Data Access

Scientists can download the data from the Data Center by following either of these links:

Atmospheric CO2 Observations from Space (ACOS)

GOSAT/TANSO-ACOS Level 2 Full Physics Standard Product

Key Data Documents

Here are the key documents describing the data products and their characteristics:

For v2.9data (New Update 11/12/12):
Data Users Guide and fields: Updated User Guide (PDF, 1.78 MB)

For v2.9data:
Data Users Guide and fields: README Document for ACOS Level 2 Standard Product v2.9 (PDF, 1.34 MB)

For v2.8 data:

Description of Data and fields: README Document for ACOS Level 2 Standard Product v2.8 (PDF, 931 KB)
Initial Data Quality Statement: Data Quality Statement ACOS v2.8 (PDF, 134 KB)
Note on Null Fields: Null Lengths Objects (PDF, 131 KB)

Updates on Data Products (Updated Periodically)

ACOS Updates from the Project Science Office -November 12, 2012

We have missed a few months of updates from the project scientist, so this is a long one! It touches on new releases of v2.9 data, key updates in the data users guide, and the plans for upcoming releases of newdata versions.

V2.9 data releases and updates

We have released additional v2.9 data – extending the record with data based on GOSAT L1B v150150 data, and replacing some earlier data that was based on GOSAT L1B v130130 data.

ACOS v2.9 was derived from several different versions of the GOSAT L1B product (v050050, v080080, v100100, v110110, v130130, and v150150). The v130130 product included an initial attempt to correct the Band-1 nonlinearity error, that is responsible for producing the observed zero level offsets in the O2 A-band spectra. Unfortunately, this correction introduced systematic errors in v130130 spectra that produced large biases in the L2 products generated from this version of the L1B product. L2 products were generated from L1B v130130 for the period extending from 20 April 2011through 18 April 2012. These products are not useful for scientific purposes. The GOSAT Project Team has recently replaced this product with L1B v150151 products. The ACOS team is reprocessing all L2 data generated during this period with v2.9, as well as continuing the forward processing with this version.

Only a preliminary assessment of the L2 data products derived from the L1B v150150 and v150151 products has been performed. These tests indicate that the L2 products retrieved from these revised L1B products are very similar to v100100 and v110110 products, typically agreeing to within ±1 ppm. The retrieved Band-1 zero level offsets are smaller for Gain H, but have not changed substantially for Gain M. These changes have slightly increased the differences between the Gain H and Gain M retrievals. These v150150 validation tests also revealed a low-amplitude “ringing” in Bands 2 and 3, which can increase the spectral residuals in these bands. This artifact has been traced to the interferogram sampling intervalnon-uniformity correction introduced in v150150. These two issues are currently under investigation.

Updated User’s guide and geolocation

We have also updated the user’s guide to provide additional information on the geolocation. The key paragraph is included here:

Geolocation Errors: The GOSAT TANSO-FTS pointing errors have continued to evolve over time. To track and correct for these variations, JAXA has used images collected a camera installed within the TANSO-FTS to image the actual bore sight. These images are compared to known topographic features (e.g. shore lines) and with images collected by the TANSO-CAI to track changes in the intended TANSO-FTS pointing. Pointing changes are documented in tables and figures that are distributed to the GOSAT usercommunity periodically. All current pointing estimates in the v2.9 product were corrected using geolocation correction tables delivered in late 2010. The pointing offsets changed abruptly in late of December 2010. The values used in v2.9 overestimate the pointing offsets through July 2011 by about a factor of 3. TANSO-FTS pointing offsets have changed more than a dozen times since then, but have mean values near those adopted for v2.9.

New Version of Science Data Products

The science team is finalizing a version update to the L2 products, which will be used for a complete reprocessing of the GOSAT data based on the v150150/v151151 L1B data. The updates of the L2 algorithm touch on many areas, including surface pressure constraint, aerosol optical depth initial guess, updated spectroscopy, revised treatment of noise treatment in retrieval algorithm, and some code changes for speed up. The preliminary plan is to complete the testing of the new software at the end of calendar year 2012 and commence with processing and delivering data to the DAAC in early 2013.

ACOS Updates from the Project Science Office -Jan 27, 2012

The V2.9 data has almost all been delivered to the DAAC, so we wanted to get in touch with our users with the latest update. The focus of this note is to provide recommendations for data usage, especially related to L1B v130130, H and M gain, and ocean vs land. We will also present some options for accessing the data. Please remember that complete users information can be found in the data user’s guide. See link at http://oco.jpl.nasa.gov/ocodatacenter/

Our key recommendations on data usage:

· There is an offset between retrieved XCO2 values using the GOSAT H- and M-gain settings. We recommend that caution be used if using the H-gain and M-gain data together. In fact we suggest using only H-gain data for most science analyses.

· It appears that the L2 ACOS products generated using the GOSAT Level 1B data version 130130 (starting on April 19, 2011) produce results that are inconsistent with those generated using previous L1B data, for both land and glint observations. These data should be used with caution if included in a science analysis with data prior to April 19, 2011.

· For glint retrievals, it was noted that the retrieved surface pressure (compared to ECMWF) decreased by about 100 to 200Pa compared to a year earlier for L1B v100100 and v110110 observations made on or after August 2010, but before the introduction of v130130 in April 2011. These changes were, to first order, dependent on latitude and this may have introduced a positive bias of about 0.4 ppmv to 0.8 ppmv in the glint XCO2 relative to the land results due to a 'shortened' O2 column. Users should take this information into accountwhen they perform analysis with the V2.9 data product. We are still investigating the root cause of this change and developing inproved screening for glint data.

In addition, note that Wunch et al (2011) have published a paper describing a bias correction approach. An equation has been developedwith V2.9 data, and appears in the appendix of that paper. But, please notethat the equation was developed with only land data, and does not apply to ocean measurements.

Tips on Data Access (also in the Data Users’ Guide):

There are a number of ways that users can download the data. One simple way is to follow this link.

http://disc.sci.gsfc.nasa.gov/acdisc/data-holdings/acos-data-holdings

Alternatively, the Mirador service can be used. – providing a web interface or command-line download

The GES DISC provides basic temporal, advanced (event), and spatial searches through its search and download engine, Mirador (http://mirador.gsfc.nasa.gov). Mirador offers various download options that suit users with different preferences and different levels of technical skills. Users can start from a point where they don’t know anything about these particular data, its location, size, format, etc., to quickly find what they need by just providing relevant keywords, like “ACOS”, or “CO2”.

Here is a direct link to the v2.9 ACOS science products on this site:

http://mirador.gsfc.nasa.gov/cgi-bin/mirador/collectionlist.pl?search=1&keyword=acos_l2s+2.9

Here are 2 methods to download the v2.9 collection:

1) Mirador Webpage

- Clicking the link above will display the collection for 2.9. Beneath the collection name, click the link “View Files”; this link will display all the files for v2.9. From here, click in the checkbox(es) to select the file(s) of interest. Click one of the buttons at the top to add the file(s) to the Cart. Doing this will update the page to show the data set collection name. On the Shopping Cart page, click the “Checkout” button. This will display the Download Data pagewith instructions on how to download the selected products.

2) Command-line

See READ ME (PDF, 24 KB)

ACOS Updates from the Project Science Office -Nov 3, 2011

The ACOS v2.9 product will be released in early November 2011. A comprehensive, yet concise guide to the data has been written, and is required reading for the science users of the data. This Data User’s Guide can be found on at the DAAC Goddard Earth Sciences Data and Information Services Center (http://daac.gsfc.nasa.gov/acdisc/documentation/ACOS.shtml) and on the OCO website (http://oco.jpl.nasa.gov/ocodatacenter/).
The Data User’s Guide is intended to provide an overview of the v2.9 Atmospheric CO2 Observations from Space (ACOS) data product, key features and issues, preliminary validation information, recommendations on data usage, as well as background on the Greenhouse Gases Observing Satellite (GOSAT) mission measurements and the ACOS algorithm. The later sections provide the reader with information on filename conventions and a detailed guide on the format and fields in the hdf product.
The version 2.9 ACOS XCO2 data product has undergone a preliminary validation using roughly 15 months of data. Validation against TCCON data shows the v2.9 XCO2 has smaller biases at most TCCON stations than v2.8, and significantly reduced scatter as compared to v2.8. In some cases, the scatter has been reduced by as much as a factor of two. The overall bias in the v2.8 data was approximately 7 ppm and much of that has been removed in the v2.9 retrievals (see Figure 1). The mean global bias is now about 0.13 ppm (1.97 ppm standard deviation), though that number does have a small seasonal variation.
The users must read the data user’s guide, specifically to understand a number of key issues specific findings and recommendations that relate to proper use of the data:
- H and M gain differences and our recommendations for use
- Identification of glint data through the use of the surface_type field, rather than the glint flag
- Advanced screening criteria that are recommended instead of the master_quality_flag
- The use of version130130 L1B data since April 19, 2011, although the consistency of this L1B version with earlier versions has not been evaluated
- Details of the validation results based on the TCCON measurements

The key algorithm changes are described in the Data User’s Guide, including some description of the impact on the product. Some of the more important ones were:
- Added geometric correction factors to L1B geolocation
- Applied a zero-level offset correction in the A-band to reduce bias in O2 fits that depends on signal level
- Rescaled O2 A-band cross sections with a constant factor of 1.025 in order to reduce a surface pressure bias of 10hPa
- Added ILS interpolation
- Empirical noise has been applied to both the ocean and land scenes
- Cloud screening is applied to glint and land data

ACOS Updates from the Project Science Office -Oct 14, 2011

As described in an earlier update, we are evaluating the B2.9 data presently, and will release it publicly in the near future. There are example files and some details of the hdf format changes available at the DAAC. (see the yellow box near the top of http://daac.gsfc.nasa.gov/acdisc/documentation/ACOS.shtml)
In this update, I include a brief description of the key changes that were made to the algorithm between B2.8 and B2.9. As you can see in the description below, the surface pressure and XCO2 biases are significantly reduced in the 2.9 data. The material below address both L1B and L2 changes.
The following is a summary of the key L1B changes made in B2.9 compared to B2.8.
· Updated time-varying calibration coefficients received from JAXA in November 2010
· Time dependent correction had been only applied toradiances but not the noise estimate in B2.8. As a result, fewer and fewer retrievals passed the quality flag overtime in version B2.8. In B2.9, it is applied to both, so there is no longer a time dependence in the number of data passing the quality flag.
· Custom glint flag calculation: In B2.8, the JAXA define glint flag was used to select data for processing as glint. Due to an error in that flag, this resulted in missing glint data, with more and more data missing over time. We now implement a custom flag for glint, and have full coverage.
· Added solar distance to L1B product
· Added L1B noise elements with _l1b suffix
· Added geometric correction factors to L1B geolocation
Changes to the L2 products are as follows:
· Significantly affecting the retrieval results
o Applied zero-level offset correction in the A-band to reduce bias in O2 fits that depend on signal level. By adding a flat zero-level offset term in the state vector and fitting for it, many systematic biases were eliminated. However, this has caused differences in M and H-gain to increase; this issue is being investigated and will be addressed in B2.10.
o Rescaled O2 A-band cross sections with a constant factor of 1.025 in order to reduce a surface pressure bias of 10hPa. Mean surface pressure is now unbiased & retrieved aerosols optical depths are somewhat lower; XCO2 is in better agreement with TCCON. The scaling of the O2-A band increased the XCO2 values by approximately 4 to 5 ppm.
o Added ILS interpolation. Previously, the ILS tabulated data was used, now we perform an interpolation on that ILS data. The impact of this change is that XCO2 is now 1.5ppm closer to TCCON.
o Glint noise treatment: In B2.8, the ocean scenes were assigned the noise as supplied byJAXA, and had very large values of chi-squared, whereas the land scenes had an empirical noise applied, which was more consistent across bands and resulted in chi-squared values close to 1. In B2.9, the empirical noise has been applied to both the ocean and land scenes.
o Cloud screening applied to glint data in preprocessing
· Within the code
o Static input data moved to a single HDF file
o Upgraded LIDORT version to 3.5T
o Reworked Jacobian calculations to use automatic derivatives
· Instrument capability
o Added support for FTS Instrument in up-looking mode
o Added support for OCO-2 instrument mode
· Speed improvement
o Used only two-streams in the Low Streams Interpolator (LSI) part of the radiative transfer code when a low number of streams is required (was 4 previously).

ACOS Updates from the Project Science Office - Aug 25, 2011

As we have discussed in previous notes, the ACOS B2.8 XCO2 data have been analyzed and found to have significant biases. An approach for bias correction has been developed and was discussed here previously.

More recently, Wunch et al have published a paper discussing the initial validation of the ACOS data and provided an update to the B2.8 bias correction equation. We will not include all the details here, but the reference can be found at: http://www.atmos-chem-phys-discuss.net/11/20899/2011/acpd-11-20899-2011.pdf

In addition, the ACOS algorithm team has been testing a number of updates to the retrieval algorithm. The main elements that are being tested are the retrieval of a zero level offset, improvements to the ILS representation, and scaling of the spectroscopy of the A-band. These updates are expected to make significant improvements to the bias seen in the retrievals of XCO2 and surface pressure. The ACOS team is currently evaluating this new dataset (labeled B2.9). The plan is for the team to generate one year of test data and complete the evaluation by October 31, 2011. The findings of that evaluation and any issues seen in the data will be documented, and then both the data and documentation will be released to the user community through the Goddard Earth Science ACDISC. The planned date for release of the initial B2.9 data is November 14, 2011.

ACOS Updates from the Project Science Office - May 12, 2011

This information is from the analysis of the ACOS team, with significant
input from Debra Wunch, Chris O'Dell, Brendan Fisher, along with others.

The ACOS team has analyzed our B2.8 data product, and determined that
there is a bias in the XCO2 values, relative to the TCCON validation data.
In analysis that followed, we have identified some of the key
characteristics of the errors, and suggested formulations for bias
correction.

In our recent analysis, a number of factors were included in the
formulation. The current formulation is based on four variables. These
variables are: blended albedo, surface pressure difference, airmass, and
the signal level in the O2 band. The signal level and airmass are
included because they have been shown to explain a large fraction of the
variance, and statistically, they are both useful in the correction. The
blended albedo is a combination of two albedos, and it is included in the
correction because it has been observed that in the ACOS data, and
simulations using the ACOS algorithm, there is dependence of XCO2 error
(difference from TCCON for ACOS and input truth for simulations) and the
blended albedo. The surface pressure difference (difference of retrieved
value and ECMWF prior) is included because we see a consistent difference,
(most likely explained by spectroscopy errors), and if the surface
pressure is incorrect, the total column of air is incorrect, which will
propagate to an error in XCO2. So, the four key variables have been
included in a multivariate formulation of the correction equation.

The formulation is:
XCO2_corrected = XCO2/0.982 - (11.3±0.2)(blended albedo-mean(blended
albedo)) + (0.123±0.006)(dP-mean(dP)) + (2.8±0.2)(airmass-mean(airmass)) +
(0.48±0.03)(signal_o2 mean(signal_o2))*10^7

where dP is in hPa and the means are global over the whole dataset
blended albedo = 2.4*Albedo_AO2 - 1.13*Albedo_SCO2
dP = P_surf – P_ecmwf

As for the airmass, it can be calculated from the data on the file:
airmass = 1./cos(!dtor*data.sounding_zenith)+
1./cos(!dtor*data.solar_zenith)

Where !dtor is an idl function that contains a conversion factor for
converting degrees to radians

Maps of the correction - delta XCO2 (i.e. XCO2_corrected - XCO2_raw) are
attached (labeled 6a). These are binned monthly average maps of the delta
XCO2, and show these general characteristics: The spatial characteristics
change from month to month, with the maximum positive change (of 10 to 12
ppm) shifting to the north from southern latitudes from June through
August, with little latitudinal gradient in Sept, and by October, the
maximum changes are in the northern most latitudes of measurement. There
are bands of very small correction, which also shift north and south over
the year. A year of maps are provided for users to review, and consider as
they interpret the data.

Previously suggested correction:
In initial analysis, it was noted that the difference between the
retrieved XCO2 and the TCCON XCO2 is highly correlated with airmass and/or
signal level. These two parameters are themselves highly correlated. After
detailed analysis, we can find little reason to think that the airmass is
incorrect, but there is evidence for an error in the calibration which
results in a radiance level or signal dependent bias. Thus, one of the
first formulations for a correction to the XCO2 bias in the ACOS product
was as follows:

XCO2_corrected = XCO2/0.982 - 0.4*(signal_o2 - mean(signal_o2))*10^7

Maps (labeled 6b) were created by Brendan Fisher to show the spatial and
temporal characteristic of this correction. Overall, the correction shows
some spatial structure it is generally positive, except over the Sahara,
and is of the magnitude of a few to 10 ppm. It becomes larger in the
winter, as the signal term of the correction equation becomes smaller.

Single Parameter Delta Maps (Zip File, 976 KB)
Multiple Parameter Delta Maps (Zip File, 1 MB)