Methane mapping, emission quantification, and attribution in two 1 European cities; Utrecht, NL and Hamburg, DE 2

. Characterizing and attributing methane (CH 4 ) emissions across varying scales is important from environmental, 14 safety, and economic perspectives, and is essential for designing and evaluating effective mitigation strategies. Mobile real-15 time measurements of CH 4 in ambient air offer a fast and effective method to identify and quantify local CH 4 emissions in 16 urban areas. We carried out extensive campaigns to measure CH 4 mole fractions at the street level in Utrecht, The Netherlands 17 (2018 and 2019) and Hamburg, Germany (2018). We detected 145 leak indications (LIs, i.e., CH 4 enhancements of more than 18 10% above background levels) in Hamburg and 81 LIs in Utrecht. Measurements of the ethane-to-methane ratio (C 2 :C 1 ), 19 methane-to-carbon dioxide ratio (CH 4 :CO 2 ), and CH 4 isotope composition ( δ 13 C and δ D) show that in Hamburg about 1/3 of 20 the LIs, and in Utrecht 2/3 of the LIs (based on a limited set of C 2 :C 1 measurements), were of fossil fuel origin. We find that 21 in both cities the largest emission rates in the identified LI distribution are from fossil fuel sources. In Hamburg,

the lower emission rates in the identified LI distribution are often associated with biogenic characteristics, or partly 23 combustion. Extrapolation of detected LI rates along the roads driven to the gas distribution pipes in the entire road network 24 yields total emissions from sources that can be quantified in the street-level surveys of 440 ± 70 t yr -1 from all sources in 25 Hamburg, and 150 ± 50 t yr -1 for Utrecht. In Hamburg, C2:C1, CH4:CO2, and isotope-based source attributions shows that 50 26 -80 % of all emissions originate from the natural gas distribution network, in Utrecht more limited attribution indicates that 27 70 -90 % of the emissions are of fossil origin. Our results confirm previous observations that a few large LIs, creating a heavy 28 tail, are responsible for a significant proportion of fossil CH4 emissions. In Utrecht, 1/3 of total emissions originated from one 29 LI and in Hamburg >1/4 from 2 LIs. The largest leaks were located and fixed quickly by GasNetz Hamburg once the LIs were 30 shared, but 80 % of the (smaller) LIs attributed to the fossil category could not be detected/confirmed as pipeline leaks. This 31 issue requires further investigation.  Myhre et al., 2013). In addition to its direct radiative effect, 38 CH4 plays an important role in tropospheric chemistry and affects the mixing ratio of other atmospheric compounds, including 39 direct and indirect greenhouse gases, via reaction with the hydroxyl radical (OH), the main loss process of CH4 (Schmidt and 40 Shindell, 2003). In the stratosphere CH4 is the main source of water vapor (H2O) (Noël et al., 2018), which adds another aspect 41 to its radiative forcing. Via these interactions the radiative impact of CH4 is actually higher than what can be ascribed to its 42 mixing ratio increase alone, and the total radiative forcing ascribed to emissions of CH4 is estimated to be almost 1 W m -2 , ≈ 43 60 % of that of CO2 (Fig 8.17 in Myhre et al., 2013). Given this strong radiative effect, and its relatively short atmospheric 44 lifetime of about 9.1 ± 0.9 yr (Prather et al., 2012), CH4 is an attractive target for short-and medium-term mitigation of global 45 climate change as mitigation will yield rapid reduction in warming rates. 46 CH4 emissions originate from a wide variety of natural and anthropogenic sources, for example emissions from 47 natural wetlands, agriculture (e.g. ruminants or rice agriculture), waste decomposition, or emissions (intended and non-48 intended) from oil and gas activities that are associated with production, transport, processing, distribution, and end-use of 49 fossil fuel sector (Heilig, 1994). Fugitive unintended and operation-related emissions occur across the entire oil and natural 55 McKain et al. (2015) discussed how inventories may underestimate the total CH4 emission for cities. Also, an analysis of 56 global isotopic composition data suggests that fossil related emissions may be 60 % higher than what has been previously 57 estimated (Schwietzke et al., 2016). A strong underestimate of fossil fuel related emissions of CH4 was also implied by analysis 58 of δ 14 C-CH4 in pre-industrial air (Hmiel et al., 2020). These emissions do not only have adverse effects on climate, but also 59 represent an economic loss (Xu and Jiang, 2017) and a potential safety hazard (West et al., 2006). While CH4 is the main 60 component in natural gas distribution networks (NGDNs), composition of natural gas varies from one country or region to 61 another. In Europe the national authorities provide specifications on components of natural gas in the distribution network 62 ( 84   loss from temporarily installed natural gas appliances during big festivals can be the major source of CH4 emissions from such   85   events, while these emissions have not been included in inventory reports for urban emissions.   86 Here we present the result of mobile in-situ measurements at street level for whole-city surveys in two European 87 cities, Utrecht in the Netherlands (NL) and Hamburg in Germany (DE). In this study, we quantified LIs emissions using an

108
The G4302 instrument is a mobile analyzer that provides atmospheric mole fraction measurements of C2H6, CH4, and 109 H2O. The flow rate is 2.2 L min -1 and the volume of the cell is 35 ml (operated at 600 mb, thus 21 ml STP) so the cell is flushed 110 in 0.01 s, which means that mixing is insignificant given the 1 s measurement frequency of the G4302. The additional 111 measurement of C2H6 is useful for source attribution since natural gas almost always contains a significant fraction of C2H6, 112 whereas microbial sources generally do not emit C2H6 (Yacovitch et al., 2014). The G4302 runs on a built-in battery which 113 lasts for ≈ 6 h. The instrument can be operated in two modes at ≈ 1 Hz frequency for each species: the CH4-only mode and the 114 CH4 -C2H6 mode. In the CH4-only mode the instrument has a reproducibility of ≈10 ppb for CH4. The factory settings for CH4 115 and C2H6 were used for the water correction. In the CH4 -C2H6 mode the reproducibility is about 100 ppb for CH4 and 15 ppb 116 for C2H6. For Utrecht surveys (see SI, Sect. S.1.2, Figure S2a), the G4302 was not yet available for the initial surveys in 2018, 117 but it was added for the later re-visits (see SI, Sect. S.1.2, Table S1). For Hamburg (see SI, Sect. S.1.2, Figure S2b Table   119 S2). The time delay from the inlet to the instruments was measured and accounted for in the data processing procedure. The 120 Coordinated Universal Time (UTC) time shifts between the Global Positioning System (GPS) and the two Picarro instruments 121 were corrected for each instrument in addition to the inlet delay (see SI, Sect. S.1.2, Table S1 and Table S2). The clocks on 122 the Picarro instruments were set to UTC but showed drift over the period of the campaigns. We recorded the drifts for each 123 day's survey and corrected to UTC time. The data were also corrected for the delay between air at the inlet and the signal in 124 the CH4 analyzers. This delay was determined by exposing the inlet to three small CH4 pulses from exhaled breath, ranging 125 from 5-30 seconds, depending on the instrument and tubing length. We averaged the three attempts to determine the delay for 126 each instrument and used the delays for each instrument. Individual attempts were 1 to 2 s different from each other. For the 127 G4302 the delay was generally about 5 s and for the G2301 it was about 30 s; the difference is mainly due to the different flow 128 rates. The recorded CH4 mole fractions were projected back along the driving track according to this delay.

129
One-quarter inch Teflon tubing was used to pull in air either from the front bumper (0.5 m above ground level) to the 130 G2301 or from the rooftop (2 m above ground level) to the G4302. To avoid dust into the inlets for both instruments, Acrodisc® 131 syringe filter, 0.2 µm was used for G2301 and Parker Balston 9933-05-DQ was used for G4302. The G2301 was used for 132 quantification and attribution purposes and the G4302 mainly for attribution. After data quality check, a comparison between 133 the two instruments during simultaneous measurements showed that all LIs were detectable by both instruments despite 134 difference in inlet height (see SI, Sect. S.1.3, Figure S3). A comparison between the two instruments during simultaneous 135 measurements showed that all LIs were detected by both instruments despite difference in instrument characteristics and inlet 136 height. In the majority of cases CH4 enhancements for each LI from both instruments were similar to each other. We note that 137 there is likely a compensation of differences from two opposing effects between the two measurement systems. The inlet of 138 the G2301 was at the bumper, thus closer to the surface sources, but the rather low flow rate and measurement rate of the 139 instrument lead to some smoothing of the signal in the cavity. Because of the high gas flow rate, signal smoothing is much 140 reduced for the G4302, but the inlet was on top of the car, thus further away from the surface sources (see Table S3 in SI, Sect. 141 S.1.3). The vehicle locations were registered using a GPS system that recorded the precise driving track during each survey.   Table S5).
(https://www.stedin.net/)) and Hamburg (GasNetz Hamburg (https://www.gasnetz-hamburg.de)) confirmed that full pipeline 163 coverages are available beneath all streets. Therefore, the length of roads in the study area of Utrecht    Table S1 and Table S2). Measurement time periods and survey areas were chosen to select favorable traffic 177 and weather conditions and to avoid large events (e.g., construction; see SI, Sect. S.1.5, Figure S4), which normally took place 178 between 10 -18 LT. Average driving speeds on city streets were in the range of 17 ± 7 km h -1 in Utrecht and 20 ± 6 km h -1 in 179 Hamburg. 180 As part of our driving strategy, we revisited locations where we had observed enhanced CH4 readings (see SI, Sect. 181 S.1.7, Figure S5). Not all recorded CH4 mole fraction enhancements are necessarily the result of a stationary CH4 source. For 182 example, they could be related to emissions from vehicles which run on compressed natural gas, or vehicles operated with 183 traditional fuels but with faulty catalytic converter systems. Later we will discuss how to exclude or categorize these  Table S1 and Table S2). It is possible that pipeline leaks 198 that were detected during the initial survey were repaired before the revisit, and the chance of this occurring increases as the 199 time interval between visits gets longer. In addition to the mobile measurement of C2H6 and CO2 for LIs attributions purposes, samples for lab isotope analysis 202 of δ 13 C-CH4 and δ 2 H-CH4 (hereinafter δ 13 C and δD respectively) were collected during the revisits at locations that had 203 displayed high CH4 enhancements during the first surveys. Depending on the accessibility and traffic, samples were either 204 taken inside the car (see SI, Sect. S.1.8, Figure S6a) using a tubing from the bumper inlet, or outside the car on foot using the 205 readings from the G4302 to find the best location within the plume (see SI, Sect. S.1.8, Figure S6b). All the samples taken in 206 the North Elbe study area and from most of the facilities were collected when the car was parked, but the samples inside the results in a range of mole fractions that allow source identification using a Keeling plot analysis (Keeling, 1958(Keeling, , 1961. Fossil 217 CH4 sources in the study areas of this paper (inside the ring for Utrecht and north Elbe in Hamburg) refers to emissions 218 originating from natural gas leaks.   urban studies, and while this algorithm doesn't include background extraction for CO2, we chose commonly adopted method 240 of background determination for this component. These background signals were subtracted from the measurement time series 241 to calculate the CH4 and CO2 enhancements. For C2H6, the background was considered zero as it is normally present at a very 242 low mole fraction; between ∼0.4-2.5 ppb (Helmig et al., 2016), and is lower than the G4302 detection limit.   (Table 1).

268
The spatial extent of individual LIs was estimated as the distance between the location where the CH4 mole fraction  Figure S10). As mentioned above, in our campaign-type 286 studies not all streets were visited twice, so this criterion was dropped from the CSU algorithm. Instead, we used explicit 287 source attribution by co-emitted tracers.

288
The emission rate per km of road covered during our measurements was then scaled up to the city scale using the 289 ratio of total road length within the study area boundaries derived from OSM to the length of streets covered, and converted 290 to a per-capita emission using the population in the study areas based on LandScan data (Bright et al., 2000). Note that in this  estimates. We applied a standard point source GPDM (Turner, 1969) to quantify methane emissions from these larger facilities.

306
A flowchart describing the steps taken during quantification from facilities in given in SI (Sect. S.2.5., Figure S11). We note 307 that emission quantification using GPDM with data from mobile measurements is prone to large errors (factor of 3 or more ) 308 (Yacovitch et al., 2018) especially when the measurements are carried out close to the source. In this study, we also report the 309 data obtained from larger facilities, since rough emission estimates from facilities can be obtained in the city surveys. Caulton

319
In Eq. (3), C is the CH4 enhancement converted to the unit of g/m 3 at cartesian coordinates x, y, and z relative to the 320 source ([x y z] source = 0), x is the distance of the plume from the source aligned with the wind direction, y is the horizontal axis 321 perpendicular to the wind direction, z is the vertical axis. Q is emission rate in g s -1 , u (m s -1 ) is the wind speed along the x-322 axis, and σy and σz are the horizontal and vertical plume dispersion parameters (described below), respectively.

398
After the LIs were analyzed and quantified, the measurements of C2H6, CO2, and isotopic composition from the air 399 samples were used for source attribution. We characterize the observed LIs as of fossil origin when they had a concomitant 400 C2H6 signal between 1 % and 10 % of the CH4 enhancements and when the isotopic composition was in the range -50 to -40 401 ‰ for δ 13 C and -150 to -200 ‰ for δD. A LI was characterized as microbial when there was no C2H6 signal (<1 % of the CH4 402 enhancements larger than 500 ppb), δ 13 C was between -55 ‰ and -70 ‰ and δD was between -260 and -360 ‰ (  re-visits were carried out several months after first detection, and the LIs were still confirmed (e.g. see SI, Sect. S.1.7, Figure   417 S5).

418
The distribution of CH4 LIs across the cities of Utrecht and Hamburg is shown in Figure 2. As shown in Table 2

438
As mentioned above, the observed total emission rates observed on roads in urban environment in the two cities are 439 relatively similar when normalized by the total amount of km covered, 0.64 L min -1 km -1 for Utrecht and 0.4 L min -1 km -1 for 440 Hamburg (Table 2). Using these two emission factors, the observed emission rates (≈110 t yr -1 in Utrecht and ≈180 t yr -1 in 441 Hamburg) were up-scaled to the entire road network in the two cities, ≈ 650 km in Utrecht and ≈ 3,000 km in Hamburg Table S9 and Table S10). The results cluster mostly in three groups, which are characterized   Table 3 shows the emission rate estimates from the larger facilities in Utrecht and Hamburg. CH4 plumes from the 484 WWTP ( Figure 6 and in SI, Sect. S.1.6., Table S5) were intercepted numerous times during the city transects, and the error 485 estimate in Table 3 Figure 7 shows an example of a fit 487 of a Gaussian plume to the measurements from the Utrecht WWTP. The derived distance to the source was 215 ± 90 m, the 488 hourly average wind speed was 3.5 ± 1.1 m s -1 and the wind direction was 178 ± 5 degrees (see SI, Sect. S.1.6, Table S5).

489
The total emission rate of the WWTP in Utrecht was estimated at 160 ± 90 t yr -1 .

525
Four different approaches were combined in Hamburg for emission source attribution, which allows an evaluation of 526 their molecular consistency. Figure 5 shows that measurements of the C2:C1, δD, and δ 13 C provide a very consistent distinction 527 between fossil and microbial sources of CH4. Except for one outlier with a very enriched δ 13 C and δD contents and no C2H6 528 signal, all samples that are classified as "microbial" and depleted in δ 13 C and δD signatures contain no measurable C2H6.

529
Samples that are characterized as "fossil", based on δ 13 C and δD signatures, bear a C2H6 concomitant signal. This strengthens 530 the confidence in source attribution using these tracers. The fossil δ 13 C signature of bag samples from natural gas leaks in 531 Hamburg (δ 13 C = -41.9 ± 1.0 ‰) is higher than recent reports from the city of Heidelberg, Germany (δ 13 C = −43.3 ± 0.8 ‰ 532 (Hoheisel et al., 2019)). This shows that within one country, δ 13 C from NGDNs can vary from one region to another. These can be used at larger scale, but with the instrument we used we were not able to clearly attribute sources with CH4 544 enhancements of less than 500 ppb. Isotopic analysis by IRMS can attribute sources for smaller LIs (down to 100-200 ppb) 545 but is clearly more labor intensive, and it would be a considerable effort to take samples from all LIs observed across an urban 546 area. Overall, C2H6 and CO2 signals are very useful in eliminating non-fossil LIs in mobile urban measurements and with 547 improvements in instrumentations, analyzing signals of these two species along with evaluation of CH4 signals can make 548 process of detecting pipeline leaks from NGDN more efficient.

549
In Hamburg, most of the LIs were detected in the city center (Figure 1). This means that the LI density is higher than 550 the average value in the center, but much lower than the average value in the surrounding districts and residential areas. Many   Figure S14). This shows that annual natural gas 600 consumption per capita in the US is about 30 % and 40 % higher than in Utrecht and Hamburg respectively. The emission per 601 km of pipeline in Utrecht is between 0.45 -0.5 L min -1 km -1 and in Hamburg is between 0.2 -0.32 L min -1 km -1 . In the US, 602 based on 2,086,000 km km of local NGDN pipeline (Weller et al., 2020), this emission factor will be between 0. these locations (≈ 20 %). A recent revisit (January 2020) to these locations confirmed that no LIs were detected at 9 out of 641 these 10 locations. For the 10 th location a smaller LI was detected in close proximity, and GasNetz Hamburg confirmed that 642 this was a leak from a steel pipeline. The whole pipeline system on this street dates back to the 1930s and is targeted for 643 replacement in the near future.

644
In summary, about 20 % of the LIs including the two largest LIs that were attributed to a fossil source were identified 645 as NGDN gas leaks (see SI, Sect. S.4.2, Figure S18), and were repaired by GasNetz

653
The average C2:C1 ratio for LIs with a significant C2H6 signals across Hamburg was 5.6 ± 3.9 %. For the spots where 654 the LDC found and fixed leaks this ratio was 3.9 ± 2.6 %. Thus, some of the locations where CH4 enhancements were found 655 were influenced by sources with an even higher C2:C1 ratio than the gas in the NGDN. One confirmed example is the very 656 high ratio found in exhaust from a vehicle as shown in Figure S12 (see SI, Sect. S.2.6). The abnormal operation of this vehicle 657 is confirmed by the very high CH4:CO2 ratio of 5.5 ppb:ppm (SI, section S2). This is more than 20 times higher than CH4:CO2 658 ratios of 0.2 ± 0.1 ppb:ppm observed during passages through the Elbe tunnel, a ratio that agrees with previous studies (SI, 659 section S2).

660
Repairing gas leaks in a city has several benefits for safety (preventing explosions), sustainability (minimizing GHG 661 emissions) and economics. Gas that is not lost via leaks can be sold for profit, but gas leak detection and repair is expensive 662 and is usually associated with interruptions of the infrastructure (breaking up pavements and roads). Also, as reported above, 663 and in agreement with the studies in US cities, for small LIs the underlying leaks are often not found by the LDCs, possibly 664 because their equipment is less sensitive and aimed for finding leak rates that are potentially dangerous.

665
Our measurements in Hamburg demonstrate that in particular smaller LIs may originate from biogenic sources, e.g. 666 the sewage system, and not necessarily from leaks in the NGDN. In this respect, attribution of LIs prior to reporting to the 667 LDCs may be beneficial to facilitate effective repair. Figure    facilities shows that these emissions may be equivalent to total CH4 emissions from NGDN leaks in urban environments. In 702 order to analyze discrepancies between spatial explicit measurement-based estimates as presented here with reported annual 703 average national emissions by sectors a coordinated effort with national agencies is necessary to address the lack of publicly 704 available activity data (e.g., pipe material) disaggregated from the national-level (e.g., at the city-level).   738  739  740  741  742  743  744  745  746  747  748  749  750  751  752  753  754  755  756  757  758  759  760  761  762  763  764  765  766  767  768  769  770  771  772  773 774 775 LI Location Colour (Figure 1, Figure 2, and Figure S14) High >7.6 >40 >1.7 Red Medium 1.6-7.59 6 -40 0.3 -1.7 Orange Low 0.2-1.59 0.5 -6 0.0 -0.3 Yellow