Study of the daily and seasonal atmospheric CH 4 mixing ratio variability in a rural Spanish region using 222 Rn tracer

The ClimaDat station at Gredos (GIC3) has been continuously measuring atmospheric (dry air) mixing ratios of carbon dioxide (CO2) and methane (CH4), as well as meteorological parameters, since November 2012. In this study we investigate the atmospheric variability of CH4 mixing ratios between 2013 and 2015 at GIC3 with the help of co-located observations of 222Rn concentrations, modelled 222Rn fluxes and modelled planetary boundary layer heights (PBLHs). Both daily and seasonal changes in atmospheric CH4 can be better understood with the help of atmospheric concentrations of 222Rn (and the corresponding fluxes). On a daily timescale, the variation in the PBLH is the main driver for 222Rn and CH4 variability while, on monthly timescales, their atmospheric variability seems to depend on emission changes. To understand (changing) CH4 emissions, nocturnal fluxes of CH4 were estimated using two methods: the radon tracer method (RTM) and a method based on the EDGARv4.2 bottomup emission inventory, both using FLEXPARTv9.0.2 footprints. The mean value of RTM-based methane fluxes (FR_CH4) is 0.11 mg CH4 m−2 h−1 with a standard deviation of 0.09 or 0.29 mg CH4 m−2 h−1 with a standard deviation of 0.23 mg CH4 m−2 h−1 when using a rescaled 222Rn map (FR_CH4_rescale). For our observational period, the mean value of methane fluxes based on the bottom-up inventory (FE_CH4) is 0.33 mg CH4 m−2 h−1 with a standard deviation of 0.08 mg CH4 m−2 h−1. Monthly CH4 fluxes based on RTM (both FR_CH4 and FR_CH4_rescale) show a seasonality which is not observed for monthly FE_CH4 fluxes. During January–May, RTM-based CH4 fluxes present mean values 25 % lower than during June–December. This seasonal increase in methane fluxes calculated by RTM for the GIC3 area appears to coincide with the arrival of transhumant livestock at GIC3 in the second half of the year.


Introduction
The impact of the atmospheric increase in greenhouse gases (GHGs) on climate change is well known (IPCC, 2013).GHG emissions, due to natural as well as anthropogenic sources, are currently estimated and reported by each na-Published by Copernicus Publications on behalf of the European Geosciences Union.
tional agency to the United Nations Framework Convention on Climate Change (UNFCCC).A better understanding of the underlying processes causing these emissions can help in the implementation of future emission reduction strategies.Methane (CH 4 ) is the second most important anthropogenic GHG that is covered by the UNFCCC.The atmospheric mixing ratio of CH 4 has substantially changed since pre-industrial times from a global average of 715 nmol mol −1 to more than 1774 nmol mol −1 (IPCC, 2013).Nowadays, the contribution of CH 4 related to anthropogenic activities in the atmosphere represents about 25 % of total additional anthropogenic radiative forcing (IPCC, 2013).However, CH 4 has a relatively short lifetime in the atmosphere (∼ 9 years) and this makes it relevant in defining immediate and efficient emission reduction strategies (Prinn et al., 2000).Particularly in Spain, man-made methane emissions are mainly due to enteric fermentation (38 %), management of manure (20 %) and landfills (36 %) (WWF, 2014;MMA, 2016).The remaining methane contributions in Spain are due to rice cultivation (e.g.Àgueda et al., 2018), coal mining, leaks in natural gas infrastructures and waste water treatment.The CH 4 emission due to enteric fermentation related to livestock is directly linked to the number of animals of each type or breed of cattle, their age, their diet and environmental conditions (MMA, 2016).Spanish CH 4 emissions due to enteric fermentation were estimated to be 11 704 Gg CO −eq 2 (MMA, 2016).In order to estimate GHG emissions, bottom-up (based on fuel consumption and anthropogenic activity data) and topdown methods (based on atmospheric observations and modelling) are both widely applied and the scientific community has focussed on reducing their related uncertainties and understanding systematic inconsistencies (e.g.Vermeulen et al., 2006;Bergamaschi et al., 2010;NRC, 2010;Jeong et al., 2013;Hiller et al., 2014).Top-down methods usually require both high-quality and long-term GHG observations.European projects, such as InGOS (www.ingos-infrastructure.eu,last access: March 2018), and infrastructures, such as ICOS (www.icos-infrastructure.eu, last access: April 2018), aim to offer atmospheric CO 2 and non-CO 2 GHG data and data products to better understand GHG fluxes in Europe and adjacent regions.
Unfortunately, in some European regions, such as Spain, there is still a significant lack of high-quality atmospheric GHG observations.The Catalan Institute of Climate Sciences (IC3) has been working since 2010 within the ClimaDat project in setting up a network of stations in national parks for continuous measurements of mixing ratios of GHGs, tracers and meteorological parameters (www.climadat.es,last access: April 2018).The IC3 network mainly aims to monitor and study the exchange of GHGs between the land surface and the lower atmosphere (troposphere) in different ecosystems, which are characterized by different biogenic and anthropogenic processes, under different synoptic conditions.
In this study we analysed the time series of atmospheric CH 4 mixing ratios measured at the IC3 station in Gredos and Iruelas (GIC3) between January 2013 and December 2015.The main aim was to investigate the main drivers that influence the daily and seasonal variability of methane concentrations in this mountainous rural southern European region.The GIC3 station is located on the Spanish plateau, an area mainly characterized by livestock activity and where transhumance is still practiced (Ruiz Perez and Valero Sáez, 1990).This is an ancestral activity consisting of the seasonal movement of livestock over long distances to reach warmer regions during the winter together with a return to the mountains in summer where pastures are greener and more suitable for grazing activities (Ruiz Perez and Valero Sáez, 1990;López Sáez et al., 2009).The livestock leaves the GIC3 region to go to southern Spanish regions during the cold period.The enteric fermentation due to digestive processes in animals could thus be a significant CH 4 source in this area.The Unión de Pequeños Agricultores (UPA, 2009) reports that between 2004 and 2009 an average of 800 000 transhumant animals were hosted in Spain and 40 000 (5 % of total) were counted in the province of Ávila (extension: 8048 km 2 ), where the GIC3 station is located.According to the available literature, in this area 85 % of livestock still performs transhumance, with 500 stockbreeders moving every winter from the Gredos national park (GNP) to warmer areas of Spain, such as Extremadura (Ruiz Perez and Valero Sáez, 1990;López Sáez et al., 2009;Libro Blanco, 2013).Generally, this mobility of cattle and associated CH 4 emissions (i.e. a major regional CH 4 source) cannot easily be included in country-wide (annual) inventories because it has not yet been properly quantified and reported by nations.The present study aims to highlight the utility of 222 Rn as a tracer to retrieve independent GHG fluxes on a monthly basis using atmospheric 222 Rn and CH 4 data.This work represents a first step towards a better characterisation of transient sources, such as transhumant livestock for CH 4 , which could help to improve national emissions inventories.Finally, it offers new CH 4 data for an under-sampled area which will help in the improvement of the regional and global methane budgets.
GIC3 is a new atmospheric station so its location, the surrounding region and the instrumentation used at this station are described in the methodology section of this paper.In the first part of the results section both the daily and seasonal changes in CH 4 mixing ratios observed at the GIC3 station have been analysed in relation to 222 Rn and PBLH variability.In the second part, the nocturnal CH 4 fluxes and their monthly variability have been estimated by the RTM, following Vogel et al. (2012), and using an emission inventory for CH 4 (EDGARv4.2).Both flux estimation methods have been applied using the same source region as modelled by the atmospheric transport model FLEXPARTv9.0.2.The possible influence of large cities surrounding GIC3 and of seasonally changing meteorological conditions on the retrieved CH 4 fluxes has also been investigated.Finally, the difference in CH 4 fluxes between the cattle season, when livestock is present in the GIC3 region, and the no-cattle season, when the transhumant cattle have migrated to the south of Spain, calculated using the RTM, has been estimated.

Study site: Gredos and Iruelas station (GIC3)
The Gredos and Iruelas station is located in a rural region of the Spanish central plateau (40.35 • N, 5.17 • E; 1440 m above sea level -a.s.l.), as shown in Fig. S1 of the Supplement.GIC3 is located on the west side of the GNP, which has a total extension of 86 397 ha.The mountains of the GNP form the highest mountain range in the east-west-orientated central mountain system.The GNP has a, predominantly, granitic basement and is thus covered by soil with high activity levels of 228 U (Nazaroff and Nero, 1988).The average 222 Rn flux in this area is about 70-100 Bq m −2 h −1 (e.g.López-Coto et al., 2013;Karstens et al., 2015), which is almost twice the average radon flux in central Europe (Szegvary et al., 2009, López-Coto et al., 2013;Grossi et al., 2016).The vegetation in the GIC3 area is stratified according to altitude and the main land use practice is a mixture of agro-forestry exploitation (EEA, 2013) Livestock farming is one of the main economic activities in the area around the GIC3 station (Ruiz Perez and Valero Sáez, 1990;López Saéz et al., 2009;MMA, 2016;Hernández, 2016).In the GNP the seasonal migration of livestock starts between November and December, when they travel to the south of the Iberian Peninsula, and they do not return until late May-mid-June (Ruiz Perez and Valero Sáez, 1990).In Fig. S2, a map of the main Spanish transhumant paths is presented.Unfortunately, no specific reports of cattle mobility data are so far available for the GIC3 area.
Besides livestock activities, there are three small-sized to medium-sized water reservoirs and four medium-sized to large cities in the wider area surrounding GIC3.The water reservoirs as well as several facilities present in the cities, e.g.landfills or waste water treatment plants, represent CH 4 sources which could also influence methane concentrations observed at the GIC3 station under specific synoptic conditions.The water reservoirs are located in the west and northwest area of GIC3: (i) The Gabriel and Galan reservoir with an extension of 4683 ha (40.25 • N; −6.13 • E; 80 km away from GIC3), (ii) Santa Teresa with an extension of 2663 ha (40.60 • N; −5.58 • E; 42 km away from GIC3), and (iii) Almendra with an extension of 7940 ha (41.25 • N; −6.26 • E; 120 km away from GIC3).The metropolitan area of Madrid, which comprises about 6.3 million inhabitants, is situated approximately 120 km to the east of GIC3.Valladolid, located 150 km to the west of GIC3, is reported to have approximately 416 000 inhabitants, while smaller cities like Salamanca (84 km to the north-west) and Ávila (55 km to the north-east) only have 229 000 and 59 000 inhabitants, respectively.More information about these four cities is reported in Table S1 of the Supplement.

Air sampling
Atmospheric CH 4 , CO 2 and 222 Rn concentrations have been continuously measured since November 2012 at the GIC3 station (air inlet 20 m above ground level (a.g.l.) tower).CH 4 and CO 2 are measured with a frequency of 0.2 Hz using a G2301 analyser (Picarro Inc., USA).Hourly atmospheric 222 Rn concentrations are measured using an atmospheric radon monitor (ARMON) (Grossi et al., 2012(Grossi et al., , 2016)).A schematic diagram of the measurement set-up used at the GIC3 station is shown in Fig. S3.
The Picarro Inc. G2301 analyser is calibrated every 2 weeks using four secondary working gas standards, which are calibrated at the beginning and end of their lifetime against seven standards of the National Oceanic and Atmospheric Administration (NOAA) (calibration scales are WMO-CO 2 -X2007 and WMO-CH 4 -X2004 for CO 2 and CH 4 , respectively).A target gas is analysed daily for 20 min in order to check the stability and quality of the instrument calibration.For the length of the study, the instrument repeatability for CH 4 was 0.80 nmol mol −1 , the long-term reproducibility was 0.36 nmol mol −1 and the observe bias was 0.81 nmol mol −1 .Previous values were calculated according to the definitions of the World Meteorological Organization (WMO, 2009).The ARMON instrument was installed at the GIC3 station in collaboration with the Institute of Energetic Techniques of the Universitat Politècnica de Catalunya (INTE-UPC).The ARMON is a self-designed instrument based on α spectrometry of 218 Po, collected electrostatically on a passive implanted detector.The monitor has a minimum detectable activity of 150 mBq m −3 (Grossi et al., 2012).The performance of the ARMON was previously tested against a widely used 222 Rn progeny monitor and good results were observed (Grossi et al., 2016).
The responses of both the ARMON and G2301 analysers are influenced by the air sample humidity level.Water correction factors for both instruments are empirically determined and corrected following Grossi et al. (2012) and Rella (2010), respectively.

Sample air drying system
The instruments used at the GIC3 station require a total flow of 2.5 L min −1 of sample air dried to a water concentration lower than 1000 ppm to perform simultaneous measurements of GHG and 222 Rn concentrations.In the GIC3 inlet system, as shown in Fig. S3, the sample air is passed through a Nafion ® membrane (Permapure, PD-100T-24MPS) that exchanges water molecules with a dry counter-current air flow.The counter-current air flow is dried in a two-step process, first through a cooling coil in a refrigerator at 3 • C and a pressure of 5.5 barg, and then a cryotrap is used at −70 • C and a pressure of 1.5 barg.Multiple cryotraps are selected with electrovalves in order to increase the autonomy of the system to about 2 months.The typical water content of sample air inside the instruments is between 100 and 200 ppm.

Meteorological observations
Meteorological variables are continuously measured at the GIC3 tower.The canopy around the tower is below 20 m.The area surrounding the GIC3 station is hilly as shown on the topographic map of Fig. S1.The tower is equipped with (1) two-dimensional sonic anemometer (WindSonic, Gill Instruments) for wind speed and direction (accuracies of ±2 % and ±3 • , respectively); (2) humidity and temperature probe (HMP 110,Vaisala) with an accuracy of ±1.7 % and ±0.2 • C, respectively; (3) barometric pressure sensor (61302V, Young Company) with an accuracy of 0.2 hPa (at 25 • C) and 0.3 hPa (from −40 to +60 • C).All the accuracies refer to the manufacturer's specifications.

Planetary boundary layer height (PBLH)
Planetary boundary layer height (PBLH) data used in this analysis have been extracted from the operational high-resolution atmospheric model of the European Centre for Medium-Range Weather Forecasting (ECMWF-HRES) (ECMWF, 2006) for the period of interest (January 2013-December 2015) for the GIC3 area.This model stores output variables every 12 h (at 00:00 and 12:00 UTC) with a temporal resolution output of 1 h and with forecasts from +00:00 to +11:00 h.The horizontal spatial resolution of the model is about 16 km.In the ECMWF-HRES model the calculation of the PBLH is based on the bulk Richardson number (Ri) (Troen and Mahrt, 1986).As regards the reliability of mod-elled PBLH data, Seidel et al. (2012) have shown that data limitations in vertical profiles introduce height uncertainties that can exceed 50 % for shallow boundary layers (< 1 km), but are generally < 20 % for deeper boundary layers.In addition, they compared radiosonde observations with re-analysis and climate models and showed that these latter two produce deeper layers due to the difficulty in simulating stable conditions.The RTM is a well-known method (e.g.Hammer and Levin, 2009) and it has been used in this study, following Vogel et al. (2012), in order to obtain observation-based estimates of nocturnal CH 4 fluxes at GIC3.The RTM uses atmospheric measurements of 222 Rn and measured, or modelled, values of 222 Rn fluxes together with atmospheric mixing ratios of a gas of interest, i.e.CH 4 , in order to retrieve the net fluxes of this gas (e.g.Hammer and Levin, 2009;Grossi et al., 2014).This method is based on the assumption that the nocturnal lower atmospheric boundary layer can be described as a wellmixed box of air (Schmidt et al., 1996;Levin et al., 2011;Vogel et al., 2012).The boundary layer is considered homogeneous within the box and with a time-varying height.

CH
No significant horizontal advection is considered due to stable atmospheric conditions (Griffiths et al., 2012).In this atmospheric volume the variation of the concentration of any tracer (shown with the subindex i) with time C i (t) will be proportional to the flux of the tracer F i (t) and inversely proportional to the height of the boundary layer h(t) (Eq.1; e.g.Galmarini, 2006;Griffiths et al., 2012;Vogel et al., 2012;Grossi et al., 2014).
Applying Eq. ( 1) for both 222 Rn and CH 4 , Equation ( 2) is obtained, with a dimensionless conversion factor c derived from the observed slope of the concurrent concentration increase in both gases: observing the concentration increase in two gases that fulfil the above assumptions, here CH 4 and 222 Rn.If the flux of 222 Rn is known, then the flux of CH 4 can be calculated (Levin et al., 2011).A description of the specific criteria used to implement the RTM can be found in detail in Vogel et al. (2012).Grossi et al. (2014) previously applied the RTM for the first time at the GIC3 station using only a 3-month dataset and with a constant (in time and space) 222 Rn flux of 60 Bq m −2 h −1 .Here, in order to apply the RTM to retrieve a time series of CH 4 fluxes (FR_CH 4 ) during 2013-2015 at the GIC3 station and to compare these results with those obtained using a bottom-up inventory for methane (FE_CH 4 ), we used the following extensive set-up: 1.A nocturnal window between 20:00 and 05:00 UTC was selected for each single night analysis in order to utilise only accumulation events in which atmospheric concentrations of both CH 4 and 222 Rn had a positive concentration gradient due to positive net fluxes under stable boundary layer conditions.
2. A data selection criterion based on a threshold of R 2 ≥ 0.8 for the linear correlation between 222 Rn and CH 4 was used to reject events with low linear correlation between the atmospheric concentrations of both gases.(Stohl, 1998).Local radon flux data were calculated as explained in the following paragraph, while the footprints obtained are described in Sect.2.4.3.
The radon flux model of the University of Huelva (from now on named the UHU model) employed in this work has been described in detail by López-Coto et al. (2013).By using this model, a time-dependent inventory was calculated for the period 2011-2014 by employing several dynamic inputs, namely soil moisture, soil temperature and snow cover thickness.These data were obtained directly from Weather Research and Forecasting (WRF) simulations (Skamarock et al., 2008).A domain of 97 × 97 grid cells centred on Spain with a spatial resolution of 27 × 27 km 2 and a temporal resolution of 1 h was defined.The 222 Rn flux data calculated using this model were only available until November 2014 due to a lack of WRF simulations.In order to obtain data for this period when modelled 222     , 2010). Figure 1 shows the EDGAR inventory grid map extracted for Spain.
The influence of the emissions associated with the cities surrounding the region of GIC3 was also modelled using this inventory to better understand their impact.In Table S1 the coordinates of the upper and lower corners of the areas used to describe the location of the metropolitan areas over the EDGAR inventory are reported.

Footprints
The Lagrangian particle dispersion model FLEXPARTv9.0.2 has been extensively validated and is nowadays widely used by the scientific community to calculate atmospheric sourcereceptor relationships for atmospheric gases and organic particles (e.g.Stohl, 1998;Stohl et al., 2005;Arnold et al., 2010;Font et al., 2013;Tohjima et al., 2014).FLEXPART allows computation of the trajectories of virtual air parcels arriving at the receptor point, i.e. the GIC3 station, at a specific time.FLEXPART has been applied here to calculate 24 h backward trajectories of 10 000 virtual air parcels starting at 00:00 UTC for each night of the period 2013-2015.Each back trajectory simulation was run with a time-step output of 3 h.Meteorological data from the operational ECMWF-HRES model with a resolution of 0.2 • were used as input fields for the FLEXPART modelling.The FLEXPART output domain resolution was 0.2 • .The domain was set at 25 • N, 40 • W for the lowest left corner and 65 • N, 10 • W for the upper right corner.A nested output domain of 0.05 • resolution was defined at 37 • N, 12 • W for the lowest left corner and 43 • N, 0 • E for the upper right corner.The FLEXPART model accounts for both the vertical and horizontal position of the virtual air parcels and their residence time in each grid cell.This information allows the influence of atmospheresurface exchange to be estimated on the observed concentrations if air parcels are within the boundary layer.A maximum height of 300 m a.g.l. has been selected for the footprint analysis following Font et al. (2013).The average nocturnal footprint for the period 2013-2015 is presented in Fig. 2. The footprints obtained for the nested FLEXPART domain were combined with the EDGAR inventory map for CH 4 emissions (EDGAR, 2010) and with the UHU 222 Rn flux inventory map (López-Coto et al., 2013), separately, in order to obtain the time series of modelled CH 4 and effective 222 Rn fluxes.The resulting mean flux F i (S, t n ), for each gas i, at the receptor S (GIC3 station) and for each night t n , with n ranging over the 3-year period, is thus given by Eq. ( 3): where t ranges between the 24 h of back-trajectories, F i (x, t) denotes the flux of a given grid cell x at time t derived from the EDGAR or UHU inventory map, separately.The weighting factor of each grid cell w(x, T ) is calculated using the FLEXPART footprint for each night t n over the 3-year period and it has been calculated by normalising the residence time of each grid cell over the nested domain and during the 24 h back-trajectories (T ), as given by Eq. ( 4): 3 Results In this section we present the results of the daily and seasonal atmospheric CH 4 variability at GIC3 station analysed using a record of 3-year hourly CH 4 and 222 Rn time series.
Unfortunately, due to problems in the air sample system, data for 11 % of the time period are not available, mainly in the summer of 2013.Grossi et al. (2016) presented a complete characterisation of the main meteorological conditions and 222 Rn behaviour at the ClimaDat stations including GIC3, and we will use these previous results to interpret atmospheric processes and the variability of CH 4 mixing ratio, as well as to understand the dominating wind regimes for CH 4 flux data analysis (Fig. S4 presents the monthly wind regimes observed at the GIC3 station both for day-time and night-time).

Statistics of the daily and seasonal atmospheric CH 4 variability
The 3-year hourly time series of atmospheric CH 4 mixing ratios measured at GIC3 shows a median value of 1904.5 nmol mol −1 with an absolute deviation of 29.6 nmol mol −1 .The box plots in Fig. 3 present the medians of the atmospheric CH 4 mixing ratios and 222 Rn concentrations measured at the GIC3 station over the dataset on an hourly (left panels) and a monthly (right panels) basis.Monthly means have been calculated separately for day-time (07:00-18:00 UTC) and night-time (19:00-06:00 UTC).
The maximum hourly median methane mixing ratio measured within the 3-year observation period is 1921.1 nmol mol −1 and is observed at 03:00 UTC, whereas the minimum hourly median value of 1889.9 nmol mol −1 is observed at 13:00 UTC.The absolute standard deviation of the hourly median is 16.97 nmol mol −1 .The hourly median daily amplitude at this station, between the minimum and the maximum, is 31.18nmol mol −1 .CH 4 concentrations usually start decreasing at GIC3 in the morning at around 07:00 and 08:00 UTC and begin to increase again in the afternoon at around 17:00 and 18:00 UTC.Night-time CH 4 concentrations present an absolute standard deviation of 60 nmol mol −1 , while for day-time concentrations it is 30 nmol mol −1 .The same pattern is observed in the daily cycle of atmospheric 222 Rn (Grossi et al., 2016).Monthly daytime and night-time medians of CH 4 mixing ratios and 222 Rn concentrations show different patterns, as seen in Fig. 3b,  d.The night-time monthly medians of methane mixing ratio measured in the months between June and December look higher than those measured between January and May.Night-time monthly medians of measured222 Rn concentration are highest between July and August.

Daily and seasonal PBLH variability
Figure 4 shows the hourly median (a) and the monthly median (b) variability of the PBLH data extracted from the ECMWF-HRES model for the grid containing the GIC3 station.On a daily basis the hourly median of the PBLH reaches its minimum during night-time between 23:00 and 07:00 UTC.The PBLH starts to increase at around 08:00 UTC, reaching its maximum between 14:00 and 16:00 UTC and then decreases again after 17:00 UTC.On a monthly basis, the day-time monthly median PBLH reaches its minimum during the winter months of January and December, while it reaches its maximum in the summer months.The highest night-time monthly medians for the PBL heights are observed in winter.The day-time monthly PBLH medians present a quite symmetric distribution (around July as a centre line), similar to the night-time monthly 222 Rn medians (Fig. 3d).For each median (black bold line) the 25th (Q1; lower box limit) and 75th (Q3; upper box limit) percentiles are reported in the plot.The lower whisker goes from Q1 to the smallest non-outlier in the data set, and the upper whisker goes from Q3 to the largest non-outlier.Outliers are defined as > 1.5 IQR or < 1.5 IQR (IQR: interquartile range).

Comparison between CH 4 and 222 Rn variability
A comparison of the daily and seasonal variability of the atmospheric concentrations of 222 Rn and CH 4 in relation to changes in height of the PBL at the GIC3 station (2013)(2014)(2015) is presented in Figs. 5 and 6, respectively.
The daily evolution of hourly means of the 222 Rn atmospheric concentrations (Fig. 5a) implies that on a daily timescale, when 222 Rn flux can be considered fairly constant (e.g.López-Coto et al., 2013), PBLH variations drive the increase or decrease in the atmospheric 222 Rn concentrations.In this sense, 222 Rn seems to be an excellent predictor of PBLH (and vice versa) on a daily timescale.Looking at the hourly means of the atmospheric CH 4 mixing ratios (Fig. 5b), we can observe that methane decreases as the PBLH increases, as was observed for 222 Rn.However, between 12:00 and 18:00 UTC higher values in CH 4 mixing ratios relative to the values observed between 10:00 and 12:00 UTC are observed, which have similar PBLH conditions and could indicate some daily variability in the methane fluxes.
To interpret the monthly variability, the daily amplitude for each gas, i.e. 222 Rn for radon and CH 4 for methane, was calculated in order to subtract the influence of the changing daily background contribution measured at the GIC3 station.The term 222 Rn is defined as the difference between average night-time concentration data (19:00-06:00 UTC) versus average day-time (07:00-18:00 UTC) concentration data (Eq.5), and CH 4 has been calculated accordingly.
Figure 6 reveals that monthly amplitudes increase in summer, when the day-time PBLH increases very strongly due to vertical mixing (see Fig. 4).This general tendency is found both for 222 Rn and CH 4 concentrations.Concentration am- plitudes of 222 Rn are higher in autumn than in winter under the same PBLH conditions (Fig. 6a).This could indicate that some process other than PBLH is driving this difference in the 222 Rn concentrations.In Fig. 7 we observe how the monthly 222 Rn flux calculated by the UHU model (presented in Sect.2.4) changes.
In agreement with Grossi et al. (2016), we find a lower 222 Rn flux at GIC3 during winter due to snow cover events and low temperatures, which prevent 222 Rn diffusion from the soil.The 222 Rn flux then increases almost two-fold and three-fold during the autumn and summer months, respectively.This is due to drier soil conditions and the high gradient of temperature in the surface atmospheric layer which facilitates the escape of 222 Rn from the pores of the granitic soil (Nazaroff and Nero, 1988).This seasonality of the 222 Rn flux could be the main cause of the increased atmospheric 222 Rn under the same PBLH conditions.Monthly variations of CH 4 shown in Fig. 6 (bottom panel) display no clear simple correlation with PBLH.The CH 4 appears to be higher between the months of June and December irrespective of the corresponding PBLH values.

Variations of CH 4 fluxes
So far, daily variations for both CH 4 mixing ratio and 222 Rn concentrations can be mainly explained in relation to the accumulation or dilution of gas concentrations within the PBL.However, the hysteresis observed for the CH 4 mixing ratio of Fig. 5b seems to indicate a small change in the methane source between 12:00 and 18:00 UTC.
Monthly 222 Rn variability can be understood when we account for seasonal 222 Rn flux changes.Unfortunately, existing emission inventories (EDGAR, 2010;MMA, 2016)   In order to understand the impact that temporal changes of CH 4 emissions may have on monthly mean atmospheric CH 4 mixing ratios, we have applied two different methodologies, as explained in Sect.2.4.1 and 2.4.2, and we have compared their resulting fluxes: FR_CH 4 and FE_CH 4 , respectively.Figure 8 presents the effective 222 Rn flux time series used for the application of the first methodology (RTM), together with the raw 222 Rn flux calculated by the UHU model and its seasonal climatology.
Figure 9 presents the time series of CH 4 fluxes estimated at the GIC3 station and T i (grey shaded rectangles) indicates the time when transhumant livestock returns to the GNP after spending the winter in the south of Spain (Tapias, 2014;Rodríguez, 2015).The green shaded areas indicate the periods, between June and December, when transhumant livestock typically stays in the GIC3 region (Ruiz Perez and Valero Sáez, 1990;López Sáez et al., 2009;Libro Blanco, 2013).Data coverage in the second part of the time series (2014-2015) is higher than in the first period (2013-2014) because the simultaneous availability of 222 Rn and CH 4 data was higher.The mean of FR_CH 4 fluxes over the dataset is 0.11 mg CH 4 m −2 h −1 with 25th and 75th percentiles of 0.07 and 0.14 mg CH 4 m −2 h −1 , respectively.The mean of FE_CH 4 fluxes over the dataset is 0.33 mg CH 4 m −2 h −1 with 25th and 75th percentiles of 0.28 mg CH 4 m −2 h −1 and 0.36 mg CH 4 m −2 h −1 , respectively.FR_CH 4 fluxes are constantly lower than FE_CH 4 fluxes, although this discrepancy decreases during some periods, as we will investigate later.FEC_CH 4 fluxes obtained with the EDGARv4.2inventory by considering only the contribution of the cities that are located around the GIC3 station, in agreement with the masks presented in Table S1, had a total mean value over the dataset Figure 10 shows monthly box plots of FE_CH 4 and FR_CH 4 fluxes.Shaded areas are coloured according to the main local wind directions reaching the GIC3 station at night.This classification is based on the results presented in Fig. S2, where monthly windrose plots for the GIC3 station between 2013 and 2015 are shown.We can observe that there is no variability in monthly FE_CH 4 flux values.In contrast, FR_CH 4 flux results show an increase in CH 4 fluxes between June and December that seems to be independent of the seasonally changing dominant wind directions.This increase is also uncorrelated with seasonally changing 222 Rn fluxes (Fig. 7).The seasonal change of CH 4 fluxes between the first and the second half of the year at GIC3 could indeed be related to variations in local CH 4 emissions.The period between June and December represents the time of year when transhumant livestock returns to the GNP.
The contribution of cities is only visible during certain months and it seems to be related with winds coming from the east in the direction of the Madrid urban area (see Fig. S2).
The disagreement observed between FE_CH 4 and FR_CH 4 fluxes in the months between June and December (Figs. 9 and 10), when the transhumant livestock is in the GIC3 area, may be due to different reasons: 1.A possible underestimation of the 222 Rn flux outputs from the UHU radon flux model could occur, which would lead to lower FR_CH 4 fluxes (Eq.2).As ex- 2. The methodology used within the EDGAR for the spatial disaggregation of national sector emission over the country could lead to a distribution of CH 4 emission in the GIC3 region higher than true levels leading to an overestimation of the FE_CH 4 .
3. The fixed height of 300 m used for the calculation of nocturnal footprints could introduce a bias.However, this value is well within the range of nocturnal PBLH values calculated with data extracted from the ECMWF-HRES model.Furthermore, the calculated FLEXPART footprints were used both for FR_CH 4 and FE_CH 4 calculations and this should not affect the relative differences between their values.
When applying a 40 % increase for the local 222 Rn source, as suggested by Karstens et al. (2015), we can re-calculate FR_CH 4 emissions as FR_CH 4 _rescale.The box plot of the monthly medians of FE_CH 4 , FR_CH 4 and FR_CH 4 _rescale are compared in Fig. 11.The mean of FR_CH 4 _rescale fluxes over the dataset is 0.29 mg CH 4 m −2 h −1 with 25th and 75th percentiles of 0.17 mg CH 4 and 0.34 mg CH 4 m −2 h −1 , respectively.FR_CH 4 _rescale is in agreement with FE_CH 4 fluxes during the months between June and December, when the transhumant livestock remains in the GIC3 area (cattle season).
To highlight seasonal differences, FE_CH 4 , FR_CH 4 and FR_CH 4 _rescale fluxes are aggregated into two box plots in Fig. 12, according to the no-cattle season (January until May), when there is no livestock in the GIC3 area, and cattle season (June until December).According to these data during the no-cattle season, FR_CH 4 fluxes present a mean value of 0.09 CH 4 m −2 h −1 with a standard deviation of 0.15 mg CH 4 m −2 h −1 .During the cattle season, the mean value of FR_CH 4 fluxes is 0.12 CH 4 m −2 h −1 with a standard deviation of 0.05 mg CH 4 m −2 h −1 .The mean value of FR_CH 4 _rescale fluxes is 0.24 mg CH 4 m −2 h −1 during the no-cattle season with a standard deviation of 0.39 mg CH 4 m −2 h −1 and it is 0.30 mg CH 4 m −2 h −1 during the cattle season with a standard deviation of 0.12 mg CH 4 m −2 h −1 .The corresponding values for FE_CH 4 fluxes are 0.31 mg CH 4 m −2 h −1 for the no-cattle season and 0.32 mg CH 4 m −2 h −1 for the cattle season.

Discussion
The present results show the different influences that meteorological conditions (PBLH and wind direction) and regional sources may have on the variability of atmospheric CH 4 concentrations observed at the GIC3 station.The 222 Rn observations have been used, together with modelled PBLH data, to better understand the reasons for the variability of the atmospheric CH 4 concentrations observed at the station for different times scales.The use of 222 Rn as a tracer to calculate independent fluxes of GHGs has been shown in order to help with the improvement of emission inventories on a regional scale.

Daily variability of atmospheric CH 4 concentrations
The daily cycle of atmospheric CH 4 mixing ratios (Fig. 3a) measured at GIC3 shows significant changes between daytime and night-time periods.The large increase in nocturnal CH 4 mixing ratios can mainly be explained by the decreased height of the planetary boundary layer (Fig. 4a), which is supported by a similar behaviour of 222 Rn concentrations (Fig. 3c).Indeed, CH 4 , as well as 222 Rn, reaches its maximum concentration when the PBLH is below 300 m a.g.l.
during the night, while their atmospheric concentrations decrease with the increase in the PBLH during day-time.
The correlation of PBLH and 222 Rn (and CH 4 ) in Fig. 5 indicates that 222 Rn fluxes do not strongly vary on daily timescales or, at least, not to a degree that can influence their atmospheric concentration variability.CH 4 fluxes seem to change on a daily timescale.Average afternoon CH 4 concentrations are slightly enhanced compared to those from the morning for similar PBLH values (Fig. 5b).They show a hysteresis behaviour which could indicate local emissions increase or that a systematic transport of CH 4 enhanced air masses, not rich in radon, occurs at GIC3.Some studies (e.g.Bilek et al., 2001;Wang et al., 2015) have found strong emission increases from dairy cows after feeding in feedlots, while McGinn et al. (2010) only found small diurnal increases in CH 4 emissions between 11 and 17 h for grazing cattle.Unfortunately, no detailed information about the feeding cycle of the GIC3 livestock is available, but grazing should be considered the predominant form of livestock management in transhumance.However, Figs. 9 and 10 together with Fig. S4 show the influences of eastern winds, coming from the Madrid direction, on the CH 4 fluxes.

Seasonal variability of atmospheric CH 4 concentrations
To understand the drivers of monthly changing concentrations of CH 4 , we need to account for PBLH local meteorology, changing regional emissions and changing background concentrations of CH 4 at GIC3.Median monthly mixing ratios for day-time and night-time (Fig. 3b) are discussed alongside CH 4 (Fig. 6), which allows us to subtract seasonal and synoptic background variations.This enables us to focus on the impact of PBLH for individual days that are then averaged to investigate how CH 4 changes on a monthly basis.The observed variability of CH 4 (Figs.3b and 6) cannot be explained only in terms of changes of the PBLH.Monthly averages of CH 4 (and night-time monthly CH 4 box plots, Fig. 3b) present their maximum values between June and December, and their minimum values during the rest of the months irrespective of the height of the PBL.From co-located 222 Rn concentration observations we learn that an increase in the average monthly fluxes (Fig. 7) can compensate the effect of increased dilution in the deeper summer PBL on the observed concentrations (Fig. 6a), yielding similar atmospheric 222 Rn concentrations.The increase in the modelled 222 Rn flux in the GIC3 region from the winter to autumn season and the following decrease can coherently help to explain the variation observed in monthly 222 Rn.Thus, the comparison between CH 4 and 222 Rn suggests that there may also be a monthly variability in the sources of CH 4 which should help to understand monthly atmospheric mixing ratios variability.This has been further confirmed by our FR_CH 4 flux estimates, as seen in Figs. 9, 10 and 11.Of course, the FR_CH 4 flux estimates are limited to nighttime due to the RTM hypothesis.FR_CH 4 fluxes show a total mean value 33 % lower than FE_CH 4 fluxes over the data set.When 222 Rn fluxes are rescaled according to Karstens et al. (2015), this difference is drastically reduced to 10-15 %.RTM-based CH 4 fluxes show an increase of 25 % during the second semester of the year on a monthly basis.This increase coincides with the period of the year when transhumant livestock resides in the GIC3 region.Although no exact information is available on the number of animals present only in the GIC3 area, during this period of enhanced ruminant emissions, the difference between CH 4 fluxes based on RTM and the EDGAR inventory is reduced from 73 to 65 % for FR_CH 4 and from 27 to 9 % for FR_CH 4 _rescale The difference during the no-cattle season is likely due to the constant annual emission factor of CH 4 emission used within the bottom-up inventory which, of course, cannot yet reflect transhumance activity.The likely explanation is that all emissions from the aforementioned animals has been constantly allocated to this region, which is why FE_CH 4 is also larger than FR_CH 4 _rescale during months when they are www.atmos-chem-phys.net/18/5847/2018/Atmos.Chem.Phys., 18, 5847-5860, 2018 not present.The RTM analysis performed here suggests that transhumance could be a relevant process to the understanding of sub-annual CH 4 emissions in the region and can affect the spatial distribution of CH 4 sources within a country.Our study indicates that the choice of 222 Rn model has an important impact on annual total emissions calculated, while seasonal and short-term patterns are preserved.

Conclusions and outlook
To gain a full picture of the Spanish (and European) GHG balance, understanding of CH 4 emissions in different regions is a critical challenge, as is the improvement of bottom-up inventories for all European regions.Our study uses, among other elements, GHG, meteorological and 222 Rn tracer data from one of the eight stations of the new ClimaDat network in Spain, which provides continuous atmospheric observations of CH 4 and 222 Rn in a region of Europe.The present study underlines the fact that this data, combined with retrieved PBLH data and atmospheric transport modelling (FLEXPARTv92) can help to understand the main causes of temporal variability of GHG mixing ratios and can offer new insights into regional emissions by identifying the impacts of changing sources, e.g.emissions from transient livestock.
These first promising results should lead to further application of this RTM to other GHG time series from the Cli-maDat network and potentially in continent-wide networks such as ICOS that routinely perform co-located GHG and 222 Rn observations.Particularly, the usefulness of the RTM has been shown, while also highlighting the need to improve this method, especially in regard to (i) validation of the 222 Rn flux maps applied within the RTM and (ii) standardization of the footprint calculation.
Although the transhumant livestock seems to be the likely reason for the seasonal changes observed in the FR_CH 4 fluxes at the GIC3 station, other sources could also contribute to this seasonality, such as waterbodies or other natural emissions.These previous sources are not included in the EDGAR inventory, but they could be detected by the RTM.However, those sources would not be able to fully explain the sudden onset of increased RTM-based CH 4 fluxes but would rather contribute to a slow increase in warmer months.Further research applying isotopic analysis of CH 4 mixing ratios measured at the GIC3 station for the different seasons should be carried out, as well as transects of the regions to assess the impact of natural sources on CH 4 mixing ratios.In addition, no precise data on transhumant activity in Spain is available to date, but our study suggests the existence of a link between regional CH 4 fluxes and highlights the need for more information on transhumance activity which could be taken into account in future emission inventories of this region (and Europe).In addition, our results show that urban emissions can be transported and could influence the atmospheric compo-sition of remote rural areas over several hundred kilometres under specific synoptic conditions.Data availability.The datasets are available at http://www.climadat.es/es/datos/.
Competing interests.The authors declare that they have no conflict of interest.
4 fluxes 2.4.1 CH 4 fluxes based on FLEXPART footprints and the radon tracer method 3. An effective radon flux influencing the GIC3 station each night from 2013 to 2015 was calculated by coupling local radon flux data, obtained using the output for the local pixel containing the GIC3 station of the model developed by López-Coto et al. (2013), with the footprints calculated by the ECMWF-FLEXPART model (version 9.02) Rn flux data were not available, from December 2014 to December 2015, a seasonal and monthly climatology was calculated by using the UHU data set model for the years 2011-2014.Karstens et al. (2015) compared the 222 Rn flux values calculated over Europe by their model to UHU values and to long-term direct measurements of 222 Rn exhalation rates in different areas of Europe.They found a generally 40 % higher222 Rn exhalation rate on their map than estimated by the UHU map over Europe.This previous result has been taken into consideration within the present study to better interpret the obtained data.
2.5 CH 4 fluxes based on FLEXPART footprints and the EDGARv4.2inventory grid map Bottom-up CH 4 fluxes influencing the GIC3 station were estimated by using the footprints calculated by the ECMWF-FLEXPART model (obtained as described in Sect.2.4.3) and the Emissions Database for Global Atmospheric Research (EDGAR) version 4.2 (EDGAR, 2010).The EDGAR inventory, developed by the European Commission Joint Research Centre and the Netherlands Environmental Assessment Agency, includes global anthropogenic emissions of GHGs and air pollutants by country on a spatial grid.The EDGAR version used in the present study provides global annual CH 4 emissions on a 0.1 • (11 km) resolution for the year 2010.All major anthropogenic source sectors, e.g.waste treatment, industrial and agricultural sources (e.g.enteric fermentation) are included, whereas natural sources (e.g.wetlands or rivers) are not.The spatial allocation of emissions on 0.1 • by 0.1 • grid cells in EDGAR has been built up by using spatial proxy datasets with the location of energy and manufacturing facilities, road networks, shipping routes, human and animal population density, and agricultural land use.UN-FCCC reported national sector totals are then removed with the given percentages of the spatial proxies over the country's area (EDGAR

Figure 2 .
Figure 2. Average nocturnal FLEXPART footprint for the 2013-2015 period (residence time t is on the logarithmic scale).

Figure 3 .
Figure 3. Box plots of hourly (a, c) and monthly (b, d) atmospheric CH 4 mixing ratios (a, b) and 222 Rn concentrations(c, d) measured from January 2013 to December 2015 at the GIC3 station.For each median (black bold line) the 25th (Q1; lower box limit) and 75th (Q3; upper box limit) percentiles are reported in the plot.The lower whisker goes from Q1 to the smallest non-outlier in the data set, and the upper whisker goes from Q3 to the largest non-outlier.Outliers are defined as > 1.5 IQR or < 1.5 IQR (IQR: interquartile range).

Figure 4 .
Figure 4. Box plots of hourly (a) and monthly (b) PBLH data extracted from the ECMWF-HRES model for the period January 2013-December 2015 at the GIC3 station.For each median (black bold line) the 25th (Q1; lower box limit) and 75th (Q3; upper box limit) percentiles are reported in the plot.The lower whisker goes from Q1 to the smallest non-outlier in the data set, and the upper whisker goes from Q3 to the largest non-outlier.Outliers are defined as > 1.5 IQR or < 1.5 IQR (IQR: interquartile range).

Figure 5 .
Figure 5. Relation between hourly means of atmospheric CH 4 (b) and 222 Rn (a) concentrations measured during 2013-2015 at the GIC3 station and ECMWF data of PBLH for the same area and for the same time interval.

Figure 6 .
Figure 6.Relation between monthly means of concentration amplitudes of CH 4 (b) and 222 Rn (a) measured during 2013-2015 at the GIC3 station and monthly ECMWF data of PBLH for the same area during same time interval.

Figure 7 .
Figure 7. Monthly 222 Rn flux means calculated by the UHU model and climatology for 2013-2015 at the GIC3 station.Coloured circles indicate the same months as in Fig. 6.

Figure 8 .
Figure 8.Time series of local 222 Rn flux calculated by the UHU model (black line; López-Coto et al., 2013) for the GIC3 area, 222 Rn flux seasonal climatology (blue line) and effective 222 Rn flux calculated on the basis of FLEXPART footprints (red dots).This last series was used within the RTM.

Figure 9 .
Figure 9. Results of night-time FR_CH 4 fluxes (mg CH 4 m −2 h −1 ) (red circles) obtained at the GIC3 station from January 2013 to December 2015 compared with night-time FE_CH 4 fluxes obtained using bottom-up inventory emissions (grey line), and calculated FEC_CH 4 fluxes from contributions from surrounding cities (green circles).The weeks T i represent the period of 2014 (21-27 June) and 2015 (20-26 June) when transhumant livestock returned to the GIC3 area after spending the winter in the south of Spain and concurrent with the availability of FR_CH 4 fluxes data.Shaded green regions represent the periods when transhumant livestock remain in the GIC3 area.

Figure 10 .
Figure10.Box plots of monthly CH 4 fluxes (mg CH 4 m −2 h −1 ) calculated for the GIC3 area using the RTM (red) and the EDGAR inventory (total in yellow; contribution of cities in green).Coloured areas indicate main wind directions for specific months.For each median (black bold line) the 25th (Q1; lower box limit) and 75th (Q3; upper box limit) percentiles are reported in the plot.The lower whisker goes from Q1 to the smallest non-outlier in the data set, and the upper whisker goes from Q3 to the largest non-outlier.Outliers are defined as > 1.5 IQR or < 1.5 IQR (IQR: interquartile range).

Figure 11 .
Figure 11.Box plots of monthly CH 4 fluxes (mg CH 4 m −2 h −1 ) calculated for the GIC3 area using the RTM (red), the EDGAR inventory (yellow) and RTM using the 222 Rn flux comparison factor found by Karstens et al. (2015) (grey).Coloured areas indicate main wind directions for specific months.For each median (black bold line) the 25th (Q1; lower box limit) and 75th (Q3; upper box limit) percentiles are reported in the plot.The lower whisker goes from Q1 to the smallest non-outlier in the data set, and the upper whisker goes from Q3 to the largest non-outlier.Outliers are defined as > 1.5 IQR or < 1.5 IQR (IQR: interquartile range).

Figure 12 .
Figure12.Box plots of FE_CH 4, FR_CH 4 and FR_CH 4 _corr fluxes (in mg m −2 h −1 ) calculated for theGIC3 area during the "warm" season (June-December, yellow box) and the "cold" season (January-May, grey box).For each median (black bold line) the 25th (Q1; lower box limit) and 75th (Q3; upper box limit) percentiles are reported in the plot.The lower whisker goes from Q1 to the smallest non-outlier in the data set, and the upper whisker goes from Q3 to the largest non-outlier.Outliers are defined as > 1.5 IQR or < 1.5 IQR (IQR: interquartile range).