Version History - CT2011
 
  • Click here for a comparison of current CarbonTracker flux estimates with earlier releases.
  • Click here to view links to earlier releases.

Release Notes for CarbonTracker 2011
  1. Product
    • Released on May 11, 2012
    • CarbonTracker 2011 is derived from a suite of eight different inversions, each using a different set of prior fluxes. See ensemble data assimilation documentation for more information.
    • CT2011 flux estimates are available from 2000 through 2010.
    • Documentation is updated for CT2011
    • The flux comparison tables page has been updated to include comparisons with four previous CarbonTracker releases (CT2010, CT2009, CT2008, and CT2007B).
    • Access to previous CarbonTracker web releases is now available.
  2. Observations
  3. Fossil Fuels
  4. Fires and Biosphere
  5. Ocean
    • Two air-sea CO2 exchange models were used as prior flux estimates, to form part of the inversion suite. See oceans module documentation for more information.
  6. Data Assimilation and Transport Modeling
    • To incorporate multiple prior flux models, the background covariance formulation of CT2011 has been retuned. While CarbonTracker's performance against observations has improved by using mulitple priors, the formal error bars on CT2011 flux estimates have increased significantly. See ensemble data assimilation documentation for more information.
    • Global resolution remains at 3° longitude x 2° latitude, with transport over North America modeled at 1° x 1°. We treat the 25-layer to 34-layer discontinuity at the end of 2005 in the same manner as was done for CT2007B and later releases (see release notes for CT2007B, below).


Release Notes for CarbonTracker 2010
  1. Product
    • Released on Friday, February 1, 2011
    • CT2010 flux estimates are available from 2000 through 2009
    • Documentation is updated for CT2010
    • The flux comparison tables page has been updated to include comparisons with three previous CarbonTracker releases (CT2009, CT2008, and CT2007B).
    • Access to previous CarbonTracker web releases is now available.
  2. Observations
  3. Fossil Fuels
    • 2008 and 2009 country and global totals (by fuel type) are extrapolated from the 2007 CDIAC statistics using energy consumption statistics from the BP Statistical Review of World Energy 2010.
  4. Fires and Biosphere
    • In CT2010 we use a hybrid of CASA-GFED2 and CASA-GFED3 modeling prodcts. Fire emissions are imposed from the newer CASA-GFED3, whereas terrestrial NEE first-guess fluxes are derived from the earlier GFED v2 product. This year the CASA-GFED team has transitioned to a model (version 3.1) driven by MODIS fPAR. The previous revision (version 2) of this model used AVHRR NDVI, scaled to represent MODIS fPAR. We have found that the seasonal cycle of NEE is reduced by about 10% in CASA-GFED3.1 compared to CASA-GFED2. CarbonTracker will require re-tuning of its optimization system to operate with this new model. This is a research problem we will be addressing over the next release cycle. However, CASA-GFED3.1 is vitally important because it is the only means we have to assess fire emissions for 2009. We have chosen a hybrid approach for CT2010, combining elements of the old and new models. While the fire emissions come from CASA-GFED3.1, terrestrial NEE first guess fluxes derive from CASA-GFED2. The NEE fluxes have real estimated interannual variations for 2007 and 2008, but since CASA-GFED2 analysis stops in 2008, first-guess fluxes in 2009 are computed from the climatological NEE of CASA-GFED2. This is not unprecedented. For several previous releases, we have used climatological first-guess fluxes for the final year in the analysis cycle.
  5. Ocean
    • As in previous releases, ocean prior fluxes in CT2010 are based on Jacobson et al. (2007) ocean inverse estimates rather than Takahashi et al. (2002). Please see documentation for more information.
  6. Data Assimilation and Transport Modeling
    • Global resolution remains at 3° longitude x 2° latitude, with transport over North America modeled at 1° x 1°. We treat the 25-layer to 34-layer discontinuity at the end of 2005 in the same manner as was done for CT2007B (see release notes for that release, below).
    • After CT2010 simulations were completed, a bug was discovered in the ensemble Kalman filter code. This involves adjusting simulated observations for certain sites two times, when they should be adjusted only once. Analysis of the bug suggests that the impact on CarbonTracker results is minimal. We are currently reprocessing results to check explicitly on the effects of fixing this error.

Release Notes for CarbonTracker 2009
  1. Product
    • Released on Friday, 13 November, 2009
    • CT2009 flux estimates are available from 2000 through 2008
    • Documentation is updated for CT2009
    • The flux comparison tables page has been updated to include results from three CarbonTracker releases (CT2009, CT2008, and CT2007B).
  2. Observations
  3. Fossil Fuels
    • 2007 and 2008 country and global totals (by fuel type) are extrapolated from the 2006 CDIAC statistics using energy consumption statistics from the BP Statistical Review of World Energy 2009.
    • Within-country emissions patterns have been upgrades to use data from EDGAR version 4, which no longer uses population density as its sole proxy for emissions distribution.
  4. Fires and Biosphere
    • We continue to use the GFED v2 product. This has been updated through the end of 2008. In the near future, the GFED team is expecting to change its source of NDVI data from NOAA polar orbiting satellites to a similar MODIS product, and this will become the GFED version 3 release. We will evaluate this new product when it becomes available, but CT2009 still uses the GFED v2 estimates.
  5. Ocean
    • As in CT2008, ocean prior fluxes in CT2009 are based on Jacobson et al. (2007) ocean inverse estimates rather than Takahashi et al. (2002). Please see documentation for more information. As in CT2008, we estimate air-sea fluxes for 30 ocean regions.
  6. Data Assimilation and Transport Modeling
    • Increase of global resolution from 6° longitude x 4° latitude to 3° x 2°. Transport over North America remains at 1° x 1°. We treat the 25-layer to 34-layer discontinuity at the end of 2005 in the same manner as was done for CT2007B (see release notes for that release, below).
  7. Responses to the 2008 CarbonTracker science review In response to the 2008 scientific review of the CarbonTracker program, several refinements and improvements have been made:
    • The flux maps page now shows results presented at the ecoregion scale, in addition to the standard 1° x 1° scale results. CarbonTracker natively estimates fluxes for an entire ecoregion, and the 1° x 1° within-region patterning we present is derived from a first-guess flux model. Since we don't optimize fluxes at the finer 1° x 1° scale, it is in some ways more formally correct to also present fluxes without the modeled within-region patterning.
    • We present both our first-guess and our final flux estimates, so that the effects of CarbonTracker optimization (the "analysis increments") can be determined.
    • We have created a new ecoregion documentation page to describe in more detail the role of ecosystems and Transcom regions in the CarbonTracker optimization scheme.

Release Notes for CarbonTracker 2008
  1. Product
    • Released on 31 October 2008
    • CT2008 flux estimates are available from 2000 through 2007
    • Documentation is updated for CT2008
    • The flux comparison table has been updated to include fluxes from CT2007, CT2007B, and CT2008.
  2. Observations
    • New daytime average data have been added for the Environment Canada (EC) site at Egbert, Ontario, Canada (EGB_06C3, beginning April 2005).
    • New daytime average data have been added for the Lawrence Berkeley National Laboratory (LBNL) site at Southern Great Plains, Oklahoma, United States (SGP_03C3, for the period 2003 through 2004).
    • New data from the Commonwealth Scientific and Industrial Research Organization (CSIRO) network of discrete sampling sites in the high southern hemisphere have been added. Sites include Cape Grim, Tasmania, Australia (CGO_02D0); Cape Ferguson, Queensland, Australia (CFA_02D0); Macquarie Island, Australia (MQA_02D0); Mawson Station, Antarctica, Australia (MAA_02D0); Casey, Antarctica, Australia (CYA_02D0). All records began in the early 1990s.
    • Time averaging of quasi-continuous measurements from the Mauna Loa Observatory (MLO_01C0) has been changed to 0-4 LST from 12-16 LST.
    • Time averaging for observatory sites has been changed to exclude erroneous hours for sites close to the international dateline.
  3. Fossil Fuels
    • 2005, 2006 and 2007 country totals (and the global totals) are extrapolated from 2004 CDIAC statistics based on energy consumption statistics from BP Statistical Review of World Energy 2008.
    • Monthly resolution seasonality of fossil fuel emissions, which existed only in North America in previous releases, has now been added to Eurasia (30-60N), based on preliminary seasonal emissions from EDGAR, partitioned by energy use sector. This imposes a much greater global seasonality in fossil fuel emissions than existed in CT2007B.
  4. Fires and Biosphere
    • We continue to use the GFED v2 product. This has been updated through the end of 2007. In the near future, the GFED team is expecting to change its source of NDVI data from NOAA polar orbiting satellites to a similar MODIS product, and this will become the GFED version 3 release. We will evaluate this new product when it becomes available, but CT2008 still uses the GFED v2 estimates.
  5. Ocean
    • As in CT2007B, ocean prior fluxes in CT2008 are based on Jacobson et al. (2007) ocean inverse estimates rather than Takahashi et al. (2002). Please see documentation for more information. As in CT2007B, we estimate air-sea fluxes for 30 ocean regions.
  6. Data Assimilation and Transport Modeling
    • No significant changes. We treat the 25-layer to 34-layer discontinuity at the end of 2005 in the same manner as was done for CT2007B (see release notes for that release, below).

Release Notes for CarbonTracker 2007B
  1. Product
    • Released on 20 December 2007
    • CT2007B flux estimates are available from 2000 through 2006 (the previous CT2007 release covered through 2005)
    • Documentation is updated for CT2007B
    • A flux summary table has been added comparing CT2007 and CT2007B total fluxes
  2. Observations
    • New daytime average data have been added for Sable Island, Nova Scotia, Canada (SBL_06C3, beginning June 2003)
    • New nighttime average data have been added for Niwot Ridge, Colorado, United States (NWR_03C3, beginning August 2005)
    • New nighttime average data have been added for Storm Peak Lab, Colorado, United States (SPL_03C3, beginning September 2005)
    • CO2 mole fractions are now based on flask pair averages rather than the first flask of a pair only.
    • Data for Fraserdale, Ontario, Canada (FRD_03C3) have been reprocessed based on the Environment Canada submission to WMO.
    • Time averaging for FRD_03C3 has been shifted by one hour based on new information about reported times.
    • Time averaging for observatory sites has been changed to exclude erroneous hours for sites close to the international dateline.
    • A few recent flask data points have had their Quality Control flag values revised and are now excluded.
    • LEF data for 2000-2003 is revised based on the use of Licor-1 values rather than Licor-2. Two CO2 sensors were present at the 396m level of LEF, but it was determined that the Licor-1 was more reliable.
  3. Fossil Fuels
    • Fossil fuel estimates in the previous release (CT2007) were based on global emission inventories from CDIAC only until 2002. For 2003 and 2004, global totals were linearly extrapolated using the average percentage increase of the previous 10 years, and for 2005, emissions from 2004 were simply repeated. Spatial patterns were based on linearly interpolated and extrapolated EDGAR emissions estimates for 1995 and 2000. For CT2007B, global totals are used from CDIAC through 2004, and 2005 and 2006 data are extrapolated based on energy consumption statistics from BP for those years. The coarse spatial patterns are now derived from CDIAC and BP country statistics, and only the within country patterns come from EDGAR. Both global totals and regional trends are different, especially after 2002. Please see documentation for more information.
  4. Fires
    • A new GFED v2 release has been implemented and runs to the end of 2006, correcting an issue in which fires for 2005 were not available.
  5. Biosphere
    • The biosphere fluxes are consistent with the new GFED v2 release and complete through the end of 2006.
  6. Ocean
    • Ocean prior fluxes are now based on Jacobson et al. (2007) estimates rather than Takahashi et al. (2002). Please see documentation for more information.
  7. Data Assimilation and Transport Modeling
    • CarbonTracker inverts for 30 ocean regions in CT2007B rather than 11. These finer regions are a superset of the 11 TransCom ocean regions, and can be simply aggregated to the coarser regions.
    • A new ecoregion mask was created with an improved land/ocean mask near the coasts, causing shifts in the total area of some coastal ecoregions.
    • In CT2007, the month of December 2005 was run without the 1x1 degree zoom region. For that month, transport was simulated at 3x2 degrees in that region. In CT2007B, the 1x1 zoom ECMWF meteorology was available throughout the end of October, 2006. As a result, the present release has 1x1 degree analysis through October 28, 2006, and 3x2 meteorology for the remaining eight weeks.
    • At the end of 2005, ECMWF changed the vertical resolution in its model from 60 to 91 layers. For use in the TM5 offline transport model, the 60-layer model results are reduced to 25 layers. Similarly, the 91-layer results are reduced to 34. As a result, the transport model has a 25-to-34 level discontinuity at the end of 2005. We have integrated over this transition by interpolating the CO2 fields to the higher resolution, using a mass-conserving algorithm.

Release Notes for CarbonTracker 2007
  1. This is a link to the CarbonTracker 2007 archive.
  2. Released on 28 February 2007.
  3. CarbonTracker 2007 has a number of caveats for the year 2005 listed below. These shortcomings will most likely lead to revised flux estimates for this year in our next CarbonTracker release (2008). Most importantly, we expect the total uncertainty on the 2005 estimate to become smaller even than the estimate of 2004, as more observations become available for assimilation.
  4. The CO2 measurements obtained from the Meteorological Service Canada (MSC) currently go through February 2005. This record will be expanded for CarbonTracker 2008.
  5. The month December of 2005 was run without the United States 1x1 degree zoom region in the transport model. This is due to a change in model resolution in the parent model from ECMWF in February of 2006. We expect future CarbonTracker releases to be able to integrate past this moment in time as we develop the software to deal with this discontinuity.
  6. The 2005 fossil fuel emissions are identical to those used in 2004. Although economic statistics are available to extrapolate the fossil fuel emissions from previous years, we did not include this information in version 2007. We expect better fossil fuel emission estimates for 2005 to be included in CarbonTracker 2008
  7. The 2005 fire emissions are the climatological average of those used in 2000-2004. The new GFED2 release (weekly fluxes through 2005) came out too late to be incorporated in CarbonTracker 2007
  8. Annual mean uncertainties for CarbonTracker 2007 are calculated as the annual average of the 52 weekly estimated covariances. This estimate does not include any temporal covariances in the fluxes which would substantially reduce uncertainty. We feel that until we have implemented a proper mechanism to propagate such information through time, we can not objectively quote the true uncertainty on the annual mean that would result from the assimilation of all the observations. Our current number is supposed to represent a conservative estimate of the real uncertainty to be compared to previous flux estimates.
  9. Uncertainty maps on the fluxes show smaller uncertainties in the tropics than in the extra-tropics. This is counter-intuitive as the tropical carbon cycle is least well understood and most poorly constrained by observations. In recognition of this, CarbonTracker 2007 uses strong covariances on ecosystems in the tropical regions. The variance on each region is lowered to limit the resulting total uncertainty per continent. Since our maps only show variances per region, and not covariances, this fact is not reflected in the mapped product.
  10. Fossil fuel emissions can occur over regions characterized as ocean. This is partly due to real emissions from international shipping, and partly due to emissions occurring in coastal land regions that are assigned to the ocean in our coarse 1x1 degree aggregation scheme. The same is true for fossil fuel emissions over non-optimized regions such as ice, polar deserts, and inland seas. This problem shows up only when aggregating to larger land, ocean, or continental regions for interpretation, but has no influence on the assimilation process itself.