Urban land cover type determines the sensitivity of carbon dioxide fluxes to precipitation in Phoenix, Arizona

Urbanization modifies land surface characteristics with consequent impacts on local energy, water, and carbon dioxide (CO2) fluxes. Despite the disproportionate impact of cities on CO2 emissions, few studies have directly quantified CO2 conditions for different urban land cover patches, in particular for arid and semiarid regions. Here, we present a comparison of eddy covariance measurements of CO2 fluxes (FC) and CO2 concentrations ([CO2]) in four distinct urban patches in Phoenix, Arizona: a xeric landscaping, a parking lot, a mesic landscaping, and a suburban neighborhood. Analyses of diurnal, daily, and seasonal variations of FC and [CO2] were related to vegetation activity, vehicular traffic counts, and precipitation events to quantify differences among sites in relation to their urban land cover characteristics. We found that the mesic landscaping with irrigated turf grass was primarily controlled by plant photosynthetic activity, while the parking lot in close proximity to roads mainly exhibited the signature of vehicular emissions. The other two sites that had mixtures of irrigated vegetation and urban surfaces displayed an intermediate behavior in terms of CO2 fluxes. Precipitation events only impacted FC in urban patches without outdoor water use, indicating that urban irrigation decouples CO2 fluxes from the effects of infrequent storms in an arid climate. These findings suggest that the proportion of irrigated vegetation and urban surfaces fractions within urban patches could be used to scale up CO2 fluxes to a broader city footprint.


Introduction
Urbanization modifies land surface characteristics and impacts local energy, water, and carbon dioxide (CO 2 ) fluxes, particularly when large changes are made as compared to pre-existing conditions [1][2][3][4][5][6][7][8][9]. Cities are the most visible sign of global change and, despite their relatively small global areal fraction (2 to 5%), urban areas are responsible for >70% of the total CO 2 PLOS ONE | https://doi.org/10.1371/journal.pone.0228537 February 12, 2020 1 / 26 a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 emissions from anthropogenic sources [6,[10][11][12][13][14][15][16]. In arid and semiarid regions, the trend in urbanization is even more pronounced than in other climate settings, which is crucial since about 30% of the global population is currently residing in cities in arid and semiarid climates [17][18][19]. Furthermore, prior efforts have shown that urbanization in these regions significantly impacts CO 2 concentrations and emissions estimated for different land cover types [3,16,20,21]. CO 2 fluxes in urban areas are influenced by anthropogenic emissions, vegetation, and hydrometeorological factors such as precipitation and temperature. Most CO 2 fluxes in cities are controlled by fuel combustion from vehicles, industries and buildings, rather than by biological processes [6,10,[22][23][24][25][26][27]. As a result, urban areas are net sources of CO 2 to the atmosphere [6,[28][29][30], though a high degree of spatiotemporal heterogeneity is present [6,15,[31][32][33]. Furthermore, the influence of point sources of CO 2 can play a disproportionate role as compared to natural ecosystems [6,34]. Nevertheless, it is possible that urban vegetation can potentially have an important role in modulating CO 2 exchanges in cities by counteracting to some extent those positive fluxes through the photosynthetic activity of plants. This urban vegetation effect is modulated by the amount of available water, provided in arid and semiarid cities through irrigation [7,35,36], and the availability of incoming solar radiation affected by cloud cover. The competing effects of anthropogenic emissions (sources) and plant-mediated CO 2 fixation (sinks) in arid cities are not well understood at present [9,16,25,30,[37][38][39].
A number of approaches exist to estimate CO 2 exchanges in cities, including bottom-up methods using emission factors, indirect calculations through CO 2 concentrations [38,[40][41][42], and CO 2 inverse modeling [43,44]. Indirect methods such as these are often associated with large uncertainties and a lack of spatial and temporal detail [6]. As an alternative applied in this study, the eddy covariance (EC) method [45] can be used to measure CO 2 fluxes (FC) in urban areas [6,11,12,15,16,30,39,46]. However, urban EC observations have been generally limited, as compared to those efforts in natural ecosystems, with most studies undertaken in very dense urban settings or in open low-density areas of northern latitudes [6,[47][48][49][50][51][52][53]. Furthermore, arid cities have been generally underrepresented in the use of the EC method [16,25]. This paucity of studies is related to the challenging nature of urban FC observations due to deployment logistics, security concerns, and the potential disruption of activities [49,54]. As the number of EC studies in urban areas grows, however, it will be possible to assemble inventories of CO 2 flux measurements that can be compared to bottom up approaches.
Limitations in urban EC studies also imply that few efforts have been carried out to quantify the role of land cover type on FC measurements, for instance between urban parks and the high-density urban core. Relevant measurements represent a challenge due the spatial variability of urban land covers and the complex morphology of urban environments [12,16,44]. Several studies have measured CO 2 exchanges in urban areas relative their surrounding environments. For example, Bergeron and Strachan [30] compared agricultural, suburban, and urban sites near Montreal, Canada. Ward et al. [55] similarly studied three areas (woodland, suburban, and urban sites) in England, while Buckley et al. [13] compared FC measurements in suburban and urban sites in Syracuse, USA. Ueyama and Ando [15] is one of the few studies to perform a direct comparison of multiple urban patches in Japan. In Indianapolis, as part of the INFLUX experiment, an important effort to measure FC over several urban landscapes was also carried out [56,57]. However, arid cities are under-represented in terms of FC measurements with the EC method, though Song et al. [16] analyzed conditions in Phoenix, USA.
In this study, we use a mobile EC tower to measure FC and meteorological conditions in three urban settings at Arizona State University (ASU) as described by Templeton et al. [58] and similar to Soegaard and Møller-Jensen [59]. These short-term deployments are compared to a stationary (reference) EC tower in a suburban neighborhood and spanning the entire period (1 January to 30 September 2015). The three mobile sites represent different land cover types: a xeric landscaping, a parking lot, and a mesic landscaping. These sites are expected to vary in terms of their CO 2 exchanges due to variations in the amount of vegetation and anthropogenic emissions. Thus, the objectives of this study are to: (1) quantify and compare FC over different urban land cover types in relation to a location that provided reference meteorological conditions during the study period, (2) relate the observed differences to measures of anthropogenic emissions, plant photosynthetic activity, and meteorological forcing, and (3) determine the role of precipitation events and outdoor water use on modifying CO 2 exchanges across the sites.

Site descriptions
The study was carried out in four locations in the Phoenix Metropolitan Area (PMA) as described in Table 1 that were non overlapping and at most 42.8 km apart. The PMA has a population of 4.1 million [60] and is located in a hot, arid climate (Köppen classification BWh), with seasonal average temperatures of 14.1˚C, 22.9˚C, 33.9˚C, and 24.8˚C, in the winter, spring, summer, and fall. A bimodal precipitation regime is present with winter frontal storms and summer thunderstorms during the North American monsoon [61,62]. Mean annual precipitation is 204 mm yr -1 based on 1981 to 2010 data, with winter (December to January) and summer (July to September) amounts of 68.3 mm and 67.8 mm, respectively. Spring and early summer (March through June) are typically dry, accounting for only 17% of the mean annual precipitation [7,58]. The low annual precipitation leads to water limited conditions in natural ecosystems [62], requiring outdoor water use to support vegetation in urban areas [7,35,36].
The three mobile deployments and the reference site represent different urban land covers in the PMA. Fig 1 presents an aerial image of each sampling location that depicts differences in urban characteristics. These urban land covers correspond to: (a) xeric landscaping (XL) site, classified as a Local Climate Zone (LCZ) 5 [63] composed of drip-irrigated trees (palo verde, Parkinsonia florida) of 3-4 m of height, with gravel and bare soil cover, located within a setting that included a midrise (three-story) building used for office space and a paved road; (b) a parking lot (PL) site, classified as a LCZ 8 [63], characterized by pavement (asphalt) with minimal vegetation, near an intersection with high traffic and frequently contained vehicles, with a low number of 6 m palm trees and large low-rise (one-to three-story) buildings used for office space surrounded by impervious cover nearby; (c) a mesic landscaping (ML), classified as LCZ 9, consisting of a sprinkler-irrigated turf grass (approximately 2-3 days per week, 3 times per day, for 20 to 30 min each time) among sparsely built single-family homes (low-rise, one story) with sparse, undeveloped land cover nearby including sparse 6 m trees; and (d) a suburban residential area, classified as LCZ 6, consisting of medium-density single-family homes, streets, and open spaces, used as a reference site (REF). As compared to ML, the REF site has lower irrigation due to larger variations in landscaping with some yards having trees and grasses, but most containing gravel and bare soil. One of the sites (PL) is nearly devoid of vegetation, while one site (ML) has light traffic. The REF site is a stationary EC system in operation during the entire sampling period and spanning the seasonal changes in meteorological conditions to allow quantitative comparisons with the short-term deployments. All mobile deployments were within ASU (Tempe campus for XL and PL and Polytechnic campus for ML) and authorized through the ASU Facilities Department.
A land cover classification was performed for the three mobile deployments using color (0.30 m) orthoimagery from the U.S. Geological Survey (http://lta.cr.usgs.gov/high_res_ ortho). A supervised classification based on RGB signatures was done using a maximum likelihood method and classifying the urban land cover as (1) trees, (2) grass, (3) undeveloped (gravel or bare soil), (4) pavement (asphalt), and (5) buildings or concrete. Percentages of each land cover type within a unique EC footprint were derived from aggregations of 30-min interval daytime footprint estimates. The EC footprint was obtained using the analytical model of Kormann and Meixner [64] for an area of 500 m by 500 m centered at each site and a horizontal pixel resolution of 5 m selected to be less than the measurement height [65]. Following Anderson and Vivoni [66], the EC footprint [67] was calculated for each 30 min interval of turbulent daytime conditions, averaged over each daytime period and aggregated to derive a unique footprint for each deployment. The proportion of land cover in the 80% cumulative source area around each deployment can be seen in Table 2. Chow et al. [7] determined the land cover at the REF site based on a 2.4 m resolution Quickbird image for a circular region of 1 km 2 around the location. At the XL site, the 80% cumulative footprint is influenced mainly (Fig 1C) by the 3-4 m trees around the tower, with some contributions from a street to the east, a public transportation center to the north, and a minimum impact of a three-story building to the west. The PL footprint is influenced primarily by the parking surface and two nearby streets to the north and west, with limited influence from surrounding buildings. At the ML site, the irrigated turf grass around the tower is the main contributor and to a lesser extent there is an influence of sparse trees and one-story houses. The highest vegetation cover is present at ML (44.3%), while the lowest occurs at PL (6.6%). A large contrast is also present in the areal coverage of urban surfaces (buildings, concrete, and pavement), with the highest cover at PL (79.5%) and the lowest at ML (21.1%).

Eddy covariance measurements and data processing
The mobile EC platform consisted of a telescoping tower that extends to a maximum height of 15 m. High frequency measurements of FC, sensible heat (H), and latent heat (λET) fluxes were made using an infrared gas analyzer (LI-7500, Li-Cor Biosciences) to measure H 2 O and CO 2 concentrations, and a three-dimensional sonic anemometer (CSAT3, Campbell Scientific) to measure wind velocities [58]. Sensors were aligned to the dominant wind direction for each deployment, which were determined as 21˚at XL, 227˚at PL, 230˚at ML, and 259˚at REF. EC measurements were carried out at a height of 7.0 (XL), 9.0 (PL), and 8.0 m (ML) to ensure that turbulent fluxes were observed above the urban roughness sublayer. In almost all the cases, the EC measurements were above the surrounding roughness elements within the footprint [56]. The average urban canopy layer height (z h ) was 3.5, 2.8, and 5 m for the XL, PL, and ML sites, respectively, leading to estimated blending heights (1.5z h ) from [47] which were smaller than measurement heights. Thus, we assume that the mobile measurements sampled a blended, spatially-averaged signal considered as representative of the urban land cover within the small footprint [68]. As a result, the application of Monin-Obukhov Similarity Theory and the concept of stability are valid [69]. The REF site had a taller height of 22.1 m, measuring turbulent fluxes from a broader and more heterogeneous residential area [7]. Data was collected at 10 (PL, ML, and REF) and 20 Hz (XL) and processed at 30-minute intervals using EdiRe [70]. EC processing included corrections for stability and density fluctuations [69,70,71], coordinate rotation [72], removal of signal lags in gas concentrations due to the separation between the sensors [73], frequency response corrections [74], and estimates of sensible heat using the sonic temperature corrected with humidity following standard procedures [75]. Measurements were also filtered to exclude periods when precipitation was > 0.2 mm per 30 min, when winds were from the opposite direction at which instruments were mounted, when fluxes were further than 3 standard deviations from the mean, when the friction velocity criterion of u � < 0.15 m s -1 was met, and for absolute values of FC greater than 2 mg m -2 s -1 , according to the behavior of the 30-min values and to Schmid et al. [76]. Missing data due to data filtering and sensor malfunction accounted for 54.1%, 29.9%, 50.2% and 37.2% of the total half-hourly data obtained during the deployments at the XL, PL, ML, and REF sites, respectively. Most of the missing data corresponded to night-time measurements (80.5%, 70.2%, 59.6% and 74.3% at XL, PL, ML, and REF). Gap-filling procedures were not used to avoid the impacts that these methods might have on comparisons of daily values. Additional measurements at all sites included net radiation (R n ) using a four-component net radiometer (CNR4, Vaisala), air temperature (T a ), and relative humidity (RH) using a HMP155A probe (Vaisala), and precipitation (P) using a tipping-bucket rain gauge (TE525MM, Texas Electronics).

Urban carbon dioxide budget and meteorological conditions
The urban CO 2 budget varies from natural ecosystems due to anthropogenic sources. Urban FC is composed of sources, sinks, and storage changes [10,68,77] as: where F F is the CO 2 emitted from fuel combustion; F R is the release of CO 2 due to respiration by animals, humans, and vegetation; F P is the CO 2 assimilated by the photosynthesis of vegetation; and ΔS C is the net changes of CO 2 storage, generally considered to be small or negligible during fully turbulent conditions [78]. Consideration of the storage changes in the urban CO 2 budget is relatively rare [32,78]. Typically, FC is reported in grams (g) or milligrams (mg) of CO 2 per unit area per unit time, while carbon dioxide concentrations ([CO 2 ]) in the atmosphere are reported in parts per million (ppm). In practice, FC measurements in urban areas using the EC method are not able to identify the various origins of the CO 2 fluxes. Nevertheless, a FC < 0 indicates that plant uptake is larger than respiration and anthropogenic emissions (F P > F F + F R , or a net carbon dioxide sink), while a positive FC suggests a net carbon dioxide source (F F + F R > F P ). Neutral flux conditions (FC ffi 0) occur when sources and sinks are balanced (F F + F R = F P ). FC and the associated meteorological conditions for each sampling period at each site were analyzed at various time scales: (1) daily averages, (2) average diurnal cycles at 30-min resolution, and (3) total amounts during the sampling period. From the large set of measurements, we focus on P, T a , RH, and incoming solar radiation (R s ). For the EC systems, R n is obtained from measurements of the net shortwave (R s net ) and net longwave (R l net ) radiation as: where a is the albedo, with all radiation fluxes measured in W m -2 . As described in Templeton et al. [58], the surface energy balance for a simple plane facet in an urban area, under the assumptions of negligible anthropogenic heat, advection and energy storage, can be described as: where G is the ground heat flux, H is the sensible heat flux, and λET is the latent heat flux, all in W m -2 . Evapotranspiration (ET in mm day -1 ), obtained using the latent heat of vaporization (λ), is analyzed at daily and diurnal time scales. Furthermore, we estimated the evaporative fraction (EF) as a daily average and for the daytime period (at 30-min resolution) as: to provide insight into the relation between FC and the turbulent fluxes. Additional analyses were performed for subsets of days classified as 'wet' or 'dry' based on the occurrence of precipitation (P > 0.2 mm day -1 ) on the day of an event and the two subsequent days.

Analyses of controlling factors with ancillary data
We related the FC measurements to anthropogenic and biogenic processes that lead to sources and sinks of CO 2 in urban environments. According to Koerner and Klopatek [20], around 80% of the total CO 2 contribution in the PMA is due to vehicular traffic. As such, we analyzed FC separately for weekdays (Monday to Friday) and weekends (Saturday and Sunday) and related these to vehicular traffic counts for nearby streets to the deployments as well as to the areal fraction of pavement classified for each site.

Seasonal variations in meteorological and CO 2 conditions
Daily values of precipitation (total in mm), incoming solar radiation (average in W m -2 ), air temperature (average in˚C) and relative humidity (average in %) are shown in Fig 2 for  (1) small variations in the sensor types and deployment heights [7,58], (2) daily differences in precipitation and cloud cover at sites which were at most 42.8 km apart (Figs 1 and 3) the effects of land cover on surface properties, including albedo, soil temperature and soil moisture, that influence meteorological states through the surface energy balance, as discussed in Templeton et al. [58].  Table 3 [80]. Except for the early part of the year, the REF site exhibits a fairly constant FC during the period (average of 10.56 g CO 2 m -2 day -1 ) and a narrow range of fluctuations (standard deviation of 4.82 g CO 2 m -2 day -1 ). This is within the ranges of values (in g CO 2 m -2 day -1 ) for other open low-rise sites, for instance, in Melbourne, Australia from 8.49 to 33.4 [37] and in Syracuse, USA with 11.23 [13]. In contrast, the XL site had wide variations in daily FC (std. of 11.39 g CO 2 m -2 day -1 ) with magnitudes (ave. of 13.64 g CO 2 m -2 day -1 ) that were generally higher than at REF as well as higher [CO 2 ] (+44.25 ppm). This value is similar to an open mid-rise site measured in Sakai, Japan with 12.8 [15], but lower than year-round values reported in Tokyo, Japan [48], Mexico City [28], and Essen, Germany [24], at 43, 35.4 and 35.4 g CO 2 m -2 day -1 . Daily fluctuations at XL correspond to changes in vehicular traffic and plant phenology.
The PL site had consistently higher FC values as compared to the REF site (ave. 20.05 g CO 2 m -2 day -1 ), with larger daily variations (std. 6.39 g CO 2 m -2 day -1 ) and higher [CO 2 ] (+49.38 ppm). The only previous study with a large low-rise structure was found in Houston, USA [81], which reported a higher daily value (29.38 g CO 2 m -2 day -1 ) during the summer, however, other highly urbanized sites including compact low-rise [13,38,39,53], compact mid-rise [12,24,48,59,[82][83][84][85][86] and compact high-rise [50], reported values between 18.7 and 71.7 g CO 2 m -2 day -1 during the summer. The Pearson correlation coefficient (CC) of FC between XL and PL and REF was significant (CC = 0.31 and 0.37, respectively), whereas ML had a much lower, insignificant correlation (CC = -0.15),  suggesting a stronger similarity in the factors affecting FC at these two sites (see Table 4 for a comparison of CC for other variables between the mobile deployments and REF site  [59], Saint Paul, USA [87], and Montreal, Canada [30]. Other highly vegetated urban sites showed CO 2 uptake values during the summer, for example in Baltimore, USA [32] and in Nagoya, Japan [88].

Diurnal variations in surface energy, water and CO 2 conditions
Diurnal variations of carbon dioxide flux and latent heat flux are compared in Fig 4,  In contrast, the PL site ( Fig 4B) has a reduced amount of λET with the least variation during the day due to its low fraction of vegetation (6.6%), resulting in a positive FC during the day, with a peak of +0.48 mg CO 2 m -2 s -1 at 5:00 p.m. coinciding with rush hour traffic in the nearby street. This leads to a decoupling of the peaks in FC and λET, which are separated by 4 hours at PL. Peaks of FC during rush hours are typical of highly urbanized areas, with reported values between 0.35 to 1.67 mg CO 2 m -2 s -1 in compact low-rise and mid-rise areas during the summer [12-15, 82, 92] and about 0.62 mg CO 2 m -2 s -1 in a compact high-rise during the summer [50]. Interestingly, the diurnal cycles at XL and REF (Fig 4A and 4D) [13,15,16,24,28,37,47,48,92,93].
To complement this analysis, Fig 5 presents the diurnal cycles of CO 2 concentration and daytime evaporative fraction during each deployment period. Given the stronger variation in [CO 2 ] from winter to summer relative to FC (Fig 3), it is useful to directly compare the mobile deployments to the simultaneous behavior at the REF site (thin lines). Relatively small variations in [CO 2 ] occur throughout the day, with standard deviations of 9.45 ppm (XL), 17.11 ppm (PL), 8.46 ppm (ML), and 8.15 ppm (REF). Higher [CO 2 ] typically corresponds to morning traffic periods from 6:00 to 8:00 a.m. and in the evening from 6:00 to 10:00 p.m. when the diurnal accumulation of CO 2 and ceasing of plant uptake play a role as well as changes in the urban boundary layer height [94]. Relative to the REF site, PL has the highest [CO 2 ] and exhibits the strongest diurnal variations, in part due to its high fraction of urban surfaces dedicated to transportation (79.5%, Fig 1D) including the parking lot and nearby streets, particularly during the afternoon and night due to the nature of the surrounding businesses. During mid-day, pavements and buildings at PL have the lowest EF (0.18), an indication that surface energy fluxes are dominated by conduction from urban materials. In contrast, irrigated turf grass and trees at the ML site support a much higher mid-day EF (0.61), whose daytime variations match well with the observed decrease in [CO 2 ] in response to plant uptake. Notably, the higher overall magnitude of [CO 2 ] at ML relative to REF (Table 3) is likely due to differences in sampling height (8 m versus 22.1 m) as the decrease in [CO 2 ] with altitude is well known [23,95]. This is supported by the higher daytime EF at ML during a simultaneous period comparison with REF (thin line) which is consistent with more negative FC at the lower sampling height of ML (Fig 4C). In between these end-member cases, XL and REF exhibit diurnal behaviors with respect to [CO 2 ] and EF that are mixtures of plant uptake and vehicular emissions. Thus, to isolate the effects of these factors requires a more detailed view of site conditions, as described next. Additionally, [CO 2 ] dynamics are affected by diurnal changes in boundary layer conditions such as vertical mixing and advection [83,94], however, those factors are not analyzed here.

Controlling factors of CO 2 conditions
The effect of vehicular traffic is assessed in Fig 6 through comparisons of the average diurnal cycle of FC and [CO 2 ] for weekday and weekend days at each site. This is an approach that has been used in several studies to assess the impacts of traffic on FC [13,16,28,30,34,37,38,39,82,86,93,96]. For reference, local traffic counts (number of vehicles per hour) are provided as diurnal cycles for available time periods. Differences in FC between weekday and weekend periods are noted for rush hour periods (8:00 a.m. and 6:00 p.m.) at the XL, PL, and REF sites, coinciding with higher traffic counts. Similarly, [CO 2 ] exhibits higher values at these sites for weekdays when a higher traffic volume is expected, but typically only in the morning. Larger FC and [CO 2 ] differences at PL (ave. of 0.18 mg CO 2 m -2 s -1 and 3.57 ppm) between weekday and weekend days suggest that the CO 2 budget in the parking lot is controlled primarily by vehicular emissions in nearby streets. In addition, a progressive increase in FC is noted at PL during the daytime hours for all days, closely matching the rise in traffic. In contrast, all other sites are characterized by a mid-day decrease in FC and [CO 2 ], despite rising traffic counts at . Negligible differences (ave. of 0.01 mg CO 2 m -2 s -1 and -2.88 ppm) are noted between weekday and weekend days, suggesting the CO 2 budget in the well-irrigated mesic landscaping is controlled by photosynthetic uptake of CO 2 by turf grass and trees. A decrease in FC during weekends has also been found for open low-rise areas [13,16,30,37,93], open and compact mid-rise sites [28,39,82,86,96], compact low-rise locations [13,38,29] and some sparsely built areas [30]. In contrast, this effect has not been noted in highly-vegetated urban areas [34].
To isolate the vegetation controls, Fig 7 compares the average diurnal cycle of FC and [CO 2 ] for sunny and cloudy days during both weekday and weekend days with the incoming solar radiation for each category shown as a reference. This analysis allows inspecting the effect of plants on the CO 2 budget as sunny (cloudy) days promote (diminish) photosynthetic activity, whereas the value of R s should not impact other controlling factors such as traffic. Prior work has compared FC with radiation data or plant phenology to analyze the role of vegetation on urban carbon dioxide fluxes [10,13,15,30,55]. Since less than 20% of the days in a year are  Table 5 as a means to determine if significant relationships exist with vegetation development. Sunny days generally lead to lower FC, but not necessarily to lower [CO 2 ], in particular during mid-day, with average differences of -0.08 (   These results are consistent with those reported in other cities, where highly-urbanized areas are insensitive to changes in radiation, but highly-vegetated landscapes show differences between days with high or low radiation [10,15,30,55].
To summarize the controls on CO 2 conditions, Fig 8 presents

Sensitivity to precipitation and urban irrigation
The sensitivity of daily FC, [CO 2 ], and EF to precipitation occurrence is assessed in Fig 9  through comparisons between wet and dry days at each study site. Wet days include those days with P > 0.2 mm day -1 and the two subsequent days after the precipitation event to account for moist soil conditions. Significance tests (p � 0.05) are conducted between wet and dry days (labeled with � ) within each deployment as well as between each mobile site and REF for simultaneous periods (labeled with +). While storm events are infrequent (note the lower n for wet days), these lead to significantly higher EF at the XL and PL sites with relatively lower amounts of vegetation, but no effect at the irrigated turf grass of the ML site. As discussed in Templeton et al. [58], this is likely due to the mesic or well-watered conditions at ML which maintain high EF that is insensitive to additional water input from storm events. Both wet and dry days have statistically significant differences in EF between each deployment and the REF site. CO 2 concentrations vary significantly between wet and dry days at XL and PL, but not at the ML site, likely due to the negligible influence of precipitation on turf grass conditions. As expected, there are significant differences in [CO 2 ] between each deployment and the REF site due to the effect of different sensor heights. Similarly, daily FC varies significantly between the mobile and REF sites for both dry and wet days, attributed to differences in CO 2 emissions by vehicles and uptake by vegetation. However, the effect of precipitation occurrence was just evident at the PL site which had the lowest vegetation fraction. At the XL and ML sites, where a sufficient level of irrigated vegetation is present, FC does not significantly change in response to the additional water provided by storm events, though small increases in FC are present for wet days. The lower sensitivity of FC to rainfall at both the ML and XL sites suggests that plant photosynthesis occurs under well-watered conditions at these locations, whereas the use of EF as a diagnostic tool of this effect [58] identifies only the ML site as functioning as a mesic site.
To explore this further, Fig 10 describes the response of FC to precipitation input for the sequence of days after rainfall at each study site. In this analysis, all periods after every storm are analyzed by inspecting the daily FC to obtain an average value for all events, up to a maximum of 8 days after the event. Standard deviations across events for each day after a rainfall day are shown as error bars (± 1 std). Linear regressions (y = mx + b) of the averaged FC with days after a rainfall event are conducted to test the sensitivity of CO 2 fluxes to the storm event. We tested whether the slope of the linear regression (m) was significantly different from zero at p � 0.05. Daily FC variations are sensitive to storm events only at the XL site (m = -1.59, Fig  10A), whereas the PL, ML, and REF sites have daily FC that is insensitive to precipitation (m of -0.06, 0.17, and -0.4 that are not significantly different from zero). This is consistent with the average FC differences between wet and dry days (Fig 9), but yields additional information on Carbon dioxide fluxes in Phoenix, Arizona the rate of FC changes with time after a rainfall event, including insight on the transition from wet to dry days. Notably, the xeric landscaping with irrigated trees at XL had progressively more CO 2 uptake (lower FC) as time progressed after rainfall events during the winter-spring. A similar behavior occurs at the REF site during the same period as XL (Fig 10B, m = -1.43, significantly different from zero), indicating that the CO 2 uptake occurred across different urban landscapes and was likely tied to seasonal (winter-spring) conditions promoting a photosynthetic response of xeric trees. For instance, we visually noted that palo verde flowered after these winter-spring rainfall events. In contrast, the parking lot at PL exhibited a very small increase in CO 2 emissions (higher FC) after rainfall events during the early summer that was Carbon dioxide fluxes in Phoenix, Arizona also noted at REF (Fig 10B), but not at a significant level (mm byt m = 0.33). This suggests that additional water from precipitation during a period of high temperatures in May and June promotes CO 2 efflux, mainly by increasing soil respiration in bare soil areas, in a similar fashion as noted in natural ecosystems of the region [90,91,99,100]. During other times of the year, the site with ample outdoor water use (ML) does not respond to precipitation (FC remains the same), suggesting that a decoupling occurs between CO 2 fluxes and storm inputs as commonly found in mesic regions.

Conclusions
While bottom-up approaches have been used to estimate CO 2 exchanges in arid and semiarid cities, few studies have carried out direct observations in different urban patch types. Indeed, a comparison of these approaches is warranted as the number of direct observations grows. At present, there are only a small number of studies discussing the controlling factors on FC and [CO 2 ], such as vehicular emissions and plant photosynthetic activity, and their link to the proportion of these urban land covers within a site [for instance, 13,15,31,35,92,94], though the temporal variation in vegetation and anthropogenic activity has typically not been taken into account to date. In this study, we conducted turbulent flux measurements using the EC technique to obtain a detailed view of CO 2 fluxes and relate these to local meteorological conditions and urban characteristics for three short-term deployments and a stationary reference site in Phoenix, Arizona, USA. Comparisons to the suburban reference site were conducted during simultaneous periods for different seasons such that measured differences could be attributed to local variations in urban conditions. Results from the comparisons across the sites, seasons, and urban land cover types indicated the following: 1. Despite the small differences noted in meteorological conditions, the magnitude and behavior of FC and [CO 2 ] varied considerably among the sites, in manners consistent with the urban land cover type. XL, PL, and REF acted as net sources of carbon dioxide, though plant activity was able to counteract anthropogenic emissions during mid-day periods. At ML, the well-watered turf grass was a net sink of CO 2 during the summer season.
2. Diurnal variations in FC and [CO 2 ] exhibited a strong correspondence to rush hour timing and vehicular counts for sites with large fractions of transportation surfaces, depending on local traffic behavior. Statistically significant differences were noted in FC between weekday and weekend days for all sites, except where vegetation activity served as a carbon dioxide sink. Vehicular emissions led to a temporal decoupling of CO 2 and water vapor fluxes during the day.
3. Where urban irrigation supports a plant community, mid-day values in FC and [CO 2 ] showed decreases consistent with the increase in measured latent heat flux. Statistically significant differences were noted in FC and [CO 2 ] between sunny and cloudy days for most sites, except where the vegetation cover was low. A close correspondence was noted in the daily peak timing of CO 2 and ET fluxes where outdoor water use supports plant photosynthesis.
4. The sensitivity of FC and [CO 2 ] to precipitation events varied considerably among the sites in accordance with the proportion of irrigated vegetation. Where outdoor water use is abundant and frequent, CO 2 conditions are insensitive to the occurrence of precipitation (wet versus dry days) or the time since the last rainfall event. This decoupling between CO 2 fluxes and storm inputs suggests that irrigated landscapes in arid urban areas behave as mesic systems.
Based on these comparisons, key differences in the CO 2 conditions can be attributed to the vegetation fraction and built surfaces in urban patches. Two of the sampled sites can be considered as end members that are dominated either by the effects of traffic and other anthropogenic emissions (PL) or by the carbon dioxide uptake from photosynthetic activities of turf grass and trees (ML). The other two sites (XL and REF) are characterized by combinations of these land cover types and thus exhibit intermediate or mixed behavior with respect to CO 2 conditions. As noted by Templeton et al. [56], it would be desirable to conduct cross-site comparisons of this type over at a full year or longer to assess net effects of vehicular traffic and vegetation activity on CO 2 fluxes. Such a study could also quantify the seasonal variations in these factors responding to plant phenology and temporal changes in anthropogenic activities. For instance, the role played by seasonality and its interaction with irrigation is considered important in determining if plant activity can fully counteract anthropogenic CO 2 emissions during an annual period. For the periods studied here, vegetation could not counteract CO 2 emissions, leading to a net carbon dioxide source at all sites, including the mesic landscaping. Nevertheless, this cross-site comparison suggests a fruitful avenue for scaling up CO 2 conditions to larger areas by using the fraction occupied by urban vegetation and built surfaces. Following this strategy could lead to considerable improvements in bottom-up estimates of CO 2 fluxes and concentrations to better capture the anticipated spatiotemporal variability in desert cities.