2.1. The RUOA network

The University Network of Atmospheric Observatories (RUOA, https://ruoa.unam.mx) is comprised of 16 measurement stations within the Mexican territory that provide high-quality data for research and teaching. Six of these stations in both urban and natural settings include high precision GHG analyzers and all of the RUOA stations are fully instrumented to routinely monitor ancillary meteorological variables such as air temperature and relative humidity, wind speed and direction, global solar radiation and barometric pressure, among other atmospheric parameters. The two stations featured in this study, located in central Mexico, are part of this network.

2.2. The urban station

The UNAM atmospheric observatory (19.3262 N, 99.1761 W, 2280 m.a.s.l.) is situated atop of a three-story building on the eastern edge of the UNAM main campus (730 ha), in southwestern Mexico City, 13 km from the city center (Fig. 1). North and west of the observatory, the campus includes low-rise buildings interspersed with lawns and patches of native shrub. To the east, there is a major public transportation hub and a compact, low-rise, highly populated area flanking the campus (Santo Domingo). An urban natural reserve with an area of 237 ha, called Reserva Ecológica del Pedregal de San Ángel (REPSA), spreads to the south and southwest of the observatory and covers a third of the campus area. This reserve represents a relict of the vegetation that covered this portion of the city before its explosive urbanization during the 20th century (^{Lot and Camarena, 2009}). The dominant vegetation type in the REPSA is xerophilous scrub with few interspersed trees, growing on shallow and low-nutrient soil over a basaltic substrate (^{Rzedowski, 1954}). The climate is temperate sub-humid (García, 1988), with summer rains. Daily records of temperature and precipitation are kept by the Meteorological Observatory of the College of Geography of UNAM since 1963 (http://observatoriometeorologico.filos.unam.mx/2016/10/metadatos-historicos). In a period of 57 years, the annual temperature averages 15.6 ºC (min/max range from 14.3 ºC to 17.4 ºC), the temperature peaks at 26.7 ºC on average in May, the warmest month of the year (range 23.2 ºC to 30.9 ºC), and the daily minimum temperature is 3.9 ºC on average in January, the coldest month (range 1.5 ºC to 6.8 ºC). Mean annual precipitation is 870 mm (ranging from 624 mm in 1982 to 1201 mm in 2006), 90% of which falls between May and October. The annual maximum occurs in August and September (~180 mm/month) and the minimum in January, when only 5 mm of total precipitation fall on average. From RUOA hourly records, at a daily scale, wind speed peaks around 17-18 h and the minimum occurs at 5-7 h. Annually, the windiest months are March and April, with daily peak wind speeds of 3.0 ± 1.1 m/s on average, while the daily minimum peak occurs in December, with only 1.9 ± 0.9 m/s on average. The diurnal cycle of wind direction will be discussed in terms of its influence on CO₂ mole fraction in subsequent sections.

2.3. The high-altitude station

The Altzomoni (ALTZ) atmospheric observatory (19.1187 N, 98.6552 W, 3985 m.a.s.l.) is located 60 km ESE in a straight line from the UNAM site (Fig. 1), and 48 km at 280º from the city of Puebla (1.5 million inhabitants). The observatory lies in a mountainous pass between the active Popocatépetl volcano to its south, and the dormant Iztaccíhuatl volcano to its north, inside the Izta-Popo-Zoquiapan National Park and Biosphere Reserve (Fig. 1), far from immediate sources of anthropogenic emissions. The immediate surroundings of the observatory are just above the timberline (3,700-3,800 m.a.s.l.) (^{Beaman, 1962}); the vegetation cover consists of subalpine grassland dominated by dense tussock grasses and some scattered individuals of Pinus hartwegii Lindl. Pine trees become more abundant at slightly lower altitudes, and along other conifer species form dense forests in the hillsides of the volcanic ridge. From hourly meteorological records kept by RUOA since 2011, the mean annual temperature is 4.9 ºC (ranging from 4.7 ºC to 5.3 ºC); the warmest month is April, with a daily maximum temperature of 9.7 ºC on average (range from 3.9 ºC to 14.8 ºC), and the coldest month is January, with a daily minimum temperature of -0.1 ºC on average (range from -6.5 ºC to 4.8 ºC). Throughout the year, the diurnal temperature oscillation is moderate (6.1 ± 1.9 ºC). Annual precipitation averages 780 mm (from 573 mm in 2016 to 1052 mm in 2013); on an annual basis, 85% of the rain falls between May and October. June and August are the wettest months of the year (~170 mm/month) while December is the driest, with only 5 mm of average cumulative rain. Daily wind speed peaks at 10 h from October to mid-March at an average of 4.7 ± 2.7 m/s; the maximum shifts to a broader peak around 16 h during March to May, at 4.1 ± 1.7 mean wind speed, and it is displaced to the late evening (20-21 h) for the whole summer (June to September), when the it reaches 5.6 ± 2.7 m/s on average, the highest daily maxima year-round. The weather is also characterized by high levels of global horizontal irradiation (5.0 kWh m^-2), and an average of 32 frost events per year. Climate is temperate, semi-cold, sub-humid, with a mild summer (^{García, 1988}).

4.1 Record filtering, seasonality extraction and trend retrieval

The spectrometers at both locations collect raw data at ~0.3 Hz. The data were averaged and their standard deviation calculated per minute, then per hour. Calibration periods and times affected by any operator’s interference with the instrument were flagged out, which amounted to 4.6% of hours at UNAM and 15.8% at ALTZ. Daily averages were calculated on the filtered datasets. Hourly data were de- trended by subtracting the daily mean from each hourly average in order to estimate diurnal cycles for each station, free of the influence of the seasonal cycle and long-term trend.

To extract the trend and the seasonal features of the evolution of CO₂ at each station, the hourly data were filtered to include only nighttime periods representative of background conditions at ALTZ and those under well-developed turbulence conditions at UNAM. We looked for correlations between dry-air CO₂ mole fractions and local meteorological conditions (e.g. wind direction, wind speed, temperature), but no clear patterns emerged. The filtering criteria were thus based only on time-of-day and statistical variability.

Seasonal features, peak-to-trough amplitudes, and long-term trends were estimated for each station using the curve fitting described in ^{Thoning et al. (1989)} and the Python code made available by NOAA (^{Thoning, 2018}). This fitting procedure finds a function with a number of polynomial terms and a number of harmonics that approximate the long-term trend of the data and the annual cycle, respectively. We used two polynomial terms and four harmonic terms in our function fitting. The residuals were then filtered through two low-pass filters, one short- term (80 days) to smooth the data, and one long-term (667 days) to capture interannual variations not determined by the fitted function. The polynomial portion of the fitted function plus the residuals of the long-term cutoff filter give a trend series that represents the de-seasonalized upward growth of the CO₂ con- centration. The first derivative of this trend series constitutes a continuous growth rate series, i.e. the rate of change of the upward trend. The trend was also used to compute a discrete annual growth rate of CO₂, defined as the difference between the average CO₂ mole fraction for the month of December of any given year and January of the following year, and the average mole fraction recorded during the same pair of months of the preceding year (^{Dlugokencky and Tans, 2019}).

4.2. Uncertainties

The two analyzers installed at UNAM and ALTZ were purchased on 2014 and they were both manufactured by Picarro Inc. at the same period (S/N CFKADS2141 and CFKADS2142). The precision of the two instruments, calculated as the standard deviation of raw data over 1-minute intervals when measuring calibration gases, are equal to 0.03 ppm and 0.024 ppm respectively for UNAM and ALTZ. This is compatible with the precisions calculated for similar instruments deployed at ICOS sites in Europe (^{Yver Kwok et al., 2021}). Unfortunately, the instruments were operated without reference measurements from summer 2014 to December 2018, so we cannot fully characterize the repeatability and drift for this period. Regular measurements of a target gas at UNAM, performed once per day since December 2018, show a very stable repeatability of 0.02 ppm. Monthly calibrations done at UNAM since December 2018 do not indicate a significant drift. We assume that the calibration coefficients established in December 2018 in the two stations had not changed since the summer of 2014. This assumption leads to an increasing uncertainty the further we go back in time. Most G2401 analyzers have a low CO₂ drift, but a drift up to 0.1 ppm yr^-1 cannot be excluded (^{Yver Kwok et al., 2015}).

It is important to point out that the calibration scale used at UNAM does not cover the full range of measurements, especially the high concentrations observed at night. Indeed, at UNAM, 8% of the daytime observations and 22% of the night time hourly means are above 450 ppm. However, the Picarro G2401 has been evaluated at different laboratories and has exhibited remarkable linearity. ^{Yver Kwok et al. (2015)} have tested the same instrument for CO₂ concentrations up to 500 ppm, and the NOAA central laboratory uses an earlier version of the same CRDS analyzer (Picarro G2301) to certify calibration material up to 600 ppm (^{Tans et al., 2017}). Both studies show the response of the CRDS instrument to be linear. The results of the linear fit to our sequences of three calibration tanks indicate that the reduced scale does not have an impact on the linearity of the instruments’ response. The residuals of the linear regressions are within a min/max range of -0.02 to +0.04 ppm and -0.02 to +0.03 ppm for UNAM and ALTZ, respectively. The mean RMSE for all instances of the calibrations are 0.01 ppm for UNAM, and 0.02 ppm for ALTZ.

The humidifying technique of the water droplet performed on site allows the evaluation of the built-in factory water vapor correction, and the determination of a new correction, specific to the tested analyzers, with an uncertainty typically around 0.03 ppm for CO₂ (^{Laurent et al., 2019}). The built-in factory correction assessment showed an unusual CO₂ response to H₂O with a residual correction error of -0.22 ± 0.02 ppm and -0.03 ± 0.02 ppm for UNAM and ALTZ respectively, over their site-specific H₂O atmospheric range. The residual error of the specific H₂O correction determined on site was estimated to be between -0.05 ppm and +0.05 ppm for CO₂, depending on the H₂O content. Even though the new specific correction seems to improve the performance, it has not been implemented on the database yet, due to our low confidence on the unusual CO₂ response to H₂O. Indeed, the droplet test must be performed again on site in order to increase this level of confidence. Moreover, the H₂O correction resi- dual error may drift over time for CO₂. The drift magnitude depends on the H₂O content: below +0.01 ppm CO₂ per year for H₂O lower than 10 000 ppm and up to +0.03 ppm CO₂ per year for H₂O between 20 000 and 25 000 ppm (^{Laurent et al., 2019}). In consequence, with a H₂O content typically below 15 000 ppm at UNAM and 10,000 ppm at ALTZ during the dry season (November to April), the H₂O correction drift is not significant (below 0.05 ppm CO₂) for either station over the 4-year period preceding the water droplet test (December 2018). During the wet season (May to October), with a H₂O content up to 25,000 and 20,000 ppm respectively for UNAM and ALTZ, the H₂O correction drift may induce a residual error up to -0.14 and -0.11 ppm of CO₂ at these locations, respectively, on the oldest data and highest H₂O content. However, based on experience collected by testing more than 50 CRDS analyzers at LSCE (^{Laurent et al., 2019}), the factory water vapor correction assessed on UNAM and ALTZ analyzers does not show a typical drifted response to H₂O. At this stage, the overall uncertainty related to the water vapor correction for UNAM and ALTZ is estimated at ±0.2 ppm and ±0.05 ppm CO₂, respectively, over the whole period.

5.1. Daily cycles

The average daily course of CO₂ molar fractions at UNAM is presented in Figure 2A, for both weekdays (Monday to Friday) and weekends/holidays. Irrespective of the day of the week, the overall pattern is consistent with a dominant effect of the boundary layer height on the CO₂ concentration. A maximum CO₂ level is reached daily between 6 and 7 h (all times are LST), then as sun rises and warms the surface, the boundary layer grows and turbulent mixing of the atmosphere increases, as described by ^{García-Franco et al. (2018)}. The CO₂ is diluted over a larger air volume, so its concentration decreases steadily, down to its minimum at 16 h. After sunset, the increase in atmospheric stability reduces the depth of the boundary layer and produces a steady buildup of CO₂ during the night in this shallower layer.

Fig. 2 A) Diurnal cycle of hourly, detrended CO₂ measurements at UNAM for weekdays (green) and weekends/holidays (blue). The polar plots inserted show the frequency distribution of wind directions for the periods 0-8 h, 8-16 h, and 16-24 h. B) Mean diurnal difference in CO₂ molar fraction sampled at 16 m and 24 m.a.g.l between March 8^th and August 23^rd, 2019. Error bars are 95% CI. C) Box and whiskers plot of difference in molar fraction between heights. Percentiles shown are the 1st and 99th (crosses), 10^th and 90^th (cap of whiskers), 25^th, 50^th and 75^th (box). The central black dot indicates the mean. 

A further inspection of the prevalent wind direction at different times of the day confirms that the boundary layer dynamics is the main controller of the observed CO₂ cycle and offers some insight into the influence of the underlying urban surface. The polar histograms inserted into Figure 2A show a rapid change in the predominant wind direction from the NW quadrant at night (23 - 8 h) to NE-SE sector for most of the day (8 - 15 h), while from 16 h to 23 h the wind directions are more evenly distributed over the NE to WNW sector. The location of the UNAM station is such that the wind coming from the west should carry the signal of the more vegetated surface that releases CO₂ at night, while lar- ger emissions could be expected from the more urbanized and almost devoid of vegetation eastern sector during the day. However, the observed daily pattern does not show an association between the relative position of the station and the wind direction, as the concentration falls steadily during the times of the day when the predominant wind comes from the heavily populated sector. The nighttime buildup might include the CO₂ respired by the vegetation on the campus side, but the observed enhancement occurs year long and even intensifies in winter, when the vegetation is less active.

In megacities where large anthropogenic emissions dominate, CO₂ molar ratio may differ between weekdays and weekends since traffic and other human activities in general diminish noticeably during non-working days. Analyses of surface CO₂ records in Melbourne (^{Coutts et al., 2007}), Lecce (Italy) (^{Contini et al., 2012}), Mexico City (^{Velasco et al., 2014}), Helsinki (^{Kilkki et al., 2015}), Sakai (^{Ueyama and Ando, 2016}) Beijing (^{Cheng et al., 2018}) and Chennai (^{Kumar and Nagendra, 2015}), have found perceptible influences of traffic volume on daily concentration and flux patterns. Even in remote marine stations and continental mountain measurements, the weekly periodicity of the anthropogenic activity of nearby cities has been detected and used to unravel the biogenic and anthropogenic contributions to atmospheric CO₂ (^{Cerveny and Coakley, 2002}; ^{Yuan et al., 2019}). However, it is also common for the daily to seasonal variability of urban environments to be driven by the dynamics of the boundary layer depth, concealing the signal of the human activity, as has been shown in locations like Basel (^{Schmutz et al., 2016}), and in suburban and central London (^{Hernández-Paniagua et al., 2015}; ^{Helfter et al., 2011}). At UNAM, we found similar daily cycles of CO₂ molar ratio for weekdays and weekends/hol- idays (Fig. 2A). On non-working days, a peak anomaly of ~17 ± 3 ppm with respect to the daily average occurs between 6 and 7 h, while during the weekdays the peak reaches 20 ± 4 ppm. The only noticeable but transient difference between days of the week occurs in this early morning maxima for April and May (not shown), when it reaches 32.6 ± 4.0 ppm above the daily average from Monday to Friday, in contrast to 23 ± 3 ppm during weekends and holidays. On the year-round average, the mean daily amplitude (peak to trough) during weekdays is 45 ± 17 ppm, and 43 ± 18 ppm during non-working days. Weekday and weekend daily courses show considerable overlap, and there are no evident features associated to diurnal changes in traffic volume.

Our analysis indicates that there is a significant positive difference between the roof level measurements and those made 8 m above, for the period between March and August 2019 (mean difference 1.23 ppm, [1.15, 1.37] 95% CI) (Fig. 2B, 2C). This gradient reaches a maximum of 4 ppm at 6 h, coinciding with the largest CO₂ accumulation in the stable surface layer before dawn (Fig. 2B). The difference decreases rapidly as the turbulent mixing takes up in the early morning, remains stable until 17 h, and begins to gradually grow again as the sun goes down. On average, between 9 h and 17 h the difference between the two sampling levels is equal to 0.31 ± 0.20 ppm.

At ALTZ, four diurnal cycles in the de-trended CO₂ mole fraction can be distinguished as the year progresses (Fig. 3). The periods roughly correspond to dry wintertime (Jan-Mar), dry springtime (Apr-May), rainy season (Jun-Oct) and late fall (Nov-Dec), when some precipitation may still occur. The overall average daily amplitude is 7.2 ± 2.5 ppm; it goes from a minimum of 5.7 ± 2.1 ppm in the dry winter period to a maximum of 8.3 ± 2.3 ppm in the rainy season. From November to May, the daily maximum occurs in the evening, between 18 and 21 h, and the mean concentration continues to decrease the rest of the night. During the rainy season the maximum is displaced to a later period (0-5 h) and the CO₂ mole fraction is mostly constant throughout the night. A sharp decrease ensues sunrise, and the minimum is reached from 9 -11 h; slightly earlier in Apr-May (~9 h) and slightly later from June to Dec (~11 h). From June to October the afternoon increase from the minimum CO₂ mole fraction is slow up to 16 h, and then increases quickly back to nighttime values.

Fig. 3 Diurnal cycles from hourly, detrended CO₂ measurements at Altzomoni (ALTZ) for different periods of the year. Inserts are circular histograms of the wind direction for the periods delimited by the light gray vertical lines. 

In general, the seasonal changes in daily patterns of surface CO₂ at the ALTZ mountain site are consistent with a combination of the aforementioned regional boundary layer dynamics and the local effect of vegetation activity. The tussock grasses in the immediate vicinity of the station and the conifers covering the hillsides of the mountainous range capture and release carbon more actively during the warm, wet season. After the onset of the rains in May, we observe a larger daily amplitude of the CO₂ cycle, a more prolonged afternoon minimum, and a larger accumulation at night from June to October, all consistent with increased photosynthetic uptake during the day and increased respiration by vegetation at night. However, a stronger vegetation sink and a more effective vertical mixing of the surface layer are correlated on daily, seasonal and annual basis, so their relative contributions to the daily patterns along the seasons are not easily unraveled.

One interesting feature of the daily cycle of the dry season, is a small but consistent enhancement from midday to the late afternoon (11-16 h). It is more prominent in April and May, and although present during the rainy season, it is considerably shortened (12-15 h). This afternoon enhancement could be the re- sult of a change in local biogenic emissions, the transport of air parcels enriched with anthropogenic CO₂ from adjacent cities, or a mixture of these situations. It has been documented that the convective boundary layer starts to develop over Mexico City one hour after sunrise and reaches 2000 m above UNAM’s elevation around midday (^{Shaw et al., 2007}, ^{García-Franco et al. 2018}). From January to May, wind from the WNW sector dominate from 0 to 16 h (Fig. 3), consistent with an upslope flow of polluted air masses from the MCMA arriving to ALTZ at midday. In contrast, wind from the E-ESE prevails from June to October; if no recirculation occurs, the smaller increase in CO₂ from 12 to 15 h in the rainy season could be the result of lower emissions from Puebla reaching ALTZ, having being partially offset by the photosynthetic uptake of crops and forests as air traversed over the valley and the mountain slope. Granted, the anemometer used in this analysis is situated only at 5 m.a.g.l. at the station, therefore its measurements are likely influenced by the local topography and might not reflect regional circulation. However, the detection of an enhancement of other combustion-related tracers such as CO and NO_x at the Altzomoni station during the afternoon is consistent with polluted air parcels being transported from the major neighboring cities (^{Whiteman et al., 2000}; ^{Baumgardner et al., 2009}). Preliminary analyses of FTIR solar absorption measurements at Altzomoni show larger CO total columns from January to June, and transient elevated values around 15 h and 18 h in April and May only (N. Taquet, unpublished data).

At the end of the dry season, a contribution from biomass burning to CO₂ levels can also be expected from both forest fires and prescribed agricultural burnings, a common practice in the region. However, this alone would not explain the persistence of the daily enrichment well into July and August, in the mid- dle of the rainy season, because most natural and manmade fires take place in March and April, are less frequent in May, and stop after the rains start (^{Gutiérrez Martínez et al., 2015}).

5.2. Seasonal variation and long-term trend

The hourly measurements obtained at UNAM from July 2014 to August 2019 present a large variability (Fig. 4, gray). Once the invalid data -previously defined as calibration periods and times when operator interference occurred- are discarded, the hourly average of CO₂ mole fraction ranges from 392 to 577 ppm. The overall mean (± SD) of the original (not de-trended) series is 427.7 ± 16.5 ppm. Values in excess of 450 ppm constitute 17% of the dataset. The average mole fraction are in reasonable agreement with a previous study of flux measurements over a densely populated neighborhood in Mexico City, in which ^{Velasco et al. (2005)} found a mixing ratio range of 398 to 444 ppm, and an average of 421 ppm. The slightly higher mean concentration reported here could be attributed to a combination of the CO₂ added to the atmosphere between 2003 and 2018 globally (35 ppm) (^{Dlugokencky and Tans, 2019}) and the increase in CO₂ emissions that the city has experienced in the years since the ^{Velasco et al. (2005)} study took place. The official CO₂ annual emissions in 2004 were 35.8 Mt (^{SEDEMA, 2008}) and 52.5 Mt for the year 2016 (^{SEDEMA, 2018}). Additionally, sampling in the earlier study took place during a short, three-week campaign in spring which included Easter week, a holiday period for schools in Mexico (^{Velasco et al., 2005}). Our CO₂ levels are comparable to those reported for other megacities like Beijing (^{Song and Wang, 2012}) where monthly averages range from 400 to 440 ppm. Similar concentrations are also observed in Paris, Cité des Sciences, where the CO₂ average observed over the same period (2015-2019) equals 426 ppm, with a higher variability in the hourly concentrations, ranging from 388 to 754 ppm (^{Lian et al., 2019}), probably due to a more dynamic atmospheric boundary layer (^{Pal and Haeffelin, 2015}). On the other hand, the short term variability (calculated as the hourly standard deviation of minute averages) is higher at UNAM site (3.6 ppm) compared to Cité des Sciences in Paris (2.0 ppm), which probably reflects a higher exposure to local sources of CO₂.

Fig. 4 Above: Hourly (gray) and daily (blue) average of daytime (13-17 h) CO₂ mole fractions at UNAM, smoothed fitted function (red) and polynomial component of fitted function (black). Below: trend (black) and growth rate (red) for 2014-2019. 

In order to estimate the five-year trend in the UNAM series, we selected only daytime periods from 13 to 17 h, with hourly standard deviations SD < 6 ppm. These filters simultaneously exclude hours when transient and very local effects were recorded (e.g. people working close to the gas intake), and periods when hour-to-hour means differ by more than 3 ppm. Point-like local disturbances are less common at night, but CO₂ concentrations steadily rise all night long as the boundary layer stability increases; additionally, nocturnal CO₂ exhibits larger effects of seasonality compared to daytime (not shown). We excluded also the observations made at a higher level above ground starting on August 2019. Therefore, the trend, seasonal amplitude, and growth rate were computed for the period between July 2014 and August 2019 only.

The results of fitting NOAA’s CCGCRV function (^{Thoning et al., 1989}) to the filtered dataset at UNAM are also depicted in Figure 4. We estimated an annual growth rate of 2.4 ppm yr^-1 for 2014-2019, slightly lower than the global growth rate and the rate reported for Mauna Loa (both 2.6 ppm yr^-1) over the same period (^{Dlugokencky and Tans, 2019}; ^{Tans and Keeling, 2019}). The continuous growth rate shows a maximum at the end of October 2015 and a minimum at the end of July 2018 (Fig. 4). Our time series is too short to attempt a correlation between the observations and inter-annual climate drivers, but it is worth mentioning that from March 2015 and March 2016, one of the strongest ENSO events on record took place (^{Chen et al., 2017}). In Mexico, the El Niño phase of ENSO is associated with negative precipitation anomalies in the northern and central parts of the country, which can reach deficits of up to 250 mm in the southwestern area of Mexico City, causing increased drought and a higher occurrence of forest fires (^{Bravo-Cabrera et al., 2018}).

A measurement of the evolution of the concentration of CO₂ in the atmosphere that is closely related to the annual growth rate is the slope of the polynomial component of the fitted CCGCRV function (^{Thoning et al., 1989}), which we estimated also as 2.4 ± 0.5 ppm yr^-1. This value is directly comparable and identical with the yearly gain estimated by ^{Baylon (2017)} for total column CO₂ at the same site for the year 2016.

The seasonality of data at UNAM is presented in Figure 5. Annual averages were subtracted from daily data to remove the long-term trend from the seasonal change. There is an annual maximum in mid-December, likely due to the shallower boundary layer that prevails during winter (^{García-Franco et al., 2018}), as previously explained. The summer-autumn minimum occurs in mid-September, when both the dilution of trace gases in a deeper convective boundary layer and more active urban vegetation draw down of CO₂ occurs. A secondary maximum occurs consistently right before the onset of the rains; we suggest that it is likely not thermally-driven but reflects 1) a significant contribution of transported CO₂ from agricultural burnings and forest fires outside the city, which peak at this time of year, and whose effects on air quality in the city’s metropolitan area have been documented elsewhere (i.e. ^{Yokelson et al., 2007}; ^{Tzompa-Sosa et al., 2016}), and 2) the lack of photosynthetic uptake by vegetation and the dominance of respiration at the end of the dry period. A simultaneous increase in the concentration of other gases and biomass burning tracers in the particulate matter added to the atmosphere during the spring months would lend support to the biogenic origin of these enhancements. Indeed, a preliminary analysis of the CH₄ data from the Picarro in our study suggests that minor enrichments do occur in April and May for three out of five years of observations at UNAM (2015-2019). However, in general, the concentration of CH₄ in the city is highly variable and it shows large events likely of local origin throughout the year (^{Bezanilla et al., 2014}). A more comprehensive analysis of wind trajectories of the spring enhancements and a concurrent measurement of other biomass burning products is warranted for a more conclusive attribution of emissions to a specific source.

Fig. 5 Average annual course of de-trended daily means of CO₂ mole fraction at UNAM for the period 2014-2019. 

The daily amplitude shows little seasonality, with a maximum in late fall (50.5 ± 18.9 ppm; Nov-Dec) and a minimum in the dry winter (42.2 ± 14.9 ppm; Jan-Mar), the latter is not significantly different from the rainy season (43.0 ± 18.3 ppm; Jun-Oct) or the dry season average (45.5 ± 14.8 ppm, Apr-May). The average seasonal amplitude was 8.9 ppm, with little interannual variation (8.2 ppm in 2018 to 10.2 ppm in 2015). In the only other long-term CO₂ monitoring study in a tropical city that we are aware of, ^{Roth et al. (2017)} showed a larger seasonal amplitude of ~14 ppm for Singapore.

Seasonality and trend for ALTZ were computed using only nighttime data (19 to 5 h) and hourly averages with standard deviations lower than 2.0 ppm. The selected daily period is the most appropriate to avoid the influence of either local vegetation activity or polluted air masses from urban areas, and therefore suitable to monitor conditions representative of the free troposphere. The overall mean CO₂ mole fraction is 406.2 ± 5.6 ppm, slightly higher than NOAA’s global average for the same period (404.9 ppm). We found an annual average growth rate of 2.6 ppm yr^-1 from 2015 to 2019, and again, like at UNAM, the continuous growth rate shows a maximum during the second half of 2015 (Fig. 6). The annual growth rate at ALTZ is in agreement with NOAA’s global rate and Mauna Loa’s (2.6 ppm yr^-1), but is slightly higher than the annual growth rate we obtain from NOAA’s data taken at the MEX station in Veracruz (2.4 ppm yr^-1) over the same period, at almost the same latitude but further away from the influence of large cities and at a higher altitude. In short times series, however, trend estimates and computations of growth rates are very sensitive to high or low years, so meaningful comparisons are difficult to establish at this point. The slope of the first-degree polynomial of the CCGCRV adjusted curve is 2.7 ± 0.0 ppm yr^-1, close to the comparable annual increment derived from the total column measurements for the period 2013-2017 (2.6 ppm yr^-1) (Baylon, 2018).

Fig. 6 Above: Hourly (gray) and daily (blue) average of nocturnal (19-5 h) CO₂ mole fractions at ALTZ, smoothed fitted function (red) and polynomial component of fitted function (black). Below: trend (black) and growth rate (red) for 2015-2019. 

A marked seasonal cycle is discernible with a maximum between the last week of April and the first half of May, and a minimum in the last days of August and the first week of September. The cycle presents an average amplitude of 10.0 ppm, ranging from 8.4 ppm in 2015 to 10.6 ppm in 2018. In order to highlight local contributions to the seasonality of the CO₂ annual cycle, we extracted the de-trended monthly means and compared them to the typical seasonal cycle estimated from the reference Marine Boundary Layer (MBL) time series at a latitude of 19º N, for the period 2015-2019 (Fig. 7). The reference MBL record is computed from averages of weekly measurements of several long-lived atmospheric trace gases from the Cooperative Global Air Sampling Network (^{Conway et al., 1994}). A subset of the sites is selected for the CO₂ MBL reference; only remote, sea level marine sites that sample well-mixed air from onshore prevailing wind directions are considered. The time series is therefore representative of background conditions of CO₂ in the atmosphere.

Fig. 7 De-trended, monthly CO₂ mole fraction at ALTZ (blue) and from NOAA’s Marine Boundary Layer product for a latitude of 19^º N (red), averaged over 2015-2019. 

Times of seasonal maximum and minimum concentrations are coincident in both datasets, and in general both curves exhibit considerable overlap, although peak-to-trough seasonal amplitude is lower for the MBL (8.8 ppm), as data from ALTZ shows a slightly higher annual maximum in May probably due to the influence of polluted air masses. From December to February, the atmosphere at ALTZ shows a lower CO₂ mole fraction compared to the MBL. Although no systematic observations of photosynthesis by individual species have been carried out at ALTZ, casual observations indicate that the mixture of alpine grassland and conifer forests that surround the station stays more active in late fall and winter than most broad-leaved forests in mid latitudes. The evergreen vegetation would tend to prolong the carbon uptake season compared to species that shed the leaves in the cold season and become dormant, so the differences observed between the MBL signal and ALTZ are likely reflecting the influence of the regional vegetation acting as a sink for atmospheric CO₂ during winter.