.063	0.027	1.857 × 10–2	207.7	27.5	19.49	24.93	438.0
SSR								4.3 × 10–23
% Error Rate								41.1%

^a1 ppm of CH₄ at 32.7 °C and 1.01 bar is equivalent to 6.39 × 10^–4 g m^–3.

In addition to the Wolf Hollow fan previously discussed, the Cardinal mine operates another ventilation unit (the Nebo fan), which merits comparison. As illustrated in Figure S5 and Table S3 (Supporting Information), two methane plumes with enhancements ranging from 0.459 to 0.671 ppm and a C2/C1 ratio of 0.020 ± 0.005 were detected approximately 250 and 317 m southeast of the Nebo fan, where the prevailing wind intersects J. Ellis Road. These modest concentrations indicate localized emissions, consistent with the fan’s proximity to coal handling operations. The estimated methane emission rate from the Nebo fan is 227.5 ± 142.0 kg h^–1, based on SSR = 1.3 × 10^–13 and a total uncertainty of 62.4%, calculated using eq with component uncertainties of Δ­[CH₄]_max (1%), u (10%), σ _y (5%), σ _z (10%), y (5%), z (10%), θ (6.8%); and t̅ (12.8%).

Methane emissions from E4–1 (Perry County) were quantified on 12/13/2021 via two plumes located downwind to the south–southeast (Figure S6 and Table S4, Supporting Information). Optimization of crosswind distance yielded an SSR of 3.6 × 10^–28, and the resulting emission rate was 299.0 ± 150.1 kg h^–1. The total propagated uncertainty was 50.2% based on eq , with individual contributions from Δ­[CH₄]_max (1%), u (10%), σ _y (5%), σ _z (5%), y (15%), and (10%), θ (14.6%); and t̅ (22.0%).

A notable observation was made along Highway Alt 41 near the Nebo fan (Figure S7 and Table S5, Supporting Information), where an American Natural Resources (ANR) gas pipeline facility was identified as a significant source of fugitive methane. The facility’s emissions were confidently attributed to natural gas, with a high C2/C1 ratio of 1.8 and an estimated rate of 3200.0 ± 1337.6 kg h^–1, supported by a converged SSR of 3.8 × 10^–22. The associated uncertainty was 41.8% based on eq , derived from Δ­[CH₄]_max (1%), u (10%), σ _y (5%), σ _z (5%), y (5%), z (10%), θ (7.0%); and t̅ (39.8%). Methane from natural gas typically contains higher molecular weight hydrocarbons (e.g., ethane, propane, butane), resulting in elevated C2/C1 ratios. In contrast, coal-derived methane is characterized by a predominance of methane and fewer heavier hydrocarbons, leading to lower C2/C1 ratios.

GHGSat Airplane-Based Methane Analysis

Airplane-based methane surveys were moderately successful in detecting Δ­[CH₄] (above a background level of ∼1.900 ppm) at active mines, with detections in 3 out of 14 cases. Specifically, anomaly maps provided by GHGSat show methane plumes linked to exhaust fans at the Cardinal Nebo, Pride, and E4–1 mines (Figures –). The anomalies were mostly 0.200 to 0.300 ppm above background but locally ranged up to about 1.000 ppm above background at the E4–1 mine (Figure ). The Pride mine is the only location recording methane above background on successive flights and features a vertically oriented exhaust fan (Figure and Table ). No anomalies were associated with coal piles or belt portals, and none were detected at nonproductive active or abandoned mines (n = 8).

5.

Airplane methane survey of the E4–1 mine in Perry County conducted on 11/12/2021. The vertical scale shows Δ­[CH₄] (ppm over background level) and the white arrow indicates the approximate wind direction for a wind speed was 3.1 m s^–1. The black field in the left image represents an area with no measurements or imaging. North is at the top of this and subsequent GHGSat images. Adapted with permission from GHGSAT Data and Products 2022 GHGSAT.

7.

Airplane methane survey near the Cardinal mine (Hopkins County) conducted on 11/19/2021. The Δ­[CH₄] originates from a natural gas compression facility immediately southwest of the Cardinal Nebo ventilation fan. The vertical scale shows Δ­[CH₄] (ppm over background level) and the white arrow indicates the approximate wind direction. Estimated wind was 3.1 m s^–1 from the southeast. See Figure and text for further discussion. Adapted with permission from GHGSAT Data and Products 2022 GHGSAT.

6.

Airplane methane survey of the Pride mine (Muhlenberg County) conducted on (a) 11/16/2021 and (b) 11/19/2021. The vertical scale shows Δ­[CH₄] (ppm over background level) and the white arrow indicates the approximate wind direction. On 11/16/2021 the prevailing wind (2.9 m s^–1) was from the southwest and dispersed the plume to the northeast. Subsequently, on 11/19/2021 the wind (1.7 m s^–1) shifted from the southeast and dispersed the plume to the west. Adapted with permission from GHGSAT Data and Products 2022 GHGSAT.

1. Summary of GHGSat Airplane-Based CH₄ Measurements with Comparison to Aeris Pico Source Rate and GHGRP Data.

target	county	date GHGSat	time	latitude	longitude	GHGSat source rate (error) (kg h–1)	date Aeris Pico	Aeris Pico source rate (error) (kg h–1)	available GHGRP (kg h–1)
E4–1	Perry	11/12/2021	12:49:04	37.2229	–83.1548	415 (±252)	12/13/2021	299.0 (±150.1)	214.0 (±18.4)
Pride	Muhlenberg	11/16/2021	15:07:50	37.2959	–87.0548	100 (±58)	N.A.	N.A.	252.6 (±133.5)
Pride	Muhlenberg	11/16/2021	15:17:58	37.2959	–87.0548	132 (±78)	N.A.	N.A.
Pride	Muhlenberg	11/19/2021	11:24:59	37.2959	–87.0548	79 (±45)	N.A.	N.A
Cardinal Nebo	Hopkins	11/16/2021	15:22:28	37.3848	–87.6137	716 (±355)	04/21/2022	227.5 (±142.0)	360.0
Cardinal Wolf Hollow	Hopkins	11/18/2021	14:26:57	37.3947	–87.5730	346 (±151)	06/30/2022	438.0 (±180.0)	524.0
Straight Creek	Bell	N.A.	N.A.	N.A	N.A.	N.A.	06/14/2022	64.8 (±22.5)	67.7
ANR Pipeline	Hopkins	11/19/2021	11:12:34	37.3832	–87.6192	353 (±256)	04/21/2022	3200.0 (±1337.6)	0.06

^aUnited States eastern time zone.

^bGPS coordinates of the plume.

^cObtained from GHGRP database records.

^dFrom data in Figure .

^eEstimated with eq for data in Table S1 (Supporting Information).

Terrain roughness, defined as the standard deviation of elevation within a 1 km × 1 km window, was significantly higher at mines where methane emissions remained undetected (mean = 22.3 m, standard deviation = 2.84 m) compared to detected sites (mean = 12.2 m, standard deviation = 3.73 m). This suggests that greater topographic complexity may hinder plume coherence or sensor visibility from the airplane. Factors such as dense vegetation cover, surface reflectance, and the relative altitude between the source and the sensor can create observational gaps, particularly for midrange emissions. These considerations align with the challenges observed in the present study, where terrain complexity and fan orientation affected plume coherence and sensor visibility, reinforcing the need for adaptive strategies that account for environmental variability when deploying remote sensing platforms. Additionally, methane was detected during three overflights at Pride mine, which employs a vertically oriented fan, while mines with horizontal fans (Cardinal Wolf Hollow and E4–1) were detected only once, highlighting the influence of ventilation shaft orientation. A full explanation of this methodological approach providing the above information is available in the Supporting Information.

Anomalous methane was also measured at two locations not originally provided as targets to GHGSat. These include the ANR natural gas compression facility located approximately 500 m southwest of the Cardinal Nebo fan (Figure S5, Supporting Information) and another ventilation fan (Wolf Hollow, Figure ) located about 3700 m northeast of the Cardinal Nebo fan (Table ).

The plume geometry and distribution of Δ­[CH₄] measured in the maps were used as inputs for estimating methane emissions using inverse modeling by GHGSat (Table ). Emission rates in Table ranged from 79 kg h^–1 (Pride) to 716 kg h^–1 (Cardinal Nebo) based on measurements in Figure . Due to the inherent limitations of the modeling-based approach and the sensitivity to local inputs such as wind speed and direction, these estimates are subject to significant source rate errors that range from 44% to 73% (Table ). Despite the error, model estimates in Table for E4–1, Pride, and Cardinal (Nebo) mines fall within the range of emissions calculated from the EPA quarterly data.

The airplane-based survey maps illustrate important features about methane plumes emanating from the ventilation fans. Notably, the highest Δ­[CH₄] values are often located farther from the source rather than close to it. These pockets of higher Δ­[CH₄] reflect the complex influence of turbulent diffusion, shaped by surface features and the prevailing wind direction. Second, as evidenced by the successive anomalies at Pride mine, the character of methane plumes can change within minutes and certainly between days (Figure ). These temporal and spatial variations in methane distribution align with our observations from vehicle surveys. Plume shapes range from broadly distributed (e.g., E4–1, Figure ) to narrow and linear (e.g., Pride, Figure ), with dimensions ranging from about 30 m (E4–1) to 250 m (Cardinal Nebo).

In reviewing the airplane-based methane surveys, the absence of anomalies at most target mines is surprising for several reasons. First, the airplane measurements were conducted with the spectrometer pointed directly over the target mine infrastructure. GHGSat uses a technique called instrument pointing, in which the spectrometer is pivoted to increase the time it is directed at the methane point source, thereby enhancing the signal-to-noise ratio. By contrast, vehicle measurements along public roads were laterally offset from sources by hundreds of meters.

Second, the EPA quarterly methane emissions reports show that large emitters had methane molar ratios equal to tenths and hundredths of a percent near the emission point. Even with significant dilution, these concentrations are orders of magnitude greater than the subppm anomalies reported by GHGSat for mines with detected plumes. Furthermore, calculated emissions based on EPA data typically exceed the 10 to 35 kg h^–1 emission threshold reported by GHGSat. This is demonstrated in the examples from Cardinal (Figure ) and E4–1­(Figure S6) mines.

Another possible explanation for the lack of detected anomalies is the potential inactivity of some mines during the airplane campaign. However, this hypothesis can be discarded for this study, as MSHA coal production data show that all active mines, except Orchard Branch #89, reported coal production in the fourth quarter of 2021 (Table S1, Supporting Information). It is acknowledged that coal production and emissions data reported on a quarterly basis may not accurately reflect daily changes in methane output. However, for safety reasons, ventilation fans tend to run continuously, which is reflected in the EPA quarterly reports for operational days, typically listing 365 days. If the lack of anomalies were due to mine inactivity, a lower level of anomaly detection would also be expected in the vehicle surveys (Table ).

With point source imaging surveys used in this study, the challenge is to retrieve the concentration of methane in the atmospheric column and to define a near-instantaneous methane plume against a noisy background on a single pass by the airplane. Aside from instrumental considerations, detection thresholds are strongly influenced by surface factors. Many exhaust fans of the mines studied in eastern Kentucky are located adjacent to steep hillsides and vent toward the hillsides (e.g., Figure ). This positioning of the fans seemingly reduces the imaging time of the spectrometer, even with optimized instrument pointing. Moreover, exhaust fans at 82% of the active mines surveyed are oriented and vent horizontally (e.g., Figures , S5, and S6). Retrieving vertical methane column concentrations from horizontally vented plumes near the ground would seem to reduce the ability to retrieve vertical column methane molar ratios, even with high spatial resolution (small pixels). This factor may be the reason why the only exhaust fan where methane anomalies were detected on successive flyovers is the vertically oriented fan at Pride mine (Figure ).

The ability to quantify emissions from a point source over a selected spatial domain and time window depends on the instrument’s detection threshold, the fraction of the domain observed at least once in the time window, and temporal sampling. With a detection threshold of 10 to 35 kg h^–1, it is assumed that GHGSat correctly imaged the targets. This leaves temporal sampling, the probability of detecting an observed source within the time window given the number of observations in the window, as the controlling variable. Although conducted using satellite measurements, Varon et al. demonstrated that time-averaging measurements were necessary to increase the signal-to-noise ratio for detecting emissions from coal mines. Considering the previous discussion about fan position and orientation, successive flyovers to produce time-averaged measurements may improve the future observations of methane emissions.

sUAS-Based Methane Analysis

On 06/21/2022, methane surveys using the Aeris Pico spectrometer mounted on a DJI S1000 octocopter were conducted at the Straight Creek (Bell County) and D-11 Panther (Harlan County) mines. These sites were selected based on prior vehicle surveys indicating anomalous methane emissions from exhaust fans and their suitability for safe, legal sUAS deployment. At Straight Creek, Δ­[CH₄] was measured at 20–22 m above ground level. The D-11 Panther survey was unsuccessful due to extreme heat (41 °C), which compromised spectrometer accuracy. Straight Creek is situated in a narrow north–south valley, with public access via Mill Creek Road within 100 m of key infrastructure (e.g., exhaust fan, Figure a). The sparsely populated area and light traffic allowed a 10 min sUAS survey along the road, consisting of seven north–south out-and-back transects at an altitude of 21  ±  1 m. Methane enhancements were primarily observed during the transects, as shown by the peaks’ variation over elapsed time in Figure b. These peaks represent spatially distributed emissions detected along the outbound and return transects. In other words, such peaks arise from plume dispersion effects influenced by terrain and wind conditions. The consistent C2/C1 ratio of 0.009 indicates thermogenic sourced methane from coal. Survey limitations included restricted access to mine properties and physical obstacles such as powerlines, trees, and narrow road shoulders. Although initial plans considered broader sUAS deployment with site permissions, none were granted (even for abandoned mines) limiting operations to public roads.

8.

(a) Methane emissions adjacent to the Straight Creek site (Bell County, KY) were assessed on 6/21/2022, using sUAS-based measurements. Red circles indicate the transects sampling methane concentration enhancements (Δ­[CH₄]) along the property boundary. Green rectangle indicates the position of the emitting fan. Meteorological conditions during the survey included an unstable atmosphere (stability class A), influenced by vegetation cover and elevated soil moisture, which reduced surface temperature gradients and limited convective turbulence. Ambient temperature was 31.9 °C, relative humidity was 67%, atmospheric pressure was 1.00 bar, and wind was from the south (∼180° relative to the true north) at 0.9 m s^–1. (b) Measured Δ­[CH₄] (red), individual fitted Gaussian peaks (color-coded), and the total fitted plume (dash black). The fitted baseline (solid black) is ∼0.46 ppm above the background.

The peaks shown in Figure b were individually fitted using Gaussian functions, and the cumulative sum of these fitted curves closely approximates the measured Δ­[CH₄] data. Table S6 (Supporting Information) presents a summary of the parameters derived from the Gaussian fitting of major methane plumes along with the corresponding emission estimates for the Straight Creek site based on the methodology described in the Experimental Methods section.

Out of 13 total peaks detected and shown in Figure b, 10 exceeded an arbitrarily established threshold of 10 ppm·s. These include peaks 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, and 13. Summary statistics for this subset indicate a mean integrated area of 26.91 ppm·s, with a maximum of 42.05 ppm·s, a minimum of 11.87 ppm·s, a mean full width at half-maximum (FWHM) of 15.77 s, and a mean peak methane enhancement (Δ­[CH₄]) of 1.87 ppm. Among these, peaks 3, 7, 11, and 12 stand out as the most substantial, each with an area >39 ppm·s. These peaks likely represent either high methane concentrations, extended durations, or a combination of both. For example, peak 4 exhibited the highest concentration (3.21 ppm) but had a relatively short duration (7.85 s), while peak 9 had a low concentration (0.66 ppm) but a long duration (20.38 s), illustrating how duration can compensate for lower concentration in total integrated area.

The largest and most significant peaks 3, 7, 11, and 12 are used to estimate the methane emission rate from the fan emitting point. For this purpose, to estimate the mass flux of methane from the source, the integrated plume areas (ppm·s = ppm in a second) were converted into g/s units using the ideal gas law and the considerations below.

The cross-sectional area was calculated based on a Gaussian plume model under stability class A. The plume cross-sectional area (CS) intercepted by the sUAS was estimated by considering a scenario involving a straight-line sUAS flight at a constant 21 m altitude, with plume dispersion governed by atmospheric conditions. The calculation integrates principles from Gaussian plume modeling and geometric analysis. The flight and environmental parameters were: sUAS flight path, 97 m in length flown in a straight line; altitude: 21 m above ground level and constant; wind direction, from southeast to northwest hitting the rough terrain and bouncing to the west transport of the methane plume to intersect mainly the last 32 m of the sUAS’s path. The cross-sectional area was A _CS = 775.356 m², which considered the part of the transect intercepting the plume and the distances to the point source. This value is critical for interpreting sensor data and estimating methane concentrations within the plume envelope.

For the mass flux, a conversion from measured ppm units to mol/m³ was facilitated using the ideal gas law (eq ), where P = 1.00 bar (≡1 × 10⁵ Pa), T = 305.05 K, R = 8.314 J mol^–1 K^–1, and M = 16.04 g mol^–1 for CH₄. Under these conditions, 1 ppm of CH₄ is equivalent to 0.632 mg m^–3. Then, for each peak the emissions fluxes reported in Table were calculated with eq . The mean methane flux from the four peaks in Table is 64.3 ± 22.5 kg h^–1, representing emissions from Straight Creek. This rate exceeds the GHGSat detection limit shown in Figure for mine No. 77 but remains below the EPA GHGRP reporting threshold of 100 kg h^–1. Notably, Straight Creek did not report emissions to the GHGRP in 2022. The propagated uncertainty, calculated using eq , is 35.1%, corresponding to the ±22.5 kg h^–1 margin. For comparison, Straight Creek produced 148,733 short tons of coal in 2022. Based on the linear fit in Figure , this production level predicts an average methane emission rate of 67.7 kg h^–1, which falls within the experimental uncertainty of the measured value of 64.3 ± 22.5 kg h^–1.

3. Calculated Methane Fluxes (Q _e,flux) at Straight Creek Mine for Individual Peaks Based on Area Integration.

peak ID	area (ppm·s)	Q e,flux (g s–1)	Q e,flux (kg h–1)
3	39.74	17.5	62.9
7	42.05	18.5	66.6
11	39.45	17.3	62.4
12	41.27	18.1	65.3
Mean			64.3
Standard Deviation			2.0

Results of Bias Analysis

To evaluate consistency across observational platforms, bias metrics were calculated using Aeris Pico data as the reference. Mean bias (MB), normalized mean bias (NMB), and root-mean-square error (RMSE), as defined in eqs –, quantify systematic and random deviations for the data in Table S7 (Supporting Information), while regression analysis informs the agreement between GHGSat aircraft measurements, GHGRP-reported data, and Aeris Pico values.

When the ANR pipeline is included (n = 4), both GHGSat and GHGRP substantially underestimate methane emissions (Table S8, Supporting Information). GHGSat yields MB = −583.6 kg h^–1, NMB = −56.06%, and RMSE = 1446.2 kg h^–1, with a regression slope of −0.054, intercept of 513.8, and R ² = 0.198, indicating weak correlation. GHGRP shows even greater underestimation (MB = −766.6 kg h^–1, NMB = −73.6%) and higher RMSE = 1602.5 kg h^–1, though its regression results (slope = −0.123, R ² = 0.636) suggest moderate alignment with Aeris Pico.

In contrast, excluding the ANR pipeline (n = 3) alters the bias profile (Table S9, Supporting Information). In such a case, GHGSat overestimates emissions (MB = 170.8 kg h^–1, NMB = 53.14%) and shows improved correlation (R ² = 0.744), though its regression slope of −1.59 indicates an inverse relationship. GHGRP also shifts toward overestimation (MB = 44.5 kg h^–1, NMB = 13.84%) and maintains a near-unity slope (0.98), with lower RMSE = 103.6 kg h^–1 and R ² = 0.455, reflecting better proportional agreement but weaker correlation than GHGSat.

These findings underscore the impact of data processing choices (specifically the inclusion or exclusion of the ANR pipeline) on bias outcomes and regression behavior. The comparative analysis highlights the complementary strengths of each observational platform. Vehicle-based and sUAS-based methods offer high temporal resolution but are constrained by limited spatial coverage. , In contrast, aircraft-based measurements provide broader spatial integration, though they are episodic and sensitive to operational constraints such as weather and cloud cover. These platform-specific characteristics reinforce the importance of integrating diverse observational strategies to achieve robust and comprehensive assessments of methane emissions. Future work should prioritize cross-platform validation through colocated, synchronous observations and robust statistical analyses to assess agreement across data sets. This approach can reveal spatial and temporal coherence in emission patterns, while also identifying platform-specific biases and resolution constraints, which would ultimately enhance uncertainty quantification and the integration of multiplatform observations.

Environmental Implications

While coal production has flattened and even decreased in advanced countries, in developing regions (e.g., China, India, Indonesia) it continues to increase. Globally, about 9 billion tonnes of coal were produced in 2024 and in 2022 methane emissions from coal exceeded 30 million tonnes. , Mitigating methane emissions from coal production is challenging and accurately and efficiently measuring emissions is part of the challenge.

The work in this study, which uses historic data from the EPA GHGRP and new measurements from vehicle- and airplane-based spectrometers, highlights methodological challenges in characterizing methane emissions. Ground-based measurements, such as the quarterly inventory reports used by the EPA, are temporally and geographically constrained and therefore fail to capture shorter-term variation in emissions. Although remote sensing techniques like sUAS, aircraft, and satellites present potential solutions, concerns about their reliability and accuracy persist. For example, airplane-based surveys in our study detected methane anomalies at only 3 out of 14 active mines, while vehicle-based surveys were more effective, detecting anomalies at 9 out of 13 active mines with maximum anomalies reaching 90.3 ppm above background levels. Moreover, individual mine emissions data from the GHGRP suggest some of the mines (e.g., No. 77, Cardinal Nebo) are emitting methane at rates that should, in theory, be detectable with the airplane-based survey. When assessing overall methane emissions for larger coal-producing regions, the challenge of detection may become more important in the future as larger coal mines close and small mines contribute to a larger proportion of emissions.

Understanding methane emissions from underground coal mines requires not only accurate measurement techniques but also a comprehensive view of emission dynamics, safety risks, and environmental consequences. Methane emissions are shaped by geological factors (coal rank), mining operations, and the effectiveness of emission control technologies. This work, while focused on the coal sector, reflects broader patterns observed across oil and gas systems, where emissions vary over time and demand accurate, site-specific monitoring. ,,

In underground coal mines, spatially distributed hazards can emerge from explosive gas accumulations in gob areas. Addressing these risks involves the development and application of technologies for gas explosion prevention and the use of advanced sealing materials to enhance coalbed methane recovery. Incorporating spatially resolved methane emission factors into predictive models has proven effective in coal-rich regions, improving the accuracy of regional emission inventories by accounting for variability across mine types and geological zones. Complementary lifecycle analyses of methane emission intensity in oil and gas production further underscore the value of multiplatform monitoring strategies for enhancing operational safety and minimizing environmental impact.

The different outcomes in this study underscore the importance of utilizing a variety of data sources and techniques to gain a more informed and accurate assessment of methane emissions. That is, redundancy will increase robustness. Notably, the differences in the spatial patterns and characteristics of methane anomalies detected by airplane-based versus vehicle-based surveys have a direct impact on how emissions are modeled. These variations influence the interpretation of emission sources and magnitudes, as demonstrated by the discrepancies observed in modeling results at the Cardinal Wolf Hollow, and E4–1 mines. This difference is illustrated at Cardinal Wolf Hollow and E4–1 mines where modeling, using both airplane- and vehicle-based surveys, reveal significant discrepancies.

For E4–1, airplane surveys estimated emissions at 415 ± 252 kg h^–1, while vehicle surveys reported slightly lower rates of 299 ± 150 kg h^–1 (Figure S5, Supporting Information) that fall in the previous bracket, suggesting a minor underestimation in the vehicle data, likely due to CH₄ transport and plume dilution. Vehicle-based detection accounted for only 72% of the emissions captured by airplane surveys, indicating a methodological agreement. Cardinal Wolf Hollow emissions showed even better agreement between the two methods, with airplane-based estimates of 346 ± 151 kg h^–1 versus 438.0 ± 180.0 kg h^–1 for the vehicle (Figure S6, Supporting Information), highlighting the potential for consistency under favorable weather and operational conditions, as 79% of the emissions can be accounted by GHGSat relative to the vehicle. Moreover, the EPA report provided slightly larger emissions of 524 kg h^–1 for Cardinal Wolf Hollow during the fourth quarter of 2021, which correspond to only ∼1.3× more than those from the airplane and vehicle surveys, reflecting partial fundamental differences in methodologies, assumptions, and the spatial or temporal resolution of data collection.

The characteristics of observed plumes can vary with time of day, though direct evidence from this study is limited. Caulton et al. examined various factors (e.g., instrumental, source height, stability class) that contribute to uncertainty in emissions estimates and note that atmospheric factors likely are the main source of uncertainty. At the Pride mine, GHGSat detected plumes twice on the same afternoon (15:07 and 15:18 on 11/16/2021), and other detections (at Pride, Cardinal Nebo, and E4–1) also occurred in the early to midafternoon. While this suggests that plume detectability is possible under afternoon conditions, atmospheric dynamics such as buoyant turbulence may still influence plume dispersion. Early morning stability, which tends to trap emissions near the surface, could enhance detectability, particularly for weaker sources. Given the short duration of measurement campaigns and the influence of topography on low-altitude wind patterns, diurnal and site-specific variability remain important considerations when interpreting detection success across platforms.

Mine operational variables were not examined in this study due to the lack of access for onsite measurements. For future evaluations of passive measurement and detection technologies, it would be beneficial to obtain access to active mine sites. This would allow for direct comparisons between passive measurements and gas samples collected at mine portals or ventilation fans, ideally during the same time period. Additionally, conducting multiple measurements across different seasons would help account for operational variability and enhance the robustness of the comparisons.

Overall, airplane- and vehicle-based surveys offer direct measurement techniques with comparable results under some circumstances, EPA reports rely on point-source assessments that yield considerably higher estimates. To enhance the accuracy of methane emission quantification, methodological reconciliation is essential. The discrepancy between GHGRP-reported emissions and those measured via airborne or vehicular methods arises from two primary factors. First, methane undergoes rapid dilution upon release, a physio-chemical process independent of measurement techniques that results in lower detected concentrations. For example, GHGRP data at mine fans are typically tenths to hundredths of a percent, while vehicle measurements near mines reach ppm levels, and aircraft detect subppm concentrations. Second, instrumental constraints limit the characterization of transient methane plumes, as measurements are restricted to relatively short durations, typically minutes to tens of minutes. As a result, GHGRP inventory emissions consistently exceed those obtained through direct atmospheric sampling. Similarly, Fiehn et al. observed that their airborne mass balance methane estimates fell into the lower range of multiple inventory estimates for coal mines in the Upper Silesian Coal Basin in Poland. The varying detection success rates indicate that redundancy in measurement techniques is essential to reduce the risk of missing methane sources. This work highlights the urgent need for methodological reconciliation to improve the accuracy of methane emission quantification.

Glossary

Abbreviations

Δ­[CH₄]

anomalous methane levels

C2/C1

ethane-to-methane ratio

CO₂-e

CO₂ equivalent

EPA

Environmental Protection Agency

GHGRP

Greenhouse Gas Reporting Program

GHGSat

Greenhouse Gas Satellite

n

number

ppm

parts per million by volume

sUAS

small uncrewed aerial system

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsestair.5c00346.

• Additional experimental methods information; coal mine study sites; spectrometers deployed for vehicle-based methane surveys; example of flight paths for airplane-based methane surveys (Figure S1–S7); underground mines surveyed for methane emissions, methane emission parameters at additional locations, and results of bias analysis (Table S1–S9) (PDF)

The authors declare no competing financial interest.

John Seinfeld Festschrift Special Issue 2025 VSI.