Carbon dioxide (CO₂) is a strong greenhouse gas, and its radiative forcing (RF) since industrial era is reported to be 1.82 W m^–2 with an average growth rate slightly less than 0.3 W m^–2 per decade (Stocker et al., 2013). The World Meteorological Organization (WMO) Greenhouse Gas Bulletin No. 13 states that CO₂ has been the dominant component of greenhouse gases (GHGs) and contributes approximately 65% to the total global RF by long-lived greenhouse gases (World Meteorological Organization, 2017). Climate modeling is an effective approach to quantifying the radiative impact of GHGs on climate system. Currently, global or regional climate models have been developed with detailed parameterization schemes to consider the RF of GHGs. However, the spatiotemporal variations of atmospheric CO₂ are usually ignored in these models (Kiehl and Ramanathan, 1983). Atmospheric CO₂ is often prescribed to be well mixed in these models. For instance, CO₂ mixing ratio is assumed as 355 ppmv in a global climate model CAM3.0 (Collins et al., 2004), while it is set as 330 ppmv in a mesoscale model MM5 (Dudhia et al., 2004). In fact, real-time atmospheric CO₂ concentrations are constantly modulated by the magnitudes of sources and sinks of atmospheric CO₂ fluxes. For example, the averaged daytime ambient concentrations of CO₂ in residential areas vary from 495 to 621 ppmv in Delhi (Sahay and Ghosh, 2013), whereas the maximum daytime ambient concentrations of CO₂ in Nagpur are only 391 ppmv (Chaudhari et al., 2007). Although annual increments of GHGs during the period 1750–2100 have been considered in radiation schemes of some climate models such as RegCM4, seasonal and spatial variations of CO₂ concentrations have been neglected (Elguindi et al., 2010). The constant CO₂ concentrations prescribed in climate models undoubtedly influence model performance (Yang et al., 2012). Therefore, it is essential to investigate the effects of non-homogeneous CO₂ concentrations on the regional RF and temperature.

Three dimensional numerical models can capture the impacts of human activities (Cheng et al., 2013), ecosystem (Kou et al., 2015, 2017), mesoscale transport (Ahmadov et al., 2009), and chemical conversion (Nassar et al., 2010) on atmospheric CO₂. Therefore, it is possible for these models to capture the spatial and temporal variations of atmospheric CO₂. Models for simulating atmospheric CO₂ can generally be divided into global and regional models by the simulation domains. Global models such as TM5 (Krol et al., 2005), GEOS-Chem (Nassar et al., 2010), and CarbonTracker (Cheng et al., 2013) cannot identify mesoscale effects generated by land surface heterogeneity and complex topography (Ahmadov et al., 2009). In contrast, it is possible for regional models to simulate atmospheric CO₂ concentrations at high spatial and temporal resolutions. Regional models like WRF-VPRM (Ahmadov et al., 2007), RAMS-CMAQ (Kou et al., 2013), and REMO (Chevillard et al., 2002) are able to capture sharp changes in diurnal variability in CO₂ concentrations within a much shorter spatial distance than global models. For example, Ballav et al. (2012) simulated the regional transport of atmospheric CO₂ using 5 different CO₂ fluxes from ocean, fossil fuel, and terrestrial biospheres at the horizontal resolution of 27 × 27 km; Kou et al.(2015, 2017) assessed the impacts of biosphere–atmosphere CO₂ exchange on seasonal variations of atmospheric CO₂ concentrations over East Asia, using a regional model.

East Asia has a broad terrain with complex topography and diverse climates (Lau et al., 2007). Along with the rapid expansion of energy consumption, anthropogenic CO₂ emissions from human activities have significantly increased in recent years (Guan et al., 2009). Moreover, regional economic development imbalance still persists in the region, which is a main factor contributing to CO₂ inhomogeneity (Wang and Liu, 2008). Previous studies have found that the CO₂ concentrations change sharply at Shangdianzi (from 377.8 to 391.7 ppmv) and Lin’an (from 383.2 to 393.2 ppmv) due to the great influence of urbanization in the Beijing–Tianjin– Hebei region and the Yangtze River Delta region (Liu et al., 2009). Measurements from national background stations indicate that the atmospheric CO₂ concentration at Waliguan station is 383.5 ppmv, while ambient CO₂ mixing ratio over urban Nanjing reaches 406.5 ppmv (Huang et al., 2015). Given such spatial variations in CO₂ concentrations, it is of scientific significance to explore the spatiotemporal distribution of atmospheric CO₂ concentrations and the temperature effects of non-homogeneous CO₂ in East Asia.

Previous studies have employed global transport models to assess spatiotemporal variability of CO₂ concentrations (Feng et al., 2011; Ballav et al., 2016). Fewer studies have considered the effect of CO₂ spatial variations on regional temperature (Yang et al., 2012). Regional models with high spatial resolutions can be powerful in simulating the processes that influence atmospheric CO₂ concentrations, such as transport, emissions, and consumption. The primary purpose of this study is to investigate the spatiotemporal characteristics of atmospheric CO₂ by using the regional climate model RegCM4 and to assess the impacts of the dynamic non-homogeneous CO₂ concentrations on temperature in East Asia. Sources and sinks of CO₂ are treated as prescribed surface fluxes in the modified RegCM4 model, while CO₂ is transported as a tracer in this model. The model can retrieve signals from the anthropogenic emissions and the biosphere–atmosphere exchanges, and thus capture the spatiotemporal variations of ambient CO₂ concentrations. To identify and quantify the effect of non-homogeneous CO₂ concentrations on regional temperature, the radiative transfer scheme in RegCM4 is revised to consider the spatial and temporal distributions of atmospheric CO₂, by substituting the prescribed constant CO₂ concentrations with the simulated non-homogeneous CO₂ concentrations. The model and input data are described in detail in Section 2, followed by a presentation and discussion of the results in Section 3. Summary and conclusions are provided in Section 4.

The regional climate model RegCM4 is an evolution of its previous version, RegCM3, developed by the International Center for Theoretical Physics (ICTP) in Italy (Giorgi et al., 2012). The model dynamics is the same as that of the hydrostatic version of MM5 (Grell et al., 1994), which has been used for a wide range of applications in the world (Giorgi et al., 2006). RegCM4 includes a mixed convection and tropical band configuration, and modifications to the pre-existing boundary layer schemes. The radiative transfer calculations in RegCM4 are carried out with the radiative transfer scheme of the global model CCM3 (Kiehl et al., 1996), which includes contributions from all the main greenhouse gases, i.e. H₂O, CO₂, O₃, CH₄, N₂O, and CFCs. The land surface module of RegCM4 is optional, with the Biosphere–Atmosphere Transfer Scheme version 1e (BATS1e) (Dickinson et al., 1993) and a new option to use the Community Land Model, version CLM3.5 (Oleson et al., 2008).

In this study, the Holtslag PBL scheme (Holtslag et al., 1990), the Grell cumulus convection scheme (Grell, 1993), and the CLM3.5 land surface module are employed. In order to reproduce the real-time atmospheric CO₂ distribution, we have revised RegCM4 to include CO₂ simulations. CO₂ is treated as an inert chemical species, whose concentrations are modulated by atmospheric transport, and source and sink processes in the model. Atmospheric transport includes horizontal/vertical advection and diffusion. Sources and sinks of CO₂ are prescribed as surface fluxes that include fossil fuel emissions, biomass burning, air–sea CO₂ exchanges, and terrestrial biosphere CO₂ fluxes.

Advection and diffusion are parameterized by using Giorgi’s method (Giorgi et al., 2006). Anthropogenic CO₂ emissions are from the Emission Database for Global Atmospheric Research version 4.2 (EDGAR4.2). This dataset provides greenhouse gas emissions per country and on a 0.1° × 0.1° grid for all anthropogenic sources identified by the IPCC for the 1970–2012 period (Olivier, 2015). The CO₂ emissions have been aggregated into five main source sectors, including fossil-fuel combustion, fugitive emissions from fuels, cement production and other carbonate uses, feedstock and other non-energy uses of fossil fuels, waste incineration and fuel fires. Biomass-burning emissions from forest wildfires, biosphere–atmosphere exchange, and ocean flux of CO₂ are adopted from the NOAA CarbonTracker CT2011 (data available at http://carbontracker.noaa.gov). CarbonTracker uses atmospheric CO₂ observations from a global observational network and simulates atmospheric transport to provide optimized biospheric and oceanic CO₂ fluxes (Peters et al., 2007). This dataset is available on the global 3° (longitude) × 2° (latitude) grids at a 3-h time resolution.

To reveal the temporal and spatial variations of CO₂ in East Asia, revised RegCM4 with prescribed surface CO₂ fluxes is run from 1 December 2008 to 31 December 2009 with the first month as a spin-up period. Only the results in 2009 are used for analyses. The study domain is 7260 km × 4620 km, with a spatial resolution of 60 km × 60 km on a Lambert conformal map projection centered at 34°N, 117°E (Fig. 1). This domain, covering most of East Asia, has large variations in topography and land cover types. In the vertical, 18 levels are set, with a top at 50 hPa. The lateral meteorological boundary conditions are updated every 6 h using the ERA-Interim reanalysis data, which can provide an improved representation of the hydrological cycle compared to previous products, especially over the tropical regions (Uppala et al., 2008). The Optimum Interpolation Sea Surface Temperature (OISST) of the NOAA Climate Analysis Center (CAC) Weekly Optimal Interpolation dataset in the original NetCDF format is used as the SST boundary forcing. CO₂ concentrations obtained from CT2011 optimized estimation (CT2011_oi; 3° longitude × 2° latitude) are interpolated into the model grids as the initial and boundary conditions.

Two numerical experiments are carried out to further assess the effects of non-homogeneous CO₂ concentrations on longwave radiation flux and temperature. One (named EXP_CTRL) adopts the primitive radiation scheme of the RegCM4 model, which uses constant CO₂ concentrations, i.e., 385.9 ppmv for the experiment period 2009. The other experiment (named EXP_REL) is conducted by using simulated CO₂ mixing ratios, which are closer to the real-time atmospheric CO₂ distributions. The column averaged CO₂ concentrations of EXP_REL is 387.6 ppmv, which is slightly higher than that in EXP_CTRL (385.9 ppmv), although the difference between them can bring about some uncertainties in our assessment. Nevertheless, the difference is much lower than the CO₂ spatial inhomogeneity, which varies from 379.1 to 411.8 ppmv in EXP_REL. Therefore, the difference in the simulations between the two experiments is mainly caused by non-homogeneous CO₂ concentrations. Comparing these two experiments, we can assess the impact of non-homogeneous CO₂ on longwave radiation flux and regional temperature.

Figure 1 Model domain with topography (shaded, m) and locations of the observation stations (red dots). See Table 1 for station full names.

Figure options

The observed CO₂ concentration data is taken from GLOBALVIEW-CO₂ (2013), which is provided by the Cooperative Atmospheric Data Integration Project. This dataset provides pseudo-weekly CO₂ for each site by fitting the CO₂ flask or in-situ observations with a smooth curve (Masarie and Tans, 1995). Besides, the project is coordinated and maintained by the Carbon Cycle Greenhouse Gases Group of the NOAA Earth System Research Laboratory (ESRL), which is a cooperative effort among many organizations and institutions making high-quality atmospheric CO₂ measurements available.

The locations of seven observation stations in East Asia are given in Fig. 1. Among the seven stations, Guam (GMI) station is located in the western Pacific island, surrounded by ocean, far from anthropogenic sources and terrestrial ecosystems. The Mt. Waliguan (WLG) station is the global background station of WMO/GAW (Global Atmosphere Watch) in the central Eurasian continent (Liu et al., 2009). Lulin (LLN) station and Tae-ahn Peninsula (TAP) station are located in populous areas, but being isolated and far from megacities. Detailed geographical information of these stations is listed in Table 1.

Figure 2 compares the monthly mean CO₂ mixing ratios at the seven stations between the model simulations and measurements, with the detailed statistical information given in Table 2. The simulated CO₂ concentrations are generally in good agreement with the observations, particularly at UUM and GMI. CO₂ mixing ratios show strong seasonality at the inland stations (UUM, WLG, KZM, and KZD), typically with a winter or spring maximum and a summer minimum. This seasonality is mainly driven by the CO₂ flux variation between the atmosphere and terrestrial vegetation and the seasonal patterns of atmospheric mixing and transport. The annual mean CO₂ mixing ratios at TAP and KZD stations are higher than the other stations (Table 2). It is predominantly related to the strong influence of local emissions. KZM, which is geographically adjacent to KZD, shows lower CO₂ all year round due to a higher terrain elevation (Table 2). Apparently, the model performs less well in urban and coastal areas than inland areas. For example, the model generally overestimates CO₂ at LLN station where it is in urban areas of the Taiwan Island and sometimes underestimates CO₂ at TAP station located in a coastal area of Korean Peninsula. Because the input anthropogenic emissions are adopted from yearly mean EDGAR4.2 dataset, it is difficult for model to capture the local emission fluctuations in urban regions. The effects of complex topography and microscale circulation also make it hard for the model to well reproduce the seasonality of CO₂ concentrations (Pillai et al., 2011).

Figure 2 Comparison of simulated (blue) and observed (red) monthly mean CO2 mixing ratios (in ppmv) at the seven stations.

Figure options

For these seven stations, the biases between the simulated and observed monthly mean CO₂ mixing ratios range from –0.53% at KZD to 1.20% at LLN (Table 2). The difference between the simulated and observed annual amplitude of monthly mean CO₂ concentrations is less than 2.5 ppmv, indicating the seasonal variation of atmospheric CO₂ levels is well captured by RegCM4. The remote ocean station (GMI) shows the lowest seasonal variation, with the annual amplitude of 6.78 ppmv and standard deviation of 2.46 ppmv in the observations, respectively. At KZD and UUM, the standard deviations of the monthly mean CO₂ mixing ratios are higher than at the other stations, indicating remarkable impacts of terrestrial biosphere. The difference between simulated and observed standard deviations of monthly mean CO₂ mixing ratios is less than 0.4 ppmv, suggesting that RegCM4 can reasonably reproduce the monthly variations of CO₂ mixing ratio. At the mountain station (LLN), the simulated annual mean CO₂ mixing ratio (387.12 ppmv) is much higher than the observation (382.51 ppmv), while the simulated annual amplitude is much weaker. The discrepancy is probably due to the inability of the model to resolve the complex local topography and small-scale system effects.

Figure 3 depicts the simulated spatial variations of atmospheric CO₂ mixing ratios in four seasons. At the surface layer, atmospheric CO₂ varies with region. The CO₂ mixing ratio in urban areas is evidently higher than that in remote regions with a maximum discrepancy of approximately 30 ppmv. Discernible gradients of surface CO₂ concentrations can be found between southwestern and eastern China, especially in winter and spring. Surface CO₂ mixing ratios can reach 410 ppmv in parts of East China and Korea, especially in urban areas.

Figure 3 Spatial distributions of atmospheric CO2 mixing ratio (ppmv) at surface, 800 hPa, and 600 hPa in four seasons: (a) winter, (b) spring, (c) summer, and (d) autumn.

Figure options

It is also seen from Fig. 3 that the simulated surface CO₂ mixing ratios in East Asia exhibit notable seasonal variations, which agree with numerous ground-based in situ measurements (e.g., Dlugokencky and Tans, 2017). CO₂ concentrations are generally higher in winter, particularly near the surface, similar to the observations (Dlugokencky and Tans, 2017). Moreover, the seasonality of surface CO₂ concentrations varies with region, with weaker seasonal variation over oceans than over continental areas. In winter, high CO₂ mixing ratios mainly appear in the Sichuan basin, North China, Japan, and Korea, where considerable anthropogenic CO₂ is released from intensive human activities. Due to the strong influence of the biosphere, CO₂ concentrations are much lower in summer, with the maximum of 400 ppmv in the North China Plain.

Due to the influences of human activities and the ecosystem, the CO₂ concentrations in the lower troposphere (the surface to 800 hPa) are significantly affected by the underlying surface. Anthropogenic emissions make more contributions to CO₂ concentrations in urban areas than in remote regions, where CO₂ concentrations are primarily influenced by the terrestrial ecosystem (Kou et al., 2013). Atmospheric CO₂ mixing ratios in the Sichuan basin are higher than neighboring areas, which is related to the inverse temperature structure (Ge et al., 2011), and local emissions.

Different from in the lower troposphere, the CO₂ concentrations in the middle troposphere (600 hPa) tend to have smaller longitudinal variations with weaker land–sea CO₂ contrasts, resulting from lesser influence from the underlying surface (Fig. 3). The result suggests that air masses from the middle to upper troposphere are well mixed, with similar spatial distribution patterns and small CO₂ horizontal variations.

By conducting two numerical experiments, the impacts of non-homogeneous CO₂ concentrations on longwave radiation flux are assessed. Figure 4 demonstrates the difference in annual mean net upward longwave radiation under clear sky and the total sky conditions between EXP_CTRL and EXP_REL at top-of-atmosphere (TOA) and the surface. Generally, the regionally averaged difference is negative for all the four conditions, ranging from –0.31 to –0.03W m^–2. There is more longwave radiation that has reached the earth surface and atmosphere after using the simulated spatially explicit CO₂ concentrations in the radiative transfer scheme. Comparing Figs. 4a, b with Figs. 4c, d, we find that the difference of upward longwave flux at the earth surface is evidently larger than that at TOA, because the CO₂ spatial variation is larger in the lower troposphere than in the upper troposphere. The impact of non-homogeneous CO₂ concentrations on radiation flux for the total sky condition is more notable than for the clear sky condition. The altered CO₂ spatial distribution affects the energy balance of the atmosphere, changing the amount of water vapor content and clouds, and resulting in a feedback on the radiation flux.

The differences between EXP_CTRL and EXP_REL vary spatially. From Fig. 4a, it is seen that the change of clear sky TOA upward longwave flux is positive in most areas of East Asia except for eastern Sichuan and Hebei Provinces, and Japan as well. This difference is more evident under the total sky condition (Fig. 4b). The spatial distributions with positive or negative differences in Fig. 4b are similar to those in Fig. 4a, but with larger absolute values, ranging from 1 to 2.5 W m^–2. This indicates the indirect impact of cloud variation induced by non-homogeneous CO₂ concentrations on radiation flux. In Fig. 4c, the difference of surface net upward longwave radiation flux between the two experiments is relatively homogeneous over ocean areas under the clear sky condition, similar to the distribution of atmospheric CO₂ concentrations (Fig. 3). The difference in surface net upward longwave radiation flux under the total sky condition ranges from 3 to 5 W m^–2, the largest among all four conditions, which is related to the sharp CO₂ horizontal gradient near the surface (Fig. 3). Evidently, non-homogeneous distribution of CO₂ concentrations near the surface and cloud forcing are crucial factors affecting the longwave radiation flux at TOA and the surface.

Figure 4 Differences of the annual mean upward longwave flux (W m–2) under (a, c) clear sky and (b, d) total sky conditions between EXP_CTRL and EXP_REL at (a, b) TOA and (c, d) the surface.

Figure options

To further understand the impact of non-homogeneous CO₂ on longwave radiation flux, the zonal averaged annual mean energy balance is explored. Figure 5 shows the net upward longwave radiation under the total sky and clear sky conditions at TOA and the surface, and longwave cloud forcing at TOA and the surface. The difference between the clear-sky and total-sky net upward longwave radiation is defined as cloud forcing. The blue curves represent the results of EXP_REL, and the green curves show the difference between the two experiments. Figure 5 shows that the TOA net upward longwave radiation reduces with latitude, mainly due to the colder surface and atmosphere in the north of the study domain than in the south of it. Due to the accommodation of non-homogeneous CO₂ concentrations, the TOA net upward longwave radiation under the total sky condition decreases in southern East Asia over 15°–45°N, but increases in northern East Asia over 45°–50°N (Fig. 5a). The differences between the two experiments range from –1.25 to 0.5 W m^–2, with the maximum difference appearing near 37°N. Different from the total sky condition, the discrepancy of the TOA net upward longwave radiation under the clear sky condition between the two experiments is quite small, ranging from –0.25 to 0.25 W m^–2 in East Asia (Fig. 5b). As seen from Fig. 5c, TOA longwave cloud forcing increases with latitude and reaches the maximum near 32°N, which is rather unlike the total- and clear-sky TOA net upward longwave radiation. Opposite to Fig. 5a, the discrepancies of TOA longwave cloud forcing between the two experiments are positive in the regions of 15°–46°N, but negative in the regions of 46°–50°N. The maximum (1.1 W m^–2) appears near 37°N while the minimum (–0.4 W m^–2) is found at 48°N. As mentioned, the spatial variation of atmospheric CO₂ has slight impact on the clear-sky TOA upward longwave radiation. However, a non-homogeneous CO₂ distribution notably affects the spatial and temporal distributions of cloud cover, contributing to the alteration of the total-sky TOA outgoing longwave radiation.

Figure 5 The zonal averaged annual mean net upward longwave radiation under the total sky condition at (a) TOA and (d) the surface, net upward longwave radiation under the clear sky condition at (b) TOA and (e) the surface, and longwave cloud forcing at (c) TOA and (f) the surface for EXP_REL (blue, scales on the left) and for the differences between EXP_CTRL and EXP_REL (green, scales on the right). Units are W m–2.

Figure options

In contrast to the situations at TOA, the energy balance at the earth surface is more complicated. The zonal averaged surface net upward longwave radiation varies with latitude, and the relatively high longwave radiation flux appears over 30°–40°N. This is possibly related to the integrated effects of the large water vapor gradients, cloud variation, and complicated topography in these regions. Comparing the two experiments finds that the difference of the total sky surface net upward longwave radiation in East Asia is negative over 21°–40°N, where atmospheric CO₂ concentrations are much higher than in other areas due to intensive human activities (Fig. 3), with a minimum of –1.1 W m^–2 (Fig. 5d). The maximum difference between adjacent latitudes even reaches 0.5 W m^–2, probably due to the substantial CO₂ spatial variations. In Fig. 5e, the discrepancy between the two experiments under the clear sky condition ranges from –0.5 to 0.5 W m^–2, smaller than that under the total sky condition, but similar to that under the total sky condition in the spatial pattern of positive and negative areas. By considering the non-homogeneous CO₂ concentrations, surface longwave cloud forcing is enhanced across 22°–43°N, which is in agreement with net upward longwave radiation flux. Non-homogeneous CO₂ concentrations affect the longwave radiation balance at the surface, resulting in a decline in net upward longwave radiation flux in the subtropics. Furthermore, cloud forcing derived by CO₂ variation strengthens this trend.

Figure 6 indicates the absolute and the relative (defined as the ratio of heating rate differences to the heating rate from EXP_CTRL) differences of annual mean zonal longwave radiation heating rate between the two experiments at multiple altitudes. Broadly speaking, the application of non-homogeneous CO₂ concentrations in the radiative scheme exhibits an evident effect on the heating rate of longwave radiation at different altitudes, and the range of heating rate difference is from –0.13 to 0.13 K day^–1.

In Fig. 6a, positive heating rate differences mainly appear at the altitudes of 970–830 hPa, and over the regions between 32°–41°N (about 0.08 to 0.13 K day^–1) and 44°–48°N (about 0.07 to 0.09 K day^–1). It is probably due to high atmospheric CO₂ levels in these regions (Fig. 3), causing more absorption of longwave radiation. The lower troposphere over tropical areas (15°–21°N) tends to be colder with the application of non-homogeneous CO₂ concentrations in the radiative scheme, with a decline of –0.08 to –0.05 K day^–1 in the heating rate. It is consistent with the increase in the net surface upward longwave radiation in these areas (Fig. 5). The relative differences vary between –5% and 5% for most altitudes (Fig. 6b). In the surface layer, longwave radiation heating rate changed by non-homogeneous CO₂ concentrations is comparatively larger than at other altitudes. It is supposed to be a result of substantial variations in CO₂ concentrations near the surface (Fig. 3), which is induced by the combined effects of local emissions, terrestrial biosphere, and mesoscale transport.

Figure 6 (a) Difference (K day–1) and (b) relative difference (%) in the annual mean zonal heating rate between EXP_CTRL and EXP_REL.

Figure options

Previous studies have found that longwave radiation heating is a primary mechanism. The thermodynamic structure is affected by the heating rate parameterization in the radiative scheme, which, in turn, can change the atmospheric temperature and even the dynamic structure of the atmosphere in climate models (Iacono et al., 2000). The change of longwave radiation heating rate contributes to increases in surface and lower tropospheric annual mean zonal temperature (Fig. 7). The maximum increase (approximately 0.18 K) is found near the surface in the temperate zone, i.e., between the altitude of 1000–900 hPa and over the latitudes of 42°–50°N. Temperature is enhanced even from the surface to 800 hPa over the tropical zone (15°–24°N), with a range of 0.05–0.1 K. Due to atmospheric CO₂ heterogeneity, the difference of temperature between the two experiments varies spatially. The upper troposphere tends to be cold due to less longwave radiation. The cold center appears between the altitudes of 300–150 hPa and over 32°–45°N, with a maximum of –0.05 K. It is evident that non-homogeneous distribution of CO₂ concentrations affects the atmospheric temperature, and this impact varies with altitude and latitude. Considering the influence on longwave radiation, the application of simulated CO₂ concentrations in the radiative scheme enhances downward longwave radiation, especially in the lower troposphere, and then warms the earth surface and the lower troposphere, resulting in an increase in temperature near the surface.

Figure 7 Difference of annual mean zonal average temperature (K) between EXP_CTRL and EXP_REL.

Figure options

Figure 8 indicates the surface temperature difference between the two experiments for different periods. In general, surface temperature differences vary by region, especially in summer. Surface temperature increases for most parts of mainland in East Asia from a one year simulation, while the maximum increment appears in central South Korea (approximately 1.3 K) (Fig. 8a). However, for eastern Sichuan and Hebei provinces, the earth surface tends to be colder with a maximum of about –0.15 K. Seasonally, the surface temperature difference for summer (averaged for July) and winter (averaged for January) is given in Figs. 8b, c, respectively. The surface tends to be warmer in summer in northeastern China, the Yangtze River Delta, and Korean Peninsula. The maximum increase even reaches 3.5 K in Liaoning Province. In central Inner Mongolia, northern China, and Sichuan Province, surface temperature decreases by 1–4 K. However, regionally, the wintertime surface temperature difference is smaller. Most regions in East Asia exhibit a warming trend, but the temperature increment is usually less than 0.5 K except for central South Korea and the Pearl River Delta of China. For areas such as central Japan and Shandong Province, a cooling trend is shown with a range of –0.6 to –0.2 K. The maximum absolute values of the annual, summer, and winter surface temperature differences are 1.27, 4.18, and 1.10 K, respectively. Overall, the difference in summer is obviously larger than that in winter, and varies more regionally. Summer is a season when plants grow vigorously. Strong photosynthetic absorptions of atmospheric CO₂ combined with enhanced transpiration induce large decrease in temperature in vegetated areas. While in urban areas, high CO₂ concentrations and more intense summer radiation result in a warmer surface than in other seasons. The annual, summer, and winter temperature differences are all positive in the entire simulation domain, indicating that the application of non-homogeneous CO₂ concentrations in the model generally warms the earth surface.

Figure 8 Surface temperature difference (K) between EXP_CTRL and EXP_REL for (a) the annual mean, (b) the July mean, and (c) the January mean.

Figure options

In this study, regional climate model RegCM4 is revised to simulate the spatial and temporal distributions of atmospheric CO₂ in East Asia. The comparison with the ground observations suggests that the modified RegCM4 can reasonably reproduce the spatial inhomogeneity and seasonal variation of CO₂ concentrations, despite of uncertainties resulting from inconsideration of the dynamic influence of anthropogenic emissions in the model. Surface CO₂ mixing ratios are relatively higher in winter and spring. It is probably related to the source/sink transition during different phases of plant growth and the seasonal patterns of meteorological conditions. The difference in the surface CO₂ between urban and remote regions is ascribed to anthropogenic CO₂ emissions and biogenic CO₂ flux.

Furthermore, in order to assess the effect of non-homogeneous CO₂ concentrations on regional longwave flux and temperature, two simulation experiments are conducted, one with a prescribed constant CO₂ concentrations and the other with simulated CO₂ concentrations that vary spatially in the radiative transfer scheme in the model. Comparing the two experiments finds that the surface energy balance is altered when considering the spatial distribution of atmospheric CO₂ in the radiative scheme. The implementation of non-homogeneous CO₂ results in a maximum of –1.1 W m^–2 decrease in net surface upward longwave flux in subtropical regions (21°–40°N). Accumulated atmospheric CO₂ in the lower troposphere engendered by local emissions traps more longwave radiation, and consequentially enhances the downward longwave flux at the surface. With the altered earth energy balance, the lower tropospheric atmosphere tends to be warmer, with a maximum temperature increment of 0.18 K.

Our results suggest that the surface temperature simulated by current regional climate models may be biased, especially in summer, because of the hypothesis of well mixed CO₂ concentrations in the radiative scheme. The discrepancies in regional longwave flux and temperature due to non-homogeneous CO₂ concentrations found in this study indicate that the spatial and temporal distributions of CO₂ concentrations should be taken into account in future climate modeling.

Acknowledgments. The authors thank Jingxian Liu for help with the language check. We also thank the anonymous reviewers for their constructive and valuable comments on this paper.