Cities account for most anthropogenic greenhouse-gas emissions, CO2 being most important. We evaluate the net urban contribution to CO2emissions by performing a meta-analysis of all available 14 annual CO2budget studies. The studies are based on direct flux measurements using the eddy-covariance technique which excludes all strong point sources. We show that the fraction of natural area is the strongest predictor of urban CO2 budgets, and this fraction can be used as a robust proxy for net urban CO2emissions. Up-scaling, based on that proxy and satellite mapping of the fraction of natural area, identifies urban hotspots of CO2emissions; and extraction of 56 individual cities corroborates their inventory-based estimates. Furthermore, cities are estimated as carbon-neutral when the natural fraction is about 80%. This fresh view on the importance of cities in climate change treats cities as urban ecosystems: incorporating natural areas like vegetation.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Over 70% of global energy-related CO2 emissions originate from cities [International Energy Agency, 2008; Rosenzweig et al., 2010], and fossil fuel combustion―in transportation, industry, and housing―is the dominant urban CO2 source. Besides CO2sources, cities typically have green areas that are carbon sinks through photosynthetic uptake. Since over half of global population is city-based (UN,http://esa.un.org/unpd/wup), urban areas are hotspots of greenhouse gas (GHG) emissions and the focus of innovation in climate-change mitigation [Kennedy et al., 2009]. International negotiations are a key part of climate-change mitigation via GHG-emission reductions. These negotiations necessitate quantification of net GHG exchange between surface and atmosphere, especially CO2. The eddy-covariance method, the only direct way of measuring surface–atmosphere GHG exchange, is widely applied in natural surroundings. Extensive measurement networks have emerged (AmeriFlux [Baldocchi et al., 2001], EuroFlux [Valentini et al., 2000], AsiaFlux [Mizoguchi et al., 2009]), and continental-scale budgets have recently been estimated of biological GHG fluxes for Europe [Schulze et al., 2009] and global terrestrial ecosystem CO2 uptake [Beer et al., 2010]. Measurements in urban environments have only, during the past year, become extensive enough to enable synthesis of CO2exchange at numerous eddy-covariance sites [Grimmond and Christen, 2012]. Individual local-scale and city-scale budget estimates include both direct emissions from the city and contributions from vegetation: photosynthetic uptake and respirative emissions, which are seldom within inventories. We thus denote estimates from direct flux measurements asnet urban ecosystem exchange(NUE); a counterpart to net ecosystem exchange, used canonically for non-urban fluxes. NUE describes the CO2budget of urban ‘background activity’: including direct emissions from e.g. buildings and traffic; but excluding strong point sources like power stations. This is not disadvantageous, since emissions from such strong point sources are well-described in inventories (Carbon Monitoring and Action,http://carma.org), whereas NUE has remained unresolved until now.
 Apart from direct flux measurements, multiple studies have gathered inventories of annual GHG emissions from nations [Hertwich and Peters, 2009] or individual urban areas [Dodman, 2009; Kennedy et al., 2011] based on consumption statistics. Furthermore, area-specific inventories of CO2emissions have been utilized for mapping global emissions based on population densities and night-time lights [Raupach et al., 2010; Oda and Maksyutov, 2011], and CO2emissions in USA have been down-scaled to a 10 km grid scale [Gurney et al., 2009; Parshall et al., 2010]. These inventory-based methods, as indirect estimations of GHG emissions, have inherent disadvantages: fossil fuels are often not consumed where purchased, population data represent commuting inconsistently, and there is no clear literature consensus whether emissions estimates should be end-use or only from those emissions produced within city borders. Conversely, the definition of NUE is rigorous: it is anin-situ measure of the local CO2budget, excluding any city-related emissions located outside city borders (e.g. aviation, marine, electricity production, product manufacturing).
 We present a meta-analysis of a compilation of 17 annual NUE budgets from direct flux measurements from 14 urban eddy-covariance towers, and analyze the predictors of surface–atmosphere exchange. The fraction of natural area in a city (derived from satellite data) is used as a proxy for estimating regional variation of NUE in parts of North America, Europe, and eastern Asia; and those estimates for particular cities are compared against inventory-based estimates. Here natural fraction means the fraction of land area that is covered by surface types existing in nature (e.g. grass, trees, soil, sand, rock), whereas urban fraction means all non-vegetative, human-constructed elements. Additionally, a natural fraction is conjectured for a carbon-neutral city (annual NUE zero).
2. Data and Methods
 The eddy-covariance technique is tower-based and the measurements' source area is several hectares: depending on the upwind surface, measurement height, and flow properties [Vesala et al., 2008]. The technique is based on measuring simultaneous turbulent variations in wind and gas concentrations (e.g. CO2): output typically being 30-minute fluxes. The measurements used in this study (Figure S1 and Table S1 in Text S1 of theauxiliary material) represent over 16 000 days' measurements with 58% data coverage (the percentage is typical for the method). Data were processed by original authors with widely-accepted procedures, including quality-screening, the main cause of data-loss. Time series consequently were gap-filled (typical errors are below 5% [Järvi et al., 2012]) to enable annual-sum calculations. Furthermore, typical random errors are under 20% for half-an-hour fluxes [Aubinet et al., 2012; Nordbo et al., 2012], and the random error is assumed negligible for annual sums of NUE. Data coverage at each site is taken into account when analyzing predictors of NUE and non-linear least-squares optimization is used for non-linear fits.
 The natural fraction (fn) is estimated from the urban fraction (fu = 1 − fn) which can be retrieved from global satellite data. The urban fraction is derived from binary (urban/non-urban), 500-meter-resolution data from the MODIS satellite in 2001–2002 [Schneider et al., 2009, 2010]. Urban area was defined as “a place dominated by built environment”—which includes all non-vegetative, human-constructed elements like buildings, roads, and runways. The mean accuracy of the binary data exceeds 93% [Schneider et al., 2009]. Binary data were converted by us to fu by aggregating to 4 × 4 km resolution, each aggregated pixel based on 64 binary values. GHG inventories from 56 urban areas (cities, metropolitan areas, counties) in North America, Europe, and eastern Asia were compiled (Table S2 in Text S1 of the auxiliary material). Corresponding NUE estimates are subsequently retrieved using the 4 km resolution values of fu within official administrative areas (Database of Global Administrative Areas, www.gadm.org). A pixel is included if the pixel's center is within the administrative area. If beyond 50% of a pixel is water, the pixel is omitted. If administrative areas are below 1000 km2, fudata are disaggregated back to 500-meter resolution to minimize edge problems.
3. Results and Discussion
3.1. Predictors of Net Urban Ecosystem Exchange
 The compilation of direct annual NUE measurements shows that urban areas are sources of CO2 (Figure 1). Annual emissions are as high as 9.7 kg C m−2 yr−1 in London [Helfter et al., 2011], which is forty times the typical uptake of grassland [Soussana et al., 2007] and over ten times the global median terrestrial ecosystem uptake [Beer et al., 2010]. The Minnesota site, conversely, is only a very small CO2 source, since the measurements were carried out in an urban park [Hiller et al., 2011].
 The fraction of natural area (fn) in the source area of the measurements is a robust proxy for annual NUE (coefficient of determination r2 = 0.84, Figure 1), especially when considering the variety between cities' surface cover and latitude. Non-gap-filled daily CO2 fluxes (mostly from summertime) have been compared with vegetation fraction for 18 sites [Velasco and Roth, 2010] and with building fraction for 22 sites [Grimmond and Christen, 2012]. Five annual CO2 budgets were compared as a function of vegetation fraction [Helfter et al., 2011], but two of the budgets were based on measurements during one season, and the given exponential fit saturated to an unrealistic value for complete vegetation cover (2.5 kg C m−2 yr−1). Population density has a lesser correlation with NUE (r2 = 0.60, rms 1.13 kg C m−2 yr−1 , Figure S3 in Text S1 of the auxiliary material), although population density is an indicator for energy consumption, perhaps because traffic-fuel demand decreases as population density grows [Kennedy et al., 2009; Karathodorou et al., 2010]. A correlation between annual CO2 budgets and population density has not been proven before.
 The strong relationship between NUE and fn can be explained by the indirect links that fn has to many factors determining CO2 release: greater fn is consistent with a lesser road and population density, which thus limits CO2release from fossil-fuel combustion and human respiration [Moriwaki and Kanda, 2004]. Greater fncan also reduce pedestrian and building cooling-needs, e.g. via shading effects of trees [Simpson, 2002]. Vegetation itself is also a key NUE component through daytime sequestration of carbon via photosynthesis. The non-linearity of the relationship inFigure 1 comes from the dependency of population density on urban density: population grows exponentially when a dense city becomes even more compact [Pozzi and Small, 2005] since cities do not only grow horizontally but also vertically. Furthermore, dense urban living may generate less per-capita GHG emissions compared to rural living, given a similar income level, since urban areas often have lower emissions than the national average [Brown et al., 2009; Dodman, 2009] or rural areas [Parshall et al., 2010]. Standard of living is a predictor for CO2 release, since greater incomes often lead to greater consumption [Kennedy et al., 2009]. Conversely, technological advances—usually a consequence of wealth increase—can decrease the emission intensity of CO2 per unit GDP: the overall global emission intensity has decreased by 41% from 1971 to 2007 [International Energy Agency, 2009].
 A minimum fnrequirement for a carbon-neutral city (annual NUE zero) can be interpreted from the fit inFigure 1: cities are net sinks of CO2if their natural fraction exceeds about 80%. This value can be used as a first rule-of-thumb estimate in urban planning, among other indicators [Kennedy et al., 2011]. The general definition of a carbon-neutral city suffers from a scoping problem: some definitions require zero carbon emissions, others allow emissions to be balanced by sequestration or export of low-carbon goods [Kennedy and Sgouridis, 2011]. In our case, the limit for carbon-neutrality treats the city as an urban ecosystem comprisingin-situ sinks and sources of CO2 within the city boundaries. Increasing the natural land area fraction within a city is expected to decrease the CO2emissions per unit area, but this is not a general solution for climate-change mitigation: if natural area substitutes previously-occupied buildings, the per-capita emissions might increase, if the living density decreases. Conversely, if an unused urban area (e.g. abandoned car park) is transformed into a vegetated area, then there is an obvious, but small, net gain. Green roofs, to the contrary, are usually a net gain in energy savings [Sailor et al., 2012] and photosynthetic uptake of CO2.
3.2. Regional Estimates for NUE
 The strength of fnas a predictor of NUE provides a means for producing annual NUE estimates based solely on land-cover data, using the relationship inFigure 1. The global fu is calculated based on satellite observations for North America, Europe, and eastern Asia (Section 2). These areas were chosen for mapping since all-but-one of the flux sites are within these regions and thus the relationship inFigure 1 is assumed to be applicable within these regions. In Europe, high urbanization dominates around the Benelux countries, Germany, and southern England (Figure 2a). In North America, the north-east coast of USA, the Great Lakes region, Los Angeles, and Florida have conspicuously continuous and high urban fraction (Figure S4a in Text S1 of theauxiliary material). In eastern Asia, the east coast of China and the Tokyo metropolis have prominently-high urban fractions (Figure S5a in Text S1 of theauxiliary material).
 Mid-Europe and United Kingdom are regions of high NUE in Europe (Figure 2b). The summed NUE over EU25 and EU27 (EU countries prior to and after 2007) are 410 Tg C yr−1 and 414 Tg C yr−1, from which about 9% can be allocated to human respiration of the whole population. The inventory-based emissions were twice the NUE (767 Tg C yr−1 in 2006) [International Energy Agency, 2008], and NUE is almost four times the uptake of CO2 by biological fluxes (−102 ± 23 Tg C yr−1, EU25) [Schulze et al., 2009]. The inventory-based emissions and NUE are not intended to coincide, since some inventories include both strong point sources and emissions occurring outside the city borders. Furthermore, NUE includes vegetation uptake, whereas the inventories only include CO2 emissions and might include emissions of other GHGs.
 In North America, the regional distribution of NUE follows that of urban fraction (Figure S4b in Text S1 of the auxiliary material). The summed NUE is 460 Tg C yr−1 for USA48 (i.e. all states excluding Alaska and Hawaii). From this, 5% can be allocated to human respiration of the whole population (76.3 kg C yr−1 per person [Moriwaki and Kanda, 2004]) and about 5% to uptake by urban trees (23 Tg C yr−1) [Nowak and Crane, 2002]. In 2006, the inventory-based CO2 emissions from urban areas in USA were three times the NUE (1228 Tg C yr−1) [International Energy Agency, 2008].
 In eastern Asia, a large area of NUE sink is seen in north-east China where low urban fraction is observed (Figure S5b in Text S1 of theauxiliary material). Chinese urbanization is characterized by a continuous sprawl as an opposite to the more confined city structures seen in Europe (see also Figure S2 in text S1 of the auxiliary material). To the contrary, the metropolitan areas of Tokyo, Seoul, Beijing, and Shanghai arise as large CO2 sources. The NUE of Japan is 77 Tg C yr−1, which is 22% of the country-wide inventory estimate [Nojiri et al., 2012]. An estimate for the whole of China is not given, since the sites in Figure 1 are not representative of China as a whole.
3.3. NUE and Inventories of Individual Cities
 The NUE estimates of urban areas are expected to relate to inventories, though they are also expected to be systematically lower (as discussed above). Following this reasoning, a set of 56 GHG inventories from individual cities, or metropolitan areas, was collected in order to conduct an independent comparison against the corresponding NUE estimates (see Table S2 in Text S1 of the auxiliary material). The NUE estimates are lower than the inventory-based GHG emissions for all cases but Prague (slope 0.50,Figure 3). The inclusion of strong point sources in inventories can be seen for example for Rotterdam where over 60% of emissions are due to energy industries. Nevertheless, there is a clear linear dependency (r2 = 0.72, rms = 1.42 kg C m−2 yr−1), and 22 out of the 56 cities are net CO2 sinks. This result corroborates the usability of the proxy of NUE as a function of fn, and confirms its use as a robust independent check against international inventory studies.
4. Summary and Conclusions
 The urban CO2 budgets (Figure 2, Figures S4 and S5 in Text S1 of the auxiliary material) are the first continental-scale estimations from direct flux measurements. The mapping estimation treats cities as ecosystems, i.e. incorporating vegetation. The mapping is solely from the relationship between the CO2budget and the natural land fraction; the 14 eddy-covariance stations are assumed representative at continental scale—resulting in high uncertainties in estimates of annual CO2budgets. Direct validation of the continental-scale NUE parameterization is not possible due to the non-existence of another method that could provide CO2budgets of urban ecosystems. Nevertheless, the high correspondence of our continental-scale estimates with individual cities' inventory-based estimates (Figure 3) supports the new method's robustness. The inventory-based estimates exceed the parameterized NUE (median ratio 1.33), this is reasonable since NUE lacks strong point sources and includes vegetative uptake.
 Direct urban-flux measurements have become comprehensive enough to benefit decision-makers and urban planners. Additional urban CO2 budget measurements, extending beyond a year, are needed to represent diverse urban morphologies (very high or low fu) and climates (Asia, Africa, and South America). Until present only short-term urban CH4 [Gioli et al., 2012] and N2O [Famulari et al., 2010] campaigns have been under-taken, although their anthropogenic emissions are increasing [Montzka et al., 2011]; the need for long-term flux measurements is clear.
 For funding we thank the Academy of Finland Centre of Excellence program (project 1118615), the Academy of Finland project 138328, the Academy of Finland ICOS project (263149), the EU ICOS project (211574), the EU GHG-Europe project (244122) and an EU FP7 grant (ERC 227915). For the satellite surface-cover data we thank Annemarie Schneider. We acknowledge Hotel Torni for providing a platform for eddy-covariance measurements.
 The Editor thanks an anonymous reviewer for assistance evaluating this manuscript.