Methane Emissions from the Munich Oktoberfest

. This study presents the ﬁrst investigation of the methane (CH 4 ) emissions of a large festival. Munich Oktoberfest, the world’s largest folk festival, is a potential source of CH 4 as a high amount of natural gas for cooking and heating is used. In 2018 we measured the CH 4 emissions of Oktoberfest using in-situ measurements combined with a Gaussian plume dispersion model. Measurements were taken by walking and biking around the perimeter of the Oktoberfest premises (There- 5 sienwiese) at different times of the day, during the week and at the weekend. The measurements showed enhancements of up to 100 ppb compared to background values and measurements after Oktoberfest. The average emission ﬂux of Oktoberfest is determined as (6 . 7 ± 0 . 6) µ g / (m 2 s) . Additional analyses, including the daily emission cycle and comparisons between emissions and the number of visitors, suggest that CH 4 emissions of Oktoberfest are not due solely to the human biogenic emissions. Instead, fossil fuel CH 4 emissions, such as incomplete combustion or loss in the gas appliances, appear to be the 10 major contributors to Oktoberfest emissions. Our results


Introduction
Climate change is a global problem that is having a profound impact on living conditions and human societies. The present global warming is very likely due to strong anthropogenic greenhouse gas (GHG) emissions. The Paris Agreement establishes an international effort to limit the temperature increase to well below 2 • C above pre-industrial levels. A "global stocktake" will revisit emission reduction goals every five years starting in 2023. The EU aims to cut its GHG emissions by 40 % by 2030 20 and by 80 % to 95 % by 2050, compared to the 1990 level. The German climate action plan (Klimaschutzplan 2050) contains similar goals, i.e. to cut at least 55 % of German GHG emissions by 2030 and at least 80 % by 2050.
emissions. Weller et al. (2018) evaluated the ability of mobile survey methodology (von Fischer et al. (2017)) to find natural gas leaks and quantified their emissions. Yacovitch et al. (2015) measured CH 4 and ethane (C 2 H 6 ) concentrations downwind of natural gas facilities in the Barnett shale region using a mobile laboratory. A couple of years later, Yacovitch et al. (2018) investigated the Groningen natural gas field, one of Europe's major gas fields, using their mobile laboratory in combination 25 with airborne measurements. Luther et al. (2019) deployed a mobile sun-viewing Fourier transform spectrometer to quantify CH 4 emissions from hard coal mines. Other studies laid a special focus on city and regional emissions of fossil fuel CH 4 . McKain et al. (2015) determined natural gas emission rates for the Boston urban area using a network of in-situ measurements of CH 4 and C 2 H 6 and a high resolution modeling framework. Lamb et al. (2016) quantified the total CH 4 emissions from Indianapolis using the aircraft mass balance method and inverse modeling of tower observations, and distinguished its fossil 30 fuel component using C 2 H 6 /CH 4 tower data. Wunch et al. (2016) used total column measurements of CH 4 and C 2 H 6 recorded since the late 1980s to quantify the loss of natural gas within California's South Coast Air Basin. Most recently, Plant et al. (2019) reported aircraft observations of CH 4 , CO 2 , C 2 H 6 , and carbon monoxide (CO) of six old and leak-prone major cities along the East Coast of the United States. They found emissions attributed to natural gas are about a factor of 10 larger than the values provided by the EPA inventory. 35 Large folk festivals are also likely sources of anthropogenic emissions of air pollutants, such as nitrogen oxides (NO x ), CO, particulate matter (PM2.5, PM10), Sulfur dioxide (SO 2 ), etc. Huang et al. (2012) investigated the impact of human activity on air quality before, during, and after the Chinese Spring Festival 2009, the most important festival in China. They used potential source contribution function analysis to illustrate the possible source for air pollutants in Shanghai. Shi et al. (2014) measured concentrations of particulate matters and polycyclic aromatic hydrocarbons (PAHs) during the Chinese New Year's 5 Festival 2013 and estimated the source attributions from cooking, vehicle, and biomass and coal combustion. Kuo et al. (2006) investigated PAH and lead emissions from cooking during the Chinese mid-autumn festival. Nishanth et al. (2012) reported elevated concentrations of various air pollutants such as ozone (O 3 ), NO x , and PM10 after the traditional Vishu festival in South India. Nevertheless, up to now, festivals have not been considered a significant source of CH 4 emissions and accordingly, to the best of our knowledge, CH 4 emissions from large festivals have not yet been studied.

10
Oktoberfest, the world's largest folk festival with over 6 million visitors annually, is held in Munich. In 2018, during the 16 days of Oktoberfest, approximately 8-million liters of beer were consumed. For cleaning, dish washing, toilet flushing, etc., about 100-million liters of water were needed. The use of energy added up to 2.9 million kWh of electricity and 200,937 m 3 of natural gas, 79 % of which is used for cooking and 21 % for heating (München (2018a)).
The measurements during our 2017 Munich city campaign indicated Oktoberfest as a possible source for CH 4 for the first 15 time (Chen et al. (2018)). For a better source attribution and a quantitative emission assessment, we have investigated the CH 4 emissions from Oktoberfest 2018 by carrying out mobile in-situ measurements and incorporating a Gaussian plume dispersion model. These measurements and modeling approaches are described in section 2. The results of these investigations show that Oktoberfest is an anthropogenic source of CH 4 that has not been accounted for until now. We have compared the determined total emission flux with bottom-up estimates of biogenic emissions from human, and also present the daily cycle 20 of the emissions. In addition, the week and weekend variations are shown. From these findings we can draw conclusions about the origins of the Oktoberfest CH 4 emissions, which are presented in section 3.

Method
We conducted a mobile survey around the perimeter of Oktoberfest to obtain the CH 4 concentration values around the festival area (Theresienwiese) and incorporated a Gaussian plume model consisting of 16 different point sources to determine the CH 4 25 emission strength.

Measurement approach and instrumentation
The measurements include both CH 4 and wind measurements. The sensors and the way they are used are described in the following.

Concentration measurements
Mobile in-situ measurements were conducted to quantify CH 4 enhancements. To this end, two portable Picarro GasScouters G4302 for measuring CH 4 and C 2 H 6 were used. The sensor is based on the cavity ring-down measurement principle (O'Keefe and Deacon (1988)), using a laser as a light source and a high-finesse optical cavity for measuring gas concentrations with high precision, which is 3 ppb for CH 4 mode with 1 s integration time (Picarro (2017)). We applied a moving-average filter with 5 a window size of 10 s and a step size of 5 s to the 1 s raw measurements. Since the data are averaged over 10 s, the precision is improved to 1 ppb. To distinguish between fossil-fuel related and biogenic emissions, the instrument can be switched to CH 4 /C 2 H 6 mode and measure C 2 H 6 with a precision of 10 ppb for an integration time of 1 s.
Since we were not allowed to enter the festival area due to safety concerns, the measurements were carried out by walking and biking many times around the perimeter of Oktoberfest next to the security fences, wearing the analyzer as a backpack.

10
The measurements were taken on several days during and after the time of the festival to compare the differences in emission strength and distribution. Additionally, to observe the hourly dependency of the emissions, the measurements were distributed over the course of the day. In the end, we covered the period between 8:00 a.m. and 7:00 p.m. (local time) hourly.
For the study, two identical GasScouters G4302 were deployed. One instrument was provided by TNO and the other by Picarro Inc. The former was used in the first week while the latter was used in the second week of Oktoberfest as well as the 15 time after the festival. Although the measurement approach is based on determining the enhancements and not on comparing absolute concentration values, the two instruments were calibrated at the beginning of the campaign.

Wind measurements
In addition to the gas concentrations, wind measurements are vital for estimating the emissions of Oktoberfest using atmospheric models. To this end, a 2D ultrasonic wind sensor (Gill WindObserver II) was placed on a roof close by (48.148°N, 20 11.573°E, 24 m agl.). These wind measurements were utilized for the emission estimates.
To assess the uncertainty of the wind measurements, we compared these measurements with the values reported by an official station of Germany's National Meteorological Service (Deutscher Wetterdienst, DWD). The DWD station (48.163°N, 11.543°E , 28.5 m agl.) is located about 2.8 km away. As this distance is about the radius of the Munich inner city, we assumed that the difference between the two stations is representative for the uncertainty of two arbitrary measurement points in the downtown 25 area, which is also home to Oktoberfest.

Modeling approach
To quantify the emissions of Oktoberfest, we used the measured concentration values as input for an atmospheric transport model. Figure 1. The pre-processed measurement signal (dotted line, moving average with window size 10 s and step size 5 s) is shown along with a low pass filtered version (blue line), which is used to obtain the single plumes (green and red area). The signal in the center is not detected as a plume, as the enhancement is not high enough. The round shown was recorded by bike and took 750 s (12.5 minutes).

Selection Algorithm
For our modeling approach, the plumes of individual surveys (hereafter referred to as "rounds") around the Theresienwiese were evaluated. In total, we completed 94 rounds (69 during and 25 after Oktoberfest). For every round the individual plumes were determined by analyzing a low-pass filtered version of the measurement time series. A Kaiser window (Kaiser and Schafer (1980)) was utilized for the low-pass filtering.

5
Once the signal was filtered, a signal section between two adjacent minima was defined as a plume signal if it had an enhancement of more than 5 ppb. We chose this threshold to be equal to the combined uncertainty of the instrument (3 ppb) and background (4 ppb) (cf. section 2.2.6). This process is illustrated in Figure 1.
When the initial plume selection phase was completed, the identified plumes were further analyzed. As the path of a measurement around the Oktoberfest premises was predefined by the security fence, the location of each point on that route can be 10 converted into a fixed angle, which simplifies the comparison between the measurements and the model. For that purpose, a center point of the Theresienwiese was defined (cf. green dot in Figure 2, 48.1315°N, 11.5496°E). With the help of this point, an angle was assigned to all measurement and model values. This angle was defined similarly to the wind angles, meaning that 0°is in the north and 90°is in the east.
In order to decide whether a measured plume is attributable to emissions from Oktoberfest, a forward model uses the 15 measured wind direction (with uncertainty) to calculate at which angles a plume from Oktoberfest should occur. As can be seen in Figure 3, only plume 1 was selected because the angle range of this plume (green) largely overlaps with the accepted angle range (grey) computed by the forward model of this plume. In contrast, plume 2 (red) has no overlap with the range computed by the forward model; hence, plume 2 was discarded. Additionally, the standard deviation of the wind direction over the time the plume was recorded is taken into account. If the standard deviation is higher than 24°, the plume is not considered, as our approach requires stable wind conditions. Those 24°represent the measurement uncertainty in the wind direction (cf. 2.2.6) and are therefore well suited as a lower limit for filtering out too variable wind conditions. The selection algorithm described above is visually summarized in Figure

Baseline determination
As one measurement round can take up to one hour (when walking), the atmospheric conditions can vary during that time period, which will result in a changing background concentration. Therefore, the baseline for determining the concentration enhancements cannot be calculated solely using a constant value.
In our approach, we assume that the baseline during one round is either rising or falling and that there is a linear behaviour.

5
Such a straight line is clearly defined by two points. For that reason, the time series for each round was divided into two equally sized bins (first and second half). For each half, we determined the lowest 10 % quantile. Afterwards, the mean values of the 10 % smallest concentration values of each bin were used to define one straight line, which was used as the background for that specific round (cf. Figure 5). The uncertainty of that baseline was determined using the CH 4 concentration deviations of the 10 % smallest values from the baseline.

Gaussian plume model
The framework of our modelling approach is based on a Gaussian plume model, which is described in Pasquill (1966Pasquill ( , 1969Pasquill ( , 1979; Gifford (1976); Briggs (1973); Hanna et al. (1982), and widely used in studies for assessing local source emissions  2017)). It is a steady-state model that simulates the processes of diffusion and the transport of emitted trace gases from a point source. The gas disperses such that its concentration distributions fit well to Gaussian curves in vertical and horizontal directions.
For a point source emitting continuously with strength Q (unit: mol s −1 ) at effective height H above the ground and uniform wind speed, the expression for the time-averaged concentration < c(x, y, z) > (unit: mol m −3 ) is given by the formula below: with x, y and z describing the downwind distance, horizontal/cross-wind distance to the x axis and the height above the ground, respectively. u is the time-averaged wind speed, σ y (x) is the standard deviation of the concentration in the crosswind direction and σ z (x) is the standard deviation of the concentration in the vertical direction. These dispersion coefficients describe the spreading of the plume increasing with the downwind distance from the source x. 15 To determine the dependency of σ y and σ z on x, diffusion experiments were carried out (Haugen et al. 1958), which resulted in Pasquill's curves (Pasquill (1979); Gifford (1976)). Smith (1968) worked out an analytic power-law formula for the relationship between σ y , σ z and x. Briggs (1973) combined the aforementioned curves and used theoretical concepts to produce the widely used formulas given in Hanna et al. (1982).
During the measurement periods, the surface wind was lower than 4 m s −1 and the insolation was strong to moderate.
Therefore, stability class A or B was chosen according to the Pasquill turbulence types (Gifford (1976)).
Based on the recommendations by Briggs for urban conditions (Briggs (1973); Hanna et al. (1982)), the relationships be-5 tween the dispersion parameters and the downwind distance are described as: Those relationships were used in our study.

Multiple Gaussian plume model
The concentration measurements using the backpack instrument were performed close to the festival area (< 500 m), which is why the emissions of Oktoberfest cannot be seen as a single point source. For this reason multiple point sources were used. All these point sources were modelled using Gaussian plumes before they were superimposed. The spatially superimposed plumes were detected as a continuous plume signal in our measurement. Later on, these plume signals were utilized for the emission 15 assessment.
Since the emission sources of Oktoberfest were unknown, the locations with the highest density of visitors and with the highest energy consumption were chosen as main sources for the model. Those locations are represented by the 16 biggest beer tents (> 1,000 seats) on the festival premises (cf. red dots in Figure 2). To achieve a good correlation between the model and reality, these 16 tents were not treated equally in the final model. Instead, they were linearly weighted according to their 20 maximum number of visitors. Therefore, the biggest tent (about 8,500 visitors) has, in the end, a more than eight times higher influence on the total emission number than the smallest one (about 1,000 visitors).

Forward modelling approach
The aforementioned multiple Gaussian plume model was used in a forward approach to compare the measured and modeled concentration signals with each other. For that, a predefined route around Oktoberfest was used (cf. yellow route in Figure 2) 25 to determine the concentrations for each angle.
The actual shape of the concentration vs. angle graph c(α) for every selected plume i is considered for the determination of the emission number of Oktoberfest E Okt,i (cf. Figure 6, blue curve). The optimization procedure can be expressed mathematically as follows: where M represents the model. The emission number E i was varied until the areas underneath the modelled and measured curves are the same, and thus the sum of the absolute difference between the model and measurement is minimized.
Practically, we computed the forward model using the averaged wind information at this time and a prior emission number E prior of 3 µg s −1 m −2 and compared it with the measurement curve. In case the shape looks similar (high cross-correlation coefficient), a scaling factor is applied to the prior emission number and varied until the forward model matches the mea-5 surements. This procedure is illustrated for one exemplary plume signal in Figure 6. There, the prior modelled concentrations (orange) are smaller than the measured concentrations (blue). Therefore, the model has to be multiplied with a scaling factor until the areas underneath the modelled and measured curve are the same (yellow). By multiplying the scaling factor k scaling,i with the E prior , the emission number of Oktoberfest E Oktoberfest,i for every plume signal i can be determined as: 10

Uncertainty assessment
To determine the uncertainty of the final emission numbers, we considered the uncertainties of our input parameters. These include uncertainties in the wind and concentration measurements as well as uncertainties in the determined baseline. These input parameters were modelled as Gaussian distributions each. Afterwards, the emission number was determined by running our modeling approach 1,000 times using those four parameters (wind speed, wind direction, measured CH 4 concentration, 15 background concentration) as input. In each run, slightly different input values were chosen randomly and independent from each other out of those four distributions.
The concentration measurement uncertainty is indicated by the manufacturer Picarro to about 1 ppb for an averaging time of 10 s. This value was used as the standard deviation of the modelled input distribution.  The baseline approach described in section 2.2.2 introduces a further error which has to be considered as well. The background concentrations were modelled as a Gaussian distribution where its standard deviation was calculated from the CH 4 10 concentration deviations between the 10 % smallest values of each bin and the baseline shown in Figure 5.
The used parameters for the uncertainty assessment are summarized in Table 1.

Concentration mapping
The measured CH 4 concentrations were plotted for each round on a map of the Oktoberfest premises, to show that there is

Emission number
The average emission number of the Oktoberfest 2018 E Okt, avg is determined by averaging the emission numbers of the N plume signals E Okt,i during the complete Oktoberfest time period (including the weekdays and weekends), accordingly: To make the final emission number more robust and to determine an uncertainty, the basic approach of Eq. 6 was improved.

5
Instead of just using the actual measured data, an uncertainty range was applied to the four main input parameters, each using Gaussian distributions (cf. section 2.2.6).
For every plume signal i, 1,000 samples of randomly chosen input datasets from the aforementioned normal distributions of the input parameters were used to determine 1,000 slightly different emission numbers E Okt, i, k . Using Eq. 6, an average emission number for each realization E Okt, avg, k was calculated: The average emission number including an uncertainty assessment can be obtained by determining the mean µ Okt and standard deviation σ Okt of those 1,000 realizations: The result for the total emission number of Oktoberfest 2018 is shown in Figure 8 and has a value of E Okt,total = µ Okt ± σ Okt = (6.7 ± 0.6) µg/(m 2 s).
To verify whether those emissions were caused by the Oktoberfest, Figure 8 also shows the emissions determined for the time after Oktoberfest (from October 8 th through October 25 th ). This number ((1.1 ± 0.3) µg/(m 2 s)) is significantly smaller than 20 the one during Oktoberfest but still not zero. It indicates that the emissions are caused by Oktoberfest, and the disassembling of all the facilities, which takes several weeks, still produces emissions after Oktoberfest.
After grouping the emission numbers into the two categories, weekday (in total 59 valid plumes) and weekend (26 valid plumes), two separated distributions are visible in Figure 9. The average emission for the weekend ((8.5 ± 0.7) µg/(m 2 s)) is higher than the averaged emission for the weekdays ((4.6 ± 0.9) µg/(m 2 s)), almost by a factor of two. To interpret this 25 result, the visitor trend of Oktoberfest was investigated. This trend is based on the officially estimated numbers of visitors (muenchen.de (2018)) and was linearly interpolated (see Figure 10). Besides the daily trend, it also shows the mean values of the week-and weekend days (dotted lines). As the number of visitors at Oktoberfest was significantly higher on a weekend day (≈ 1.4) than on a weekday (≈ 0.75) (cf. Figure 10), a higher number of visitors results in higher emission, which indicates the CH 4 emissions are anthropogenic.

Daily emission cycle
To assess the daily cycle of the CH 4 emissions, the emission numbers of the plume signals E Okt,i,k are grouped into hourly bins.
Then, for each bin an average emission number E Okt,hour,k is calculated. Afterwards, these numbers are averaged for the 1,000 realizations to obtain robust results including an uncertainty estimate: µ Okt, hour = 1 1000 1000 k=1 E Okt, hour,k , In Figure 11, the variation of the hourly emission mean (µ Okt, hour ) is shown as a blue line. The grey shaded area shows the uncertainty (σ Okt,hour ) of the emission numbers within that hour. The daily emission cycle shows an oscillating behavior overlaid on an increasing trend towards the evening.
The linear increasing trend is in agreement with Figure 10, which shows a linearly increasing visitor amount throughout the 10 day, confirming the anthropogenic nature of the emissions. The oscillating behavior indicates that the emissions are related to time-dependent events, such as cooking, heating and cleaning, which tends to have the peaks in the morning, noon and evening.

Biogenic human CH 4 emissions
To address the question whether the people themselves caused the emissions or whether the emissions were caused by processes related to the number of visitors, such as cooking, heating, sewage, etc., we took a closer look at human biogenic emissions. Most of the previous studies define a methane producer as a person that has a breath CH 4 mixing ratio at least 1 ppm above ambient air (Polag and Keppler (2019)). Keppler et al. (2016), however, used laser absorption spectroscopy to confirm that all humans exhale CH 4 . In that study, the mean of the breath CH 4 enhancements above the background from all test persons (112 persons with an age range from 1 to 80 years) is 2316 ppb and the values vary from 26 ppb to 40.9 ppm.
To take a wider range of literature into account, we have considered the values reported in Polag and Keppler (2019). The 5 authors provided a summary of various studies of human CH 4 emissions in Table 1 and section 3.2, and used these results to calculate average human CH 4 emissions, which are 2.3 mmol d −1 via breath and 7 mmol d −1 via flatus. We multiplied these values with the 300,000 persons that visit the Oktoberfest premises (≈ 3.45 · 10 5 m 2 ) every day. This represents an upper limit of people who are at the Theresienwiese at the same time, as most visitors do not stay all day long. Please note the average emission numbers are not factor weighted by ethnicity, age and sex, because we do not have the respective statistics 10 for Oktoberfest. The expected CH 4 emission from the human breath and flatulence in total was calculated as: Although, we assumed the maximum possible number of visitors, the calculated biogenic component is 22% of the emission we determined for Oktoberfest. Therefore, the emissions are not solely produced by the humans themselves, but by processes that are related to the visitor number.

Emissions from sewage
Besides the direct biogenic human emissions, CH 4 emissions from sewer systems are also possible sources. These emissions are a product of bacterial metabolism within waste water, whose emission strength depends particularly on the hydraulic retention time (Liu et al. (2015); Guisasola et al. (2008)) which represents the time the waste water stays in the system. This time decreases with a higher amount of waste water, as the flow increases in such a case.
At Oktoberfest, the amount of waste water is very high as the 100-million liters of water consumed and the 8-million liters of beer have to flow into the sewer system at some time (München (2018b)). Therefore, the retention time in the sewer system underneath the Theresienwiese is quite low, which makes high CH 4 emissions from sewage unlikely. Furthermore, the waste 5 water consists mainly of dirty water and urine, which does not contain many carbon compounds that are necessary to produce CH 4 .

Fossil fuel CH 4 emissions
The biogenic emissions can likely not fully explain the determined emission number of Oktoberfest. Therefore, fossil-fuel related emissions have to be considered as well. According to the weekday/weekend emission comparison (cf. Figure 9) and 10 the daily emission cycle (cf. Figure 11 compared with Figure 10), there is, in general, a visitor-dependent linear increase of CH 4 emission throughout the day that is superimposed with time-dependent events such as cooking, cleaning or heating. These events can cause CH 4 emissions, as about 40 % of the used energy at Oktoberfest is provided by natural gas that is used for cooking (79 %) and heating (21 %).
As the human biogenic CH 4 emissions have already been excluded due to too small values, leakages and incomplete burning 15 in the gas appliances provide a possibility to explain the emissions. Ethane is a tracer of thermogenic CH 4 , and can be used to indicate a natural gas related source (Yacovitch et al. (2014); McKain et al. (2015)). For that reason, we deployed a portable instrument that is designed to measure CH 4 but that is also capable of measuring C 2 H 6 . Due to the aforementioned safety reasons, the distance between the measurements and the closest point source (tent) was 50 m to 250 m. Therefore, the CH 4 concentration was relatively low (max. 100 ppb). According to the Munich municipal utilities, the C 2 H 6 /CH 4 ratio of natural 20 gas used in Munich is about 3 % (München (2018c)). This results in an C 2 H 6 concentration lower than 3 ppb, assuming that all of the measured CH 4 is sourced from natural gas. Such a small concentration value is lower than the detection limit of the GasScouter (about 3 ppb with 10 s integration time), which is why we were not able to determine the C 2 H 6 /CH 4 ratio of the measured gas.
Nevertheless, it is possible to determine an upper bound for the loss rate of natural gas if one assumes that all the emissions 25 are fossil-fuel based.
The natural gas consumption at Oktoberfest 2018 sums up to 200,937 m 3 . Therefore, the total weight of the consumed CH 4 at Oktoberfest yields M gas,total = 0.668 kg m −3 · 200,937 m 3 = 1.34 · 10 5 kg.
In this study, the CH 4 flux of Oktoberfest has been determined to 6.7 µg/(m 2 s). If we assume that the emission is continuous 30 throughout the day (about 11 h opening time per day) and homogeneous throughout the entire Oktoberfest premises, the total amount of CH 4 lost to the atmosphere would be: M CH4,loss,max = 6.7 µg m 2 s · (16 d · 11 h d · 3600 s h ) · 3.45 · 10 5 m 2 = 1.46 · 10 3 kg.  The CH 4 share of the natural gas in Munich is on average about 96 % (München (2018c)). If we assume all of the CH 4 emissions are fossil-fuel related, the maximum loss rate can be determined as: M CH4,loss,max M CH4,total = 1.46 · 10 3 kg 1.34 · 10 5 kg · 96 % = 1.1 %.
This loss rate of 1.1 % is smaller than the gas leaks reported in the literature, such as a 2.7 % loss rate for the urban region of Boston (McKain et al. (2015)) or 2.3 % for the U.S. oil and gas supply chain (Alvarez et al. (2018)).  Table 2, the converted values are shown. Furthermore, one can see that the different inventories have different spatial resolutions. Therefore, the fluxes are averaged over areas that represent not only the Oktoberfest premises but also additional urban districts. Nevertheless, it can be seen that the determined Oktoberfest emissions are significantly higher than all the three considered inventories. Therefore, festivals such as Oktoberfest can be significant CH 4 sources, although they are just present for a limited time of the year, and should be included in the inventories.

Conclusions and Outlook
In this study, the methane emissions at Oktoberfest 2018 in Munich were investigated. This is the first study that deals with the methane emissions of a big festival. We concentrated on Oktoberfest as it is the world's largest folk festival and a methane source that had not yet been taken into account in the emission inventories.
Combining the in-situ measurements with a Gaussian plume dispersion model, the average emission of Oktoberfest was 5 determined to be (6.7 ± 0.6) µg/(m 2 s) (1 σ standard deviation). A comparison between weekdays (4.6 µg/(m 2 s)) and weekend days (8.5 µg/(m 2 s)) shows that the emission strength at the weekend was almost twice as high compared to during the week. It demonstrates that a higher number of visitors results in higher emissions. However, the daily emission cycle has an oscillating behavior that cannot be explained by the number of visitors. These results suggest that CH 4 emissions at Oktoberfest do not come solely from the human biogenic emissions, which was according to our calculation 5 times smaller than the emissions 10 determined for the Oktoberfest. It is more likely that fossil-fuel related emissions, such as incomplete combustion or loss in the gas appliances, are the major contributors to Oktoberfest emissions.
Due to safety reasons, we were not allowed to enter the festival premises with the instrument. Therefore, the distance from the measurement points to the suspected sources on the festival terrain was large, which resulted in low CH 4 and C 2 H 6 concentrations. Latter ones were even below the detection limit of the instrument. This limited the possibilities to attribute the 15 emissions to specific sources. To improve this aspect, several additional approaches are possible for future studies. As we are not aware of a more sensitive portable C 2 H 6 analyzer, discrete air sampling using sample bags within the tents for C 2 H 6 and isotopic CH 4 measurements is an option. Furthermore, the measurement of isotope ratios, such as δ13C and δD are useful options to improve the source attribution. For other festivals, researchers might get closer to the sources, which we were not permitted to do at Oktoberfest.

20
The method introduced in this paper is comparatively straightforward; it can be applied widely to discover and quantify overlapping methane sources: groups of small cow barns, uncovered heaps in landfills, or wetlands made of groups of ponds and swamps, etc.
In summary, this study uses Oktoberfest as an exemplary event to show, for the first time, that large festivals can be significant CH 4 emitters. Therefore, these events should be included in future emission inventories. Furthermore, our results provide the 25 foundation to develop reduction policies for such events and a new pathway to mitigate fossil fuel CH 4 emissions.
Data availability. All raw data can be provided by the authors upon request.  , 112, 1941-1946, 2015.