Journal Archive

GOSAT (also known as Ibuki), is the first successful satellite designed specifically to measure the concentrations of atmospheric carbon dioxide and methane (the two greenhouse gases making the largest contribution to climate change) with good sensitivity near the surface. GOSAT has a 666 km sun-synchronous orbit and completes one orbit in ~100 minutes providing global coverage in approximately 3 days (Kadygrov et al., 2009). The GOSAT Thermal And Near-Infrared Sensor for carbon Observation (TANSO) consists of two units: the Fourier Transform Spectrometer (FTS) and the Cloud Aerosol Imager (CAI) (Kuze et al., 2006). The TANSO-FTS has three bands in the Short Wave InfraRed (SWIR) region (0.76, 1.6, and 2.0 μm) and a wide Thermal Infrared (TIR) band (5.5-14.3 μm) with a circular ~10.5 km instantaneous field of view at nadir (Yokota et al., 2009). The retrieval of greenhouse gases from FTS spectra excludes the cloudy pixels by screening using the images from CAI, which results in a significant reduction of data (Kadygrov et al., 2009). The retrieved concentration is obtained from maximum a posteriori (MAP) method with a priori information from a radiative transfer model (RTM) and the pre-processed measured spectra. The overall retrieval algorithm for GOSAT CO₂ and CH₄ products is described at Yoshida et al. (2011, 2010).

We used the GOSAT CO₂ level 2 (L2) products from column abundance retrieved from Short Wave Infra-Red (SWIR) radiance spectra (version 01.1×) based on retrievals led by the National Institute for Environmental Studies (NIES). The details of the products including the retrieval processes and observation results are at http://www.gosat.nies.go.jp/index_e.html.

We used GOSAT (FTS SWIR, L2) CO₂ products from versions 01.11-0.1.13. Fig. 2 shows the CO₂ column-averaged dry air mole fraction or concentration in each season. This quantity is commonly referred to as XCO₂ and is determined by taking the ratio of the retrieved CO₂ column density to the dry air column density which is retrieved using the O₂ A-band (0.76 μm).

Fig. 2. 
The seasonal variability in column-averaged CO₂ concentration (xCO₂) from GOSAT L2 products (April 2009 to January 2010, unit: ppmv).

Continental observations of XCO₂ are made in nadir mode and show a pattern of seasonal variation similar to that from the NOAA-ESRL-GMD (http://www.esrl.noaa.gov/gmd/ccgg/trends/#global), where the global marine surfacemean CO₂ level in January 2004 is 377 ppmv. The oceanic XCO₂ values are derived from GOSAT measurements using sun glint mode (Yoshida et al., 2011) and are mostly between 375-380 ppmv (Fig. 2). The hemispheric gradient of CO₂ concentration is observed, but is found to be less than the gradient of in-situ or ground measurements, which is expected for column-averaged observations. The relatively low CO₂ level below 375 ppmv over the terrestrial biosphere in the northern hemisphere summer (i.e., North America and Siberia) is evident in Fig. 2. Despite these reasonable distributions in GOSAT data, the limited spatial coverage due to observations excluded by cloud and aerosol filtering is one of the major limitations of the product.

We compared the total CO₂ column densities (unit: 10²¹ molecules/cm²) between GOSAT and GEOS-Chem data from April 2009 and January 2010, which is shown in Table 1. The comparison between total columns does not consider the averaging kernels of GOSAT data which limits our results from being a truly quantitative validation. We do not include May, November, and December in 2009 since there were serious data exclusions (<5% available) due to the cloud and aerosol interference. Table 1 shows the monthly variation of total CO₂ columns over 6 continental regions (East Asia, Europe, North America, Africa, South America, and Australia). In general, the total CO₂ column varies between 7.0-9.0×10²¹ molecules/cm² and both of the data clearly reproduce the seasonal variation of CO₂ (Table 1). GEOS-Chem columns typically have a positive bias of 0.6-1.5% (Table 1). The only exception of GEOS-Chem negative bias occurred over East Asia in April 2009 and January 2010 where the GOSAT CO₂ level is slightly higher than the model (Table 1), with the GOSAT columns showing the largest seasonal variability over this region (4.4×10²⁰ molecules/cm²) while GEOS-Chem has a much smaller seasonality (2.4×10²⁰ molecules/cm²). Considering the generally underestimated GOSAT data, these higher GOSAT CO₂ concentrations over East Asia in the cold seasons are exceptional. The possible reasons are (1) the anthropogenic CO₂ emission over China in the model is still underestimated during these months due to use of an annual mean fossil fuel and cement inventory (the strong NO_x seasonality by fossil fuel use over China is also well shown in satellite-based NO₂ data (Lamsal et al., 2010)); (2) Biomass burning emissions for 2009-2010 were not available in GFED2 so, regional interannual differences in those years could lead to lower CO₂ in the simulation; (3) anomalously low photosynthetic uptake in the region due potentially to climate perturbations like the onset of El Ni~no in spring 2009; (4) Model transport errors over Asia or neighboring regions; (5) Measurement interference by Asian dust over East Asia (Yokota et al., 2009).

The column concentrations of the model and GOSAT are largest over Europe (8.02 and 7.96×10²¹ molecules/cm², respectively) revealing the latitudinal gradient of CO₂ that shows the higher CO₂ concentration in the high northern latitudes. GEOS-Chem has a positive bias by ~0.8% but the correlation is relatively poor (R²=0.33) which may be due to the very limited GOSAT data sampled over the higher latitudes in the cold season or imperfections in the CO₂ flux inventories, thus the estimation of the bias over Europe is more uncertain.

The correlation between GOSAT and GEOS-Chem columns are highest over North America (R²=0.72) which may indicate relatively more accurate CO₂ flux information over this region than that of the rest of the world. GEOS-Chem has ~1.0% bias and almost the same magnitude of seasonal variation between the two data sets (3.9×10²⁰ molecules/cm²), which is driven primarily by vegetation activities since GEOS-Chem here used the annual mean fossil fuel emission inventory.

We excluded the region of Saharan desert during the investigation of Africa because the desert storms seriously interfere with the CO₂ retrievals. The seasonal trends show a typical southern hemispheric pattern and the bias is as large as 1.4% (Table 1). The CO₂ concentration over Africa is lowest (7.5-7.6×10²¹ molecules/cm²).

In South America, the GEOS-Chem trends do not reproduce the southern hemispheric CO₂ trends which is shown by GOSAT data (Table 1) and the correlation is poor (R²=0.21) with the large bias (~1.3%). This could be attributed to seasonal cycle of CO₂ fluxes coming from the CASA run or the residual annual climatology, which recent inverse modeling work suggested overestimated South American biospheric emissions for 2006 (Nassar et al., 2011). The tropical biosphere shows lower seasonality (0.9×10²⁰ molecules/cm²), but we also have to consider the significant number of missing data due to the clouds and aerosols over the Amazon region during the wet season. The data coverage of GOSAT is relatively better over Australia and modeled and GOSAT data show a typical southern hemispheric CO₂ trend (Table 1) with the smaller seasonality (1.0×10²⁰ molecules/cm²). The GEOS-Chem bias is closer to the global mean (~1.0%) over Australia. The spatial variance of CO₂ data over Australia is smallest (1.5×10²⁰ molecules/cm²), which implies that there are no strong sinks or sources in this region.

The spatial distribution of the differences between GOSAT and GEOS-Chem CO₂ total columns in summer and winter seasons are shown in Figs. 3 and 4. In July 2009, the GEOS-Chem total columns of CO₂ are slightly higher, particularly over the ocean, which gives systemic overestimation by ~8.8×10¹⁹ molecules/ cm². The global mean difference here is ~1.0% and the spatial pattern is reasonably comparable between two data sets with the correlation (R²=0.6). This difference is entirely consistent with the ~1.0% bias determined by from the 7 months of observations assuming equal-weighting of the continents (Table 1). The value is slightly larger than the mean GOSAT uncertainties (<1%) reported by Yoshida et al. (2011).

Fig. 3. 
The monthly average of GOSAT CO₂ column density in July 2009 (top), the corresponding GEOS-Chem data (middle) and the difference between GEOS-Chem and GOSAT CO₂ column density (GEOS-Chem-GOSAT) (bottom: unit: 10²¹ molecules/cm²).

Fig. 4. 
Same as Fig. 3, but in January 2010.

Previously, a global GEOS-Chem CO₂ simulation was compared with the data from 74 GLOVALVIEW-CO₂ sites by Nassar et al. (2010). Although direct comparisons of this type have limitations due to the representation error caused by the size mismatch between the model domain and sites, the GEOS-Chem CO₂ generally agreed well with GLOBALVIEW-CO₂ data (Nassar et al., 2010). With the same limitation but different initial conditions, we focused on the comparison of the monthly mean GEOS-Chem CO₂ with ground-based measurements from 12 sites, available from the World Data Center for Greenhouse Gases (WDCGG) (WMO, 2009) from January 2004 to December 2009. The sites selected represent a range of latitudes and different regions, with most being remote background sites. However, the two Korean sites (Taeahn (36.72°N, 126.12°E) and Gosan (33.17°N, 126.1° E)), Jungfraujoch (45.5°N, 8°E), and Sable Island (43.9°N, 60°W) are somewhat closer to the industrial areas of China, Europe, and North America. The model comparison with the ground-based data is important to help infer the quality of the GOSAT CO₂ products since the vertically-integrated GOSAT column data cannot be compared directly with the ground-based “point” measurements. The fact that the model simulates a complete 3-D field allows it to be compared with both measurement approaches.

During the comparison, we first calculated the GEOS-Chem biases which have a range from -1.5% to -0.5% and corrected the model bias by adding the mean bias (3.6 ppmv) for the 12 sites. Correction with a fixed value does not change the correlation between the model and observation data. The red lines in Fig. 5 indicate the model data with the bias correction and black dots shows the observations. As shown in Fig. 5, the GEOS-Chem simulations represent the seasonal cycle of the observation timeseries very well, resulting in a fairly high correlation range (0.66≤R²≤0.99).

Fig. 5. 
The comparison of CO₂ data between 12 ground-based measurement sites from WDCGG (black) and GEOS-Chem (red). The model data have been adjusted here by correcting them with the average GEOS-Chem bias from 12 sites (3.6 ppmv).

The large gradient of the seasonal variability of CO₂ with latitude is shown in Fig. 5. The comparison for the remote stations shows high correlations with the GEOS-Chem data (R²>0.90). Jungfraujoch, Taeahn, and Gosan had relatively lower correlations (0.66-0.75), which likely relates to their inland locations and proximity to industrial sources or natural terrestrial flux regions, which are challenging to model, or from the impact of representativeness errors. The sites of East Asia (Taeahn and Gosan) have a large seasonal variability (>15 ppmv) due to the strong continental influence, which is not captured very well by the model (Fig. 5). The high correlation at Sable Island may imply that the CO₂ fluxes or model transport over North America are better represented in the model than for other continents.

The GEOS-Chem data without the correction on Taeahn and Gosan show 0.6%-1.6% negative bias that is particularly larger during the winter and spring season. That large discrepancy may be due in part to the fact that the GEOS-Chem simulation used the annual mean anthropogenic CO₂ emission inventory that averaged out the seasonal anthropogenic CO₂ trends of China, resulting in the seasonal difference in the bias (~11 ppmv in the winter and ~2.5 ppmv in the summer for the Korean sites) and due in part to the lower initial condition for the CO₂ (January 2004) than applied in Nassar et al. (2010).

Based on the comparison with those global ground measurements, GEOS-Chem data reproduce the observed CO₂ data well with a systemic bias (3.6 ppmv or ~1.0%) and we can infer that the GOSAT data have a negative bias of ~2.0%, which is consistent with the recent validation study of Morino et al. (2011) that compared the NIES XCO₂ with ground-based FTS XCO₂ measurements from the Total Carbon Column Observing Network (TCCON, Wunch et al., 2010) and found a low bias of 8.85±4.75 ppm (2.3±1.2%). Another major L2 GOSAT dataset has been developed by the NASA-led Atmospheric Carbon Observations from Space (ACOS) team using a different retrieval algorithm (O’Dell et al., 2011) on the same observations. Wunch et al. (2011) compared the ACOS-GOSAT retrievals to TCCON and found a low bias of 1.8%, which they attribute to multiple sources. Retrieval of GOSAT data by another team (Butz et al., 2011) corrected for a 3% bias in the O₂ A-band, which they identify as the primary source of the XCO₂ bias in cloud-free observations, such that after the correction, the XCO₂ bias for cloud-free observations was only -0.05% relative to TCCON. It should be noted that a different empirical correction approach is used to correct the TCCON ground-based FTS XCO₂ data for presumed biases (Wunch et al., 2010). If the source of the GOSAT XCO₂ bias overall is predominantly due to the O₂ A-band, as suggested by Butz et al. (2011) based on an analysis of cloud-free observations, then overall GOSAT CO₂ column densities should only have minor biases perhaps due to retrievals with low levels of cloud and aerosol, that were not excluded based on the CAI data and other screening methods. Until the entire cause of the GOSAT bias can be definitely confirmed, evaluation and comparison of GOSAT L2 CO₂ products using a number of different methods (TCCON, model-based approaches, etc.) will be important.

We estimated each source/sink contribution to the global atmospheric CO₂ budget using a tagged CO₂ simulation in which the each source/sink is treated as an individual tracer. The tagged simulation is useful since the geographically defined state vector can be applied for inverse modeling (Nassar et al., 2011) to constrain the surface CO₂ fluxes and the spatial distribution and trends of each source/sink’s contribution can be understood. This calculation of the source/sink contribution to atmospheric CO₂ can provide useful information for greenhouse gas reduction targets and strategies for policy-makers. Here the tagged simulation estimated the individual source/sink contribution to the global atmospheric CO₂ concentration (ppmv) for 6 recent years (January 2004 to December 2009, Figs. 6-10). The tagged simulation results show the spatial and vertical distributions of each contribution. Table 2 represents the accumulated source/sink contributions over the 6-year period in terms of the global CO₂ budget and annual trends.

Fig. 6. 
The column-averaged CO₂ fraction (unit: ppmv) contributed by fossil fuel combustion and cement production from a tagged GEOS-Chem simulation for January 2004 and December 2009.

Fig. 7. 
Same as Fig. 6, but for the contribution from biomass burning.

Fig. 8. 
Same as Fig. 6, but for the contribution from aviation.

Fig. 9. 
Same as Fig. 6, but for the contribution from the terrestrial biosphere (net terrestrial exchange).

Fig. 10. 
Same as Fig. 6, but for the contribution from oceanic exchange.

Fig. 6 represents the global atmospheric CO₂ contribution by fossil fuel and cement production. The globally-averaged accumulated contribution is about 19.5 ppmv for the 6 year period (2004-2009). There is a significant latitudinal gradient in the contribution that is particularly strong due to the large emissions from E. Asia and E. US despite inter-hemispheric transport (Fig. 6). The emission of CO₂ from biomass burning is largest in the tropics (Africa, Amazonia, Indonesia) and the contribution to atmospheric CO₂ is thus stronger within the tropical regions (>6.0 ppmv for 6 years, Fig. 7). GEOS-Chem used a 3-D emission inventory for aviation that is dominant in the intercontinental airline contrails such as over the N. Atlantic and N. Pacific regions. The influence is shown in Fig. 8. The contribution spreads throughout troposphere and low stratosphere due to the flight paths of commercial aircraft.

The tagged CO₂ simulation calculated the contribution of the residual annual terrestrial exchange (-13.3 ppmv/6 years), which is higher in the northern hemisphere due to the larger fraction of the continents and hence terrestrial vegetation (Fig. 9). Since these fluxes were based on inversions using the standard 11 Transcom land regions (Baker et al., 2006) they contain very limited information on sources/sinks at sub-continental scales. Regional scale information of net ecosystem exchange based on measurements needs to be applied to more precisely to understand the atmospheric CO₂ budget. Fig. 10 represents the spatial distribution of the oceanic contribution that is mostly influenced by the sea surface sinks at the high latitudes (>3.0 ppmv).

The mean global trend of CO₂ based on the measurements has been growing (Keeling et al., 1995) from ~1.5 ppmv/year during the 1980s to the early 1990s (Conway et al., 1994) to ~1.8 ppmv/year from 1993 to 2005 (Matsueda et al., 2008). The recent satellite CO₂ retrievals from the NASA’s Atmospheric InfraRed Sounder (AIRS) have been recently validated with aircraft data and the annual mean growth rate in the middle troposphere is 1.98 ppmv/year from 2002 to 2008 (Olsen et al., 2008).

The GEOS-Chem generally reproduces the observed trend of global atmospheric CO₂ (2.08 ppmv/year) and the simulated global CO₂ increase from 2004 to 2009 is shown in Fig. 11. This interannual trend is mainly driven by the national fossil fuel combustion inventory (3.2 ppmv/year, Table 2), which includes a contribution from cement manufacture of ~2-5%. Net terrestrial and oceanic exchange of CO₂ are the main sinks (-2.2 ppmv/year and -0.6 ppmv/year, respectively, Table 2). Thus it is clear that a large reduction to the human contribution to the carbon cycle is required to limit the current global atmospheric CO₂ increase. However, there are still many unknowns that could be better understood regarding the atmospheric CO₂ budget, which could help to achieve emission reduction targets. Carbon exchange in the terrestrial biosphere (and to some extent the ocean) tends to have relatively large spatiotemporal variability, which is one of the key questions to better estimate the global CO₂ budget. The influence of the atmospheric transport including El Ni~no/Southern Oscillation (ENSO) events also needs to be quantitatively understood.

Fig. 11. 
The global column-averaged CO₂ concentration simulated by GEOS-Chem model from 2004 to 2009, demonstrating the seasonal cycle and annual increase.