Methane emissions from contrasting urban freshwaters: Rates, drivers, and a whole‐city footprint

Global urbanization trends impose major alterations on surface waters. This includes impacts on ecosystem functioning that can involve feedbacks on climate through changes in rates of greenhouse gas emissions. The combination of high nutrient supply and shallow depth typical of urban freshwaters is particularly conducive to high rates of methane (CH4) production and emission, suggesting a potentially important role in the global CH4 cycle. However, there is a lack of comprehensive flux data from diverse urban water bodies, of information on the underlying drivers, and of estimates for whole cities. Based on measurements over four seasons in a total of 32 water bodies in the city of Berlin, Germany, we calculate the total CH4 emission from various types of surface waters of a large city in temperate climate at 2.6 ± 1.7 Gg CH4/year. The average total emission was 219 ± 490 mg CH4 m−2 day−1. Water chemical variables were surprisingly poor predictors of total CH4 emissions, and proxies of productivity and oxygen conditions had low explanatory power as well, suggesting a complex combination of factors governing CH4 fluxes from urban surface waters. However, small water bodies (area <1 ha) typically located in urban green spaces were identified as emission hotspots. These results help constrain assessments of CH4 emissions from freshwaters in the world's growing cities, facilitating extrapolation of urban emissions to large areas, including at the global scale.

lation of methanogenesis in sediments and, thus, increased emission of methane (CH 4 ) across the water-atmosphere interface . This suggests that urban freshwaters could act as an important source of CH 4 to the atmosphere (Gonzalez-Valencia et al., 2014;Martinez-Cruz et al., 2017;Wang et al., 2018). Empirical data on CH 4 emissions from urban freshwaters are scarce, however, and have not been included in global emission estimates (Bastviken, Tranvik, Downing, Crill, & Enrich-Prast, 2011;IPCC, 2013), nor in systematic assessments of CH 4 evasion from all potential sources in cities (Hopkins et al., 2016;Ware et al., 2019). In fact, most studies on freshwaters assessing urban CH 4 emissions were limited to a single type of water body and a single season (López Bellido, Peltomaa, & Ojala, 2011;Wang et al., 2018;Zhang et al., 2014;Zhang, Huang, Yang, Li, & Dahlgren, 2016), with only one recent investigation in a tropical megacity considering multiple surface waters and temporal patterns (Martinez-Cruz et al., 2017). Equivalent information is lacking from urban freshwaters in temperate climates, where seasonality is more pronounced than in the tropics.
Information available on individual urban water bodies suggests that the drivers behind CH 4 emissions are similar to those in rural, forest, and other natural areas (Martinez-Cruz et al., 2017;Yu et al., 2017). All else being equal, shallow waters, which are typical of urban areas (McEnroe, Williams, Xenopoulos, Porcal, & Frost, 2013), are likely to emit more CH 4 per surface area, because the travel times of CH 4 bubbles generated by ebullition events and rising from the sediment to the water surface are likely to be shorter, limiting CH 4 oxidation by methanotrophy in the oxic water column (Bastviken, Cole, Pace, & Tranvik, 2004;Holgerson, 2015). The small size of most urban water bodies also suggests that land use in the surroundings and associated inputs of organic matter, nutrients, and contaminants can strongly influence water quality and ecosystem properties. Large supplies of labile organic matter, whether from the catchment or through intense primary production boosted by nutrient availability, coupled with subsequent oxygen depletion are both conducive to methanogenesis (Segers, 1998). This points to a high potential of urban freshwaters to produce and emit CH 4 to the atmosphere, unless toxic substances curb biological activity.
In view of the importance and large gaps in information on rates and drivers of CH 4 emissions from urban freshwaters, the aims of this study were to (a) determine CH 4 fluxes at different times of the year from a range of contrasting urban freshwaters; (b) identify drivers of CH 4 emissions from the different types of water bodies; and (c) integrate this information to provide an initial flux estimate from a metropolitan area as a potentially important component of global urban CH 4 emissions from freshwaters. Based on the limited information available to date, we predicted rates to be particularly high in small, shallow, and nutrient-rich standing waters with sediments rich in organic matter.
When canals for transportation and ditches for sewage and rainwater collection are added, the surface river network reaches a total length of about 560 km (SenUVK, 2018). River flow is slow because of the low terrain slope (0.01%; Knappe et al., 2005), locks, and weirs. Multiple wastewater treatment plants (WWTP) within the city discharge treated effluents into the urban freshwater network (Heberer, 2002).
Four categories of surface waters were distinguished: lakes, ponds, rivers (including canals), and streams (including ditches).
Lakes were classified as water bodies ≥1 ha according to a lake inventory for Berlin (SenUVK, 2005). Rivers and streams were differentiated by width (rivers >5 m). Seven locations were randomly selected from each of the four categories. Four additional running water sites were also included because of particularly high nutrient (NO − 3 , NH + 4 , total phosphorus [TP]) and dissolved organic carbon (DOC) concentrations recorded in a monitoring program over the five previous years (SenUVK 2009(SenUVK -2014. However, CH 4 emissions at these sites were found not to differ significantly from those of the randomly selected sites and were thus treated as rivers (H1-2) or streams (H3-4), depending on size. Thus, a total of 32 sites (Figure 1; Table S1) were each sampled four times, in spring (April-May), summer (July-August), and fall (September-October) 2016, and in winter (February-March) 2017 just after ice out because of unusually cold weather late in the season.

| CH 4 emissions
Floating chambers were deployed at one selected point in each water body to estimate rates of total, diffusive, and ebullitive CH 4 fluxes to the atmosphere. The chambers were anchored but several meters of rope and tubing allowed for some free movement. The position in lakes was randomly chosen along the contour line of average water depth to avoid potential bias caused by taking measurements at the deepest point (Schilder et al., 2013). Since the bathymetry of ponds was unknown, the central point (not necessarily the deepest) was used in those cases; this was less critical than for lakes because water depth in ponds varied much less. In running waters, chambers were deployed within 2 m from the shore (Grasset, Abril, Guillard, Delolme, & Bornette, 2016).
Cylindrical floating chambers (area: 0.071 m 2 ; headspace volume 5.4 L) were used in lakes and ponds to determine CH 4 emission rates. Slightly wider and shorter but otherwise similar chambers (0.126 m 2 ; headspace volume 16.8 L) were used in streams and rivers. The chamber headspace was connected in a closed loop to an ultra-portable greenhouse gas analyzer (UGGA 24P and 30P; Los Gatos Research) before deploying a single chamber three times at each location to measure CH 4 headspace concentrations every second for 15 min (Pirk et al., 2015). Chambers were opened between series of measurements and equilibrated with the surrounding air. All fluxes were measured between 8 and 12 a.m. to minimize any possible influence of systematic diel variations. Atmospheric pressure and wind speed 1 m above the water surface were simultaneously determined using a portable weather station (Kestrel 4000; Nielsen-Kellerman).
Total CH 4 flux (F) to the atmosphere was calculated as: where ∆C is the concentration change in the headspace of the static chamber (ppm v ), ∆t is the chamber deployment time (s), V is the volume (m 3 ) of the chamber headspace, A is the area of the static chamber (m 2 ), R is the universal gas constant (8.3143 m 3 Pa mol −1 K −1 ), P is atmospheric pressure (Pa), and T is air temperature (K) during the measurement. All concentration data were plotted to visually identify whether any sampling errors or ebullition events occurred.
When initial values deviated from the atmospheric concentration measured before deploying a chamber, the first data points were re- Ebullition events were recognized by sudden steep concentration increases, which were occasionally followed by a decline. Only concentration increases with an r 2 > .7 were taken into account to compute diffusive fluxes (Martinez-Cruz et al., 2017;Sepulveda-Jauregui, Martinez-Cruz, Lau, & Casper, 2018). Ebullition flux was calculated as the difference between the total and diffusive flux.
To assess the reliability of the calculated fluxes from the chamber technique, other commonly adopted methodologies were used in tandem with the flux measurements by the chamber technique.
Specifically, CH 4 concentrations of surface waters were used to calculate diffusive fluxes following the thin boundary layer (TBL) methodology (see Supporting information), and inverted funnels deployed above the sediment for a week were used to calculate ebullition fluxes (see Supporting information). To standardize the ice-cover period among the different water bodies, we defined the start as the date where the minimum daily temperature dropped below 0°C for 3 days in a row and the end as the date when mean daily temperature rose above the freezing point for at least 1 week. Total annual emissions from each type of water body were estimated by multiplying the seasonal total emission from each type of water body (mg CH 4 /m 2 ) by the respective surface area of all water bodies in the city of Berlin assigned to that water body type. The total CH 4 emission footprint of Berlin's surface waters was then calculated as the sum of the annual emissions by each of the four types of water bodies. Estimates of variation (i.e., uncertainties) were obtained by applying error propagation rules at each step.
The filters were dried and weighed, and a weighed portion was subsequently used for elemental analysis (Vario EL; Elementar Analysensysteme GmbH) to determine POC. The filtrate was stored in acid-washed and precombusted glass vials with a polytetrafluoroethylene-lined screw cap for later measurements of DOC and dissolved inorganic carbon (DIC) on a TOC analyzer (TOC-V; Shimadzu). A second GF75 filter produced in the same way was used for spectrophotometric analysis of chlorophyll a (chl a) after hot ethanol extraction (Jespersen & Christoffersen, 1987). Soluble reactive phosphorus, NO We further characterized dissolved organic matter (DOM) by absorbance and fluorescence spectrophotometry (Aqualog).
Fluorescence spectra were recorded in a 1 cm quartz cuvette at excitation wavelengths ranging from 250 to 600 nm at 5 nm increments and emission wavelengths of 250-650 nm measured at 1.77 nm increments. These optical measurements were performed within 48 hr after sampling. The resulting data yielded the following indicators of DOM quality (Table S2)

| Land use
The total area of each type of water body and of four categories of land use (forest and natural areas, green space, agricultural land, paved areas) within a 50 m wide strip along the shores of each site were calculated using Quantum GIS (Development Team), based on land-use data freely available from the Senate Department for the Environment, Transport and Climate Protection of Berlin. Historical reviews and personal communication with citizens and authorities complemented the database to determine whether a given water body was natural or man-made and whether it had any other distinct anthropogenic features.

| Data analysis
All statistical analyses were performed with R version 3.2.2 (R Development Core Team, 2010). Linear mixed models were used on log-transformed data to test for differences in total CH 4 emissions among seasons, types of water bodies, and the interaction of both, taking into account the repeated-measures nature of the data. Tukey post hoc tests were used for pairwise comparisons. Wilcoxon signed rank test was used to compare estimates of diffusive flux by the TBL and chamber method, as well as to compare ebullitive flux assessed with the funnel traps and the chamber method.
To explore possible controls of total CH 4 emissions, the large number of variables recorded to characterize the water bodies was first condensed by a principal component analysis (PCA). The analysis was based on water temperature, a range of water chemical variables (conductivity, pH, alkalinity, DO, TP, NH  Table   S4). No significant differences were found among the other seasons.
Ponds showed the highest emission (503 ± 699 mg CH 4 m −2 day −1 ) in all seasons (Figure 2), with fluxes significantly exceeding (p < .05) those from rivers (123 ± 285 mg CH 4 m −2 day −1 ) and streams (118 ± 348 mg CH 4 m −2 day −1 ) but not from lakes (159 ± 473 mg CH 4 m −2 day −1 ). Within each of the four types of water bodies, seasonal differences were only significant between summer F I G U R E 2 Seasonal changes in (a) daily mean air temperature in Berlin Tempelhof recorded by the German Meteorological Office, with the light gray area representing a period of ice cover on the larger lakes and the dark gray areas representing the sampling periods, and (b) Total methane (CH 4 ) emissions from four types of urban water bodies. Box plots show the median (horizontal line), interquartile range (box limits), highest and lowest values within 1.5 times the box size from the median (whiskers) and outliers (points) and winter in lakes, ponds, and rivers (p < .05), whereas streams never showed any significant difference among seasons.
Total CH 4 emission derived from all chamber measurements indicated a higher contribution of ebullition (80%). Although the relative contribution of ebullition varied among types of water bodies (Table 1; Figure S2). Estimates of ebullition and diffusive fluxes derived from different methodologies also showed some differences. Ebullition fluxes estimated by 1 week deployments of funnels accounted for an average of 62% of the emissions at those sites where ebullition was observed (N = 12), compared to 51% based on measurements at the same sites made with the chamber technique (Table S3). Ebullition fluxes determined with the two techniques were positive correlated (Spearman's ρ = 0.73; p < .01). There were no significant differences in ebullition fluxes among individual water bodies within each type. In contrast, diffusive fluxes estimated by the two methods were significantly different for lakes (p < .001), ponds (p = .024), rivers (p < .01), and streams (p < .001). However, despite these differences, the values obtained with the different methods were in a broadly similar range for most of the observations. Taking into account the calculated areas of the different types of surface waters in the city of Berlin (Table 1), the annual total CH 4 emission estimated by the chamber method was 2.6 ± 1.7 Gg CH 4 .
Lakes alone contributed almost two-thirds to the total emissions, due to the large total lake area, while streams contributed the least (Table 1).
The first four axes of the PCA to characterize the 32 investigated water bodies in terms of water chemistry and land use accounted for 58% of the total variability. PC1 and PC2 clearly separated the four types of water bodies (Figure 3a,c), with PC1 separating running from standing waters mainly based on differences in land use (green space, paved, or agricultural) and the DOM spectral ratio (S R ), and PC2 separating larger from smaller water bodies based on conductivity and solute concentrations (e.g., NH  appear to be largely driven by differences among lakes ( Figure S1), which produced similar relationships with emission data when lakes were analyzed alone. No such patterns emerged for the three other types of water bodies analyzed alone. Ponds were the only exception in that low DO concentrations in surface water were weakly related to CH 4 emission (r 2 = .19; p = .04).   (Grimm et al., 2008), and stormwater and sanitary infrastructure (Smith, Kaushal, Beaulieu, Pennino, & Welty, 2017). This conclusion is supported by the hypereutrophic conditions reported for all water bodies analyzed by Martinez-Cruz et al. (2017).

| D ISCUSS I ON
Our budget calculation is based on measurements of total flux including both diffusion and ebullition made with floating chambers. This enabled a first approximation of total annual emissions, for a large metropolitan area encompassing a wide range of different water bodies. Expanding the coverage of these measurements at different scales, both spatial (within and among water bodies) and temporal (diel to interannual), would reduce the uncertainties associated with the estimates available at present. In addition, a comparison with alternative methods can help constrain and validate these estimates. Therefore, we also computed diffusive fluxes by the commonly employed TBL approach and determined ebullitive fluxes at selected sites by deploying funnel traps for 1 week.
The TBL approach makes several assumptions, particularly on piston velocities (k) depending on wind speed, which makes this method vulnerable to biases, especially in aerodynamically rough and heterogeneous urban environments. This could be one reason for several discrepancies observed between the two methods used to derive diffusive fluxes in our study (Table S3). The use of anchored rather than freely drifting chambers could also have contributed to the observed differences in running waters, mainly because unnatural water turbulence created by the chambers could unnaturally increase fluxes (Lorke et al., 2015). However, the typically slow flow of the lowland streams and ditches in Berlin makes it unlikely that this error was large. Ebullition fluxes assessed with inverted funnels deployed for 1 week produced remarkably similar results as our short-term measurements of ebullition, despite the documented high stochasticity and spatial heterogeneity of ebullition (Wik, Crill, Varner, & Bastviken, 2013). This suggests that the results of our short-term chamber measurements were broadly realistic across sites.
Although lower than in Mexico City, the calculated total annual emission per km 2 from Berlin's freshwaters (49 Mg CH 4 km −2 year −1 ) is more than twice that of the global average (22 Mg CH 4 km −2 year −1 ) reported by Bastviken et al. (2011) for 4.6 million km 2 of global freshwater surfaces. The fraction of urban areas contributing to freshwater surfaces globally is unknown, but our rates for Berlin, like those for other urban freshwaters (Table 2), were higher than both the average calculated for lakes and ponds at northern latitudes (Wik, Varner, Anthony, MacIntyre, & Bastviken, 2016) and values for streams and rivers globally (Stanley et al., 2016). This could suggest that urban areas in general contribute disproportionally to CH 4 emissions from freshwaters. Given that there are >500 urban centers worldwide with >1 million inhabitants each and that urbanization trends continue (UN, 2016), emissions of CH 4 from urban areas may be sufficiently considered in large-scale estimates.
An extremely rough estimate assuming 3 Gg of CH 4 emitted annually by each of the 500 most densely populated cities in the world results in a total annual emission of 1.5 Tg CH 4 , but emissions from the total urbanized area globally are evidently much larger. A related question is whether surface waters also contribute significantly to the total CH 4 footprint of metropolitan areas. Currently, the answer to this question is speculative, too, because other sources of CH 4 have not been quantified. However, a recent estimate of 20,000 Tg of CO 2 emitted by the city of Berlin in 2012 (Reusswig, Hirschl, & Lass, 2014) suggests that even the high total CH 4 fluxes from Berlin's surface waters would contribute little to the total greenhouse gas emissions from the city, equivalent to 0.004% in CO 2 equivalents.
High variability of CH 4 emissions rates in space and time is common (Deemer et al., 2016;DelSontro, McGinnis, Sobek, Ostrovsky, & Wehrli, 2010) and also apparent in our dataset on surface waters in Berlin. Despite this high variability both within and across water bodies, PCA could differentiate between standing and flowing waters, and subsequent regression analyses identified water chemistry and the  predominant land use near each site as factors influencing CH 4 emissions ( Figure 3). Ponds, in particular, were identified as hotspots of CH 4 emissions in Berlin, with the annual average emissions four times higher than from lakes, streams, and rivers (Table S3). This information is important, not least because anthropogenic ponds are neglected water bodies in terms of CH 4 emissions both in cities and other landscapes . For example, Berlin has a detailed inventory of all lakes and their water quality is regularly assessed in monitoring programs. In contrast, no systematic information is available on ponds, despite the fact that these small water bodies are increasingly recognized as important urban habitats (Hassal, 2014).
Although emissions did not significantly differ when lakes and ponds were statistically treated as categories, a significant negative relationship emerged between log-transformed CH 4 emission flux and lake and pond surface area (r 2 = .46; p = .01), corroborating a previously observed pattern of increasing CH 4 flux to the atmosphere with decreasing size of water bodies (Bastviken et al., 2004;Holgerson & Raymond, 2016;Wik et al., 2016).
Nevertheless, even though ponds had high emissions, their contribution to the overall emission budget is low in comparison to lakes, which account for more than 50% of the total emission from freshwaters in Berlin (Table 1) In contrast, urban ponds are mostly associated with green spaces throughout the city where they are likely to receive anthropogenic inputs resulting, for example, from feeding of waterfowl, fertilizer application, or pet waste (Hobbie et al., 2017).
The particularly high variability in emissions rates that we observed from running waters was not clearly related to riparian land cover or other characteristics. High emission rates characterized some stream sites experiencing diffuse nutrient inputs from agriculture (S3 and S6) or some highly engineered streams (paved riparian areas, channelization; S7, S2), but this was not universally true for other water bodies showing similar characteristics (S1 and S4  Garnier et al., 2013), and may be due to the fact that our study sites were not located directly downstream of WWTP outlets.
The relation between oxygen concentration and total CH 4 emission was also weak (r 2 ≤ .12), although oxygen concentrations varied widely across sites (Table S4). When interpreting these data, it must be borne in mind, however, that our measurements in surface water are not necessarily good proxies of conditions conducive to methanogenesis in sediments. Furthermore, differences in chemical characteristics and land use had little explanatory power; only their combination produced a clear relationship while substantial scatter still remained. Clearly, a multitude of factors create complex environmental conditions in urban freshwaters that make it a challenge to tease apart individual drivers of CH 4 emissions from these systems.
Overall, however, the variables with the highest explanatory power in our study (i.e., POC, chl a, and DO) all point to trophic state as a determinant of CH 4 emissions from urban freshwaters. This is in line with results of Martinez-Cruz et al. (2017) and DelSontro, Beaulieu, and Downing (2018) and is also reflected by the conspicuous peaks in DOC and chl a in autumn (Figure 3; Table S4) when the emissions from ponds were highest. This result and our finding that ponds act as hotspots of CH 4 fluxes to the atmosphere are important contributions toward robust assessments of CH 4 emissions from whole cities and extrapolation to large areas, including global estimates.

ACK N OWLED G EM ENTS
We are grateful to the many students and technicians for their invaluable assistance during extensive fieldwork, especially to C.N.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest. Export of dissolved methane and carbon dioxide with effluents from municipal wastewater treatment plants. Environmental