Improved simulation of group averaged CO 2 surface concentrations using GEOS-Chem and fluxes from VEGAS

Introduction Conclusions References


Introduction
An improved understanding of the carbon sources and sinks at a global scale is essential to predict the future rate of atmospheric CO 2 increases and to plan an international CO 2 management strategy (Tans et al., 1990).But these fluxes remain quantitatively Figures uncertain.The full range of results in past studies spans budgets with northern terrestrial uptake of 0.5 to 4 Pg C yr −1 , and tropical terrestrial emissions of −1 to 4 Pg C yr −1 (Stephens et al., 2007;Peylin et al., 2002;Gurney, 2004).Some studies show increasing sinks in tropical forest plots (Baker et al., 2004).Piao's results show that rising temperatures may already decrease the efficiency of terrestrial carbon uptake in the Northern Hemisphere (Piao et al., 2008), while larger net sinks were found over northern and southern continents (Feng et al., 2011).
It is not possible to measure all CO 2 sources and sinks in the every part of the globe.Fortunately any geographical distribution of CO 2 sources and sinks is reflected in the spatial and temporal variations of CO 2 concentration patterns in the atmosphere.
The CO 2 in atmosphere is an unbribable witness of the surface flux.In early studies concentration differences between monitoring sites have been used as a constraint to infer net fluxes (Enting et al., 1995;Kaminski et al., 1998).The mean annual meridianal/longitudinal gradient observation is compared with model values (Bousquet et al., 1999;Kaminski et al., 1998).Latitudinal distribution of the sources and sinks of CO 2 from the concentration gradient has been discussed.The samples are grouped into latitude bands to derive the sources and sinks (Tans et al., 1989(Tans et al., , 1990)).Some inverse technique researches adjust the CO 2 surface flux via minimizing the distance between the modeled/optimized values and the observational data at each station (Enting, 2002;Peylin et al., 2002;Bousquet, 2000;Baker et al., 2006;Gurney et al., 2002;Rodenbeck et al., 2006).
However one persistent problem in using model-observation comparisons for this goal relates to the issue of compatibility.Observations at a single site reflect all underlying processes of all scales that usually cannot be fully resolved by model simulations at the grid points nearest the site due to lack of spatial or temporal resolution or missing processes in models.In this article we proposed a new technique to evaluate the regional surface fluxes by checking group averaged differences between model simulation and observations, rather than the difference at every single observational site.To compare and validate the climatological and yearly-varying CO 2 fluxes using the Figures

Back Close
Full modeled CO 2 results of GEOS-Chem model, several sites in one region are grouped according to the regional characteristics of temporal distribution of seasonal cycle derived from a new atmospheric CO 2 observation dataset from GLOBALVIEW-CO 2 2010.The difference of observation and modeled results of the average of stations in one group can reflect the uncertainties of the flux in the region where the grouped stations are.
Where and when the CO 2 emissions from fossil fuels are absorbed by land ecosystems and oceans is a major issue for global carbon cycle.It is significant for using highly accurate inventories in the optimization of CO 2 fluxes with inverse method.There are three parts which needed to be taken into account, emissions from fossil fuel, net ecosystem exchange of the terrestrial biosphere and ocean-atmosphere carbon exchange.Optimized estimates of surface source and sink are produced by some different bottom-up and top-down ways.GEOS-Chem atmospheric transport model has been widely used in the assimilation of CO 2 and inverse of CO 2 flux.It was used to evaluate the influence of reduced carbon emissions on the distribution of atmospheric CO 2 and described in early studies (Suntharalingam, 2004(Suntharalingam, , 2005)).Nassar made modifications to the GEOS-Chem CO 2 simulation including improved temporal variability in the national surface fossil fuel inventory, the addition of surface CO 2 shipping, 3-D domestic and international aviation CO 2 emissions, and 3-D chemical production of CO 2 (Nassar et al., 2010).The balanced biosphere flux is based on a 3-h NEP for 2000 from CASA model (Olsen, 2004), which has always been used as the prior flux in the GEOS-Chem.It is available for the simulation of global CO 2 concentration using new net land-atmosphere carbon exchange with the development of Dynamic Global Vegetation Model (DGVM).The spatial trends in Net Biome Production (NBP) have been produced by different DGVMs (http://www-lscedods.cea.fr/invsat/RECCAP/V2/Data%20Policy%20Trendy.pdf).
All DGVMs are consistent with the global land carbon budget (Sitch et al., 2008).VEGAS model is developed to simulate the net primary productivity and described by Zeng (Zeng et al., 2005;Zeng, 2003).In this paper the land-atmosphere fluxes from Introduction

Conclusions References
Tables Figures

Back Close
Full VEGAS is introduced into GEOS-Chem model to replace all the current inventories except fossil and ocean flux.The outline of this paper is as follows: Sect. 2 introduces the data.Section 3 of this paper group observation stations in one region and demonstrate the temporal and spatial variability in CO 2 .Section 4 presents the group averaged differences between the observation and the model results with fluxes from CASA (VEGAS).We present conclusions in Sect. 5.

GLOBALVIEW CO 2 data
GLOBALVIEW-CO 2 (GLOBALVIEW-CO2, 2010) is an update product of the Cooperative Atmospheric Data Integration Project.While the project is coordinated and maintained by the Carbon Cycle Greenhouse Gases Group of the National Oceanic and Atmospheric Administration, Earth System Research Laboratory (NOAA ESRL), gaps in the data are filled by extrapolation from marine boundary layer measurements.Flask samples of whole air enable highly accurate and precise measurements of atmospheric CO 2 concentrations (Conway et al., 1994).It can be downloaded from http://www.esrl.noaa.gov/gmd/ccgg/globalview/co2/co2download.html.This update includes extended records derived from observation made by 21 laboratories from 14 countries.The data product includes extended records for the period 1 January 1979 to 1 January 2010.We choose one lowest record if there are several records at different altitude for the same site, and 108 records are kept.For one record, the negative value denotes the smoothed value of CO 2 concentration in this month is less than the benchmark trend values.Figures

Back Close
Full

VEGAS data
Net ecosystem exchange (NEE, which = net primary productivity (NPP) -heterotrophic respiration (RH)) is simulated by DGVMs.Simulated land-atmosphere fluxes are between −1.52 and −2.75 Pg C yr −1 for the 1990s, DGVMs simulate a greater land carbon uptake, which is in agreement with IPCC estimates (Sitch et al., 2008).The land fluxes are defined as the sum of photosynthesis, ecosystem respiration and biomass burning.The terrestrial carbon model VEGAS is described in Zeng (2003).It was run at 2.5 • × 2.5 • resolution and forced by the observed precipitation and temperature.The monthly NBP flux as net land-atmosphere carbon exchange is regridded offline to the GEOS grids (2 • × 2.5 • ) in this work, which is equal to the magnitude of NEE.
A diurnally varying NEP flux is constructed from gross primary production (GPP) and ecosystem respiration (Re) in the CASA model (Olsen, 2004).The CASA NEP output is used as NEE in GEOS-Chem.The comparisons of monthly land-atmosphere fluxes from VEGAS and CASA are concluded as Fig. 1.The difference of spatial distribution is seen as Fig. 2 (January) and Fig. 3 (July).It is evident that the largest sinks in July of VEGAS is smaller than that of CASA about 500 Tg, which is distributed in the region of Asia, South America boreal, and South American tropical.The sources of VEGAS from January to April is lower than that of CASA, which is distributed in the region of South American tropical and Northern Africa.
The original emission inventories (ORI) in GEOS-Chem is shown in 3 Grouped CO 2 observations

Determining groups of observational sites
To represent the regional characteristics of CO 2 , several stations are grouped according to the regional characteristics of temporal distribution of seasonal cycle.The stations are grouped based on two factors: atmospheric mixing regimes and surface characteristics (land or ocean).Atmospheric mixing regimes can be reflected by seasonal pattern and amplitude of the seasonal trends of each station.
All groups in land show similar patterns.The minimum value of CO 2 appears in summer and fall, and the maximum value appears in spring and winter.The difference between minimum and maximum values is much more than 6 ppm for most stations in land.The Earth was divided into 11 land and 11 ocean regions in TransCom3 project (Gurney et al., 2002).The 11 TransCom land regions are used except the boundary of two land regions.The latitude is defined as the division for most two adjacent land regions in this work.The stations in each land region are grouped.
The differences of seasonal pattern and amplitude of stations in Ocean regions are apparent.We group the stations in ocean regions as follows: The stations that have similar phase and magnitude range are grouped.We also require that all the sites in one group should be next to each other in order to approximately reflect the same type of sink/source.Then the ocean is divided into 15 regions and the stations in Ocean regions are grouped to 15 groups.All 108 stations (see Table A1) are classified into 26 groups and 72 sites in 15 ocean regions.The map of all grouped sites is shown in Fig. 4.

Seasonal patterns of stations in land regions
All groups in the land show a similar pattern.The seasonal patterns of stations (more than one station) in 5 regions are shown in Fig. 5.We can know that the minimum value for land groups appears in summer and fall, and the maximum value appears Introduction

Conclusions References
Tables Figures

Back Close
Full in spring and winter from Fig. 5. Seasonal cycles of atmospheric CO 2 are caused primarily by the terrestrial biosphere moving from being a net source of carbon to the atmosphere (mainly in winter) to becoming a net sink (mainly in summer), where net carbon uptake or release is determined by the balance between photosynthesis and respiration.Seasonal cycle of atmospheric CO 2 in the Northern Hemisphere (NH) is in phase with the ecosystems (Randerson et al., 1997).The difference of the magnitude for the amplitude, minimum values, maximum values of all groups in the NH can be an important constraint for further improving our understanding of the surface fluxes in the NH.
The stations in the Pacific Ocean North of 5 • S are classified into 5 different groups (O1, O2, O3, O6 and O7), and the stations in the Atlantic Ocean are classified into 2 groups (O8, O9).Though the seasonal trends of ocean regions north of 5 • S are similar as that of the land groups in Northern Hemisphere, there are different amplitudes as shown in Fig. 6.The amplitude of groups O1, O6 is larger than 10 ppm, and the amplitude of O2, O7 is much less than that of northern regions, while the amplitude of group O3 is much less than 6 ppm.The amplitude of group O9 is less than that of group O8.
The amplitude typically decreases moving southward, since the Southern Hemisphere has less mid-latitude vegetation to seasonally absorb and release CO 2 (Randerson et al., 1997).We require that all the sites in one group should be next to each other in order to approximately reflect the same type of sink/source.To separate these regions with different amplitudes of concentration of CO 2 , it is helpful to distinguish when and where the sources and sinks are.
The South Pacific region between 5 • S and 35 • S is divided into two subregions (O4 and O5) according to the different seasonal trend of stations in these regions.Though Introduction

Conclusions References
Tables Figures

Back Close
Full the amplitude is smaller than 1.4 ppm, the seasonal patterns of the groups are clear in these regions.The sites in South Pacific region between 5 • S and 15 • S (group O4) show a special trend.The minimum appears in September or October.There are two obvious increase phases from April to June and from October to December, which indicate that there may be sources in these months.The sites of group O5 locate in South Pacific region between 15 • S and 35 • S shows another special trend, the maximum value of group O5 appears in January, the minimal value appears in May, which indicates there may be sinks from January to May and from June to September in this region.
The seasonal patterns are more complicated in the Indian Ocean South of 35 • S.
They are classified into two groups with different seasonal patterns (O12 and O13).
The average pattern is totally different from other oceans.The North Indian Ocean O12 shows a consistent decrease from February to November.The stations in South Indian Ocean O13 show chaos in the first half year and show a consistent increase in the second half year.The South Atlantic is divided into 2 regions (O10, O11) without average pattern, one station with quite different seasonal pattern in each region.The concentrations of CO 2 of stations in ocean south of 5 • S are mainly influenced by the oceanic sources and sinks, and the amplitudes of seasonal cycles are not more than 2 ppm (O4, O5, O10, O11, O15), which are much smaller than that of the NH.General negative values in seasonal cycle denote sinks, positive values denote sources.It is evident that the seasonal variations are positive in NH winter (January) and negative in NH summer (August).The seasonal variations are positive in austral winter (August) and negative in austral summer (January) in the south of 35 • S, such as group O15 (Fig. 6).Seasonal signals observed in all subtropical regions of the NH and SH show that the CO 2 concentration decreases southward in summer and vice versa in winter (Metzl et al., 2006).Group O15 exhibits obvious increase from February to September and decrease from September to February in the next year.It is opposite in shape to that of NH; on the other hand, there are sources in austral winter and sinks in austral summer, which is similar as that of NH.To some extent, it is coherent for the whole Figures

Simulation results and comparison with observations
We use the GEOS-Chem model (http://acmg.seas.harvard.edu/geos)to describe the relationship between 3-D atmospheric CO 2 concentrations and surface CO 2 fluxes.
We carried out the simulations with original emission inventories (ori) and new emission inventories (new).CO 2 was simulated at a horizontal resolution of 2

Comparisons of group averaged CO 2 concentrations for land regions
The CO 2 seasonal patterns and amplitudes simulated by model runs with two emission inventories were different.The largest discrepancy between model results and observations for runs with original emission inventories is 17.5 ppm, about 4.5 % of observation values.The discrepancy for runs with new emission inventories is below 8.4 ppm, about 2.2 %.The largest discrepancies for both runs appear in the region L11, which indicates there may be large uncertainties for CO 2 surface fluxes in Europe.Introduction

Conclusions References
Tables Figures

Back Close
Full The model results with new emission inventories have smaller discrepancy than the simulations with original emission inventories for North American boreal (L1) from January to September, which is smaller than 1 % of observations.It's higher than observations about 2-5 ppm from December to November, which suggests that more sinks in North America may be required for this period.
There are good agreements between model results with new emission inventories and observations for North American Temperate and Eurasian boreal and Eurasian Temperate region (L2, L7 and L8).The biggest discrepancy is 4.4 ppm, 1.1 % of observations, which appears in December and better than the results with original emission inventories of 7.2 ppm for region L2.The largest discrepancy is decreased from 15.0 ppm to 4.4 ppm for region L7 by using the new emission inventories.The discrepancy is below 3.8 ppm for runs with new emission inventories, while 8.9 ppm for runs with original emission inventories in region L8.
The modeling results can not be compared with observations because of scarcity of observations for South America, Africa, and Australia.The new emission inventories can be used as good prior fluxes in the forward model and be adjusted in future inverse model from the above comparisons of 5 land regions.

Comparisons of group averaged CO 2 concentrations for ocean regions
The trends of CO 2 concentration over ocean regions are also influenced by the change of emission inventories in land.The largest discrepancy in ocean regions between the model results with new inventories and observations is below 7.8 ppm, 2.1 % of observations, which is lower than the discrepancy of results from original inventories (14.3 ppm).There are great improvements in the region south of 15 • S (O5 and O11) for the model results with new inventories.It can be deduced that the sources and sinks are improved in the South American Temperate though there is no direct observations in this region.
The largest discrepancy for runs with new inventories appears in April for Indian Tropical (O12).It has a persisting decrease from April for the observation, while there is Introduction

Conclusions References
Tables Figures

Back Close
Full a minimum value in April for both model results.Both model results have discrepancy from January to April for South Indian Temperate (O13).It is very complex for the seasonal trend of the observation stations in South Indian Temperate.Unfortunately there is scarce of observations in the adjacent land regions.Some more observations are very necessary for these regions in future.
There is still large positive bias for North Atlantic Temperate (O8, O9) from July to September.More sinks may be required in this region or the surrounded land region.It's difficult to simulate the concentration of South Pacific Tropics (O4), which has more complex seasonal cycles compared to other regions (Fig. 6).It is obviously that the ocean emission inventories are needed to be adjusted to match the observations in the ocean region.In future work, the net land and ocean fluxes will be improved based on the discrepancy of the group averaged values between model results and observations.

Comparisons of the root-mean-square difference
To further demonstrate the difference of regional model results between original biospheric fluxes (includes land-atmosphere fluxes from CASA) and new biospheric fluxes from VEGAS, we compare the root-mean-square difference (RMSD) between modeled and observed concentration at the observation sites in one group along with the group averaged values (Fig. 9).The RMSD of group averaged value between model results with fluxes from VEGAS and observation is reduced by 0.24-0.63ppm for 5 land regions.
The concentrations of CO 2 in ocean regions are influenced by the change of emission inventories in land.It is clear that the RMSD of group averaged value between model results using fluxes from VEGAS and observation is less than the results using fluxes from CASA by 0.15-0.53ppm for North East Pacific, South Pacific and Southern Ocean (O1, O4, O5, O12, O13 and O15).There is little improvement for North Pacific and Northern Ocean (O2, O6, and O14).It is convenient to evaluate the regional model results according to the comparisons of group averaged values.Introduction

Conclusions References
Tables Figures

Back Close
Full

Conclusions
We grouped several sites in one region according to the phase and amplitude of seasonal cycles of observations.The group averaged measurement values of CO 2 concentration contain less small scale "noise" that models often cannot resolve and are used to evaluate the regional model results.The differences of group averaged values between observations and model results reflect the uncertainties of the flux in the region where the grouped stations are.
We compared group averaged values between model results with two landatmosphere flux from CASA (VEGAS) and observations.Results show that the modeling group averaged values of CO 2 concentrations with fluxes from VEGAS have improvements in most regions.There is still large uncertainty in Atlantic and North Atlantic, Indian Ocean, and South Pacific Tropics.This implies possible large uncertainties in the fluxes there.
The differences of group averaged values between observations and model results will be used to estimate the error of regional fluxes and optimize the regional fluxes with inversing methods in future work.

Summery of emission inventory
The original (new) CO 2 flux used in this study is 7.   (Feng et al., 2011;Nassar et al., 2011).
It is hard to evaluate the inverse results by the total value.It's very useful to compare the group averaged values using these inventories to evaluate inverse results in future work.Introduction

Conclusions References
Tables Figures

Back Close
Full  Full Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Fig. 1, which includes the balanced biosphere from CASA, biofuel burning, biomass burning and residual annual biospheric flux (Nassar et al., 2010) and is replaced with fluxes from VEGAS in this work.Obviously the sinks of VEGAS are smaller than the original results in GEOS-Chem especially from June to August.The summary of fluxes is in Appendix Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | global in winter or summer.The seasonal variations typically decrease from positive to negative values from NH to SH in January, and increases from negative to positive values in August.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 8 Pg C (7.8 Pg C, fossil emission inventories), −1.4 Pg C (−1.4 Pg C, net air-sea fluxes), −2.3 Pg C (−1.9 Pg C, net air-land fluxes) for 2006.The original(new) global annual net CO 2 emissions for 2006 is 4.1 Pg C (4.5 Pg C).The simulated seasonal cycles shows big discrepancies though there is a small difference between the total net fluxes from new emission inventories and original emission inventories in GEOS-Chem.There are also little differences between the total fluxes from other inversion results.JENA S99V3.2 data (3.78Pg C) is available Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Tans, P. P., Fung, I. Y., and Takahashi, T.: Observational constraints on the global atmospheric CO 2 budget, Science, 247, 1431-1438, 1990.Zeng, N.: Glacial-interglacial atmospheric CO 2 change -the glacial burial hypothesis, Adv.Atmos.Sci., 20, 677-693, 2003.Zeng, N., Mariotti, A., and Wetzel, P.: Terrestrial mechanisms of interannual CO 2 variability, Discussion Paper | Discussion Paper | Discussion Paper |

Table A1 .
Stations used in this paper.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |