Volume 122, Issue 6 p. 3317-3329
Research Article
Free Access

Biospheric and anthropogenic contributors to atmospheric CO2 variability in a residential neighborhood of Phoenix, Arizona

Jiyun Song

Corresponding Author

Jiyun Song

School of Sustainable Engineering and the Built Environment, Arizona State University, Tempe, Arizona, USA

Department of Applied Mathematics and Theoretical Physics, Centre for Mathematical Sciences, University of Cambridge, Cambridge, UK

Department of Architecture, University of Cambridge, Cambridge, UK

Correspondence to: J. Song,

[email protected]

Search for more papers by this author
Zhi-Hua Wang

Zhi-Hua Wang

School of Sustainable Engineering and the Built Environment, Arizona State University, Tempe, Arizona, USA

Search for more papers by this author
Chenghao Wang

Chenghao Wang

School of Sustainable Engineering and the Built Environment, Arizona State University, Tempe, Arizona, USA

Search for more papers by this author
First published: 15 March 2017
Citations: 22

Abstract

Urban environment contributes significantly to the global carbon cycle with complex governing mechanisms due to the combined biospheric and anthropogenic contributors. In this study, we analyzed the patterns of boundary layer CO2 flux and concentration for a residential neighborhood in Phoenix, Arizona by using the eddy covariance technique and a single column atmospheric model. Atmospheric stability, anthropogenic emission, and biogenic effect are found to be key determinants to atmospheric CO2 variability. In a diurnal cycle, two CO2 flux peaks coincide with morning and afternoon peak traffic hours, exemplifying the influence of traffic emissions. In the annual cycle, maximum CO2 concentration is found in winter, mainly due to additional emission from the combustion of natural gas combined with the effect of poor dispersion. On the other hand, the minimum CO2 concentration is found in the spring and is attributable to the strong convective mixing and active vegetation uptake. In addition, prominent hysteresis has been found between the atmospheric CO2 concentration and air temperature with a “plait-shaped” pattern in the diurnal cycle and an “oval-shaped” loop for the seasonal variability.

Key Points

  • We studied the variability of atmospheric CO2 flux and concentration in a semiarid city through both measurement and modeling
  • We found that the minimum atmospheric CO2 concentration occurs in spring, which is unique for desert climate
  • We found a prominent hysteresis between atmospheric CO2 concentration and air temperature

1 Introduction

The atmospheric carbon dynamics of urban areas are strongly affected by land-atmospheric exchanges generated by the anthropogenic and biospheric processes [Churkina, 2008]. Urban areas cover approximately 2% of global land surface area but contribute ~70% of anthropogenic CO2 emissions via human respiration, fossil fuel consumption (such as road traffic, industries, heating, and electricity), cement production, etc. [UN-Habitat, 2011]. Changes concomitant with continuous urbanization, such as the land use land cover change, population growth, increase number of vehicles, and air conditioners, add further complexity to the variability of urban atmospheric CO2 concentration [Coutts et al., 2007; Grimmond, 2007]. For example, the spatial distribution of natural and engineered land covers influences the spatial pattern of atmospheric carbon sources and sinks [Wentz et al., 2002]. Specifically, anthropogenic emissions often take place over built terrains, while biospheric processes such as photosynthesis and respiration occur over natural landscapes (i.e., vegetation and soil) [Sailor and Lu, 2004; Crawford et al., 2011].

In addition, the use of engineering materials and the complex morphology significantly alter the energy and hydrological cycles in cities [Arnfield, 2003; Grimmond, 2007; Wang et al., 2013]. The modified thermal and hydrological processes over built terrains regulate the overlying urban boundary layer (UBL) dynamics including the evolution of UBL heights, thermal stratification, and spatial scalar profiles [Song and Wang, 2016a], which in turn impact the spatiotemporal patterns of atmospheric CO2 variation in the urban environment. One example is that the transportation network in a city adds to the environmental CO2 flux in the atmospheric surface layer via traffic emission, whereas the enhanced surface heating over paved roads, especially over those of asphalt, raises the height of the overlying boundary layer [Song and Wang, 2015a], tending to reduce the CO2 concentration in the UBL. The land-atmosphere interactions in urban areas therefore further complicate the quantitative analysis of CO2 patterns. To quantify the urban atmospheric CO2 patterns, short-term field experiments were conducted to analyze the spatial patterns of CO2 across cities [Idso et al., 1998, 2001]. To measure the long-term variations of CO2 for a neighborhood-scale area, eddy covariance (EC) approach has been widely adopted in recent years for investigating land-atmosphere interactions of CO2 in built-up areas [Grimmond et al., 2002, 2004; Coutts et al., 2007; Crawford et al., 2011; Liu et al., 2012].

The eddy covariance (EC) method has been first applied to quantify the exchange of CO2 between terrestrial ecosystem and atmosphere [Baldocchi et al., 2000, 2001; Baldocchi, 2003; Falge et al., 2002; Turnbull et al., 2015]. This method has its advantages in capturing the canopy-scale net CO2 exchange across the canopy-atmosphere interface for a relatively large spatial area (i.e., flux footprint ranging between 100 m and 1000 m) within a wide spectrum of time scales ranging from hours to years [Schmid, 1994; Baldocchi, 2003]. However, the accuracy of this method is largely influenced by atmospheric conditions and underlying land surface conditions. This method is most applicable to steady atmospheric conditions over flat terrain covered by homogeneous vegetation within an extended upwind distance [Grimmond et al., 2002; Baldocchi, 2003].

Recent years have seen drastic increasing research interests of applying this EC method in urban areas where most of anthropogenic CO2 emission occurs [Grimmond et al., 2002; Grimmond and Christen, 2012; Decina et al., 2016]. In contrast to natural terrains (forests, agricultural lands, or wetlands), there are enormous challenges to make meaningful flux observations of CO2 in urban environments due to its wide range of land surface characteristics with different surface roughness, different radiative and hydrothermal properties, and diverse land use land cover types (commercial, industrial, suburban, etc.) [Grimmond et al., 2002]. Therefore, experimental sites for sitting the EC towers in urban areas should be carefully selected with data quality control, in order to conform to the assumptions of EC method (i.e., steady atmospheric condition and homogeneous terrain). Specifically, urban EC towers are often installed in the downwind of a fairly extensive and homogeneous suburban surface [Grimmond et al., 2002; Moriwaki and Kanda, 2004; Coutts et al., 2007; Liu et al., 2012].

In addition, the EC sensing instruments are normally equipped on tall towers and capable of recording micrometeorological variables, CO2 concentration, and flux at high frequency (commonly 10–20 Hz). One limitation of EC measurements is that the towers are usually located within the inertial sublayer, mostly lower than 100 m [Bergeron and Strachan, 2011; Chen et al., 2011; Grimmond et al., 2004; Järvi et al., 2012], and rarely reach several hundred meters [Liu et al., 2012]. To measure the variability of CO2 concentration at higher elevations above the inertial sublayer, airborne instruments such as lidar are often used [Nakazawa et al., 1997; Niwa et al., 2011; Ramanathan et al., 2015]. However, ground-based EC measurements are temporally abundant with continuous measurements but spatially sparse since the distribution of EC towers is sparse, while airborne measurements such as remote sensing images have large spatial coverage but are intermittent. Despite that some significant research progresses have been made in monitoring and documenting CO2 emissions by using ground based approaches as well as remotely sensed techniques, quantification of carbon dynamics within urban domain at high temporal and spatial resolutions remains a large challenge due to lack of observation and complexity of carbon sources [Gurney, 2014; Decina et al., 2016]. In particular, the study of temporally continuous patterns of CO2 variability in the urban boundary layer (UBL), within a typical depth of 1 km, has been inadequate [Crawford et al., 2016] and remains an open challenge.

To address the outstanding challenges in characterizing the variability of atmospheric CO2 concentration in the UBL, we will statistically analyze the EC measurements of CO2 concentration and flux and simulate the CO2 concentration in the overlying UBL via a single column model. In particular, for our study site, the EC tower is sited in a suburban residential area with homogeneous single-story building design and works as the first long-term urban flux tower in a large arid city to the best of our knowledge. The unique contribution of this study is to analyze the characteristics of atmospheric CO2 concentration and flux for an arid city and to reveal the relationships between CO2 concentration and meteorological variables within a desert climate. In addition, the results of analysis help to further our understanding of the fundamental questions: (1) how will anthropogenic and biospheric contributors influence the atmospheric CO2 variability correspondingly? (2) What are the meteorological determinants that will affect the CO2 variability? (3) What are the differences of CO2 characteristics in the inertial sublayer (lower part of UBL) and mixed layer (upper part of UBL)? Note that this study is mainly focused on building the link between atmospheric dynamics (i.e., stability and meteorological forcing) and variability of atmospheric CO2 concentration via both field measurements and numerical modeling techniques, which are different from the detailed inventory of urban CO2 flux sources via top-down and bottom-up approaches as reported in previous studies [e.g., Christen et al., 2011; Velasco et al., 2014; Ward et al., 2015; Turnbull et al., 2015].

2 Data and Methods

2.1 Study Area and Data Collection

In this study, we choose a typical residential low-rise neighborhood in a semiarid city, i.e., Phoenix in Arizona as our test bed. Phoenix is the 13th largest city in the United States with a population of about 4.3 million [U.S. Census Bureau, 2013]. Located in the heart of Arizona Sun Corridor and the northeast of Sonoran Desert, Phoenix has a subtropical desert climate with sparse precipitation all year-round and abundant sunshine with extreme hot summers with daily mean temperature of 35°C and maximum temperature exceeding 43°C as well as mild winters with mean temperature of 10°C [Georgescu et al., 2012; Hedquist and Brazel, 2014]. Since 1945, this area has experienced extensive land use changes, specifically from agricultural land and desert remnant to residential areas, and emerged as a hub of urban environmental study [Chow et al., 2012; Wang et al., 2016; Yang et al., 2016].

An EC tower with a height of 22.1 m has been deployed at west Phoenix (33°29′1.9″N, 112°08′33.4″W) by the local Central Arizona-Phoenix Long-Term Ecological Research project with continuous measurements of radiation, air temperature, humidity, wind velocity, carbon dioxide concentration, carbon dioxide flux, etc. at 10 Hz. The surrounding 1 km2 area around the EC tower is low-density urban residential space with 26.42% building surface, 22.02 roads/asphalt, 4.61% tree, 10.01% grass, 36.83% bare soil, and 0.11% water/pool based on a 2011 Quickbird image with a resolution of 2.4 m [Chow et al., 2014a]. This residential area largely consists of single-story and single-family houses with small backyard size. The backyard space of 75% houses is covered by bare soil, while about 20–25% housing backyard has Bermuda grass (Cynodon dactylon) yards [Chow et al., 2014a]. Besides, there are scattered trees around, including Velvet Mesquite (Prosopis velutina), Indian Rosewood (Dalbergia latifolia), and Arizona Ash (Fraxinus velutina). The landscaping conditions are mostly dry and are intervened by the residents' casual watering with garden hoses [Chow et al., 2014a].

In this study, a yearlong data set recorded by the EC tower throughout 2012 is used for subsequent analysis. The raw 10 Hz data were processed through the EdiRE software with signal despiked and then converted to 30 min averaged data set. The CO2 concentration is measured by infrared gas analyzer from LI-COR Biosciences (LI-7500). The CO2 flux measurement data series were generated and corrected by accounting for density effects due to heat and water vapor transfer [Webb et al., 1980]. More detailed information of the instrumentation, data retrieval and processing, and data quality control can be found in Chow et al. [2014a].

2.2 Source Area of the EC Tower

To assess the source area of the EC tower, we adopted a typical analytical footprint model proposed by Kormann and Meixner [2001]. The crosswind distributed scalar flux footprint in Kormann and Meixner [2001] is given by
urn:x-wiley:2169897X:media:jgrd53728:jgrd53728-math-0001(1)
where φ is the flux footprint (i.e., vertical flux per unit point source); x, y, and z are the space coordinates of point source; σ is the dispersion length; ξ is the flux length scale; μ is a constant and can be obtained by μ = (1 + m)/r, where r is a shape factor (r = 2 + m − n); and m and n are the exponents of the wind velocity power law and the eddy diffusivity power law, respectively.

The 50% and 90% footprint source areas of the EC tower at daytime and nighttime in four seasons are calculated and mapped over a high-resolution orthoimagery of the study area, respectively (see Figure 1). The radii of the 90% source area are about 250 m and 800 m at daytime and nighttime, respectively. The source area at daytime is much smaller than that at nighttime, due to the presence of turbulence and unstable atmosphere.

Details are in the caption following the image
The location of the EC tower site (marked by a red star) and its coverage of 50% (blue) and 90% (red) source areas at daytime and nighttime for four seasons in the year 2012.

2.3 Evolution of CO2 Concentration in the Urban Boundary Layer

The evolution of CO2 concentration in the UBL is modeled by a single column atmospheric model (SCM) based on a modified K-theory [Noh et al., 2003; Hong et al., 2006; Song and Wang, 2015a, 2015b, 2016a, 2016b]. The governing turbulence diffusion equations in the SCM for a generic scalar atmospheric variable c in the UBL can be expressed by [Song and Wang, 2015a]
urn:x-wiley:2169897X:media:jgrd53728:jgrd53728-math-0002(2)
where t is the evolution time; z is the elevation above the land surface; w is the vertical wind speed; c could be temperature, humidity, or CO2 concentration; and urn:x-wiley:2169897X:media:jgrd53728:jgrd53728-math-0003 is the vertical kinematic flux of heat (if c is temperature), moisture (if c is humidity), and carbon dioxide (if c is CO2 concentration) correspondingly. The surface flux urn:x-wiley:2169897X:media:jgrd53728:jgrd53728-math-0004 at the lower boundary of the UBL is provided by EC measurements, while the upper boundary (entrainment) is prescribed as urn:x-wiley:2169897X:media:jgrd53728:jgrd53728-math-0005 with βc as the empirical fraction, specifically βc = 0.1–0.3 for temperature [Noh et al., 2003], βc = −1.0–0 for humidity [de Arellano et al., 2004], and βc = −0.2–3.0 for CO2 concentration [de Arellano et al., 2004]. Specifically for Phoenix site, we used 0.2 as the entrainment rate of heat flux and 0.95 as the entrainment rate of moisture flux and CO2 flux based on previous studies [Song and Wang, 2015a, 2015b, 2016a, 2016b]. Detailed parameterizations for UBL height, eddy diffusivity, velocity scale, and CO2 flux within the UBL for both daytime and nighttime are described in Table 1 [Deardorff, 1971; Yu, 1977; Hong et al., 2006; Hong, 2010; Ouwersloot and Vilà-Guerau de Arellano, 2013]. It is noteworthy that there is no measurement of CO2 concentration profiles in the UBL for our study area to the best of our knowledge. To validate this SCM, we compared the measurements and simulations of specific humidity profiles in the UBL (see Figure 2), given the similarity in the mechanism of atmospheric transport of CO2 and vapor, both as passive scalars [Stull, 1988].
Table 1. Parameterization Schemes for Daytime and Nighttime Boundary Layersa
Variables Daytime Nighttime
UBL height urn:x-wiley:2169897X:media:jgrd53728:jgrd53728-math-0006 [Ouwersloot and Vilà-Guerau de Arellano, 2013] urn:x-wiley:2169897X:media:jgrd53728:jgrd53728-math-0007 [Deardorff, 1971; Yu, 1977]
Eddy diffusivity urn:x-wiley:2169897X:media:jgrd53728:jgrd53728-math-0008 [Hong et al., 2006] urn:x-wiley:2169897X:media:jgrd53728:jgrd53728-math-0009 [Hong, 2010]
Velocity scale urn:x-wiley:2169897X:media:jgrd53728:jgrd53728-math-0010 [Hong et al., 2006] wm = u* [Hong, 2010]
Vertical kinematic fluxes urn:x-wiley:2169897X:media:jgrd53728:jgrd53728-math-0011 [Hong, 2010] urn:x-wiley:2169897X:media:jgrd53728:jgrd53728-math-0012 [Hong, 2010]
  • a Notes: zh is the boundary layer height; w is the vertical velocity scale; u* is the friction velocity; γθv is the lapse rate in the free atmosphere; θv is the virtual potential temperature; c is the generic scalar variable such as temperature, humidity, and CO2 concentration; K is the eddy diffusivity; γ is the nonlocal mixing term; Pr is the Prandtl number; k is the von Karman constant; and subscripts “0,” “b,” “s,” “m,” “h,” and “e” denote at the initial stage, corrected by incorporating moisture, on the surface, in the mixed layer, at the boundary layer top, and in the entrainment, respectively.
Details are in the caption following the image
Comparison of simulated and measured atmospheric profiles of specific humidity for two time points, i.e., (a) 16:44 P.M. (local time) on 2 July 2013 and (b) 16:37 P.M. (local time) on 9 July 2013 at NOAA Earth System Research Laboratory Phoenix site.

3 Results and Discussion

3.1 Diurnal and Seasonal Variabilities of CO2 Concentration and Flux

Atmospheric stability is a key determinant that influences diurnal and seasonal variabilities of urban atmospheric CO2 concentration by modifying the efficiency of atmospheric mixing [Coutts et al., 2007; Crawford et al., 2011]. In a diurnal cycle, the atmosphere is unstable at daytime due to strong convection driven by thermals arising from land surface as well as the turbulent shear production generated by synoptic scale motion, while it is stabilized at nighttime since the land surface has been sufficiently cooled [Stull, 1988]. In an annual cycle, solar radiation and wind speed are the governing variables that strongly regulate the dynamic atmospheric stability (denoted by, e.g., Richardson number) by controlling the surface-air temperature deficit and the horizontal advection, respectively [Golder, 1972; Stull, 1988; Woodward, 1998]. Specifically, the seasonal variation of solar radiation leads to the temporally differential heating of land surface and the variability in daytime and nighttime lengths. For example, the land surface is least heated in winter with the weakest solar radiation and the shortest daytime duration, resulting in the most stable UBL in winter on average. The diurnal and seasonal evolutions of atmospheric stability have strong implications on the diurnal and seasonal variabilities of atmospheric CO2 concentration, which are explained in the subsequent paragraphs. Wind shear, on the other hand, influences the atmospheric dynamic stability and advection through its contribution to turbulent kinetic energy [Stull, 1988]. It is found that the daily averaged atmospheric CO2 concentration has a significant decreasing trend with the daily averaged wind speed (Figure 3), indicating that high wind speed facilitates CO2 dispersion.

Details are in the caption following the image
The relationship between daily averaged atmospheric CO2 concentration [CO2] and wind speed, u in the year 2012.

The averaged diurnal variations of CO2 flux and concentration for the year 2012 are shown in Figures 4a and 4b, respectively. The two peaks of the diurnal surface CO2 flux (around 5 µmol m−2 s−1) occur at around 9:00 A.M. and 17:00 P.M. local time coinciding with the morning and afternoon peak traffic hours [Chow et al., 2014b]. Similar impact of traffic on diurnal patterns in other cities such as Melbourne, Australia, and Beijing, China was reported in previous study as well [Coutts et al., 2007; Song and Wang, 2012]. Without intensive traffic load, the CO2 flux at weekends has lower peaks with a decrease of 1 µmol m−2 s−1. The lowest CO2 concentration of around 380 ppm occurs at around 16:30 P.M. in the afternoon for both averaged weekday and weekend, due to sufficient dispersion by strong mixing and vegetation uptake [Idso et al., 1998, 2001]. On the other hand, there are, in general, two peaks of CO2 concentration occurring at midnight and morning peak traffic hour, respectively. The midnight peaks in weekdays and weekends are similar, with a magnitude of about 406 ppm, mainly due to vegetation respiration and poor dispersion under stable atmospheric conditions. Another concentration peak at morning traffic hour differs substantially at weekday (416 ppm) and weekend (405 ppm), probably owning to different travel patterns [Wentz et al., 2002]; i.e., people have more flexible travel time at weekends, so the traffic concentration is mitigated.

Details are in the caption following the image
Diurnal variation of (a) CO2 flux and (b) CO2 concentration in weekday and weekend averaged over the whole year 2012.

The seasonal variability of atmospheric CO2 flux is shown in Figure 5a. For each season, there are two peaks in the diurnal cycle of CO2 flux, which happen in the morning and afternoon, respectively. The maximum value of morning peaks (8.5 µmol m−2 s−1) occurs in the winter, which could be mainly attributable to additional natural gas combustion. Monthly residential natural gas consumption data in 2012 in Arizona were obtained from U.S. Energy Information Administration (http://www.eia.gov/state/?sid=AZ) as shown in Figure 5b. The statewide residential natural gas consumption is found to be the largest in the winter with monthly mean consumption of 169 × 106 m3. Since majority of the source area of the CO2 measurement constitutes of residential area, the seasonal natural gas consumption in the study area is expected to closely follow the statewide residential consumption trend. The minimum value of morning peaks occurs in the summer, which is possibly resulted from stronger atmospheric turbulence and higher UBL height due to stronger surface heating [Song and Wang, 2015a] as well as less traffic volumes due to extreme heat stress [Chow et al., 2014b]. Also note that there is a local peak at around 20:00 in the summer, which might be due to (1) more respiration of vegetation and soil resulted from the more intensive watering of backyards during the extreme hot summer season and (2) later sunset at around 19:30–19:45, i.e., a later transition from convective to stable atmospheric boundary layer schemes.

Details are in the caption following the image
Averaged diurnal variations of CO2 flux in four seasons, i.e., spring (March, April, and May), summer (June, July, and August), fall (September, October, and November), and winter (December, January, and February).

Apart from anthropogenic contributors such as traffic and natural gas combustion, the variability of CO2 flux is also strongly influenced by vegetation. To analyze the effect of vegetation, we obtained the information of normalized difference vegetation index (NDVI, a numerical indicator of the abundance of live green vegetation) in the spatial range of 0.25 km2 (i.e., ~90% source area at daytime for each season) around the EC tower from MODIS Collection 5 land products in the year 2012 with 16 day intervals [Oak ridge National Laboratory Distributed Active Archive Center, 2008]. The relationship between averaged daily minimum CO2 concentration which always occurs at daytime and averaged NDVI with 16 day intervals throughout the year 2012 is shown in Figure 6a. It can be seen that the daily minimum CO2 concentration will be decreased with NDVI, implicating the enhanced effect of vegetation uptake with higher NDVI in reducing atmospheric CO2 concentration through photosynthesis.

Details are in the caption following the image
(a) The relationship between averaged daily minimum CO2 concentration (min (CO2)) and averaged NDVI with 16 day intervals for the year 2012; the monthly variation of (b) CO2 concentration (ppm), (c) wind speed, and (d) air temperature in the year 2012.

It is also noteworthy that photosynthesis of live green vegetation is restricted to a certain ambient temperature range for normal biological activity. For each type of vegetation, there is an optimal temperature (15°C to 30°C for most plants) related to maximum efficiency of photosynthesis [Hew et al., 1969; Mooney et al., 1982]. By increasing ambient temperature, the photosynthesis rate will increase when environmental temperature is below the optimal temperature and will decrease when environmental temperature exceeds the optimal temperature. In particular, under extreme heat conditions, the heat-stressed leaves may have a lower rate of photosynthesis due to stomatal closure for water conservation [Xue et al., 2011]. The NDVI in the 90% daytime source area (Figure 1) has mean values of 0.181, 0.157, 0.166, and 0.178 in the spring, summer, fall, and winter, respectively. The highest seasonally mean NDVI occurs in the spring, possibly due to the temperate temperature with a mean of 25°C. The lowest seasonally mean NDVI occurs in summer, probably due to the extreme heat stress since there are long durations of extreme hot period with air temperature exceeding 40°C.

The contour plot of diurnal variations of CO2 concentration for each month in 2012 is shown in Figure 6b. It is found that the highest atmospheric CO2 concentration of 430 ppm occurs in winter night, which could be related to a combined influence of (1) natural gas combustion for heating and (2) poor dispersion due to stable atmosphere with weak vertical mixing and low wind speed (Figure 6c). The lowest CO2 concentration of 370 ppm occurs in spring afternoon, likely due to strong vertical mixing; maximum efficiency of vegetation uptake with the highest mean NDVI in March, April, and May (see Figure 6a); and the optimal temperature range (with a mean value of 25°C) for photosynthesis of most plants (see Figure 6d). In addition, the CO2 concentration in the afternoon is smaller in spring and winter than in summer and fall, which could be attributed to the differences of vegetation uptake efficiency, indicated by greater local NDVI in spring and winter (Figure 6a).

It is noteworthy that the observed minimum CO2 concentration in spring is unique for the selected residential area in a desert city, since most of previous studies found that minimum CO2 concentration occurs in summer for their study areas [Coutts et al., 2007; Crawford et al., 2011; Liu et al., 2012]. In our study area, summers see a mean air temperature of about 35°C and a peak temperature of more than 40°C. This leads to most vegetation in the study area having lower photosynthesis rate in summers due to extreme heat stress. On the other hand, the CO2 concentration in the fall is higher than in the spring, despite both seasons having similar temperature ranges and anthropogenic stresses. Figure 7a shows the monthly relationship between CO2 concentration and air temperature, which is averaged over 12 months in the year 2012. As seen from Figure 7a, when air temperature is 20°C, the CO2 concentrations are 380 ppm and 414 ppm for the spring and fall, respectively. Greater CO2 concentration in fall could be attributed to defoliation, since the NDVI in fall (0.166) is smaller than that in spring (0.181) within the local footprint area. Therefore, the CO2 uptake by vegetation is reduced along with decreased leaf areas in the fall.

Details are in the caption following the image
(a) The hysteresis effect between monthly averaged CO2 concentration and air temperature for the whole year 2012 and the ensemble-averaged diurnal hysteresis of normalized atmospheric CO2 concentration [CO2] and normalized air temperature Ta (b) for winter and (c) for summer. The blue and red line sections around the scatters in Figures 7b and 7c represent for 0.5 standard deviation of normalized CO2 concentration and air temperature, respectively. The arrows in Figures 7a–7c indicate loop directionality, meaning the x axis variable (i.e., normalized Ta) is increasing/decreasing when the arrow points to the right/left.

In addition, though it is clear that anthropogenic and biospheric contributors severally influence the urban CO2 patterns and their contributions are qualitatively independent, it remains challenging to separate these factors quantitatively due to the lack of high-resolution observation, mismatch of source areas (for example, natural gas combustion contributes as “point” sources, while vegetation impact presents itself as “areal” sinks/sources), inaccuracy of inventory method (for anthropogenic stressors), etc. [Gurney, 2014]. One plausible way to address these challenges is the use of isotopic techniques, such as using dual carbon and oxygen isotopic tracers to quantify the combustion of fossil fuel (gasoline, natural gas, etc.) and respiration of plants and soils based on mass balance approaches [Pataki et al., 2003; Pang et al., 2016].

3.2 The Hysteresis Effect

Another intriguing phenomenon about atmospheric CO2 variability is the hysteresis effect (due to phase lag) between CO2 concentration and air temperature in both diurnal and seasonal cycles. For better clarification, both CO2 concentration and air temperature are normalized by
urn:x-wiley:2169897X:media:jgrd53728:jgrd53728-math-0013(3)
where X can be CO2 concentration or air temperature and the subscripts min and max denote daily minimum and maximum values, respectively. In a diurnal cycle, the hysteresis effect has a unique “plait-shaped” loop with three subloops (indicated by red, blue, and green scatters) in both winter and summer as shown in Figures 7b and 7c, respectively. The physics behind the subloops at two ends is possibly related to the transition between day and night, with the diurnal cycle of carbon emitting-uptaking analogous to that of the cyclic loading-unloading in the study of elastoplasticity [Kim et al., 2004] as well as that of the diurnal heating-cooling in the study of surface energy budgets [Gao et al., 2010; Sun et al., 2013]. The diurnal hysteresis effect between atmospheric CO2 concentration and air temperature could be resulted from (1) the hysteresis effect between soil CO2 concentration and soil temperature [Zhang et al., 2015], (2) the diurnal effect of vegetation as carbon source and sink alternatively [Crawford et al., 2011], and (3) the diurnal variation of atmospheric stability schemes [Stull, 1988]. The two loops at both ends are smaller in the winter (Figure 7b) than those in the summer (Figure 7c), especially for the one at right (shown by green scattered points). The right loop in winter becomes indiscernible, likely due to the fact that the UBL is more stable in winter with smaller surface sensible heat flux and the transition from day to night is faster. On the contrary, it takes longer in the summer for the UBL to be stabilized during nocturnal cooling, leading to a more prominent residual loop at the right end. However, the “plait-shaped” loop with three subloops in diurnal cycle will transform to a simple “oval-shaped” loop in monthly cycle which is averaged over the whole year 2012 (see Figure 7a), likely due to the average effect. This monthly hysteresis effect is probably due to the compound effect of air temperature on vegetation photosynthesis and atmospheric stability, which in turn explains the seasonal variability of CO2 concentration.

3.3 Characteristics of CO2 Concentration in the Urban Boundary Layer

The above analysis results are all based on the EC observations in the urban surface (specifically inertial layer). To look into the carbon dynamics in the overlying UBL (~1000 m), we applied the SCM (introduced in section 2.3) to simulate the diurnal variation of UBL CO2 concentration with the observed surface CO2 flux as the lower boundary condition. Specifically, as a case study, a weekly surface CO2 flux of Phoenix from 13 to 19 June 2012 was collected from the EC tower site as presented in Figure 8a. The simulated vertically averaged CO2 concentration in the mixed layer (~90% upper part of the UBL) is compared with the observed surface CO2 concentration (~10% lower part of the UBL) for the study period as shown in Figure 8b. Similar as surface CO2, higher CO2 concentration appears at nighttime than at daytime in the mixed layer, mainly due to the atmospheric stability. On the other hand, the magnitude of mixed-layer CO2 is always lower than that of surface CO2 with a mean difference of 45 ppm in the test period. This is physical since the free atmosphere works as an infinite reservoir and can absorb CO2 of the mixed layer through assumed upward entrainment CO2 fluxes [Pino et al., 2012]. Note that this modeling approach only provides a preliminary effort to vertically extend the research framework of atmospheric CO2 and is limited by its deficiency in capturing horizontal effects and real-time entrainment processes.

Details are in the caption following the image
Case study of UBL CO2 simulation for a week from 13 to 19 June 2012 with (a) measured CO2 flux rising from urban surface as model input and (b) comparison of simulated CO2 concentration in the UBL and measured CO2 concentration in the USL.

4 Concluding Remarks

In this study, we analyzed the surface-atmosphere exchange of CO2 through the EC observation and atmospheric modeling. Atmospheric stability and anthropogenic and biospheric processes are found to be key determinants to the atmospheric CO2 variability in both diurnal and seasonal cycles. Stable atmosphere (such as at nighttime or wintertime) significantly impedes CO2 dispersion than unstable atmosphere (such as at daytime or summertime). As a major CO2 source, anthropogenic emissions affect the diurnal flux peaks through traffic volumes and the winter flux peak through intensive natural gas consumption for heating. The vegetation uptake, on the other hand, largely reduces the daytime CO2 concentration through net photosynthesis. It is noteworthy that the lowest CO2 concentration related to the highest vegetation uptake efficiency occurs in spring rather than in summer as reported in other related researches, which is unique for arid or semiarid cities under excessive thermal stress in summer. In addition, we extended the urban CO2 research framework from urban surface to the entire urban boundary layer and found that boundary layer CO2 concentration is always lower than surface concentration due to the dilution of air from free atmosphere.

It is noteworthy that the results reported here are subject to limitation inherited in the annual EC data set with a source area covering a residential area. The CO2 variability patterns may vary if large variability of source areas is included, such as an industrial area or a desert remnant. These limitations can be released if long-term data set over different land use land cover types is available, for example, by obtaining data from remote sensing lidar measurements, or by extensive analysis of data set obtained from a network of flux towers over different landscapes, e.g., the AmeriFlux network [Massman and Lee, 2002]. The latter option imposes other practical challenges such as data quality control. With more accurate quantifications of urban carbon dynamics, mesoscale land-atmosphere models coupling with numerical weather predictions offer a promising method in further characterizing, e.g., the urban contribution to regional or global carbon cycles and its implications to urban-carbon-climate repercussions.

Acknowledgments

This work is supported by the U.S. National Science Foundation (NSF) under grants CBET-1435881 and CBET-1444758. The measurement data of atmospheric CO2 concentration and flux are available from Central Arizona-Phoenix Long-Term Ecological Research (https://sustainability.asu.edu/caplter/).