Downstream natural gas composition across U.S. and Canada: implications for indoor methane leaks and hazardous air pollutant exposures

Previous research has shown that natural gas (NG) leaks from residential appliances are common, affecting greenhouse gas emission inventories and indoor air quality. To study these implications, we collected and analyzed 587 unburned NG samples from 481 residences over 17 North American cities for hydrocarbons, hazardous air pollutants, and organosulfur odorants. Nearly all (97% of) gas samples contained benzene (between-city mean: 2335 ppbv [95% CI: 2104, 2607]) with substantial variability between cities. Vancouver, Los Angeles, Calgary, and Denver had at least 2x higher mean benzene concentrations than other cities sampled, with Vancouver exhibiting a nearly 50x greater mean benzene level than the lowest-concentration city (Boston). We estimate that current U.S. and Canadian emissions inventories are missing an additional 25 000 [95% CI: 19 000, 34 000] and 4000 [95% CI: 3700, 5200] lbs benzene yr−1 through downstream NG leakage, respectively. Concentrations of odorants added for leak detection varied substantially across cities, indicating a lack of standardization. Houston, for instance, had 5x higher mean tert-butyl mercaptan levels than Toronto. Using these odorant measurements, we found that methane emissions as high as 0.0080–0.28 g h−1 and indoor benzene enhancements 0.0096–0.11 ppbv could go undetected by persons with an average sense of smell, with large uncertainties driven by smelling sensitivity, gas composition, and household conditions. We also observed larger leaks (>10 ppm ambient methane) in ∼4% of surveyed homes, confirming that indoor leakage occurs at varying degrees despite the presence of odorants. Overall, our results illustrate the importance of downstream NG composition to understand potential emissions, exposures, and odor-mediated leak detection levels. Given methane’s global warming potency, benzene’s toxicity, and wide variation in smelling abilities, our findings highlight the deficiencies regarding the sole reliance on odorization to alert and protect all occupants from indoor leaks.


Introduction
Multiple studies have measured natural gas-sourced methane (CH 4 ) leaks in buildings, despite the presence of odorants added to gas for leak detection [1][2][3][4].Lebel et al observed natural gas (NG) leakage from 50 of 53 stoves measured, with more than three-quarters of observed emissions taking place while stoves were off [1].Storage water heaters also emit at least 75% of total methane emissions while idle [2], illustrating that ignoring idle fugitive emissions will underestimate appliance emissions.Saint-Vincent and Pekney calculated that 144 Gg yr −1 methane could be emitted from residential homes in the United States (U.S) [5], with higher leakage observed during the heating season [6].Since CH 4 is odorless and leaks can occur near people, federal regulations (49 CFR 192.625) [7] require that NG contain enough odorant to be detectable by people with a 'normal sense of smell' at 1% CH 4 by volume in air (one-fifth of the 5% v/v lower explosive limit; LEL)-implying that if gas were odorized only at this standard, relatively large leaks or smaller leaks in an enclosed area (i.e.yielding indoor concentration <10 000 ppmv CH 4 ) could evade a normal sense of smell.Given methane's role as a powerful greenhouse gas and the reliance on odorants for detecting residential leaks, surprisingly little data are available on odorant concentrations in residential NG.
Three recent studies reported midstream and downstream NG measurements of trace nonmethane volatile organic compounds (NMVOCs), including the known human carcinogen benzene and other hazardous air pollutants (HAPs) [8][9][10].These studies show that NMVOCs are co-emitted in NG leaks, with public health relevance given leak locations indoors in proximity to occupants.NMVOC concentrations in unburned NG collected in California were generally an order of magnitude greater than in Boston, with considerable withincity variation observed in the North San Fernando and Santa Clarita Valleys of California [8].Applying NG stove leakage rates from Lebel et al [1], Lebel et al [8] showed that steady-state off leakage alone could appreciably enhance indoor benzene levels in modeling scenarios with high benzene leakage and low ventilation.However, to our knowledge, no research has assessed if leaks that elevate indoor benzene concentrations would also produce indoor odorant concentrations that could be detected by smell.
Further complicating the efficacy of odorization is the documented variation in human smelling ability to the organosulfur compounds commonly used as NG odorants [11][12][13].A recent critical review showed that reported odor detection thresholds for a typical NG odorant spanned up to four orders of magnitude (i.e. 10 000x) across the general public [11].Furthermore, age is an important determinant for gas leak detection with an estimated 31% of adults aged ⩾70 years misidentifying odorized NG in the 2012 U.S National Health and Nutrition Examination Survey [12,13].And although few studies have examined the effects of COVID-19 on NG detection, early results suggest that patients who have recovered from COVID-19 lose some ability to detect NG odorants [14][15][16][17][18].As the only mandated mechanism for rapid gas leak detection, the efficacy of NG odorization is understudied particularly in light of scrutinized methane emissions budgets and coemitted NMVOCs including benzene [8].
Here, we report gas mole fractions (concentrations) of major NG constituents, HAPs, and sulfurbased odorants from 587 residential NG samples collected across 17 North American cities.We couple results with a literature review of human odor thresholds for common odorants.Our primary aims are: (1) characterize NMVOC and odorant content of NG throughout the U.S and Canada; and (2) quantify theoretical indoor gas leaks that could evade detection and their associated methane emissions and indoor benzene enhancements.

Collection process
We collected 257 samples of unburned NG from stoves in three Canadian and seven U.S. cities, conducted quality control, then combined with 335 samples previously collected from six Californian cities [8] and throughout Greater Boston [9] (table 1).Our total dataset covers 17 metropolitan areas, representing an estimated 41% and 35% of the U.S. and Canadian populations living with gas stoves, respectively.Populations were estimated from gas stove prevalence data from the 2020 Residential Energy Consumption Survey [19] and Natural Resources Canada [20], and population data from the 2020 US Census [21] and Statistics Canada [22].
We used similar sampling methodologies as our previous campaigns [8,9], contracting commercial, nationally accredited external labs to process samples.We sampled gas directly from the stove's gas outlet into lab-provided 1.0-1.4l Silonite-lined canisters using PTFE tubing.HAP content was analyzed via U.S. Environmental Protection Agency's (EPA) TO-15 method, which uses gas chromatography and mass spectroscopy.Samples were analyzed within a 30 d hold time, except for Chicago and Indianapolis samples which were reanalyzed with 69-73 d hold time following a failed initial analysis by one commercial lab.A stability study on subsequently-collected samples indicated that 31-85 d hold time had no noticeable effect on reported concentrations of benzene, toluene, or ethylbenzene (mean percent differences < 15%), though some impact for m & p-xylene, o-xylene, and hexane (mean percent differences: 29, 24, and −35%) (figure S1).Due to this potential bias, xylenes and hexane concentration data from Chicago and Indianapolis were excluded from analysis.

Data quality assurance
At each city we collected a blank prior to turning on the gas and a duplicate NG sample.Ambient blanks were collected from ambient kitchen air without attaching hosing; field blanks were connected to stove outlet without turning gas on.All field blanks were returned clean (i.e.target compounds were nondetects), except for hexane in Denver and Vancouver (146 and 441 ppb; both <0.35% of respective citymean concentrations).Verification of sufficient collection of NG rather than ambient air was verified from nitrogen and oxygen content analyzed by a modified ASTM D-1946 method [23].Samples with especially low methane and high nitrogen were identified via k-means clustering, as conducted previously [8,9].Based on clustering results, a cutoff of >60% methane was applied, removing five samples to yield 252 final samples.Isotopic analysis also verified NG origin (note S1).

Organosulfur odorant analysis
Method ASTM D-5504 was used to analyze 84 samples for 21 individual sulfur-based compounds including common odorants, totaling 33% of newly collected samples and 14% of total samples.A subset of samples per city had odorant analysis due to: (1) cost constraints and (2) a priori expectation of similar odorant levels within each city.Odorant concentrations reported from Boston samples were not included as they were obtained via EPA Method TO-15, which is considered less precise for these compounds than ASTM D-5504 [9].

Residence leak survey
Prior to sample collection, residential site visits entailed a brief leak detection survey including ambient air of the residence entryway, kitchen area, stove, oven, and appliance connection points (table 6) using either a Bascom Turner Gas Rover (model VGI-221) or a Sensit Technologies Portable Methane Detector (model HXG-2D).These surveys were not designed to quantify emissions but rather to emulate a general leak detection survey for near-LEL concentrations that would be performed by licensed appliance technicians (e.g.gas utility employees) or firefighters responding to a reported leak.

Statistical analysis of NG content Trace gas composition
We examined variation of major HAPs (benzene, toluene, ethylbenzene, xylenes [BTEX] and hexane) and odorants across cities via the intraclass correlation coefficient (ICC), which reports the percentage of total variance explained by a categorical variable.Given the substantial variation observed between cities, analyses were conducted within-city Table 2. Literature-derived human odor thresholds for the commonly observed NG odorants TBM, IPM, and DMS.Detection and recognition thresholds (i.e.DT50 and RT50) in this context generally represent the odorant chemical concentration presented to a test panel of varying numbers of participants corresponding to detection performance half-way between chance (no better than guessing) and 100% correct.Our literature review yielded limited results for a 100% odor detection threshold (DT100) and RT50, thus we scaled the DT50 value from Nagata [41], applying the DT50:RT50 [42] and DT50:DT100 ratios reported elsewhere [43].
Odorant DT50 from Nagata [41] RT50:DT50 from Moschandreas et al [42] Estimated RT50 using ratio from Moschandreas et al [ then pooled.We estimated between-city means and uncertainty via bootstrap sampling, wherein for each of 1000 bootstraps each city was re-sampled, then each city was averaged.We also assessed the feasibility of predicting benzene using co-mingled hexane, which is more commonly reported and dominants the downstream C6+ fraction.A small number of samples analyzed for benzene, toluene, hexane and tert-butyl mercaptan (TBM) were below the limit of detection (LOD); given the prevalence of these compounds across all cities (table S2), we substituted LOD/2 for the non-detects, following U.S. EPA recommendations [24].For less-prevalent isopropyl mercaptan (IPM), we substituted LOD/2 in cities with at least two positive detections in other samples, and zero otherwise (for Calgary, Pittsburgh, and Toronto; see table S7).
We examined the influence of sample size on uncertainty via a simulation study on benzene, the focus of this study, (figure S4) and found that increasing the sample size from 10 to 20 gas samples reduced the width of the 95% confidence interval by 30% on average.Given the long-tailed distribution of trace gas concentrations, especially benzene in Los Angeles, smaller sample sizes may lead to underestimates of mean concentration, as illustrated by asymmetric confidence intervals.For benzene, this issue is most prominent in Los Angeles, which had a much longertailed distribution of benzene concentrations than other sampled cities (e.g.Los Angeles had the highest kurtosis of 24, whereas the second-highest kurtosis was 10, in San Diego; table S2).In our simulation study, 64% of 10-sample subsets and 43% of 20sample subsets in Los Angeles underestimated the mean observed in the full sample, whereas the resampled cities with smaller tails showed much smaller trends in the same direction (figure S4).
We estimated the associated total emissions of entrained BTEX and hexane associated with downstream NG leaks for the U.S and Canada.We first estimated the typical HAP:methane ratio as countryspecific population-weighted averages of city-specific ratios, using estimated gas stove populations and bootstrap sampling.These ratios were multiplied by the methane emissions attributed to downstream NG leaks, including estimated residential leaks, according to the U.S. EPA Greenhouse Gas Reporting Program [25] and the Environment Canada Greenhouse Gas Emissions Inventory [26].The Canadian inventory reports in units of 100 year CO 2 equivalents, so methane and CO 2 emissions were decoupled using their fugitive emissions factors for appliances and their global warming potential.

Characterizing leaks that evade odor detection Odorant literature review
We conducted a literature review via Web of Science to identify published, scientific estimates of detection and recognition thresholds for the most prevalent odorants in our samples: TBM, IPM, and dimethyl sulfide (DMS) (see note S2 for methods; table 2 for results).

Determination of human odorant detection and recognition thresholds and odorant activity values
We considered three odor thresholds: DT 50 for median detection ability, a statistically derived concentration at which 50% of a test panel in a highly controlled setting correctly detects an odor, DT 100 for weaker detection ability, the corresponding concentration for 100% of the test panel, and RT 50 for median recognition ability, for recognizing an odor's identity.Reported DT 50 values were traceable to European Reference Odorant Mass (EROM) value defined in EN13725 [27], but due to paucity of reported values traceable to EROM, DT 100 and RT 50 were derived from values in the literature (Note S3).For each city, we characterized the odorant mixture using odorant activity value (OAV; measured odorant concentration divided by DT 50 ) [28] (note S3).

Indoor air quality and emissions modeling
Using empirical NG constituent measurements, we quantified the maximum indoor concentration and associated leakage rates for benzene and methane that could evade detection according to three odor thresholds.We assumed leaked NG had the same mixing ratios of NG constituents as pipeline NG (i.e. the collected NG samples).The maximum indoor concentration of NG that could evade detection is constrained by the odorant content and the odor threshold selected (e.g. a higher concentration of odorant could go undetected according to detection thresholds based on weaker-than-median smelling ability).We first estimated maximum indoor NG concentrations that could evade detection using cityspecific OAVs (table S10) and literature-based odor thresholds (table 2).We then estimated the associated indoor benzene and methane concentrations using city-specific average mixing ratios (equation ( 2), note S4).We then back-calculated the maximum steadystate methane emissions rates that could evade detection, using a set of ventilation and building footprint designs parameterized in CONTAM, a multizone, whole-building air model developed by the National Institute of Standards and Technology [29], (equations ( 3) and ( 4), note S4).In summary,

HAP composition
We detected HAPs in >99% of the 587 samples collected, with the most prevalent HAPs being benzene (97% of samples), toluene (97%), and hexane (99%) (table 3).Overall, we detected 25 different HAPs throughout this study (table S1), with substantial variability within and between cities. Cities high in benzene tended to be higher in other HAPs (figure 1, table S2, and figures S6(a)-(e)).Vancouver, Los Angeles, Calgary, and Denver were among the top five cities for highest mean levels of each BTEX compound and hexane.We focus on benzene given its prevalence, toxicity, and carcinogenicity as an IARC Group I human carcinogen for leukemia [30].ICC analysis indicated substantial variation of benzene within cities (62% [95% CI: 23%, 59%] of total variation (figure 1; SI note 3; table S3).The largest discrepancy was observed in comparing gas in Vancouver to Boston, whereby mean benzene levels differed by more than 50x.Benzene content was generally lower in the Northeast and higher in the West Coast, though this suggestive trend was inconsistent.For example, Los Angeles had one of the highest mean benzene concentrations whereas neighboring Bakersfield had one of the lowest (figure 1).Relatedly, 'city' explained 55% of δ 13 C methane isotope variability (bootstrap 95% CI: 39, 71%; figure S2), suggesting that isotopic signals are relatively distinct between cities and likely from distinct geological origins.This finding may have implications for methane source apportionment, particularly in urban areas where both distribution-grade NG and local hydrocarbon production are co-located.Michanowicz et al also observed seasonal variation in Boston [9], though examining seasonality was outside of the scope of this analysis.While seasonality may influence our reported city means, its magnitude is likely dwarfed by inter-city variation (figure S9) and its direction and magnitude likely vary between cities.
Hexane was a strong predictor of benzene (R 2 > 0.70; table S5) in eight cities, yet weakly associated in some cities (e.g.Calgary's R 2 = 0.2).Relatedly, Washington, D.C., reported substantially higher hexane content-almost 90x-than other cities, though benzene and other heavy consitituents were not similarly elevated, suggesting hexane was uniquely high, potentially due to a unique source.To verify these somewhat anomalous results, we performed a follow-up sampling campaign (at a later date) in Washington, D.C., which generally confirmed elevated NG hexane (see figure S3).

Odorants and implications for HAP and methane emissions
Odorant composition TBM was identified consistently across cities and households (concentration range: 1.1-5.0ppmv; figure 2; table S7).IPM was also detected in six of the eight analyzed cities (figure 2) suggesting the use of pre-mixed odorant blends.Due to its lower detection threshold, IPM was the most active odorant in Denver, Houston, and New York City (table S10).And while in Washington, D.C., DMS was the most concentrated odorant (figure 2), TBM was more active due to its much lower detection threshold (table S10).Houston and Toronto both rely on TBM as the primary odorant, yet the former reported a 500% greater mean TBM concentration.Likewise, neighboring New York City and Washington, D.C. appear to use different odorant products all together.NG odorant concentrations in end-use gas varied greatly-sometimes even registering as non-detect in samples that met QA/QC checks (n = 4), potentially indicating odorant fade.When calculating the efficacy of odorization in individual samples, the OAV ranged from 1740-57 000 indicating the high variability of   functional odorant content delivered to customers.
Assuming measured odorant mixtures and concentrations in end-use gas reflect odorization practices by companies, these large observed differences suggest a lack of industry-wide standardization.Overall, measured odorant NG content aligns with what little previous data exists-with reported concentrations ranging from <1 to ∼10 ppmv [33][34][35]-yet this is the first study to examine NG odorants at this scale.

Odorant detection and recognition threshold literature review
We found high variability in previously reported odor thresholds for TBM, IPM, and DMS, each varying by two to five orders of magnitude (table S9).These variations are suspected to be due to systematic variability in testing protocols and materials, data interpretation, and reporting, and random variability in smelling ability between and within individuals, even within the same panel on separate days [36][37][38].More recent studies suggest that lower reported thresholds are due to improvements in chemical purity, concentration control, and sniff testing protocols [37,39].A recent preprint derived a very low detection threshold for TBM (2.35 × 10 −5 mg m −3 ) from a panel of 41 participants but noted that within the panel individual sensitivity ranged by 1450-fold [40], indicating the importance of individual variation.

Characterizing leaks that evade odor detection: indoor benzene concentrations
For potential benzene exposures from leaking NG, we found that NG-sourced indoor benzene could as high as 0.0096-0.11ppbv could evade average smelling ability (DT 50 ), depending on gas composition and household setting.Higher concentrations would be able to evade thresholds for recognition (0.022-0.27 ppbv) or sub-average ability (0.052-0.39 ppbv; table 4).Given the observed variance in odorant levels within each city, it is notable that a larger range of potential concentrations were observed when modeling with concentrations from individual samples, including potentially very high levels for samples where odorant concentrations were below detection limits (figure S11).Overall, these simulations indicate that on average, indoor NG leaks should be detectable and recognizable via average smelling ability prior to reaching benzene levels above California EPA Office of Environmental Health Hazard Assessment (OEHHA) 8-h REL of 1 ppbv for benzene [44].Notably, the OEHHA REL, like many short-term thresholds, ignores carcinogenic outcomes.The World Health Organization estimates that lifetime exposure of 0.53 ppbv benzene is associated with an excess lifetime risk of 1 cancer case per 100 000 people [45].
Characterizing leaks that evade odor detection: methane concentrations and emissions from modeled indoor gas leaks Average odorization levels were sufficient for detecting leaks when considering the federal regulation requiring detection one-fifth of methane's LEL (49 CFR 192.625) [46].Generally, we observed a safety margin of 10x-100x (depending on city and odor threshold applied; table S11) beyond the CFR.
Using CONTAM, we found that methane leaks as large as 0.008-0.28g CH4 h −1 could evade average smelling ability (DT 50 ) (highest and lowest emissions scenarios in table 5; all scenarios in Tables S11-14).Using a set household size and ventilation rate, leaks in Toronto could grow the largest while evading detection-5x larger than the city with the smallest such leaks, Houston.Larger residence sizes and higher ventilation rates can dilute leaks, allowing relatively larger leaks to evade detection compared to leaks in smaller residences with poorer ventilation.For context, in a previous study we found a mean methane emissions rate of 0.0579 [95% bootstrap CI: 0.0363, 0.0840] g CH4 h −1 during steady-state off among 53 stove/oven appliances [1].We were unable to smell any of the 50 measurable leaks during sampling; however, we do not have odorant data for the cities sampled in that study.Nonetheless, their highest observed emissions rate (0.405 g CH4 h −1 ) would have yielded a 'recognizable leak' in all manufactured home scenarios and even in most single-family home scenarios with average ventilation (0.44 ACH).

Model limitations
First, while we selected odor thresholds based on traceability to EU standards [27,47], other sources have reported much higher odorant detection thresholds (table S9), including scientific literature (0.9-0.00626 ppbv for TBM), odorant manufacturers (0.1 ppbv for TBM-IPM mixture) [48] and the American Gas Association [49] (1.6 ppbv for 'mercaptans,' ∼250x higher than the threshold we applied).In applying the 1.6 ppbv detection threshold, six of the eight cities could have undetectable leaks above 1 ppbv benzene, with a maximum of 15 ppbv benzene in Denver, indicating the sensitivity of results on selected odor threshold (table S11).Second, our primary odor threshold DT 50 should not be viewed as an all-encompassing protective threshold as is commonly used in the public health field.Rather, this threshold value reflects the point at which 50% of a test panel in a highly controlled setting correctly detected this odor.Therefore, leak results in relation to DT 50 applies to only half the population (assuming the original test panel is representative of the population) and fails to encompass the portion of the population that are below average in odor acuity.While we attempted to include more encompassing thresholds (DT 100 and RT 50 ), they are less certain and rely on cross-study deductions.Third, the emissions model included simplifying assumptions and would not necessarily apply to any individual residence or occupant.Fourth, leak detection or recognition does not automatically lead to mitigation-barriers may include awareness of leakassociated hazards, mitigation costs, or low agency for occupants such as renters.

Residence leak survey results
Our sampling team identified numerous odorless stove burner, oven, and kitchen appliance gas line connection leaks, supporting Lebel et al's finding of highly prevalent stove leaks [1].We identified 13 ambient-type NG leaks (defined as CH 4 >5x above outdoor background or >10 ppm) whereby methane concentrations were steadily elevated throughout the interior living space, ranging from 12-244 ppmv (table 6).These leaks were observed in 7 of 11 cities, resulting in a North American leak rate of ∼1 in every 25 homes sampled (4.0%).In Indianapolis, the research team identified a highly elevated ambient methane level (persistently ∼244 ppmv) throughout the living-space, requiring emergency action.This very large leak was immediately recognized by the research team by odor, however the study participant (and tenants of the other three building apartments units) reported not being able to smell the leak.Overall, leaks encountered in the field were not originally anticipated and should be viewed with caution due to relatively poor detection instrument accuracy (±10 ppmv) compared to researchgrade instruments [50].However, our leak detection equipment matches equipment often used by licensed appliance technicians (e.g.gas utility workers) and firefighters.Despite these limitations, our findings generally corroborate that leaks, sometimes substantial, occur despite the presence of odorants in residential NG and align with Lebel et al's finding that fugitive appliance emissions have a longtailed distribution that should be incorporated into risk assessments.

Conclusions and policy considerations
Our sampling results demonstrate that benzene and other HAPs are prevalent throughout residential NG delivered to consumers across North America.Given the wide variation of benzene and odorant content observed in end-use NG, regulators and consumers would benefit from open access to NG composition data to better understand potential emissions, exposures, and odor-mediated gas leak safety levels.Our results indicate that benzene and odorant content can be better understood with randomized and targeted source-specific monitoring at distribution network level, with more frequent monitoring to characterize long-tailed trace gas distributions.Based on the prevalence of indoor NG leaks from stoves [1], water heaters [2], and other gas appliances [3,4], systematic efforts to reduce leakseither through improved leak detection and mitigation or reducing usage altogether-can reduce exposures, yielding indoor air quality and public health benefits.Given methane's potency as greenhouse gas and co-emitted HAPs' health risks, the efficacy of NG odorization as a low-level leak detection mechanism has been understudied.Our empirical measurements and simulation modeling indicate that for a person with average smelling ability, typical odorant levels were sufficient to detect leaks large enough to pose an explosive hazard or exceed California's acute REL threshold for benzene (excluding combustionbased benzene emissions).Nevertheless, leaks that can evade median smelling ability can also yield low, yet persistent benzene concentrations and methane emissions rates between 0.01-0.95g CH 4 h −1 contributing substantially to greenhouse gas budgets.Furthermore, population-level disparities in smelling acuity likely exist mediated by older age [12,13], and infections (e.g.COVID-19) [14][15][16][17][18].This indicates cause for concern regarding the sole reliance on 'normal sense of smell' to protect from NG leaks.
Gas leak detection and recognition can be fortified by improving odorization standards and expanding beyond olfaction.NG odorization policies in Canada and the U.S. could consider more stringent odorization standards similar to EU countries [47,51].Standardization could entail increasing the minimum odorization level as similarly promulgated by Massachusetts [52] who requires a 0.15% combustible gas odorization threshold compared to the ∼1% federal regulation.Finally, readily available gas leak detectors should be required in all buildings with gas appliances akin to smoke alarms or carbon monoxide detectors.Some gas utilities are now providing gas leak detectors, without charge, that automatically alert authorities of any hazardous leak [53].Realtime, highly sensitive methane sensors could alert occupants of smaller leaks that could pose climate and indoor air quality issues.Such detectors may be particularly effective for individuals suffering from odor insensitivity, or settings where high ventilation might dilute leaks, e.g., commercial kitchens.

Figure 1 .
Figure 1.Distribution of benzene in natural gas by city and ranked by city-mean concentration (black horizontal bar).Each point represents an NG sample.Note the discontinuous scale to capture the elevated measurement observed in Los Angeles.Benzene non-detects were substituted with LOD/2.

Figure 2 .
Figure 2. Distribution of organosulfur odorants measured in natural gas by city.For TBM and IPM, samples below the detection limit were assumed to be 1 /2 limit of detections, except Calgary and Pittsburgh, and Toronto where we observed only zero or one detections of IPM.

Table 1 .
Natural gas constituent data collected in this study and previous studies by city.

Table 3 .
Distribution of top hazardous air pollutants measured in NG delivered to residential customers across 17 North American cities.
a Summary statistics for hexane are presented with and without D.C. data due to extremely high hexane concentrations measured in Washington, D.C. and verified through repeat sampling.

Table 4 .
Maximum indoor benzene concentrations associated with natural gas leaks that would evade odor detection at various human odor thresholds.

Table 5 .
Kitchen appliance emissions rates that could evade odor detection/recognition according to varying odor detection thresholds (i.e., undetectable leaks).Units are in g CH4 h −1 .

Table 6 .
Detected indoor-ambient and kitchen appliance natural gas leaks identified via CH4 detection instrument-based survey a .
c Gas connection lines were not surveyed in Boston residences.