Multiple previous studies have attempted to design algorithms to detect fugitive emissions of natural gas, all of which focus on locating enhancements in CH4, the major component of natural gas. These algorithms define an enhancement based on whether CH4 mixing ratios are higher than a certain value (Phillips et al., 2013), above a certain percentile in measured readings (Hopkins et al., 2016, Chamberlain et al., 2016) or by using an outlier detection model (Keyes et al., 2020).

The paper upon which our methodology is based (von Fischer et al. 2017), defines Observed Peaks (OPs) as CH4 enhancements >110 % of a 2.5 min rolling background of the mean CH4 concentrations two minutes before and after each measured point. Additionally, OPs must not cover a distance greater than 160 m. Enhancements occurring within 5 s of each other are grouped together. Mobile surveys are repeated multiple times and Leak Indications (LIs) are determined by grouping OPs that occur within 20 m of one another and determining which of these grouped clusters contain OPs from more than one mobile survey. The LIs are then quantified into emission rates in Lmin-1, using an equation derived from the results of a controlled release experiment, shown in Eq. (1). 1(release rate/Lmin-1)=0.1178+0.08267×M-0.005175×A+0.08626×K where:

M is the maximum CH4 reading

In Maazallahi et al. (2020), it was proposed that the existing methodology categorised certain burning emissions as fugitive emissions. To counter this, an additional stage using CO2 ratios with CH4 was introduced to filter out burning emissions.

Source attribution was also used in Fernandez et al. (2022), using isotopic measurements of CH4 in addition to ethane : methane (C2:C1) ratios.

Most recently in Tettenborn et al. (2025), the approach was changed further, adapting the quantification equation to be based on peak area as opposed to peak height, resulting in the quantification equation shown in Eq. (3), 3(release rate/Lmin-1)=exp⁡(1.292×ln⁡(peak area)-2.377) where ln⁡(peak area) is the mean ln(peak area) of all OPs within the LI cluster.

Variations of this algorithm have been used in many major cities across the USA and Canada (Ars et al., 2020; Weller et al., 2022), Europe (Defratyka et al., 2021; Fernandez et al., 2022; Wietzel and Schmidt, 2023; Vogel et al., 2024) and Asia (Joo et al., 2024, Ueyama et al., 2025, Umezawa et al., 2025). This paper attempts to detect smaller enhancements of methane by adapting detection and clustering parameters to be specific to the limitations of the instrumentation used. The paper also explores the effect of introducing a source attribution filter at the OP stage of the algorithm and how this affects the number and the magnitude of LIs.

The Wolfson Atmospheric Chemistry Laboratories (WACL) Air Sampling Platform (WASP) detailed in Wagner et al. (2021) is the base for these measurements. The sampling inlet for the WASP is located at the front of the van on the driver's side, meaning that the vehicle will sample the centre of the road regardless of direction of travel. Since publication of Wagner et al. (2021), the WASP has been updated to include a Quark-Elec QK-AS07-0183 for GPS readings. For the measurements surrounding natural gas, the WASP was equipped with a Los Gatos Microportable Greenhouse Gas Analyser (MGGA) for measurements of CH4 and CO2, Iterative CAvity enhanced Differential optical absorption spectrometer (ICAD) for measurements of NOx (NO2+NO), and an Aerodyne Tuneable Infrared Laser Direct Absorption Spectrometer (TILDAS) laser trace gas analyser for measurements of ethane (C2H6) (Yacovitch et al., 2014). Measurements of C2H6 were calibrated using a three point calibration of a high standard (17.5 ppb), medium (2.5 ppb) and a zero, where calibration standard concentrations were confirmed via GC-MS. For each mobile survey a calibration was performed before and after the mobile survey itself, a linear regression was performed to find the slope and intercept of the calibration concentrations versus measured concentrations. The average of the two calibrations was taken to account for instrument drift during the mobile survey and the resulting equation, Eq. (4), was used to apply a correction to C2H6 concentrations, 4C2H6corrected=C2H6uncorrected⋅m+c Where:

m = Gradient of calibration concentration vs. mean response averaged over the two calibrations

York is a city in the north-east of England with a population over 200 000. A driving campaign took place over two separate weeks in May and June 2024 resulting in 18 mobile surveys of a “flower petal” route, shown in Fig. 2, staying within the outer ring roads of the A64 and A1237 and focused primarily on sampling residential areas of the city. The majority of the roads sampled on the route were only driven in one direction, but due to the position of the sampling inlet this allowed the middle of the road to be sampled regardless of the direction of travel. The route was driven 18 times as, in order to capture >90 % of emissions, a route should be driven at least 5–8 times over separate days (Luetschwager et al., 2021). The route was chosen as it covers multiple different neighbourhoods within York, but was not intended to be used to compare measurements to the cities emissions inventory as it only covers a small fraction of the total miles of road within the York urban area, 27 mi of a total 507 administered by the local authority (Department for Transport, 2025).

The original algorithm, used in Weller et al. (2019), was adapted following the findings in Sects. 2.1.1 and 2.1.2. OPs were clustered within 1 s as opposed to 5 s. With a faster instrument response, it was expected that the measurements would more readily distinguish between two separate enhancements that occurred spatially close to one another. By clustering over a time of 5 s, assuming an average speed of 20 mih-1 (8.9 ms-1), this would mean the potential to cluster together enhancements 44.5 m apart, whereas a cluster time of 1 s would at most be clustering enhancements 8.9 m apart, the reason for this change was discussed in Sect. 2.1.1. Enhancement criteria was also changed, instead of an enhancement needing to be more than 110 % of the baseline, it must be 105 % of the baseline. This allows detection of smaller enhancements, this change was discussed in Sect. 2.1.2. LI determination occurred after identifying the source type of each OP, ensuring LI analysis occurred only on OPs of the thermogenic source type, to further reduce the chance of comparing long standing thermogenic fugitive emissions with possible nearby single occurrence pyrogenic or biogenic emissions.

To obtain a quantification equation specific to the equipment used in York, a controlled release experiment was conducted at the Bedford Aerodrome over 5 d in May 2024. A MiniCRF was deployed to manage releases of CH4 and ammonia (NH3), while a MidiCRF was deployed for releases of C2H6. In total, there were 41 releases lasting an average of 30 min each. Releases consisted of varying amounts of CH4 (0.2–70.48 Lmin-1), C2H6 (0–7.01 Lmin-1) and NH3 (0–7.87 Lmin-1) to reflect a range of CH4 emission sources, including natural gas and farm emissions. Releases were from a mixture of linear vertical releases, a multi emission point ring, multi point source emissions and single point releases, occurring at heights ranging from ground level to 3 m elevation. Over the course of the experiment wind speeds were measured using four Gill Met Pak Pro instruments deployed at 3, 6, 9 and 12 m elevation, winds were recorded as 1 min vector averages. Average wind speed over the 5 d was 3.87 ms-1 with wind speeds ranging from 0–9.75 ms-1. During each release, an initial period was spent locating the plume before sampling the plume at set distances for 10 repeats, the platform then moved further away in distance for another set of 10 repeats. This continued until the plume was either lost, or a lack of driveable ground was left available. It was noted that larger releases were detectable further away, however, as the data from the controlled release was intended to be used in quantifying under-road and near-road fugitive emissions of natural gas, a maximum distance of 30 m from the point of release was applied for data analysis to reflect the maximum road widths typically found within a city like York (Essex Planning Officers Association, 2018).

Of the 41 releases conducted in the controlled release, only 26 releases were able to be used for data analysis due to several reasons, including that some releases did not have detectable enhancements. Within these 26 releases, 3525 CH4 enhancements were detected over distances between 5.8 and 382.1 m from the release point; the majority of releases detected further away from the release point were from higher emission rate releases. When enhancements were filtered to a maximum distance of 30 m from the release point, this resulted in 1226 enhancements from 23 releases. Density plots of the number of detected enhancements against distance from source are shown in Fig. 3 for all detected enhancements and Fig. 4 for enhancements detected within 30 m from the source.

Source determination using ethane : methane (C2:C1) ratios has been shown to be effective, due in part to the knowledge that C2H6 is present in measurable quantities in thermogenic gas but not biogenic gas (Fernandez et al., 2022). These ratios can be used in order to determine the source of a CH4 emission. Demonstrated in Yacovitch et al. (2014), Lowry et al. (2020), Defratyka et al. (2021), and Fernandez et al. (2022), C2:C1<0.005 may be associated with biogenic sources, >0.005 to <0.09 are thermogenic and >0.1 are considered pyrogenic or combustion. Ideal examples of these relationships are shown in Fig. 7. In order to calculate these ratios, CH4 and C2H6 values must first be aligned in time, due to them being measured on separate instruments, the criterion for time alignment was discussed in Sect. 2.4.2. Additionally, enhancements are removed where the R2 of CO2:CH4 is greater than 0.9 to ensure no combustion sources are wrongly assigned as thermogenic.

18 mobile surveys were conducted across the route of York, the raw data was taken from 10 Hz files for CH4 (MGGA) and C2H6 (TILDAS) and time averaged to 1 Hz data to be of the same response time as the WASPs other internal components (e.g. GPS), a colour map of the measured CH4 concentration is shown in Fig. 8. The data was then processed to remove measurements taken when vehicle speeds were 0 or >40 mih-1 as well as removing data within the area of WACL, as calibrations and other instrument tests were conducted in this location. A rolling 2.5 min median background of CH4 was then applied and enhancements were determined as any CH4 measurement taken that was greater than 1.05 times the calculated background. The enhanced readings were then clustered such that any elevated reading within one second of another were assumed to correspond to the same enhancement. These enhancements were then spatially averaged such that 468 OPs were detected over the course of the 18 mobile surveys.

For each of these OPs the maximum C2H6 value was found from the time of maximum CH4 to 5 s prior. The two instruments' data were then aligned for each OP such that time of maximum CH4 measurement was equal to time of maximum C2H6 measurement. A linear regression was then taken of values from 5 s prior to the maximum methane measurement to 5 s after and a source type was assigned such that C2:C1<0.005 is associated with biogenic sources, >0.005 to <0.09 are thermogenic and >0.1 are considered pyrogenic or combustion.

Of the 468 OPs, 177 (37.8 %) were found to be thermogenic in origin. All thermogenic OPs were then spatially clustered using a 30 m threshold, with the resulting clusters filtered to ensure that each cluster contained OPs occurring on at least two separate mobile surveys, removing any OPs occurring from an event observed during only one mobile survey. The remaining clusters were then averaged into LIs such that the latitude and longitude were calculated as a weighted spatial average, resulting in 24 thermogenic LIs from the 177 thermogenic OPs. Leak rate was determined using the equation present in Sect. 2.4.1 using the mean ln⁡(peak area) of all OPs within each LI cluster. The smallest leak rate was determined to be 0.01 Lmin-1 and the largest being 4.13 Lmin-1, when assigned to bins 2 were classified as small (2–6 Lmin-1) and 22 were very small (0–2 Lmin-1). When the source determination step was omitted, it resulted in 58 LIs with leak rates ranging from 0.01–4.70 Lmin-1, when assigned to bins 9 were small (2–6 Lmin-1) and 49 were very small (0–2 Lmin-1).

While 177 of the 468 OPs were determined to be thermogenic, 39 were assigned as biogenic (8.3 %), 199 were pyrogenic (41.8 %) and 53 were not able to be assigned a source type. NOx:CO2 ratios were investigated for the pyrogenic OPs using the same methodology used for C2:C1 source assignment. 115 of the 199 pyrogenic OPs were able to be analysed in this way, 87 of these 115 OPs (75.7 %) had a NOx:CO2 ratio <0.88×10-3. This implied that the majority of pyrogenic emissions did not originate from traffic, but were more likely emissions from domestic heat and power generation (such as emissions from domestic boilers, Cliff et al., 2025).

Emissions from pyrogenic and biogenic sources were compared to thermogenic emissions at the OP stage on a mobile survey by mobile survey basis due to the high unlikelihood of pyrogenics and biogenics being persistent emission sources, the number of times each source type was detected per mobile survey is shown in Fig. 11.

Thermogenics were the most frequently located source type on 13 of the 18 surveys, with mobile surveys 7, 18, 19, 21 and 22, finding pyrogenic emissions related to heating and cooking were the most frequently occurring source type.

The main alterations to this methodology from that presented in Weller et al. (2019) (and other studies that were based off this method) was the decrease in enhancement criteria from 1.1 times the baseline to 1.05 times the baseline, a decrease in the clustering time window from 5–1 s and the addition of a source determination stage as a core step in the algorithm, as opposed to previous iterations that either had no source determination stage, or one that came later in the analysis. Table 1 shows the effect of each of these changes on the resulting detection of OPs and LIs.

These results show the new methodology could locate more LIs. Binning into the leak rate categories of very small (0–2 Lmin-1), small (2–6 Lmin-1), medium (6–40 Lmin-1) and high (>40 Lmin-1) showed that of the 24 LIs in the new source filtered methodology, 2 were small and 22 were very small. For the 58 LIs of the new non filtered methodology, 9 were small and 49 were very small. For the 27 LIs detected in the original unaltered methodology 10 were small and 17 were very small. Finally, for the 6 LIs detected when applying the source determination step to the original unaltered methodology, 1 was small and 5 were very small. This shows the original methodology, requiring an enhancement of 1.1 times the baseline with 5 s time clustering, misses a large proportion of LIs that the newer methodology, requiring an enhancement of 1.05 times the baseline with 1 s time clustering, detects. A large proportion of these missed LIs occur in the very small category as expected with a smaller enhancement criteria. Source filtering shows that regardless of criteria used, less LIs will be detected with this included in the method. This suggests previous methodologies that do not use this stage may be mischaracterising some thermogenic enhancements as being permanent, as they may instead be detecting methane enhancements of differing source types that occur within the same vicinity of one another.

As previously described, the quantification equation used within this body of work is based on the Tettenborn et al. (2025) approach of using peak area to calculate the leak rate of LIs. However, previous works have used the quantification equation present in Weller et al. (2019) which quantifies release rate based on peak height. This campaign's results were reprocessed using each of these previous quantification equations in order to compare the effects of the updated parameters in the York quantification equation to the original, present in Tettenborn et al. (2025) but also to explore the difference in quantified leak rates from a peak height approach. As previously mentioned, the results of the York quantification approach resulted in the 24 LIs being assigned to leak rate bins such that 2 were small and 22 were very small, the Tettenborn et al. (2025) equation results in 1 medium, 1 small and 22 very small and the Weller et al. (2019) equation results in 1 small and 23 very small. The specific leak rates of LIs calculated with these three equations are presented in box-plots in Fig. 12.

This shows that both the peak area approaches result in a much larger range of calculated leak rates from the LIs than from the peak height approach present in Weller et al. (2019). This suggests that the instrumentation used to detect CH4 enhancements may result in low, wide peaks as opposed to higher, sharper peaks, thus explaining why leak rates are weighted much lower from this method. The Tettenborn et al. (2025) equation appears to be mostly consistent with the equation determined from the York methodology, however there is slightly higher weighting of leak rates with the Tettenborn et al. (2025) equation, resulting in the 24 LIs changing from the assignments of 5 small and 19 very small to 1 medium, 5 small and 18 very small.