Analysis of the Inuence of Sources on the Spatial Variation in Atmospheric Methane Concentrations in the City of Tandil, Argentina

There is an overall trend in urban methane (CH 4 ) emissions due to the presence of several sources; however, differences exist between cities, and therefore further local research should be undertaken. The present study analyzes the spatiotemporal variation in atmospheric CH 4 concentrations during a year at ten sampling sites in the urban core of a medium-sized city. The mean annual atmospheric CH 4 concentrations varied between 2.02 ppm and 5.45 ppm; the maximum concentrations were found in a site close to a wastewater treatment plant (WWTP), presenting a signicant increase toward the summer. In the rest of the sites, the maximum concentrations were recorded in the coldest months due to the inuence of combustion sources dependent on natural gas (NG). An exploratory regression analysis was performed, in which the variables “homes connected to the gas network” and “distance from compressed NG stations” each explained 66 and 65% of the spatial variation of the atmospheric CH 4 concentrations at the 9 sites (excluding that one nearest the WWTP). The results show the need to prevent NG leaks in all urban areas to reduce the emissions of this potent greenhouse gas, which, at the same time, will provide economic benets for the sectors involved.


Introduction
The task of reducing greenhouse gases (GHGs) emissions su ciently to stabilize their atmospheric concentrations and therefore mitigate climate change is a great global challenge (Erickson 2017).
Methane (CH 4 ) becomes relevant because its atmospheric concentrations continue to increase, making it the second most important anthropogenic GHG in terms of climate forcing, after carbon dioxide (CO 2 ) (Saunois et al. 2020). CH 4 emission reduction is technically challenging: it demands identifying each source, quantifying its emission ux, and developing effective emission reduction methodologies (Nisbet et al. 2020). CH 4 is emitted by a variety of processes, including natural and anthropogenic ones, and by a variety of sources, xed and diffuse, non biogenic and biogenic, thermogenic and pyrogenic ones (Saunois et al. 2020). The goal should be to reduce anthropogenic emissions, which we can control (Nisbet et al. 2019).
Cities generate 70% of anthropogenic GHGs emissions, a fraction that is growing with global urbanization (Hopkins et al. 2016). Because of their density, e ciency, and adoption of innovations and new technologies, cities can provide solutions for reducing emissions (Hoornweg et al. 2011). The effectiveness of these remedial actions depends on accurate knowledge of the many sources of GHG in each city (Ars et al. 2020). In many cities, estimating the source contribution when so many emission sources coexist, can be challenging. Besides, according to Nisbet et al. (2020), the location of emissions may be complex because what initially appears to be a point source may actually comprise many small sub-sources.
In general, the urban sources of GHGs can be classi ed in energy (electricity, heating), industry (processes, product use), transportation (road, non-road, navigation, take-offs, landings, and aircraft ying over urban areas), and waste (land lls, wastewater treatment) (Marcotullio et al. 2013  GHGs emissions are signi cantly associated with population size, density, growth rates, and per capita income (Marcotullio et al. 2013 The aim of this study is to identify possible CH 4 sources in a medium-sized city in order to explain the spatial variability of atmospheric CH 4 concentrations. For it, periodical measurements were performed in several strategically selected urban sites, and then, using spatial analysis tools, the relative contribution of CH 4 from each source was evaluated.

Materials And Methods
The sampling sites were selected (Fig. 1a), ensuring a uniform distribution over the urban area that would include different urban densities, residential and commercial uses of soil (Figs. 1b and 1c), and CH 4 sources (Fig. 2).
For the meteorological characterization of the study period, data regarding air temperature, relative humidity, atmospheric pressure, visibility, and wind velocity were obtained daily from the National Meteorological Service, Tandil-AERO weather station. In addition, information on NG monthly consumption-CNG, residential, commercial, and public consumption by different sectors of the city was provided by the only NG supplier (CAMUSSI S.A).

Data analysis and statistics
Basic descriptive statistical analyses were performed to assess temporal and spatial variation in atmospheric CH 4 concentrations measured in each sampling site. An ANOVA Test and a Fisher's LSD test were performed to examine temporal differences between atmospheric CH 4 concentrations measured in each site and analyze the differences between sampling sites (mean values with the same letter are not statistically different). The relative incidence of biogenic and non biogenic sources in the atmospheric CH 4 concentrations at each site was measured using the Pearson correlation analysis with ambient temperature and with NG sectoral consumption. For all the analyses a p-value of 0.05 was considered to assess signi cance. Infostat Statistical software was used for all statistical analyses.

Exploratory Regression
Spatial analysis tools (ArcGIS 10.5®) were used to identify the independent variables that account for the mean CH 4 values obtained. First, georeferencing of the sampling sites was performed. To assess the in uence of the environment on each sampling site, a buffer zone (area of in uence) was created around each one (Hoek et al. 2008). We selected a buffer radius of 300 m, following the recommendations by several authors (Briggs et al. 2000;Henderson et al. 2007).
Based on the last National Census of Population and Housing in the city of Tandil (INDEC 2010), we calculated the average of homes connected to the NG network (GN) within each census tract that included one of the buffers at the sampling sites (Fig. 2a). In addition, the xed sources that could a priori contribute to the spatial variation in atmospheric CH 4 concentrations in the urban area-CNG stations, a wastewater treatment plant (WWTP) and an arti cial lake, were represented in a vector layer, and the Euclidean distance to each sampling site was calculated. In this way, three individual maps were created: distance in meters from the CNG station (GD), from the WWTP (PD), and from the arti cial lake (LD) (Figs. 2b, 2c and 2d).
Taking into account that the dependent variable is the mean CH 4 concentrations in each site (seasonal and annual) and that the independent variables are the four layers created, an exploratory regression analysis was performed. This tool examines whether the association between variables is constant in the whole urban area or whether variations take place with the goal of looking for the Ordinary Least Squares Such a relationship can be positive or negative in sign (directly or inversely proportional) with a high or low % of signi cance of the variable (consistency in the relationships). To validate the results (passing models), the coe cients considered were the following: the adjusted R 2 (Adj R 2 ) > 0.50, the level of signi cance p < 0.05, the Variance In ation Factor (VIF, that indicates independence between the independent variables) < 7.5, the spatial autocorrelation I Global Moran with p > 0.10 and the residual normality summary Jarque-Bera with p > 0. 10   NG residential and commercial consumption are higher in winter, with statistically signi cant differences in relation to the other seasons in the case of the residential sector and to spring and summer in the case of the commercial sector (p < 0.05). This behavior was re ected in a good correlation between NG consumption and ambient temperature, with an R and p value (value in parenthesis) of -0.78 (p = 0.003) and -0.94 (p <0.0001) for the residential and commercial sectors, respectively. In general, heating degree days is currently an important determinant of the amount of energy required to heat urban buildings (Kennedy et al. 2009). In our country, residential and commercial NG consumption reaches its maximum level in the winter months (Secretariat of the Environment and Sustainable Development 2015).
Regarding CNG demand, only statistically signi cant differences were measured between summer and winter (p < 0.05), with a fairly good correlation with ambient temperature (R = -0.69, p = 0.013). A decrease in the CNG consumption in the summer is probably related to less economic activity as a consequence of the holidays (Gil 2006;Gioli et al. 2012).
Temporal variability of atmospheric methane concentrations Table 1 shows the average atmospheric CH 4 concentrations for each season and for the complete study period at each sampling site.   The best correlation between atmospheric CH 4 and CNG demand was observed in S6. This site is located near National Route 226 and two CNG stations (Fig. 2b). A good correlation was also established for other sites close to CNG stations, S2, S5, and S8, although the last one with a value of p < 0.  Tables 1 and 2, which indicated a greater relevance of non biogenic sources associated with NG consumption during the coldest months. When repeating this analysis, excluding S3, the differences in atmospheric CH 4 concentrations between sites for each season become more evident. S8 presented the highest atmospheric CH 4 concentrations with statistically signi cant differences with respect to the other sites in spring and summer, and compared with S7 and Page 10/22 S10 in the fall. In winter, the greatest atmospheric CH 4 concentrations were measured in S6, with statistically signi cant differences compared with S4, S5, S7, S9, and S10.
The spatial variation in the atmospheric CH 4 concentrations measured in the city depends on the type of dominant source ( xed, diffuse, biogenic, or non biogenic), its relative contribution, and its distance from the sampling site (Carranza et al. 2018;Helfter et al. 2016). When performing the exploratory regression analysis on the 10 sites, no variable met all the search criteria established in section Exploratory Regression for each diagnostic test. However, some ndings are worth noting. When performing it on 9 sites (excluding S3), the seasonal behavior of the sources that account for spatiotemporal variation becomes more notable (Table 3).  The independent variable GD met the search criteria of each diagnostic test in the fall and winter, whereas GN in the spring and summer, as well as the fall (Table 3). Still, both variables signi cantly correlated with atmospheric CH 4 concentrations with a value of p < 0.05 or p < 0.10 in those seasons when they failed to meet the search criteria. These results suggest that one source predominates over another according to the season. In winter, GD is the independent variable that best explains atmospheric CH 4 concentrations, while in the spring and summer, the independent variable GN accounts for them. In the fall, both of these independent variables explain the spatial variations in atmospheric CH 4 concentrations. The difference in the predominance of the sources for each season was more easily observed in the case of residential and commercial consumption of NG than in the CNG sector (Fig. 3). This behavior was re ected in Pearson correlations between atmospheric CH 4 measured in each site and the general demand on NG for each use ( were not observed for GD and PD or GN and PD (VIF < 7.5); however, the results of the Pearson correlation test between mean monthly atmospheric CH 4 and monthly consumption of NG by sector suggest that GD or GN probably best accounts for the high CH 4 concentrations measured in these sites. This is reasonably expected especially in winter when the sources dependent on NG are the most relevant and the differences of atmospheric CH 4 concentrations between the sites become smaller.

Annual mean atmospheric CH 4 concentrations
The annual mean atmospheric CH 4 concentration measured in S3 showed statistically signi cant differences compared with the other sites (p < 0.05). The second highest value, found in S8 (zone near a CNG station and with a high density of homes connected to the gas network) (Fig. 2), only presented statistically signi cant differences in relation to S3 and S10. Because of these observations, the exploratory regression analysis performed for the seasonal mean atmospheric CH 4 concentrations was repeated for the annual mean concentrations in order to explain the general CH 4 behavior in the city of Tandil. Once again, when considering the 10 sites for the exploratory regression analysis, no variable met the search criteria of each diagnostic test. However, some ndings are worth noting. When performing it on 9 sites (excluding S3), the variables GD and GN satis ed those criteria (Table 3).
Atmospheric CH 4 concentration correlated positively with GN (Adj R 2 = 0.66, p < 0.01) and negatively with GD (Adj R 2 = 0.65, p < 0.01). In Liu et al. (2019), population density had a remarkably positive correlation with CH 4 , with a correlation coe cient of 0.74 (p < 0.01). As Sailor and Lu (2004) suggest, the anthropogenic heating pro les for the urban core would be correspondingly higher as they scale with population density. In Florence (Italy), road tra c and domestic heating were responsible for only 14% of the observed CH 4 uxes, while the major residual part was likely dominated by gas network leakages  (Fig. 2b). For this reason, the CNG stations could generally contribute to the spatial variation in the atmospheric CH 4 concentration in the city.
No signi cant correlations were established with LD (p > 0.1); this variable failed all the tests (Table 3), proving not signi cant in this study. Although the urban lake was expected to acquire relevancy in the warmer months for being a biogenic CH 4 source (Ortiz-Llorente and Alvarez-Cobelas 2012), its contribution was almost nonexistent because of its xed location in an urbanized zone.

Methane concentration associated with natural gas sources
From the results of the exploratory regression, it can be observed that when 9 sites were considered, only the models with just one variable (GN and GD) were able to meet all the search criteria established in the diagnostic tests. The reason for this may be that one of these variables could best explain the temporal CH 4 behavior in one site and, at the same time, have less relative weight in another site. For instance, the signi cance of GN and GD excluding S3 in the exploratory regression analysis for the entire study period was equal to 37.5 % for both variables (Table 3). This accords with the results of the Pearson correlation test between mean monthly atmospheric CH 4  near CNG storage tanks and connecting pipes in Orange County, California; however, CH 4 increase was highly variable across the 12 different CNG stations surveyed, suggesting that fugitive leaks are responsible for these high concentrations.
In order to nd a multiple OLS model that allows quantifying the interrelationships between both sources associated with NG consumption (GN and GD) and atmospheric CH 4 concentration in the city of Tandil, it would be important to incorporate more sampling sites (Quinn and Keough 2002). These should be located not only within the urban core but also towards the periphery of the city to obtain a more precise atmospheric CH 4 concentration for the entire city by increasing the measurement sites. Population density in urban cores is usually one order of magnitude higher than for the city as a whole (Liu et al. 2019

Conclusion
To explain spatial and temporal variations in the atmospheric CH 4 concentrations in an urban site, it is important to learn about the predominance of the sources of this gas and about the nature of each source-whether it is biogenic or non biogenic, xed or diffuse. Identifying the main sources responsible for the increase in atmospheric CH 4 concentrations in a city helps to develop mitigation strategies.
Although the WWTP is an important CH 4 source, because it is a xed source, it only explained 29% of the spatial variation of the annual mean atmospheric CH 4 concentrations and the maximums registered during the summer. The variables "distance from CNG stations" and "number of homes connected to the gas network" are the ones that best explained the spatial variability of the annual mean atmospheric CH 4 concentrations (65% and 66%, respectively) in the urban core of the city of Tandil and the maximum concentrations registered during the fall and /or winter.
Although the sources associated with NG consumption (residential, commercial, and CNG) cause only minor increases in CH 4 concentrations, they are scattered in the whole urban area, and together their relative contribution therefore increases. Improving the e ciency of each system involved, from NG distribution, residential and commercial consumption to CNG use in vehicles, would not only reduce the emissions of this potent greenhouse gas and its resulting impact on the environment, but would also reduce gas losses and consequently bring economic bene ts to each of these sectors. Preventing NG leaks to reduce emissions in all urban areas should be a goal to achieve in the short term.
Based on the results shown here, we propose expanding the sampling network and also performing the same study adjusting the model using the same and other variables. Finally, the baseline atmospheric CH 4 concentration for the urban area obtained in this study, would allow estimating how the city growth would contribute to atmospheric CH 4 as a result of the relative increase of the number of sources in the periphery of the city. In this sense, the results obtained serve as an important reference for medium-sized cities in constant growth.

Declarations
Ethics approval and consent to participate  Natural gas (NG) consumption in cubic meters (m3) for each sector (residential, compressed NG [CNG], commercial, public, and various)