Y a n g L i u a

,H u i z h i L i u a,b,*,Q u n D u a,L u j u n X u a

a State Key Laboratonry of Atmospheric Boundary Layer Physics and Atmospheric Chemistry, Chinese Academy of Sciences, Beijings China

b University of Chinese Academy of Sciences, Beijing China

2. Site description and instrumentation

2.1. Site description

The EC data were measured from January 2008 to December 2012 at the 325-m meteorological tower of Beijing, China (39°58’N, 116°22’E).It is rare to have such a tall meteorological platform in the center of a densely populated megacity with EC measurements at multiple layers. The surrounding areas of the tower are high-rise residential buildings and impervious roads, which represent the characteristics of typical urban surfaces. The description of the measurement site parallels that of Liu et al. (2017) . The aerodynamic roughness length and the zeroplane displacement height were taken as 3.9 m and 16.7 m, respectively( Liu et al., 2012 ).

2.2. Instrumentation and the data processing

The EC instruments were mounted at three layers of 47, 140, and 280 m with azimuth angles of 110°, 120°, and 150°. A sonic anemometer (CSAT3, Campbell Scientific Inc., USA) and an open-path infrared gas analyzer (LI—7500, Licor Inc., USA) were used to continuously measure the 3D wind speed, the concentration of water vapor and CO 2 at the frequency of 10 Hz. The precipitation data were obtained from the Beijing Chaoyang meteorological station.

Dynamic mean and standard deviation values within a series of moving windows were applied to detect spikes. The fluxes were computed at 30-min averages. The mean vertical velocity was forced to be zero after a double rotation. The CO2fluxes were corrected for the density fluctuations ( Webb et al., 1980 ). Data marked ‘7-9’, which failed the steady state and integral turbulence test, were excluded from the study( Foken and Wichura, 1996 ). Missing data were reconstructed as follows.Linear interpolation was used when the gaps were less than 2 h. Gaps less than 2 d were replaced using the mean diurnal variations method.When gaps were longer than 2 d, the multiple imputation method was employed.

The footprint model by Kormann and Meixner (2001) was applied to investigate the influence of the surface land fraction on CO 2 fluxes.The 5-yr average accumulative footprint areas at 47, 140, and 280 m were 4.1, 31.9, and 124.4 km2. At 280 m, the footprint area was out at the 5th ring road of Beijing and contained suburban areas. The largest distances of the footprint areas at 47, 140, and 280 m were 3.1, 10.7,and 20.4 km, respectively. The largest distances were all located in the northwest, which were in accordance with the prevailing wind directions. The footprint areas were equally divided into 12 sectors, which corresponded to 30°( Table 1 ).

Table 1 Annual mean surface cover fractions of roads (2008—2012). The road fractions (units: %) were calculated from satellite maps from Google Earth.

Fig. 1. Average diurnal pattern of CO 2 flux in each season at (a) 47 m, (b) 140 m, and (c) 280 m. The bars denotes one standard deviation.

3. Results and discussion

3.1. Meteorological conditions

Beijing is located in the North China Plain region, with a temperate continental climate. Precipitation is unevenly distributed among each season. The wet season, from May to October, accounts for nearly 80%of the annual precipitation. The annual total precipitation from 2008 to 2012 was 618.5, 606.2, 559.1, 732.9, and 733.2 mm, with an average of 650.0 mm.

The prevailing wind directions at 47, 140, and 280 m in summer are from the south, southwest, and south, while in winter they are from the northwest, west, and northwest, respectively. In Beijing, most cases are under unstable and near-neutral atmospheric stratifications at 47 m (77.5%) and 140 m (68.2%) for human activities and stored heat released from the urban canopy layer.

3.2. Diurnal and seasonal variations of CO 2 fluxes at three layers

The CO2fluxes at 47 m and 140 m remained the same from 0800 to 1930 LST following the rush-hour traffic ( Fig. 1 ). In Beijing, traffic jams exist during daytime and photosynthesis from vegetation cannot obviously cut down the CO 2 fluxes. The observed diurnal pattern of CO2fluxes does not follow those measured in other cities where a pronounced two-peak pattern is found ( Bergeron and Strachan, 2011 ). The CO2flux at 280 m reached its maximum at 1200 LST and continued to 1800 LST. The duration was shorter at 280 m and the peak value was also smaller than at 47 and 140 m. At the daily time scale, traffic is the most important controlling factor of CO2flux in Beijing.

Fig. 2. Average CO 2 flux versus wind direction and its relationship with road fractions at 47, 140, and 280 m in summer.

Table 2 Annual total CO 2 fluxes at 47, 140, and 280 m and total vehicle numbers, population, and annual mean air temperature during 2008 to 2012. Total vehicle numbers and population data were from the China Statistics Yearbook( National Bureau of Statistics of China, 2014 ).

For domestic heating and fewer CO 2 sinks, the CO 2 fluxes were largest in winter at the three layers. In midlatitude cities like Tokyo and Lodz, the CO 2 fluxes in winter are 1.5 to 2 times larger than those in summer because of domestic heating ( Moriwaki and Kanda, 2004 ;Pawlak et al., 2011 ). In Singapore city, which is in a low latitude area with the absence of domestic heating in winter, the mean CO2fluxes in summer and winter are almost the same ( Velasco et al., 2013 ).

3.3. Spatial variation of CO 2 fluxes

The average CO2fluxes versus wind direction in summer are shown in Fig. 2 . The road fractions were calculated from the values in summer in 2008—2012. The largest CO 2 fluxes during the daytime were observed from the northeast—southeast direction, which corresponded to the Beijing—Tibet expressway and high-rise residential buildings. The lack of traffic and photosynthetic processes by vegetation in the park caused CO2fluxes to be much lower in the west.

The CO 2 fluxes at the three layers all presented positive correlation with road fraction. A larger fraction of roads can bring more traffic,which will lead to more CO 2 fluxes. TheR2values between the fraction of roads and CO2fluxes at 47, 140, and 280 m were 0.69, 0.57, and 0.54, respectively (P< 0.05).

3.4. Annual total and interannual variations of CO 2 fluxes

With all gaps filled, the annual total CO 2 fluxes in Beijing at 47, 140,and 280 m are given in Table 2 . The measurement at 280 m began from July 2008 and so the annual total of 2008 is not given. The urban surfaces were a net source of CO2annually, with an average of 5.78, 6.47,and 3.99 kg Cm−2yr−1at 47, 140, and 280 m, respectively. The lower annual total CO 2 fluxes in 2008 were because of the restriction of traffic and closure of factories due to the Beijing Olympic Games ( Song and Wang, 2012 ). From the year of 2011, some restrictions such as a vehicle plate number limit caused the slowing down of increasing rates of annual total CO2fluxes.

The growth rate of annual total CO 2 fluxes at 140 m from 2008 to 2010 was 7.8%, while from 2010 to 2012 it was 2.3%. The differences in annual total CO 2 fluxes between 47 m and 140 m were because of the distinct corresponding footprint areas. The footprint area at 280 m contained suburban areas. The fraction of vegetation in the footprint area was 30.6%, which was much larger than that at 47 and 140 m. Furthermore, a larger footprint area will cut down the emissions of nearby sources from roads and buildings. The mixing process during long-distance transport will weaken the CO2flux. Results in Beijing were the highest compared with other urban sites around the world ( Table 3 ).The small vegetation cover and large population density caused the Fc(CO2flux) in London to be the largest ( Helfter et al., 2010 ; Ward et al.,2015 ).

The annual total CO2fluxes increased year after year at all three layers and were positively correlated with the total vehicle numbers and total population. From 2011 the restriction on vehicles during weekdays slowed down the increasing rate of annual total CO2fluxes. In Beijing,the annual mean air temperature did not have an obvious trend from 2008 to 2012. Compared to the total vehicle numbers and total population, the mean air temperature was less relevant to CO 2 flux on the annual time scale.

3.5. Effects of meteorological parameters and anthropogenic activities on CO 2 fluxes

In Beijing the monthly average CO2fluxes and air temperature were negatively correlated ( Fig. 3 ). The combustion of fossil fuels and lack of photosynthesis by vegetation both increased the CO2fluxes. When the temperature lowers in winter, the fuel consumption by heating increases. As temperatures then increase, the disappearance of domestic heating and the uptake of CO 2 by vegetation lead to lower CO 2 fluxes. Insummer, the power consumption used for air conditioning will not cause the CO2to increase because the electricity supply in Beijing comes from other provinces like Hebei, Shanxi, and Inner Mongolia. The largest CO2uptake occurred in the middle of summer as the leaf area index of vegetation reached its peak. Therefore, the CO 2 fluxes decreased as air temperature increased. Monthly total CO2emissions have also been found to be negatively correlated with air temperature in Montreal and London ( Bergeron and Strachan, 2011 ; Ward et al., 2015 ). The linear fit functions of monthly average temperature and monthly average Fc are shown in Fig. 3 (b—f). TheR2values at 47, 140, and 280 m were 0.66,0.71, and 0.65, respectively. In Florence, theR2value was found to be 0.73, which is nearly the same as in our study ( Matese et al., 2009 ).Meanwhile, the gradient in London ( − 1.95 μmol m−2s−1/°C) was found to be larger than that in our study at all three layers ( − 0.33, − 0.49, and− 0.25 μmol m−2 s−1 /°C, respectively) ( Ward and Evans, 2013 ).

Table 3 Annual observed CO 2 flux, vegetation cover, and population density from urban sites in the literature.

Fig. 3. Monthly average CO 2 fluxes and air temperatures at (a, c, e) 47, 140, and 280 m. The error bars indicate one standard deviation. Linear fits of monthly average air temperature and monthly average CO 2 flux are given in (b, d, f).

Fig. 4. Annual net carbon exchange for the three layers in this study versus (a)plan area vegetation cover and (b) population density alongside values from the literature.

The annual net carbon exchanges for the three layers versus plan area vegetation cover are presented in Fig. 4 (a). The annual net CO2exchanges at the three layers increased as the vegetation cover decreased.Reduced vegetation cover corresponded to an increased impervious fraction of buildings and roads, which brought more emissions. Once the vegetation cover decreases to a critical value, the buildings will become taller and the density higher ( Nordbo et al., 2012 ). Consequently,CO2fluxes have an exponential relationship with vegetation cover in urban areas. In 18 urban and suburban sites, the average coefficient between vegetation cover and the CO 2 fluxes was 0.87 ( Velasco and Roth, 2010 ). The photosynthesis and respiration of vegetation will influence the carbon exchange. The higher fraction of vegetation was in accordance with a decreasing fraction of roads and population density,which will restrict the influence of traffic and human activities on CO2fluxes ( Moriwaki and Kanda, 2004 ).

The population density was linearly correlated with CO2fluxes( Fig. 4 (b)). There are large variations in the number of people (residents and tourists) in Beijing throughout the day. Such considerable uncertainty in population density weakens its validity as a predictor for CO 2 flux. The central London annual carbon release measured in Ward et al. (2015) was largest (12.71 kg C m−2yr−1), corresponding to the largest population density and low vegetation cover ( Table 3 ). A highly vegetated and low population density suburban site in Baltimore had the smallest CO 2 flux of 0.36 kg C m−2yr−1( Crawford et al., 2011 ).

4. Conclusions

The CO2exchange between the atmosphere and urban surfaces in Beijing megacity was investigated using the EC method. At daily time scales, traffic was the most important controlling factor of CO 2 flux. The monthly average CO2flux was negatively correlated with the monthly average temperature. The mean annual total CO 2 flux was larger at 140 m (6.41 kg C m−2yr−1) than at 47 m (5.78 kg C m−2yr−1) and 280 m(3.99 kg C m−2 yr−1 ). At 280 m, the footprint contained suburban areas,which led to the smallest CO 2 flux.

There is a linear relationship between the fraction of roads and the CO2fluxes. A larger fraction of roads could bring more traffic, which will produce more CO2fluxes, while theR2values were 0.69, 0.57, and 0.54 at 47, 140, and 280 m, respectively. From 2011 the control on total vehicles caused the growth rate of annual total CO2fluxes to slow down compared with 2008 to 2010. Compared to air temperature, total vehicle numbers and the population are better predictors for annual total CO2fluxes in Beijing. The decreasing fraction of vegetation will cause an increasing of annual total CO 2 flux, and there is an exponential relationship between them. The annual total CO2flux is larger when the population density is higher. Owing to the uncertainty in measuring population density, the precision between population density and urban CO2fluxes has been limited.

FundingThis research was funded by the National Key Research and Development Program of China [grant number 2017YFC1502101 ] and the National Natural Science Foundation of China [grant numbers 41905010 and 41675013 ].