Datasets:

---

LLMDH

/

OpenScience

like 0

Follow

LLM - Digital Humanities 16

ERROR: type should be string, got "https://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nThe Total Carbon Column Observing Network’s GGG2020 Data\nVersion\nJoshua L. Laughner1 , Geoffrey C. Toon1 , Joseph Mendonca2 , Christof Petri3 , Sébastien Roche4,5 ,\nDebra Wunch6 , Jean-Francois Blavier1 , David W.T. Griffith7 , Pauli Heikkinen8 , Ralph F. Keeling9 ,\nMatthäus Kiel1 , Rigel Kivi8 , Coleen M. Roehl10 , Britton B. Stephens11 , Bianca C. Baier13 ,\nHuilin Chen14 , Yonghoon Choi15,16 , Nicholas M. Deutscher7 , Joshua P. DiGangi15 , Jochen Gross17 ,\nBenedikt Herkommer17 , Pascal Jeseck18 , Thomas Laemmel19,* , Xin Lan12,13 , Erin McGee6 ,\nKathryn McKain13 , John Miller13 , Isamu Morino20 , Justus Notholt3 , Hirofumi Ohyama20 , David\nF. Pollard21 , Markus Rettinger22 , Haris Riris23 , Constantina Rousogenous24 , Mahesh Kumar Sha25 ,\nKei Shiomi26 , Kimberly Strong6 , Ralf Sussmann22 , Yao Té18 , Voltaire A. Velazco7,27 , Steven C. Wofsy4 ,\nMinqiang Zhou25,** , and Paul O. Wennberg10,28\n1\n\nJet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA\nClimate Research Division, Environment and Climate Change Canada, Toronto, ON, Canada\n3\nInstitute of Environmental Physics, University of Bremen, Otto-Hahn-Allee 1, 28359 Bremen, Germany\n4\nJohn A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, MA, USA\n5\nCenter for Astrophysics, Harvard & Smithsonian, Cambridge, MA, USA\n6\nDepartment of Physics, University of Toronto, Toronto, Canada\n7\nCentre for Atmospheric Chemistry, School of Earth, Atmospheric and Life Sciences, University of Wollongong,\nWollongong, New South Wales, Australia\n8\nSpace and Earth Observation Centre, Finnish Meteorological Institute, Sodankylä, Finland\n9\nScripps Institute of Oceanography, La Jolla, California, USA\n10\nDivision of Geological and Planetary Sciences, California Institute of Technology, Pasadena, CA, USA\n11\nEarth Observing Laboratory, National Center for Atmospheric Research (NCAR), Boulder, CO, USA\n12\nCooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, USA\n13\nNOAA Global Monitoring Laboratory, Boulder, USA\n14\nCenter for Isotope Research, University of Groningen, Groningen, the Netherlands\n15\nNASA Langley Research Center, Hampton, VA 23681\n16\nAnalytical Mechanics Associated, Hampton, VA 23666\n17\nKarlsruhe Institute of Technology (KIT), Institute of Meteorology and Climate Research (IMK-ASF), Karlsruhe, Germany\n18\nLaboratoire d’Études du Rayonnement et de la Matière en Astrophysique et Atmosphères (LERMA-IPSL), Sorbonne\nUniversité, CNRS, Observatoire de Paris, PSL Université, 75005 Paris, France\n19\nUniversité Paris-Saclay, CEA, CNRS, UVSQ, Laboratoire des Sciences du Climat et de l’Environnement (LSCE/IPSL),\nGif-sur-Yvette, France\n20\nNational Institute for Environmental Studies (NIES), Onogawa 16-2, Tsukuba, Ibaraki 305-8506, Japan\n21\nNational Institute of Water and Atmospheric Research Ltd (NIWA), Lauder, New Zealand\n22\nKarlsruhe Institute of Technology (KIT), IMK-IFU, Garmisch-Partenkirchen, Germany\n23\nNASA Earth Science Technology Office (ESTO), B22, 242, 8800 Greenbelt Rd., Greenbelt, MD. 20771\n24\nClimate and Atmosphere Research Centre (CARE-C), The Cyprus Institute, Nicosia, Cyprus\n25\nRoyal Belgian Institute for Space Aeronomy, Brussels, Belgium\n26\nEarth Observation Research Center (EORC), Japan Aerospace Exploration Agency (JAXA), Tsukuba, Japan\n27\nDeutscher Wetterdienst, Meteorological Observatory Hohenpeissenberg, 82383 Hohenpeissenberg, Germany\n28\nDivision of Engineering and Applied Science, California Institute of Technology, Pasadena, CA, USA\n2\n\n1\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n*\n\nNow at: University of Bern, Department of Chemistry, Biochemistry and Pharmaceutical Sciences and Oeschger Center for\nClimate Change Research, Bern, Switzerland\n**\nNow at: Institute of Atmospheric Physics, Chinese Academy of Sciences, China\nCorrespondence: Joshua Laughner (josh.laughner@jpl.nasa.gov)\nAbstract. The Total Carbon Column Observing Network (TCCON) measures column-average mole fractions of several greenhouse gases (GHGs) beginning in 2004 from over 30 current or past measurement sites around the world, using solar absorption\nspectroscopy in the near infrared region. TCCON GHG data have been used extensively for multiple purposes, including in\nstudies of the carbon cycle and anthropogenic emissions as well as to validate and improve observations made from space5\n\nbased sensors. Here, we describe an update to the retrieval algorithm used to process the TCCON near IR solar spectra and\nthe associated data product. This version, called GGG2020, was initially released in April 2022. It includes updates and improvements to all steps of the retrieval, including but not limited to: converting the original interferograms into spectra, the\nspectroscopic information used in the column retrieval, post hoc airmass dependence correction, and scaling to align with the\ncalibration scales of in situ GHG measurements.\nAll TCCON data are available through tccondata.org and hosted on CaltechDATA (data.caltech.edu). Each TCCON site has\n\n10\n\na unique DOI for its data record. An archive of all sites’ data is also available with the DOI 10.14291/TCCON.GGG2020\n(Total Carbon Column Observing Network (TCCON) Team, 2022). The hosted files are updated approximately monthly, and\nTCCON sites are required to deliver data to the archive no later than one year after acquisition. Full details of data locations\nare provided in the data availability section.\n\n15\n\n1\n\nIntroduction\n\nThe Total Carbon Column Observing Network (TCCON) is a network of nearly 30 ground-based, solar-viewing, Fourier\ntransform infrared (FTIR) spectrometers that report observations of column average mole fractions of CO2 , CH4 , N2 O, CO,\nHF, H2 O, and HDO in the atmosphere. The first two TCCON stations were established in 2004, with additional stations\njoining over the following years. As of July 2023, 30 sites exist. In that time, TCCON data have been used to estimate or\n20\n\nevaluate carbon fluxes (e.g. Keppel-Aleks et al., 2012; Peiro et al., 2022), for satellite validation (e.g. Wunch et al., 2017; Chen\net al., 2022; Lorente et al., 2022), for model verification (e.g. Byrne et al., 2023), and for other purposes.\nTCCON instruments measure solar spectra in the near-infrared (NIR) wavelengths; these spectra are converted into the final\ncolumn average mole fractions (henceforth denoted as “Xgas ”, e.g. “XCO2 ”) using the retrieval software GGG.1 Major versions\nof GGG are identified by the year of development. The previous version used to generate public TCCON data was GGG2014\n\n25\n\nand is described in Wunch et al. (2015). GGG2020 is the first major update applied to TCCON public data since GGG2014.\nGGG retrieves trace gas column amounts by iteratively scaling an a priori vertical trace gas profile until the best fit between\na spectrum simulated from those trace gas profiles by the built-in forward model and the observed spectrum is found. A single\ngas may be fit in more than one spectral window; for example, GGG2020 retrieves the standard TCCON CO2 product from\n1 GGG\n\nis the proper name of the software, and is not an acronym.\n\n2\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\ntwo separate windows (6220 to 6260 cm−1 and 6297 to 6382 cm−1 ). Each window is run separately and produces its own\n30\n\nposterior scaled trace gas profile, which can be integrated to generate a column density. Retrieving each window separately,\nrather than concatenating the spectral information, makes it simpler to handle non-contiguous windows that need different state\nvector elements. It also allows biases that differ between these windows to be expressed separately in the resulting output data\nand, if necessary, corrected separately. The output values (column densities and profile scaling factors) from different windows\nwith similar averaging kernels for the same target gas are combined in a weighted average during post processing.\n\n35\n\nThe post processing step includes the above window-to-window averaging alongside an empirical airmass-dependent correction, a scaling correction to tie TCCON data to the relevant calibration scales, and the conversion from column densities\nto column-average mole fractions. Airmass-dependent errors can arise from, for example, errors in the relative intensities of\nstrong and weak absorption lines for a target gas. At large solar zenith angles (SZAs), the longer light paths through the atmosphere will cause strong absorption lines to completely absorb incoming light within their core wavelengths; such lines may be\n\n40\n\nreferred to as “blacked out”. Blacked out lines cannot contribute information to the retrieval, so the retrieval must get a greater\nfraction of its information from weaker lines in the spectral window or the wings of saturated lines. If there is a different bias\nin the forward model between the strong and weak lines, it will manifest as an error in the retrieved column amounts that varies\nwith SZA and is symmetric about solar noon. Once the magnitude of this error is derived (§7.1), a post-processing correction\ncan be applied to remove it.\n\n45\n\nThe scaling factor used to tie to calibration scales is necessary because the spectroscopic parameters needed by the forward\nmodel are not in general known to the ∼ 0.25% or better accuracy needed for greenhouse gas data. However, since all TCCON\n\nsites use the same retrieval (and thus the same forward model), we use a single mean scaling factor to remove the mean bias\ncaused by errors in the spectroscopic parameters. This does implicitly assume that imperfect instrument line shape (ILS) or\n\nimperfect representation of the instrument in the forward model are either (a) consistent across sites and thus accounted for by\n50\n\nthe scaling factor or (b) random and average to zero. The scaling factors for the various gases are derived from comparisons\nbetween TCCON data and in situ vertical profiles measured by aircraft- or balloon- borne instruments (§7.3).\nFinally, the conversion from column densities to column-average dry mole fractions is done by dividing the target gas column\n(Vgas ) by the O2 column (VO2 ), then multiplying by the mean O2 mole fraction (fO2 ) in the atmosphere:\nXgas =\n\n55\n\nVgas\n· fO2\nVO2\n\n(1)\n\nGGG2020 assumes that fO2 = 0.2095 for all retrievals except those listed in §7.3.2. The advantages of normalizing to the O2\ncolumn are:\n1. It normalizes for path length. Observations at surface elevations will have smaller column densities compared to those\nfrom lower altitudes, due to the shorter vertical extent. Normalizing to the O2 column removes this effect.\n2. Because O2 and the primary TCCON gases are measured on the same detector, many biases related to the detector and\n\n60\n\npointing will be cancelled out (Wunch et al., 2011, Appendices A and B).\n\n3\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nGGG is comprised of several sub-programs, which handle these various elements of the retrieval. Each of these has been\nupgraded for GGG2020:\n– i2s: converts interferograms to spectra. Updates include identifying detector nonlinearity and better phase correction\n(§5).\n– gsetup: prepares the input files needed to run gfit (a priori meteorology and trace gas profiles, atmospheric path infor-\n\n65\n\nmation, etc.) in the required formats. Updates include the source of a priori meteorology and trace gas profiles and the\nretrieval grid (§4).\n– gfit: retrieves column densities from the spectra output by i2s. Updates include the forward model spectroscopy (§3) and\ncontinuum fitting (§6).\n– Post processing: a suite of programs that collates the output from gfit and applies any required post hoc corrections.\n\n70\n\nUpdates include the airmass correction (§7.1), window to window averaging (§7.2), and scaling to tie to in situ calibration\nscales (§7.3).\nGGG2020 data is available through tccondata.org. A repository containing the full set of publicly available data is available\nthrough CaltechDATA (Total Carbon Column Observing Network (TCCON) Team, 2022). Each TCCON site’s data record\n75\n\nhas its own unique DOI. On occasions that a site needs to reprocess and redeliver data already released to the public, the\nrevised dataset will receive a new DOI with the revision number incremented. TCCON sites are permitted to withhold data\nfrom the public archive for up to one year from acquisition. This public archive is updated approximately once per month\nwith newly delivered or released data. The TCCON data product is documented extensively through the TCCON Wiki (https:\n//tccon-wiki.caltech.edu/). Users are asked to familiarize themselves with the data use policy and license, which are available\n\n80\n\nat https://tccon-wiki.caltech.edu/Main/DataUsePolicy.\n2\n\nNew Xgas products\n\nGGG2020 introduced XCO2 mole fractions retrieved in two new windows: a strong band between 4809.74 and 4896.0 cm−1\nand a weak band between 6041.8 and 6105.2 cm−1 . We refer to these as “lCO2 ” and “wCO2 ”, respectively. These are reported\nas separate CO2 products (XlCO2 and XwCO2 ) and are not averaged together with the standard TCCON XCO2 product. Figure\n85\n\n1 shows the column averaging kernels (AKs) and CO2 absorption lines in these two windows. The lCO2 AKs increase towards\nthe surface, while, at small slant Xgas amounts (i.e. small solar zenith angle) the wCO2 AKs are greater in the stratosphere\nthan in the lower troposphere. This is because, as seen in Fig. 1b and d the CO2 absorption lines in the lCO2 band are mostly\nsaturated at the line center, while the wCO2 lines are not. In theory, when used together with the standard TCCON XCO2\nproduct (which has an AK profile that is more constant with altitude than the wCO2 or lCO2 products, see Fig. 25), this\n\n90\n\nprovides the potential to separate changes in CO2 at the surface, from those in the free troposphere or stratosphere (Parker\net al., 2023).\n4\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nTable 1. List of TCCON sites and their associated data citations as of 20 Dec 2022. Some sites (Lauder, JPL) have had different FTIR\ninstruments operating over different periods, and so are listed multiple times.\nSite ID\n\nSite Name\n\nLocation\n\nData Citation\n\nae\n\nascension01\n\nAscension Island, Saint Helena\n\nFeist et al. (2017)\n\nan\n\nanmeyondo01\n\nAnmyeondo, South Korea\n\nGoo et al. (2017)\n\nbi\n\nbialystok01\n\nBialystok, Poland\n\nPetri et al. (2017)\n\nbr\n\nbremen01\n\nBremen, Germany\n\nNotholt et al. (2022)\n\nbu\n\nburgos01\n\nBurgos, Philippines\n\nMorino et al. (2022c)\n\nci\n\npasadena01\n\nPasadena, California, USA\n\nWennberg et al. (2022c)\n\ndb\n\ndarwin01\n\nDarwin, Australia\n\nDeutscher et al. (2023a)\n\ndf\n\nedwards01\n\nAFRC, Edwards, CA, USA\n\nIraci et al. (2022b)\n\net\n\neasttroutlake01\n\nEast Trout Lake, Canada\n\nWunch et al. (2022)\n\neu\n\neureka01\n\nEureka, Canada\n\nStrong et al. (2022)\n\nfc\n\nfourcorners01\n\nFour Corners, NM, USA\n\nDubey et al. (2022b)\n\ngm\n\ngarmisch01\n\nGarmisch, Germany\n\nSussmann and Rettinger (2017a)\n\nhf\n\nhefei01\n\nHefei, China\n\nLiu et al. (2022)\n\nhw\n\nharwell01\n\nHarwell, UK\n\nWeidmann et al. (2023)\n\nif\n\nindianapolis01\n\nIndianapolis, Indiana, USA\n\nIraci et al. (2022a)\n\niz\n\nizana01\n\nIzana, Tenerife, Spain\n\nBlumenstock et al. (2017)\n\njc\n\njpl01\n\nJPL, Pasadena, California, USA\n\nWennberg et al. (2022e)\n\njf\n\njpl02\n\nJPL, Pasadena, California, USA\n\nWennberg et al. (2022a)\n\njs\n\nsaga01\n\nSaga, Japan\n\nShiomi et al. (2022)\n\nka\n\nkarlsruhe01\n\nKarlsruhe, Germany\n\nHase et al. (2022)\n\nlh\n\nlauder01\n\nLauder, New Zealand\n\nSherlock et al. (2022a)\n\nll\n\nlauder02\n\nLauder, New Zealand\n\nSherlock et al. (2022b)\n\nlr\n\nlauder03\n\nLauder, New Zealand\n\nPollard et al. (2022)\n\nma\n\nmanaus01\n\nManaus, Brazil\n\nDubey et al. (2022a)\n\nni\n\nnicosia01\n\nNicosia, Cyprus\n\nPetri et al. (2023)\n\nny\n\nnyalesund01\n\nNy-Ålesund, Svalbard, Norway\n\nBuschmann et al. (2022)\n\noc\n\nlamont01\n\nLamont, Oklahoma, USA\n\nWennberg et al. (2022d)\n\nor\n\norleans01\n\nOrleans, France\n\nWarneke et al. (2022)\n\npa\n\nparkfalls01\n\nPark Falls, Wisconsin, USA\n\nWennberg et al. (2022b)\n\npr\n\nparis01\n\nSorbonne Université, Paris, FR\n\nTe et al. (2022)\n\nra\n\nreunion01\n\nReunion Island, France\n\nMaziere et al. (2022)\n\nrj\n\nrikubetsu01\n\nRikubetsu, Hokkaido, Japan\n\nMorino et al. (2022a)\n\nso\n\nsodankyla01\n\nSodankylä, Finland\n\nKivi et al. (2022)\n\ntk\n\ntsukuba02\n\nTsukuba, Ibaraki, Japan, 125HR\n\nMorino et al. (2022b)\n\nwg\n\nwollongong01\n\nWollongong, Australia\n\nDeutscher et al. (2023b)\n\nxh\n\nxianghe01\n\nXianghe, China\n\nZhou et al. (2022)\n\nzs\n\nzugspitze01\n\nZugspitze, Germany\n\nSussmann and Rettinger (2017b)\n\nBeginning with GGG2020, experimental mid-IR data products will be available from select TCCON sites equipped with an\nInSb (indium antimonide) detector that enables measurements in the 1800 to 4000 cm−1 frequency range. Gases measured in\nthis range include, but are not limited to, O3 , N2 O, CO, CH4 , NO, NO2 , carbonyl sulfide, formaldehyde, and ethane. These\n95\n\nproducts offer the potential to extend the applications of TCCON data to new areas of research. However, it is important to\n\n5\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n104\n\n0\n\n(b)\n\nSlant XlCO2\n\nPressure (hPa)\n\n200\n400\n\n103\n\n600\n800\n\n(c)\n\n0\n\n0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75\nXlCO2 column AK (unitless)\n\n104\n\nSlant XwCO2\n\nPressure (hPa)\n\n200\n400\n\n103\n\n600\n800\n1000\n0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25\nXwCO2 column AK (unitless)\n\n0.8\n0.6\n0.4\n0.2\n0.0\n\n102\n\n(d)\nIntensity (arbitrary units)\n\n1000\n\n1.0\n\nIntensity (arbitrary units)\n\n(a)\n\n102\n\n4820\n\n1.00\n\n4840\n4860\nFrequency (cm 1)\n\n4880\n\n4900\n\n0.95\n0.90\n0.85\n0.80\n6040\n\n6050\n\n6060\n\n6070 6080\nFrequency (cm 1)\n\n6090\n\n6100\n\nFigure 1. Column averaging kernels (panels a, c) and calculated CO2 absorption lines (panels b, d) in the lCO2 (panels a, b) and wCO2\n(panels c, d) windows, respectively. The absorption lines in panels (b) and (d) are for a TCCON spectrum measured at solar zenith angle =\n39.684◦ in Jul 2004 at Park Falls, WI, USA. In panels (a) and (c), the different colors indicate AKs for different slant Xgas amounts. “Slant\nXgas ” is a measure of total absorber column along the light path. See §9.1 for details.\n\nnote that these data do not have any postprocessing corrections for airmass dependence (§7.1) or scaling to in situ data (§7.3)\napplied.\n3\n\nUpdated spectroscopy\n\n3.1\n100\n\nTelluric & Solar line lists\n\nThe telluric linelist (atm.161, Toon, 2022c) is a \"greatest hits\" compilation based heavily on HITRAN predecessor lists, but\nwith ad hoc empirical corrections performed to some lines, bands, and gases. The linelist is updated when improved linelists\nbecome available, as determined by 1) improved fits to laboratory and atmospheric spectra, 2) better consistency of retrieved\ngas amounts from different windows and bands, and 3) reduced airmass-dependence of the retrieved gas amounts. Since the\nGGG2016 version of the linelist, there have been many improvements to the H2 O and HDO spectroscopy throughout the main\n\n105\n\nTCCON region (4000 to 8000 cm−1 ).\nThe solar linelist (Toon, 2022b) is completely empirical, based on high-resolution solar spectra measured by various instruments from the ground, balloon, and space. In the 4000 to 8000 cm−1 spectral region covered by TCCON, the linelist is based\nprimarily on ground-based Kitt Peak and TCCON spectra, with additional balloon-borne MKIV spectra from 40 km altitude\nup to 5600 cm−1 . To deduce which absorption features are solar, rather than telluric, we fit out the telluric spectrum as best we\n\n110\n\ncan. Remaining dips in the residuals are solar, unless they grow with airmass, in which case they are missing tellurics. Since\nGGG2016 the improvements have been modest, adding new weak lines (< 0.1% depth) in the TCCON windows.\n\n6\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n3.2\n\nNon-Voigt lineshapes for O2 , CO2 , and CH4\n\nAbsorption coefficients calculations were improved in GGG2020. In previous versions of GGG absorption coefficients were\ncalculated using a Voigt spectral line shape. Numerous spectroscopic studies have shown that the Voigt line shape is insufficient\n115\n\nfor use with CO2 and other molecules, so a more sophisticated line shape is required to achieve the necessary retrieval accuracy.\nSo quadratic speed-dependent Voigt (qSDV) with line mixing (LM) code from Tran et al. (2013) was implemented into forward\nmodel of GGG (Toon, 2022a).\nIt was shown in Mendonca et al. (2016) that using the qSDV with first order LM and adopting the spectroscopic parameters\nfrom Devi et al. (2007b) for the CO2 lines in CO2 window centered at 6220 cm−1 and Devi et al. (2007a) for the window cen-\n\n120\n\ntered at 6339 cm−1 resulted in an up to 40% improvement to both spectral fit RMS and a reduction in the airmass dependence\nof the retrieved XCO2 . For the strong CO2 band lines, in the window centered at 4850 cm−1 , the spectroscopic parameters\nfrom Benner et al. (2016) are used with the qSDV and first order LM to calculate absorption coefficients. This resulted in\nimproving the quality of XCO2 retrievals from this spectral region. New spectroscopic studies aimed at improving CO2 absorption coefficient calculations are ongoing. Recent studies like Hashemi et al. (2020) that provide spectroscopic parameters\n\n125\n\nfor CO2 can be tested with TCCON spectra to see if the retrievals can be improved.\nTCCON CH4 is retrieved from three windows that are composed of the P, Q, and R branches of the 2ν3 CH4 band. To\nimprove the forward model of GGG the spectroscopic parameters from Devi et al. (2015, 2016) are used to calculate the\nabsorption coefficients with the qSDV with full line mixing. Unlike CO2 that uses first order line mixing requiring one extra\nparameter to be added to the linelist per spectral line, CH4 requires full line mixing. This requires spectroscopic parameters\n\n130\n\nfrom all coupled lines (i.e. a relaxation matrix) be used to calculate the effective spectral line parameters for each spectral line.\nIn previous versions of GGG absorption coefficients could only be calculated by reading in spectroscopic parameters line by\nline making it awkward to take into account full line mixing. GGG2020 has been updated to read in spectroscopic parameters\nand the relaxation matrix (supplied with Devi et al. (2015, 2016)) at the same time for spectral lines that require full line mixing.\nMore details on how this is done are provided in Mendonca et al. (2017). The improved absorption coefficient calculations for\n\n135\n\nCH4 lines for the 2ν3 CH4 band has improved the quality of the spectral fits and airmass dependence of the retrieved XCH4 .\nThe addition of full line mixing can be extended to other molecules to improve retrievals.\nTo improve the retrievals of O2 columns, which are required to calculate Xgas , spectroscopic parameters for the O2 singlet\ndelta band were retrieved by fitting cavity ring down spectra as detailed in Mendonca et al. (2019). The spectroscopic parameters derived from the cavity ring down spectra were tested on TCCON spectra where they were shown to slightly improve the\n\n140\n\nquality of the spectral fit as well as greatly decrease the airmass dependence of the retrieved O2 column. The study by Mendonca et al. (2019) is the first to show the need for a spectral line shape that takes into account speed-dependence. Since then,\nnewer spectroscopic studies such as Tran et al. (2020) and Fleurbaey et al. (2021) have shown the need to take into account\nDicke narrowing and line mixing in order to fit new cavity ring down spectra in the O2 singlet delta band. The spectroscopic\nparameters of Mendonca et al. (2019), Tran et al. (2020), and Fleurbaey et al. (2021) were used to fit TCCON O2 spectra in\n\n7\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n145\n\nTran et al. (2021). The study showed that the newer spectroscopic parameters slightly improved the quality of the spectral fit\nbut they should also be assessed on how they impact the airmass dependence of retrieved O2 columns.\n3.3\n\nEmpirical optimization of O2 line widths\n\nDuring pre-release testing, we found that a diagnostic quantity we call Xluft had a noticeable temperature dependence (Fig.\n2a). Xluft is a ratio of two ways of calculating the column of dry air (one from surface pressure and the a priori H2 O profile,\n150\n\nand one from the column of O2 retrieved in the singlet delta band—or put another way, it is the column-average mole fraction\nof dry air), and thus should not have a temperature dependence. Since dry mole fractions of O2 in the atmosphere are highly\nconstant over space and time, this implied that either temperature-dependence or the water broadening of the O2 line widths in\nthe forward model was incorrect, as the concentration of water in the atmosphere is generally correlated with temperature.\nTo disentangle the effect of temperature and water, we first examined data from the Darwin, Australia TCCON station.\n\n155\n\nDarwin is located in the tropics, and so experiences greater water columns and a narrower range of temperatures than other\nTCCON sites (Fig. 3a,b). We chose approximately 14 months of data from Darwin when the instrument was performing well,\nand processed that year three times, with water broadening set to 1.0, 1.4, and 1.8 times that of the air broadening half width.\nTo identify the optimal strength for water broadening, we examined the slope of Xluft vs. water column in 10° SZA bins\nfor each of these tests. Binning the data by SZA helps to separate the water dependence from airmass dependence. Figure 3c\n\n160\n\nshows that a water broadening of 1.4 times that of air minimized the dependence of Xluft on water.\nWith the water broadening optimized, we turned to the temperature dependence of the O2 line widths. Reducing the dependence of Xluft on temperature was the primary goal; however, we had to account for the interplay between the temperature and\npressure dependence. In particular, our concern was that changing the temperature dependence of the O2 line widths would\nintroduce or increase an SZA dependence by changing the average line widths.\n\n165\n\nOur solution was to simultaneously adjust both the temperature and pressure dependence of the O2 line widths. To find the\noptimal combination of these coefficients, we minimized a cost function of three quantities. For each quantity, we tested how\nthe results changed using a different collection of TCCON sites:\n1. The average magnitude of the Xluft vs. temperature at 700 hPa (T700) slope across various combinations of 1–3 of the\nEast Trout Lake, Lamont, and Park Falls sites.\n\n170\n\n2. The variance of the Xluft vs. SZA slopes across the Darwin, East Trout Lake, Lamont, and Park Falls sites.\n3. The variance of the magnitude of Xluft across the same sites as #2.\nOur rationale was that the temperature dependence of Xluft was the most important error to eliminate, thus minimizing its\nmagnitude took priority. T700 is taken from the a priori meteorology data and was chosen as a useful metric of synoptic-scale\nchange (Keppel-Aleks et al., 2011). We then minimized the variance in slopes of Xluft vs. SZA across different TCCON sites\n\n175\n\nbecause GGG already has a well-tested program to remove spurious SZA dependencies in the output Xgas products, so long as\nthose dependencies are the same across sites. While minimizing the magnitude of the SZA dependence itself would have been\n8\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n(a) 1.005\n\ny = 1.03 + -0.00013x\n\n102\n\n0.995\n\nCounts\n\nXluft (unitless)\n\n1.000\n\n101\n0.990\n0.985\n0.980\n\n100\n\n(b) 1.015\n\ny = 1.01 + -3.31e-05x\n\n102\n\n1.005\n\nCounts\n\nXluft (unitless)\n\n1.010\n\n101\n1.000\n0.995\n0.990\n\n260\n\n270\nT700 (K)\n\n280\n\n290\n\n100\n\nFigure 2. Correlation between Xluft and temperature at 700 hPa (a) before and (b) after optimizing the O2 line broadening in terms of\nits water, pressure, and temperature dependencies. Note that (a) is not from the previous TCCON data version (GGG2014), it is from a\npreliminary beta test of GGG2020. In both panels, the colored background is a 2D histogram, the gray diamonds mark the mean Xluft in 5\nK bins, and the black line is a linear fit to the gray diamonds. The data shown here is from the Lamont TCCON site between 2 Sep 2017 to\n30 Sep 2018. Note that the y-axis limits shift between the panels; this is because the mean magnitude of Xluft changed with the increase of\nO2 line intensities (see text) between the tests plotted in the two panels. The slope is visually comparable between the panels, since the span\nof Xluft is the same (0.025) in both panels.\n\n9\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n# spectra per bin\n\n(a)\n\nDarwin, Australia\nLamont, OK, USA\n\n15000\n10000\n5000\n0\n\n255\n\n260\n\n265\n\n6000\n5000\n4000\n3000\n2000\n1000\n0\n\n280\n\n285\n\n290\n\n0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75\n1e23\nH2O column (molec. cm 2)\nH2O +0%\n1e 26\n\n(c)\n\nSlope of Xluft vs. H2O column\n( 1/[molec. cm 2] )\n\n270 275\nT700 (K)\n\nT700\n\nH2O +40%\n\nH2O +80%\n\n[281.0 K, 285.0 K]\n\n0.5\n\n8000\n\n0.0\n\n6000\n\n# spectra\n\n# spectra per bin\n\n(b)\n\n0.5\n\n4000\n\n1.0\n\n2000\n\n1.5\n0\n\n10\n\n20\n\n30\n\n40 50\nSZA\n\n60\n\n70\n\n80\n\n0\n\nFigure 3. (a) Histogram of temperatures at 700 hPa at the Darwin (located at 12.5° S) and Lamont (at 36.6° N) TCCON sites. (b) Histogram\nof water column amounts at the same sites. (c) Slopes of Xluft vs. water column in 10° SZA bins at Darwin with water broadening of O2 set\nas equal to, 40% greater, and 80% greater than air. The grey bars give the number of spectra in each bin. The Lamont data in (a) and (b) is\nfrom the period 2 Sep 2017 to 30 Sep 2018, and the Darwin data in all bins is from 21 Jul 2015 to 30 Sep 2016.\n\n10\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nNot optimal\n\nOptimal\n\nTest parameter / original value\n\n1.50\n1.25\n1.00\n0.75\n0.50\n0.25\n0.00\n0.25\n\ns.\npe v\n\n0\n\nT70\n\nSlo\n\nA)\n. SZ\ns\nv\npe\n2 (Slo\n\nuft)\n\n2 ( Xl\n\nGGG2020\n\n(c)\n\nO2 T-dependent pres. broadening half-width\n\n(b)\nO2 air broadening half-width (cm 1/atm)\n\n(a)\n\nGGG2014\n0.055\n0.050\n0.045\n0.040\n0.035\n0.030\n0.025\n\n7800 7900 8000\nFrequency (cm 1)\n\n0.825\n0.800\n0.775\n0.750\n0.725\n0.700\n0.675\n7800 7900 8000\nFrequency (cm 1)\n\nFigure 4. Result of the O2 spectroscopy optimization. (a) The values of each criterion for each test using different values of pressure and\ntemperature broadening coefficients. The values are normalized to their values in the baseline test (before optimizing the O2 spectroscopy).\nThe points within each parameter are spread horizontally for clarity. (b) The air broadening half widths used in GGG2020 (after optimization)\ncompared with GGG2014. The mean GGG2020/GGG2014 ratio is 1.0025, so the points are barely different on this scale. (c) As (b), but for\nthe temperature broadening coefficient. The mean GGG2020/GGG2014 ratio is 0.9323.\n\npreferable, we were not certain there would be enough flexibility in the Xluft -O2 spectroscopy relationship to simultaneously\nminimize the temperature and SZA dependencies. Similarly, we minimized the variance in Xluft itself because the average\nmagnitude of Xluft depends on the strengths of the O2 lines, rather than the pressure and temperature effects on line width\n180\n\nadjusted in this initial experiment.\nTo carry out this optimization, we ran approximately one year of data from four TCCON sites (Darwin, Australia; East Trout\nLake, Canada; Lamont, OK, USA; Park Falls, WI, USA) multiple times. In each test, we scaled the temperature dependence,\npressure dependence, or both of all lines in the O2 band. We could then interpolate between these test runs to estimate the three\ncost function quantities for any pressure/temperature broadening coefficients, and from that find the combination of coefficients\n\n185\n\nthat minimized the overall cost function. Note that we did not use Darwin data to calculate the Xluft versus T700 slopes for the\ncost function, as the small range of temperatures that Darwin experiences (Fig. 3a) make it difficult to get reliable fits versus\ntemperature.\nThe results of the optimization are shown in Fig. 4. Figure 4a shows how the three criterion described above (slope of Xluft\nvs. T700, variance in slope of Xluft vs. SZA, variance in Xluft ) varied across the tests performed with different pressure and\n\n190\n\ntemperature broadening coefficients. The values are normalized to their respective pre-optimization values. We found that the\nbest combination of coefficients reduced the slope of Xluft vs. T700 by 82%, the variance in Xluft vs. SZA slopes across\nTCCON sites by 89%, and the variance in Xluft itself by 49%. The optimized air broadening half widths and temperature\ndependence coefficients for GGG2020 are shown in Fig. 4, panels b and c respectively, with GGG2014 values for comparison.\n\n11\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nThe air broadening half widths were increased by 0.25% and the temperature dependence coefficients were decreased by\n195\n\n6.77%. The effect on the Xluft vs. T700 relationship is shown in Fig. 2b, where the slope is reduced by a factor of 4 compared\nto its pre-optimization value.\nFinally, the O2 line intensities were increased by ∼ 1% to bring Xluft closer to 1. This effect is apparent in Fig. 2, where the\n\npost-optimization Xluft in panel b is near 1, but the pre-optimization values are between 0.990 and 0.995.\n4\n200\n\nImproved a priori profiles\n\n4.1\n\nModified retrieval grid\n\nIn GGG the retrieval is done on a fixed altitude grid. In GGG2014 the altitude grid had a constant spacing of 1 km with 71\nlevels between 0-70 km above sea level. In GGG2020 the grid was updated to 51 levels between 0-70 km above sea level with\nspacing increasing away from the surface following the expression:\n\nzi = i · (0.4 + 0.02 · i)\n205\n\n(2)\n\nwhere zi is the altitude of the ith level in kilometers. As the altitude grids are fixed to sea level, this does mean that some\nsites have some levels below the terrain which are not included in the integration.\n4.2\n\nMeteorological updates\n\nIn GGG2014 the a priori H2 O, pressure, density, and temperature profiles were derived from NCEP 6-hourly reanalyses. In\nGGG2020, these profiles are now derived from GEOS 5 FP-IT 3-hourly product in addition to potential temperature, potential\n210\n\nvorticity, O3 , and CO profiles. The potential vorticity profiles are used to derive equivalent latitude profiles based on the\nequation in Allen and Nakamura (2003). Equivalent latitude is used in deriving the stratospheric part of the a priori trace gas\nconcentration profiles (Laughner et al., 2023). GGG2020 will transition to the GEOS IT product when it replaces GEOS FP-IT;\nan analysis to quantify the impact of that change on TCCON Xgas products is planned.\n4.3\n\n215\n\nTrace gas profile updates\n\nGGG2020 includes a substantial redesign of the algorithm that generates the CO2 , CH4 , N2 O, HF, CO, and O3 a priori\nprofiles. Generating these profiles is now handled by ginput, a separate program from gsetup. The ginput algorithm is described\nin detail in Laughner et al. (2023). Briefly, the CO2 , CH4 , and N2 O profiles are tied to the long term records from the NOAA\nobservatories in Mauna Loa, Hawaii and American Samoa (Lan et al., 2022b, a, c), in order to ensure the growth rates of these\ngases are correctly accounted for. Individual profiles are produced based on the mean transport time between the profile location\n\n220\n\nand the Mauna Loa/American Samoa observatories and (in the stratosphere) chemical loss. HF profiles are derived from CH4\nprofiles using the HF-CH4 relationships previously identified by Washenfelder et al. (2003) and Saad et al. (2014, 2016). CO\n12\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nprofiles are drawn from the GEOS FP-IT chemical product2 (Lucchesi, 2015) with adjustments in the stratosphere to better\nmatch observations. (See Laughner et al. (2023) for details on these adjustments.)\nOne additional change compared to GGG2014 is that the a priori profiles are now given in wet, rather than dry, mole fraction.\n225\n\nThis is necessary as GGG calculates absorber number densities as the prior mole fractions times the number density of air,\nwhich is assumed to include water. The a priori profiles provided in the published data files are also in wet mole fraction. Thus,\nwhenever comparing GGG2020 a priori profiles in the published netCDF files with other sources, care must be taken to ensure\nthat the comparisons convert both profiles to the same (wet or dry) mole fractions.\n5 Improved interferogram-to-spectrum conversion\n\n230\n\nThere have been substantial code changes and streamlining of common code in i2s, the interferogram-to-spectrum conversion\nsubroutine. The main substantive improvements to the code are in the handling of detector nonlinearity, the phase correction,\nand other changes.\n5.1\n\nDetector nonlinearity\n\nThe largest signals in an interferogram generated by a Fourier transform spectrometer are found near zero-path difference\n235\n\n(ZPD), where light from all wavelengths is in phase. The signal levels drop significantly away from ZPD. If the detector\nmeasuring the interferogram has a nonlinear response, the variations in the signal near ZPD will be more distorted than in\nthe rest of the interferogram. This causes a discrepancy between the low-resolution spectral envelope and the high resolution\nspectral lines. Nonlinear detector responses can be strongly pronounced or subtle, and several improvements to i2s have been\nmade to address these situations.\n\n240\n\nWe have implemented a check early in i2s processing to remove interferograms affected by detector or signal chain saturation, an extreme form of detector nonlinearity. If the signal intensity is too large, the ZPD signal will reach the maximum\nvalue permitted by the detector electronics, and no additional light can be detected. We call this “detector saturation” and this\ncauses irreversible detector nonlinearity in which spectral information is permanently lost. To resolve the problem, detectors\nused for TCCON measurements have reduced pre-amplifier gain settings. Additionally, we must limit the number of photons\n\n245\n\nincident on the detector through reducing the field stop or aperture stop diameter or by placing an optical filter in the beam.\nMeasured interferograms that are saturated cannot be recovered, and therefore should be discarded before producing their\nspectra. Because this effect depends on sunlight intensity, saturation could occur near noon but not later or earlier in the day,\nit can be seasonally dependent, or dependent on the amount of water vapor in the atmosphere. In GGG2020, we have implemented a detector saturation check to discard any saturated interferograms based on the maximum and minimum values of the\n\n250\n\ninterferogram signal.\nWe now compute and store a detector nonlinearity diagnostic variable (“DIP”) as part of the regular TCCON data processing.\nKeppel-Aleks et al. (2007) described the solar intensity variation correction applied to the TCCON interferograms that has been\n2 We\n\nexpect to transition to the GEOS IT product when it supersedes GEOS FP-IT. However, that had not yet occurred at time of writing.\n\n13\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\npart of the TCCON processing software since 2007. In this correction, a low-pass filtered interferogram is used to re-weight\nthe original interferogram, largely removing the impacts of solar intensity fluctuations during a measurement. As part of this\n255\n\nwork, Keppel-Aleks et al. realized that detector nonlinearity becomes observable in the low-pass filtered interferogram as a\n“dip” at zero path difference (see Fig. 6b in Keppel-Aleks et al., 2007). The magnitude of this dip is a diagnostic of the severity\nof detector nonlinearity, and is now computed, stored, and reported as part of the routine TCCON processing.\nA subtle detector nonlinearity in the Sodankylä TCCON data persisted from early in their record until the problem was\nfound in 2017. The problem in the early data was resolved by applying the nonlinearity correction developed by Hase (2000)\n\n260\n\ndirectly to the interferogram before transforming it into a spectrum. This correction process and its results are described in\ndetail in Appendices A and B of Sha et al. (2020). In that paper, the authors show that the nonlinearity caused a bias in XCO2\nof about 0.5 ppm in the 2017 Sodankylä data. After 2017, the problem was resolved by optically limiting the light entering\nthe interferometer. This correction process is being applied to GGG2020 data at other sites for periods when significant nonlinearity is identified. We are in the process of incorporating the correction process as a standardized part of the interferogram-\n\n265\n\nto-spectrum processing to make this process easier to complete in the future.\nA second class of nonlinearity results in supralinear detector response, rather than sublinear response as was seen at Sodankylä. The correction procedure described in the last paragraph is not effective at correcting the supralinear behavior as it\nhas a different physical cause than the sublinear behavior. Based on tests performed at the Garmisch TCCON site, our current\nhypothesis is that this behavior results from overfilling the detector element with the light beam (Corredera et al., 2003), and\n\n270\n\nthe magnitude of the effect varies from detector to detector. Another possible cause of supralinearity in detectors can come\nfrom absorptive layers on the InGaAs active region itself (Fox, 1993), but we do not yet have evidence that this is occurring in\nour instruments.\n5.2\n\nPhase correction\n\nSampled interferograms are always asymmetrical, either because the sampling grid does not include the ZPD position, or\n275\n\nbecause the under-lying continuous igram is already asymmetrical even before it is sampled. This asymmetry causes the\nresulting, post-FFT, complex spectrum to have substantial imaginary terms. A phase correction is necessary to resample the\ninterferogram such that is it sampled symmetrically about ZPD, resulting in a computed spectrum that has the signals of interest\n\n280\n\nin the real component and only the noise is divided between both the real and imaginary component.\n√\nIf we used a power spectrum ( <2 + =2 ), avoiding phase correction, it would compute a spectrum that is entirely real, but\n\nwould retain all of the noise in the real and imaginary component of the spectrum. Therefore the final noise level in a power\n√\nspectrum would be a factor of 2 greater than in a phase-corrected and Fourier transformed spectrum. Additionally, in a power\nspectrum, saturated (zero intensity) regions would no longer be centered at zero, as any noise present is rectified and so made\nall positive. For these reasons, we compute a phase correction.\nWe use the phase correction method described by Forman et al. (1966), with a spectral domain convolution as described by\n\n285\n\nMertz (1965, 1967). The phase correction is performed using a low resolution double-sided interferogram, apodized with a cos2\nfunction, to compute the angle between the real and imaginary components of the spectrum. This angle is a smoothly varying\n14\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nfunction of wavenumber, and is called the phase curve. Its counterpart in interferogram space is called the phase correction\noperator. In regions of the spectrum with sufficient signal, the phase curve well defined, but where the spectrum is blacked out\nby water vapor, another strong absorber, or an optical component, it can become undefined. Therefore, to compute the phase\n290\n\ncorrection operator, we need to set a signal threshold so that we can compute a well-behaved phase curve across the spectral\nregion of interest. We interpolate the phase curve linearly across the blacked-out regions of the spectrum where the phase curve\nis below the signal threshold. The phase curve is interpolated to 0 at 0 cm−1 and at the Nyquist frequency (15798 cm−1 ).\nIn GGG2014, several TCCON stations showed retrievals of Xgas with systematic differences between spectra generated\nfrom interferograms collected while the scanning mirror moves away from zero path difference (“forward” scans) and while\n\n295\n\nmoving toward zero path difference (“reverse” scans). These differences are typically less than 0.5 ppm in XCO2 , but with\nlarger differences observed at the Ny Ålesund, Eureka, Paris, and Zugspitze TCCON stations. This forward-reverse bias was\ntracked down to the phase correction operator, and, more specifically, the minimum signal level threshold for which the phase\noperator is calculated. We have lowered the phase curve threshold from 0.02 (2%, in GGG2014) to 0.001 (0.1%, in GGG2020)\nof the peak spectral signal which improves the consistency between forward and reverse scans. This does not completely\n\n300\n\nresolve the problem, and we hope to develop a future version with a phase correction scheme that is independent of the signal\nlevel.\n5.3\n\nOther i2s changes\n\nWe now make better use of the entire interferogram collected by the spectrometer in i2s. In typical linear single-passed Fourier\ntransform spectrometers, we collect most of our interferometric data between zero path difference (ZPD) and the maximum\n305\n\noptical path difference (MOPD) positions of the scanning mirror. However, in order to perform a phase correction, a small\namount of data must be collected on the other side of ZPD, which we call the “short arm” of the interferometer. The “long\narm” is the section from ZPD to MOPD. In previous versions of GGG, the short arm data were discarded after the phase\ncorrection was completed. We now use the short arm data along with the long arm data to compute the spectrum. This is a\nmore efficient use of the data collected.\n\n310\n\n6\n\nContinuum fitting\n\nTCCON spectra are a combination of narrow features due to solar and telluric absorptions superimposed on the much broader\nspectral responses of the instrument and the solar Planck function (the continuum). To accurately fit the telluric features\nof interest, all other components of the spectrum must be accurately modelled simultaneously. Since TCCON spectra are\nnot radiometrically calibrated, the continuum can vary from instrument to instrument or even from day to day (if optical\n315\n\ncomponents are inserted or replaced) and therefore a general approach was needed to model the continuum. Prior to GGG2014,\nthe continuum was fitted with only two terms (mean and slope) over the <100 cm−1 wide windows used to retrieve atmospheric\ngases. To make use of wider spectral windows, it became necessary to include additional complexity in the model of the\ncontinuum, to account for optical components within the instrument (e.g., detectors, optical filters, beamsplitter, etc.) that\n15\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\ninduce curvature in the spectral response (e.g., Kiel et al., 2016b). In GGG2014, we implemented the ability to fit higher order\n320\n\npolynomials to the continuum level using discrete Legendre polynomials for test purposes, although this capability was not\nuniformly used in the GGG2014 TCCON data processing (Wunch et al., 2015). Higher order polynomials are now used widely\nin the GGG2020 spectral windows to better account for continuum shape changes between instruments and over time. The\ncontinuum curvature fitting option is not intended to fit out spectroscopic deficiencies; they will be airmass-dependent and\nso should be fixed separately. The default polynominal order in GGG2020 for each window has been chosen to capture the\n\n325\n\ncontinuum shapes of all sites in GGG2020 and reduce the spectral residuals without over-fitting the spectrum.\n6.1\n\nChannel Fringe Fitting\n\nParallel optical surfaces delay a small fraction of the transmitted beam, which subsequently interferes with the main, un-delayed\nbeam, resulting in a small periodic modulation of the spectral transmittance. This modulation has an amplitude of R2 where R\n330\n\nis the reflectivity of each surface, and a period of (2 · n · d · cos θ)−1 cm−1 , where n in the refractive index of the optic, d is its\n\nthickness (in cm) and θ is the angle to the normal.\n\nFor decades, GFIT has the capability to fit a channel fringe to determine its amplitude (as a fraction of the continuum), its\nperiod, and its phase, and then remove it from the measured spectrum during the spectral fitting. This capability was not used\nby TCCON until GGG2020, when spectral fits from some sites were noticed to exhibit the tell-tale periodicities in the residuals.\nLeft untreated, channel fringes can seriously bias the retrieved gas amounts, by an amount that can vary from instrument to\n335\n\ninstrument and even over time for a single instrument, e.g., if its temperature changes.\nAn important code change for GGG2020 was to prevent channel fringes from being mistaken for higher-order continuum\nterms. This was much less of a problem for GGG2014 when we only ever fitted a straight line to represent the continuum. But\nnow, if a particular wavelike feature in the continuum could be fitted by a higher order polynomial or by a channel fringe, this\ntends to slow down convergence as the continuum fitting and channel fringe fitting vie with each other. To prevent this, a lower\n\n340\n\nlimit was imposed on the channel fringe period that was fittable in a given window, such that it always was narrower than the\nperiodicities in the continuum fitting polynomial. So if we are fitting an N -term polynomial to the spectrum (called the number\nof continuum basis functions, or NCBF ), in a window of width w cm−1 , then the period of the fitted channel fringes must be\nless than w/(NCBF − 1).\n7\n\n345\n\nPost-retrieval data processing\n\nGGG incorporates several post-retrieval steps to (1) collate and average data (§7.2) from the individual retrieval windows into\nthe final Xgas products and (2) correct post hoc for known errors in the forward model. There are two corrections. The first is\nan airmass-dependent correction (§7.1), which aims to eliminate spurious dependence of Xgas quantities on SZA. The second\nis an in situ-based, or airmass-independent correction (§7.3), which aims to eliminate the mean bias in Xgas values arising\nfrom incorrect spectroscopic line strengths.\n\n350\n\nIn the following sections, the post processing steps are presented in the order in which they are applied in GGG2020.\n16\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n7.1\n\nUpdated airmass dependence correction\n\nIn the limit of no horizontal variation in trace gas dry air mole fraction, Xgas quantities are independent of atmospheric path\nlength, as the change in column density due to path length is multiplicative and so will cancel out between the target gas in\nthe numerator and O2 in the denominator. However, a spurious dependence of Xgas on airmass can arise from errors in the\n355\n\nspectroscopic forward model.\nGGG2020, like GGG2014, applies a post hoc correction to the Xgas values to remove airmass dependences. We calculate a\ncorrection for each Xgas value as\n\nfc =\n\n\u0012\n\nθ+g\n90 + g\n\n\u0013p\n\n−\n\n\u0012\n\n45 + g\n90 + g\n\n\u0013p\n\n(3)\n\nand use this to correct the Xgas value as\n\n360\n\nXgas,corr =\n\nXgas,raw\n1 + αfc\n\n(4)\n\nIn Eq. (4), α is a coefficient for each gas (in GGG2014) or each window (in GGG2020). In Eq. (3), θ is the solar zenith angle\n(SZA) in degrees and g and p are coefficients chosen to best represent the SZA-dependent behavior. This form was chosen to\nnormalize to a 90° window centered on (45 + g)◦ . While the basic approach is the same in GGG2020 as it was in GGG2014,\nwe made two changes to the implementation:\n365\n\n1. In GGG2014, column densities from different spectral windows used to retrieve a target gas were averaged first, then\na single airmass correction applied to each gas. In GGG2020, each spectral window is airmass corrected first, then the\nresulting Xgas values are averaged.\n2. In GGG2014, g = 13 and p = 3 for all gases. In GGG2020, different values of g and p were selected for each window.\nThe rationale for the first change is clear from Fig. 5. The standard TCCON CO2 and CH4 products are derived from two\n\n370\n\nand three spectral windows, respectively. Although the overall SZA dependence has a similar shape for all windows of a given\ngas, there are clear differences in low and high SZA behavior. Thus, we decided to apply an SZA dependent correction to\nindividual windows, rather than the average Xgas value.\nThe rationale for the second change is that we do not know a priori the best form to represent the airmass dependence in\nany given window. For GGG2020, we used data from the Darwin TCCON site for all of 2015 to choose the values of g and\n\n375\n\np for each window. We used Darwin because, as a tropical site, it sees a wide range of SZAs (useful for examining SZA\ndependence) and water columns (useful to check for water effects on the derived airmass dependence). We used 2015 data\nbecause the instrument at Darwin was well aligned during that year.\n\n17\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nXCO2 - daily medians (ppm)\n\n(a)\n\n0.5\n0.0\n0.5\n1.0\nCO2 6220\nCO2 6339\n\n1.5\n\nXCH4 - daily medians (ppb)\n\n(b)\n2\n0\n2\nCH4 5938\nCH4 6002\nCH4 6076\n\n4\n6\n10\n\n20\n\n30\n\n40 50 60\nSZA (degrees)\n\n70\n\n80\n\nFigure 5. Variation of (a) the two CO2 and (b) three CH4 windows used by TCCON with SZA without the airmass correction applied.\nIn both panels, the y-axis is column average dry mole fraction of CO2 or CH4 derived from a single spectral window, with the central\nwavenumber given in the legend. The y values have the daily median values subtracted (to remove day-to-day variability), and each point\nrepresents the median of all such values in a 5° SZA bin.\n\nTo understand how g and p were determined, we must first explain how the airmass dependent correction factor (ADCF, i.e.\nα in Eq. 4) is calculated for a given g and p. The ADCF is calculated by fitting the following function to each day’s data:\n\n380\n\nf (t, θ|c1 , c2 , c3 ) = c1 + c2 · (2π(t − tnoon )) + c3 fc\n\n(5)\n\nwhere t and tnoon are the measurement time and solar noon time (in day of year), fc is the polynomial defined in Eq. (3), and\nc1 , c2 , and c3 are the fitted coefficients. The coefficients and their errors are calculated with a weighted least squares fit using\nthe individual windows’ Xgas uncertainties as the weights. The ADCF for a given window is the error-weighted mean of all\ndays’ c3 values.\n385\n\nTo find the optimal g and p values, we derived ADCFs for five subsets of the 2015 Darwin data (data with SZA > 20°, 30°,\n40°, 50°, and 60°, all with H2 O column < 1.1 × 1023 molec. cm−2 ) for values of g between −45 and +45 and p between 1\nand 6. We then find the combination of g and p that gives the smallest standard deviation across all five subsets and choose that\n18\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n0.005\n0.010\n0.015\n0.020\n\n20\n\n25\n\n30\n\n35 40 45\nSZA min (deg)\n\n50\n\n55\n\n40\n20\n10\ng\n\n0.009474\n0.008421\n0.007368\n0.006316\n0.005263\n0.004211\n0.003158\n0.002105\n0.001053\n0.000000\n\n0\n10\n20\n30\n40\n1\n\n2\n\n3\n\np\n\n4\n\n5\n\n6\n\n1\n\n0.005\n0.010\n0.015\n0.020\n\n60\n\n20\n\n25\n\n30\n\n35 40 45\nSZA min (deg)\n\n50\n\n55\n\n40\n\nxco2_6220 ADCF std. dev.\n\n30\n\nCO2 from window centered on 6339 cm\n\n0.000\nxco2_6339 ADCF\n\nxco2_6220 ADCF\n\n1\n\n30\n\n60\n0.009474\n0.008421\n0.007368\n0.006316\n0.005263\n0.004211\n0.003158\n0.002105\n0.001053\n0.000000\n\nxco2_6339 ADCF std. dev.\n\nCO2 from window centered on 6220 cm\n\n0.000\n\n20\n10\n0\n10\n20\n30\n40\n1\n\n2\n\n3\n\np\n\n4\n\n5\n\n6\n\nFigure 6. Example of how g and p in Eq. (3) were chosen for the two TCCON CO2 windows. The left two panels are for the CO2 window\ncentered at 6220 cm−1 and the right two for the window at 6339 cm−1 . The line plots at the top show how the value of the ADCF (α in\nEq. (4)) changes as we increase the lower limit in SZA for the data fit to. Each gray line represents one combination of g and p, with the\nblack line representing the combination with the smallest standard deviation in the ADCF. The contour plots show the standard deviation\nof the ADCF across different minimum SZAs for each combination of g and p. The white star represents the combination with the smallest\nstandard deviation; it corresponds to the test show with the black line in the line plots.\n\nas the optimal combination. This approach assumes that the values of g and p (and thus the form of fc ) which best capture the\nairmass dependence of a particular window will have the smallest change in ADCF as smaller subsets of data are fit.\n390\n\nThis procedure is illustrated for the two TCCON CO2 windows in Fig. 6. In the top panels, the gray lines show the variation\nin ADCF with the minimum SZA in the subset of data fit to; each line represents one combination of g and p. It is clear that the\nvariation in ADCF is much greater for some combinations of g and p than others. The contour plots in Fig. 6 show the standard\ndeviation of ADCF for each g and p combination. In both windows, there is a clear minimum valley. The white stars in the\ncontour plots and thicker black lines in the upper panels show the g and p combination with the smallest standard deviation.\n\n395\n\nThe final step in selecting ADCFs for GGG2020 was to account for spurious temperature dependence in the Xgas values. As\nwe saw with O2 in §3.3, incorrect temperature dependence in the line widths introduces a temperature dependence in retrieved\nXgas , which could alias into the airmass dependence. To check this, we derived ADCFs from data from 18 TCCON sites, using\ntwo month long subsets of data to sample different temperatures. Figure 8 shows how the CH4 ADCFs vary with potential\ntemperature averaged between 500 and 700 hPa (θmid ) as an example. (Figure 7 shows how θmid and T700 relate to assist\n\n400\n\ncomparisons with Fig. 2.) Here, we see that the 6002 cm−1 and 6076 cm−1 windows’ ADCFs have no or little temperature\ndependence (Fig. 8b,c), but the 5938 cm−1 window has a clear temperature dependence. To compute the final ADCFs for\neach window, we used the value of the fit to this data at θmid = 310 K. 310 K was chosen as it is approximately the midpoint\ntemperature for the TCCON network, as can be seen in Fig. 8.\n\n19\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n320\n\n101\n\n280\n\n#\n\nmid\n\n(K)\n\n300\n\n260\n240\n\n240\n\n260\n\n280\n300\nT700 (K)\n\n320\n\n100\n\nFigure 7. A heatmap of the relationship between θmid and T700, taken from the Park Falls TCCON data. The red dashed line denotes the\n1:1 line.\n\nADCF\n\n(a) 0.015\n\nADCF\n\n1\n\n0.010\n0.005\n0.000\n0.005\n0.010\n0.015\ny = (0.0711 +/- 0.00729) + (-0.000261 +/- 2.35e-05)x\n0.020\n290\n295\n300\n305\n310\n315\n\n(b) 0.015\n\nCH4 from window centered on 6002 cm\n\n320\n1\n\n0.010\n0.005\n0.000\n0.005\n0.010\n0.015\ny = (-0.0085 +/- 0.00788) + (7.99e-06 +/- 2.53e-05)x\n0.020\n290\n295\n300\n305\n310\n315\n\n(c) 0.015\n\nADCF\n\nCH4 from window centered on 5938 cm\n\nCH4 from window centered on 6076 cm\n\n0.010\n0.005\n0.000\n0.005\n0.010\n0.015\ny = (0.0356 +/- 0.00698) + (-0.000134 +/- 2.24e-05)x\n0.020\n290\n295\n300\n305\n310\n315\nmid (K)\n\n320\n1\n\nBialystok\nBurgos\nPasadena\nDarwin\nArmstrong\nEast Trout Lake\nGarmisch\nJPL 01\nSaga\nKarlsruhe\nLauder 03\nLamont\nOrleans\nPark Falls\nReunion Island\nRikubetsu\nTsukuba 02\nWollongong\n\n320\n\nFigure 8. ADCFs derived from two month periods from 18 sites throughout the TCCON network versus mean potential temperature between\n500 hPa and 700 hPa over the same two month period. Each panel is one of the TCCON CH4 windows. The text inset in each panel gives\nthe intercept and slope of the robust fit through the data shown by the black dashed line.\n\n20\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nADCF\n\n0.2\n\nCO from window centered on 4233 cm\n\n1\n\n(b)\n\n0.1\n\n0.1\n\n0.0\n\n0.0\n\n0.1\n0.2\n0.3\n\n0.1\n\n0.3\n\n320\n\nwCO2 from window centered on 6073 cm\n\n1\n\n0.000\n\n0.000\n0.005\n\n0.010\n0.015\n\ny = (0.133 +/- 0.0715) + (-0.000486 +/- 0.000229)x\n290\n295\n300\n305\n310\n315\n320\n\nwCO2 from window centered on 6500 cm\n\n(d) 0.005\n\n0.005\n\n0.020\n\n1\n\n0.2\ny = (-0.448 +/- 0.139) + (0.00132 +/- 0.000446)x\n290\n295\n300\n305\n310\n315\n\nADCF\n\nADCF\n\n(c) 0.005\n\nCO from window centered on 4290 cm\n\n0.2\n\nADCF\n\n(a)\n\n1\n\nBialystok\nBurgos\nPasadena\nDarwin\nArmstrong\nEast Trout Lake\nGarmisch\nJPL 01\nSaga\nLauder 03\nLamont\nOrleans\nPark Falls\nReunion Island\nRikubetsu\nTsukuba 02\nWollongong\n\n0.010\n0.015\n\ny = (0.000425 +/- 0.00238) + (-8.95e-06 +/- 7.64e-06)x\n290\n295\n300\n305\n310\n315\n320\n\n0.020\n\ny = (0.0225 +/- 0.00412) + (-0.000104 +/- 1.33e-05)x\n290\n295\n300\n305\n310\n315\n320\nmid (K)\n\nFigure 9. Similar to Fig. 8, except for two CO windows (a, b) and two weak CO2 windows (c, d).\n\nThe magnitude of this temperature dependence varies from gas to gas: the primary TCCON CO2 windows have almost\n405\n\nno slope, while the N2 O windows have slopes of ADCF vs. θmid similar to or larger than the CH4 5938 window. We plan to\ninvestigate these temperature dependence behaviors more thoroughly in the next major GGG version and identify spectroscopic\nimprovements that will reduce or eliminate this behavior using a similar approach to that described for O2 in §3.3.\n7.1.1\n\nFitting windows excluded in GGG2020\n\nBased on the ADCF analysis, several spectral windows were excluded from the TCCON GGG2020 product. Figure 9 shows\n410\n\nthe ADCF versus θmid plots for two CO windows and two weak CO2 windows. The CO window centered on 4233 cm−1\n(Fig. 9a) has slightly stronger temperature dependence and clearly larger scatter than the 4290 cm−1 CO window (Fig. 9b). We\nsuspect this is due to water interference; the 4233 cm−1 CO window has more water lines in it than the 4290 cm−1 window.\nWe examined the spectral residuals in both CO windows to try to identify and correct the water interference, but were not able\nto reduce it to satisfactory levels. Thus, in GGG2020, the XCO product relies on only the 4290 cm−1 window.\n\n415\n\nSimilarly, the new XwCO2 product was planned to use two windows, one centered on 6073 cm−1 and another on 6500 cm−1 .\nHowever, as shown in Fig. 9c and 9d, the 6500 cm−1 window’s ADCF have more scatter and stronger temperature dependence\nthan the 6073 cm−1 window. As the 6500 cm−1 also has more water interference than the 6073 cm−1 window, we elected to\nuse only the 6073 cm−1 window.\nLastly, we also removed a number of HCl windows. TCCON used 16 windows to measure HCl in GGG2014, but like the CO\n\n420\n\nand wCO2 windows, many of these have water absorption lines in them. We can diagnose unaccounted for water interference\nby computing the ADCFs for each HCl window from Darwin 2015 data, split by the amount of water in the column. The\nresult is shown in Fig. 10. Most of the GGG2014 windows have a clear difference in ADCF with small or large water column\namounts. Based on this, we chose to only retain the 5625, 5687, 5702, 5735, and 5739 cm−1 windows. Most of the windows\nremoved clearly have a water interference. The 5754 and 5763 cm−1 windows are special cases. The 5754 cm−1 window was\n\n21\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n0.06\n\nH2O < 1.1 × 1023 molec. cm\nH2O 1.1 × 1023 molec. cm\n\nADCF\n\n0.04\n\n2\n2\n\n0.02\n0.00\n0.02\n5577 5597 5625 5683 5687 5702 5706 5719 5735 5739 5749 5754 5763 5767 5779 5790\nWindow central wavenumber\n\nFigure 10. ADCF calculated for each HCl window from 2015 Darwin data, divided by the total column amount of water.\n425\n\nrejected because its airmass dependence is slightly more negative than the retained windows. The 5763 cm−1 window was\nrejected because it exhibits a clear temperature dependence in the window-to-window scale factors (§7.2).\n7.2\n\nUpdated window-to-window averaging\n\nMany gases retrieved by GGG are retrieved in more than one spectral window. GGG retrieves the column amount in each\nwindow separately, then averages together the columns with similar averaging kernels to produce a mean value. Specifically,\nP\n2\nj sj yij /\u000fij\n430 y i = P 2 2\n(6)\nj sj /\u000fij\nwhere subscript j represents the spectral window. That is, the average value for the ith measurement (y i ) is an error weighted\n\naverage of the individual windows’ column amounts (yij , with errors \u000fij ) with a mean bias in each window removed by the\nper-window scale factor, sj . The errors \u000fij are the posterior errors in the Xgas amounts as calculated from the retrieval error.\nIn GGG2014, the sj values were determined online, using an iterative process that minimizes the differences between yij\n435\n\nand the corresponding sj y i values. While this calculates sj values that best fit the data being averaged, it means that how the\nwindows are combined depends on how much data is averaged at once—processing a month could give different results than\nprocessing a year of data, for example. Thus, while GGG2020 retains the capability to compute the sj values on-the-fly, the sj\nvalues are prescribed for standard TCCON processing.\nTo determine the standard TCCON sj values, we used a very similar approach to how we derived the ADCFs in §7.1.\n\n440\n\nSpecifically, we calculated the sj values for two month subsets of data from the same 18 TCCON sites as in §7.1 and fit these\nvalues versus θmid . As with the ADCFs, we used the values of the fit at θmid = 310 K as the final choices of sj .\n7.3\n\nUpdated in situ bias correction\n\nAs in GGG2014, the GGG2020 XCO2 , XCH4 , XN2 O , and XH2 O products are tied to standard scales by in situ aircraft, balloon,\nand/or radiosonde measurements to remove any mean multiplicative bias introduced by error in absorption line intensity.\n445\n\nUnlike GGG2014, XCO in GGG2020 is not tied to in situ measurements, due to previous work that found the difference\n22\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nbetween TCCON XCO and both NDACC (Kiel et al., 2016a) and MOPITT (Hedelius et al., 2019) XCO was approximately the\nmagnitude of the in situ correction. However, we do evaluate XCO against a subset of in situ data from AirCore only below.\nComparison of TCCON data against in situ data follows the following steps:\n1. Identify in situ vertical profiles in available data and convert to a standardized file format\n450\n\n2. Extend the profiles’ tops to 70 km altitude using the standard GGG2020 priors and to the surface by extrapolation or use\nof surface data\n3. Match profiles to available TCCON spectra\n4. Run custom retrievals using the match profiles as the a priori trace gas profile\n5. Compare integrated in situ Xgas values against matched TCCON data, accounting for TCCON vertical sensitivity\n\n455\n\nPoints 1–4 are described in detail in Appendix C. Briefly, we use profiles from:\n– the GlobalviewPLUS 5.0 CO2 (Cooperative Global Atmospheric Data Integration Project, 2019) and GlobalviewPlus\n2.0 CH4 ObsPack (Cooperative Global Atmospheric Data Integration Project, 2020) products,\n– AirCore balloon measurements (Tans, 2009; Karion et al., 2010) flown by NOAA (v20201223, Baier et al., 2021) at\nmultiple TCCON sites and by FMI/LSCE/RUG at the Sodankylä, Finland (Kivi and Heikkinen, 2016) and Nicosia,\nCyprus (Rousogenous, in prep) TCCON sites,\n\n460\n\n– the Infrastructure for Measurement of the European Carbon Cycle (IMECC) campaign,\n– Profiles over the Manaus, Brazil TCCON site (Dubey et al., 2016),\n– ARM radiosondes over the Darwin, Australia (Deutscher et al., 2010) and Lamont, OK, USA TCCON sites\nCH4 profiles have an additional correction to the stratospheric levels obtained from the GGG2020 priors, see §C3 for details.\n465\n\nWe have addressed the recent change of CO2 data from the X2007 to X2019 WMO scales, which will be covered in §7.3.2.\nDue to the relative sparsity of N2 O profiles, GGG2020 TCCON N2 O products were evaluated against surface N2 O data and a\ndifferent approach, which will be covered in §7.3.3.\n7.3.1\n\nCO2 , CH4 , CO, and H2 O in situ comparisons\n\nThe first step in comparing TCCON XCO2 , XCH4 , XCO , or XH2 O to their respective in situ profiles is to match each in situ\n470\n\nprofile to temporally proximate, good quality TCCON retrievals. For this step, we define custom quality filters. A TCCON\nretrieval is considered to be good quality in this context if:\n– Fractional variation in solar intensity (FVSI) is ≤ 0.05. This filters out observations impacted by intermittent clouds.\n23\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n– Solar zenith angle (SZA) is ≤ 80◦ . This avoids observations at large airmasses, where spectroscopic errors can be more\npronounced.\n\n475\n\n– The unscaled Xgas value is > 0 mol mol−1 . A negative retrieved value is unphysical, and the distribution of retrieved\nvalues should not be large enough to make negative values a reasonable part of it.\n– The Xgas error is < 2\u000fmedian , where \u000fmedian is the median error for that Xgas across all the spectra used for the given\ngas. This limits for observations where the observed spectra was fit reasonably well.\n– The median Xluft for a comparison is between 0.996 and 1.002. Xluft and this rational are explained near the end of this\nsubsection.\n\n480\n\nFor each in situ profile, we require 30 TCCON observations passing these quality checks within a certain window of time\naround the corresponding profile’s lowest altitude measurement. Our initial window is ±1 hour. If 30 points meeting these\ncriteria are not present within ±1 hour, we increase both the time window and the allowed Xgas error, trying the combinations\n\n(±1 hr, < 2\u000fmedian ), (±2 hr, < 3\u000fmedian ), and (±3 hr, < 4\u000fmedian ). We use the smallest of these time/error window that yields\n485\n\n30 passing TCCON observations, but if a profile does not have 30 passing TCCON observations in the (±3 hr, < 4\u000fmedian )\nrange, it is removed from the comparison.\nThe remaining in situ profiles are integrated following Wunch et al. (2010), where the integrated in situ Xgas value, zinsitu\nis calculated as:\n\nzinsitu = I(γxa , p, xH2 O ) + I(δx, p, xH2 O )\n490\n\n(7)\n\nwhere\n– p is the vector of pressure at each profile level\n– xH2 O is the vector of water dry mole fractions at each profile level\n– γxa is the TCCON posterior profile (i.e. the prior times the retrieved VMR scale factor γ)\n– δx is the difference between the in situ and TCCON posterior profiles, modified by the TCCON averaging kernel:\nδxi = ai (xinsitu,i − γxa,i )\n\n495\n\nI represents the pressure-weighted integration function:\n· dpi · Di\ni dpi Di\n\u0012\n\u0013\nMH2 O\nDi = gi · Mair · 1 + xH2 O,i ·\nMair\n\nI(x, p, xH2 O ) =\n\nwhere\n\nP\n\nxi\niP\n\n(8)\n(9)\n\n24\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n500\n\n– dpi represents the pressure thickness of layer i\n– gi represents the acceleration from gravity at layer i,\n– Mair and MH2 O represent the mean molecular masses of dry air and water, respectively.\nThe integrated in situ Xgas values are compared against the median of the TCCON Xgas values from the matched observations. The TCCON Xgas values used here have the airmass correction (§7.1) and window-to-window averaging (§7.2) applied.\n\n505\n\nBecause we expect the bias in the TCCON data to arise from incorrect absorption line strengths or broadening coefficients, it\nshould be a multiplicative bias. Therefore, we calculate an uncertainty-weighted mean of the TCCON/in situ Xgas values to\nderive the bias correction. We consider five sources of uncertainty.\n1. Measurement error in the in situ data.\n2. Uncertainty from the unmeasured portion of the free troposphere. (Will be zero if the in situ vertical profile extends\n\n510\n\nthrough the tropopause.)\n3. Uncertainty from the unmeasured portion of the stratosphere.\n4. Random error in the TCCON observations.\n5. Bias in the TCCON observations from instrument misalignment or similar hardware concerns.\nThe calculation of each term and how they are combined is detailed in Appendix C6.\n\n515\n\nThe results of the TCCON-in situ comparison are shown in Fig. 11. In this plot, the y-axes are the ratio of TCCON to in\nsitu Xgas amounts and the x-axes show Xluft , a diagnostic quantity defined as the ratio of the column of dry air derived from\nsurface pressure to the column of dry air derived from the O2 retrieval. We will return to the significant of Xluft shortly. The\nuse of TCCON to in situ ratios to derive the in situ correction is equivalent to the best fit lines forced through the origin used\nin Wunch et al. (2010), as the best fit line through the origin is essentially the mean TCCON to in situ ratio. The use of ratios\n\n520\n\ndirectly in Fig. 11 allows us to more clearly identify outliers and evaluate the correlation of the TCCON vs. in situ bias with\nother variables, such as Xluft here.\nThe ratios from Fig. 11 indicate that the mean biases are within approximately 1% of unity in all cases, with water being the\nfurthest from unity at 0.9883 (−1.17%). The differences among the CO2 products are interesting; the standard CO2 product\nis biased about 1% high before correction (which is in line with expected uncertainties for the CO2 lines), while the other two\n\n525\n\nCO2 products are much closer to unity (0.08% for wCO2 and 0.14% for lCO2 ). This suggests that the absorption coefficients\nin these latter two windows are more accurate than in the standard TCCON windows (which are centered on 6220 and 6339\ncm−1 ). However, as the wCO2 and lCO2 are more sensitive to the upper and near-surface atmosphere, respectively, it may be\nthat this reflects other factors, such as the accuracy of the a priori temperatures at those levels.\nThe CO comparison (Fig. 11e) suggests that, without scaling, the GGG2020 XCO has no significant bias with respect to\n\n530\n\nAirCore CO measurements. Figure 11e shows significant variation in the TCCON/in situ CO agreement, with individual points\n25\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n(b)\n\n(c)\n1.015\n\n1.015\n\n1.010\n\n1.010\n\n1.010\n\n1.005\n1.000\n0.995\n\nr = 1.01010\n2 r = 0.00050\n\n1.005\n1.000\n0.995\n0.990\n\n0.996 0.997 0.998 0.999 1.000 1.001 1.002\nXluft (mol/mol)\n\n(d)\n\n0.995\n0.990\n\nTCCON XCO / in situ XCO\n\n1.010\n1.005\n1.000\n0.995\n0.990\n\nr = 1.00310\n2 r = 0.00140\n\n0.980\n0.996 0.997 0.998 0.999 1.000 1.001 1.002\nXluft (mol/mol)\n\n(f)\n\n1.8\n1.6\n\n1.2\n1.0\n0.8\n0.6\n\nr = 1.00140\n2 r = 0.00070\n\nr = 1.00850\n2 r = 0.05260\n\n0.996 0.997 0.998 0.999 1.000 1.001 1.002\nXluft (mol/mol)\n\nATom\nAirCore\nCOBRA\nGSFC\nHIPPO\nHarvard\nIMECC\nLARC\nNOAA Manaus\nORCAS\nRadiosonde\n\n0.996 0.997 0.998 0.999 1.000 1.001 1.002\nXluft (mol/mol)\n\n1.4\n\n1.015\nTCCON XCH4 / in situ XCH4\n\n1.000\n\n0.996 0.997 0.998 0.999 1.000 1.001 1.002\nXluft (mol/mol)\n\n(e)\n\n1.020\n\n0.985\n\nr = 1.00080\n2 r = 0.00050\n\n1.005\n\nTCCON XH2O / in situ XH2O\n\n0.990\n\nTCCON XlCO2 / in situ XlCO2\n\n1.015\nTCCON XwCO2 / in situ XwCO2\n\nTCCON XCO2 / in situ XCO2\n\n(a)\n\n1.4\n1.2\n1.0\n0.8\n0.6\n0.4\n\nr = 0.98830\n2 r = 0.01570\n\n0.2\n0.996 0.997 0.998 0.999 1.000 1.001 1.002\nXluft (mol/mol)\n\nAnmyeondo\nArmstrong\nBialystok\nBremen\nDarwin\nLamont\nLauder 01\nLauder 02\nManaus\nNicosia\nPark Falls\nPasadena\nRikubetsu\nSodankyla\n\nFigure 11. Plots of the TCCON/in situ Xgas ratios for (a) CO2 , (b) wCO2 , (c) lCO2 , (d) CH4 , (e) CO, and (f) H2 O. In all plots, the y-axis\nis the ratio of TCCON/in situ Xgas and the x-axis is the median Xluft value for the TCCON observations in a comparison (see text for\nexplanation of Xluft ). The marker style of each comparison indicates the source of the in situ data and the color indicates which TCCON site\nthe comparison occurred at. The text inset in the lower-right corner of each plot gives the uncertainty-weighted mean TCCON/in situ ratio\nand its 2σ uncertainty. The dashed black lines mark the mean ratio. Panels a, b, and c are set to use the same y-limits; some of the error bars\nin (b) go outside the y-limits.\n\nalso having large uncertainty. This resulting 2σ uncertainty in the mean ratio is significantly larger than for the other gases, at\n0.0526. Thus, the mean TCCON/in situ CO ratio is well within its 2σ uncertainty of 1. We do acknowledge that limiting the\nCO comparisons to AirCore profiles alone may contribute to a larger uncertainty than if aircraft campaigns were included, due\nto the use of a CO-spiked fill gas in AirCores (see §2.1 of Martínez-Alonso et al., 2022). However, comparing TCCON XCO\n535\n\nto AirCore profiles was significantly more straightforward than including aircraft profiles, since the already-matched AirCore\nprofiles for CO2 and CH4 intrinsically include CO as well. Given the other reasons discussed above for not applying an in\nsitu-derived scaling to GGG2020 XCO and the process needed to match aircraft data with TCCON (see Appendix C1.1), we\nchose to accept the additional uncertainty from using AirCore profiles only. Future versions of the TCCON data product will\nreevaluate the inclusion of aircraft profiles alongside AirCore ones.\n\n540\n\nFigure 11 also provides insight into how instrumental errors affect different TCCON products. Under ideal circumstances,\nXluft (the quantity on the x-axis) should be 1; in practice, the nominal value for the TCCON network is 0.999, due to small\nresidual biases in the O2 spectroscopy. Deviations of Xluft from the nominal value indicate either (a) variable errors in spectroscopy, such as temperature or pressure broadening, or (b) instrument issues, such as a misalignment in the beam path. From\nFig. 11a, we can see that the TCCON/in situ ratio tends to be less when Xluft < 0.999 and greater when Xluft > 0.999. The\n\n26\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n545\n\nslope for Fig. 11a is 0.363. This translates to a bias in CO2 of about 0.15%, or approximately 0.5 ppm, when Xluft is 0.004 units\naway from the nominal value of 0.999 (0.15% = 0.0015 = 0.363 × 0.004). To keep this bias well below the expected 0.25%\naccuracy, we limit the comparison used here to those where Xluft is between 0.996 and 1.002 and have instituted additional\n\nquality checks of TCCON data that filter out observations when Xluft is outside the range of 0.995 to 1.003 for an extended\nperiod of time. Additionally, Xluft is now reported in the public data set alongside other Xgas retrievals.\n550\n\nWe note that the standard CO2 and the near surface-sensitive lCO2 products show the clearest dependence on Xluft . The\nreason for this is not clear at this time, though it implies a stronger dependence of these products on instrument line shape\n(ILS) compared to the other four products discussed in this section. Future versions of GGG are planned to account for errors\nin the ILS, which we hope will mitigate this bias and improve the accuracy of CO2 data when Xluft is outside the 0.995 to\n1.003 range.\n\n555\n\nThe correlation of XCO2 and XlCO2 with Xluft implies that we could develop an Xluft -based bias correction for those CO2\nproducts. Such a correction is planned for a minor update to the GGG data product. Our aim is to quantify the underlying\nphysical drivers of the XCO2 bias and use the correlation of those factors with Xluft to derive the bias correction. This would\nallow us to use the comparison to in situ data shown here as an independent verification of the bias corrections efficacy.\n7.3.2\n\n560\n\nAddressing the CO2 scale change from X2007 to X2019 and changing O2 mole fraction\n\nThe update from the previous WMO CO2 X2007 calibration scale to the new X2019 calibration scale (Hall et al., 2021)\noccurred late enough in the process of releasing GGG2020 that we were not able to incorporate it into the initial release. Given\nthe clear need expressed by the community to have TCCON data tied to the same scale as in situ data, we have since derived\nnew in situ correction factors to tie all three TCCON XCO2 products to the X2019 scale. Doing so required obtaining in situ\ndata that had been adjusted to the new scale, which we did in one of three ways:\n\n565\n\n1. The preferred approach was for the data originator to fully recalibrate their data to the new scale using the updated standards provided by the NOAA Global Monitoring Laboratory. NOAA AirCore and some NOAA ObsPack data followed\nthis approach.\n2. The second approach was for the data originator or an intermediate provider to adjust the CO2 data using the linear\ncorrection described in §9.1 of Hall et al. (2021). The remaining NOAA ObsPack data not covered by approach #1\n\n570\n\nfollowed this approach.\n3. The third approach was for us to perform the same linear correction as #2 ourselves. All other data used this approach.\nAlso recall that the profiles must be extended to 70 km altitude using the TCCON standard priors to ensure that the same\nvertical extent is captured in the in situ and TCCON column averages. As discussed in Laughner et al. (2023), the standard\npriors are derived from NOAA data at the Mauna Loa and American Samoa observatories, and so are also intrinsically tied to\n\n575\n\nWMO calibration scales. To ensure consistency throughout the in situ profiles, we used the latest available monthly average\n\n27\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nCO2 flask data on the X2019 scale as input to the priors when generating the profile extensions. Once this was complete, we\nredid the analysis described in §7.3 with the in situ profiles adjusted to the X2019 scale to generate updated correction factors.\nThe overall effect of the scale change for each of the three TCCON CO2 products is shown in Fig. 12 compared to the “raw,”\nun-bias corrected XCO2 value on the x-axis. The magnitude is about +0.15 ppm for typical current XCO2 values of 400 ppm.\n580\n\nIn the TCCON data products, there are three CO2 variables with the suffix _x2019 which are adjusted to the new X2019\nscale.\nAnother source of bias that is of similar magnitude to the effect of the scale change is the assumed O2 mole fraction. As\nshown in Eq. (1), the column-average mole fractions reported by TCCON are computed by dividing the column density of the\ntarget gas by the O2 column density, and scaling by the mean O2 mole fraction in the atmosphere. We have assumed that this\n\n585\n\nmole fraction is fixed for the initial GGG2020 data products; however, it is in fact changing over time due to various processes,\npredominantly fossil fuel combustion and the land biosphere (Keeling et al., 1998; Keeling and Manning, 2014).\nBecause the effect of ignoring the change in the global average O2 mole fraction is of similar magnitude to the X2007 to\nX2019 scale change, we decided to account for the change in O2 mole fraction over time in the CO2 products updated to the\nX2019 scale. We did not retroactively apply this correction to the X2007 XCO2 or the other Xgas products, as doing so would\n\n590\n\nchange the Xgas values and require a new data version. This correction will be applied to all Xgas values in the next GGG data\nversion.\nOur approach to account for changing O2 mole fraction takes advantage of the anticorrelation between atmospheric O2 and\nCO2 to derive the O2 mole fraction from CO2 measured by TCCON. (For our application, this assumption is sufficiently\naccurate; however, we note that this is not generally true for other applications of O2 /N2 ratio data.) Specifically, the value for\n\n595\n\nfO2 in Eq. (1) is calculated as (see Appendix E1 for the full derivation):\n\nfO2 = (α − α · fO2 ,ref − fO2 ,ref ) ·\n\nXCO2 − XCO2 ,ref\n+ fO2 ,ref\n1 − XCO2 − α · XCO2\n\n(10)\n\nwhere:\n– α = ∂NO2 /∂NCO2 = −1/0.4575, i.e. the change in the number of moles of O2 in the atmosphere for a given change in\n600\n\nthe number of moles of CO2 in the atmosphere. The choice of −1/0.4575 comes from the agreement with the measured\nchange in fO2 as shown in Fig. 13. This value is chosen to remove the effect of long term trends in the O2 mole fraction,\nand ignores synoptic-scale variations due to e.g. photosynthesis or fossil fuel emissions.\n\n– fO2 ,ref is the reference value for the mole fraction of O2 . We use 0.209341 based on the value measured by Aoki et al.\n(2019) at Hateruma Island, Japan in 2015 and adjusting by ∼ 2 ppm to approximate the global mean fO2 by using the\n\ndifference between the annual mean CO2 reported for Hateruma Island by Aoki et al. (2019) and that for the NOAA\n605\n\nglobal marine boundary layer reference. (A revised calculation accounting for possible influence of fossil fuel emissions\non Hateruma Island puts the global mean O2 mole fraction closer to 0.209347, however the 0.209341 value is what is\nused in GGG2020.)\n28\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nX2019 - X2007\nX2019+Var. O2 - X2019\nX2019+Var. O2 - X2007\n\nXCO2 (ppm)\n\n(a)\n0.0\n0.5\n1.0\n\n350\n\n400\n450\n500\n\"Raw\" XCO2 (ppm)\n\n550\n\n350\n\n400\n450\n500\n\"Raw\" XwCO2 (ppm)\n\n550\n\n350\n\n400\n450\n500\n\"Raw\" XlCO2 (ppm)\n\n550\n\nXwCO2 (ppm)\n\n(b)\n0.0\n0.5\n1.0\n\nXlCO2 (ppm)\n\n(c)\n0.0\n0.5\n1.0\n\nFigure 12. The change in TCCON (a) XCO2 , (b) XwCO2 , and (c) XlCO2 due to the WMO scale change, change in assumed O2 mole\nfraction, and the combination of both. The x-axis is the “raw” XCO2 value that has no in situ bias correction and assumes a fixed O2 mole\nfraction. The “X2019 – X2007” line shows the difference due to only the CO2 WMO scale change, the “X2019+Var O2 – X2019” shows\nthe difference due to only the change from fixed to variable O2 mole fraction, and the “X2019+Var. O2 – X2007” line shows the total change\nfrom both effects combined.\n\n29\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nReference\nBest estimate\n= 1/0.4575\n= 1/0.4000\n= 1/0.5000\n\nfO2 (mol/mol)\n\n0.209425\n0.209400\n0.209375\n0.209350\n0.209325\n\n1995 2000 2005 2010 2015 2020\nYear\nFigure 13. Comparison of fO2 values calculated using Eq. (10) for three different values of α versus a best estimate of fO2 using δ(O2 /N2\nfrom the Scripps Intitute of Oceanography (Scripps O2 Program, 2022) and NOAA global mean CO2 (Lan et al., 2023) data. The three\ncolored lines also use NOAA global mean CO2 data for the XCO2 and XCO2 ,ref values in Eq. (10). The black circle marks our reference\nvalue of fO2 = 0.209341.\n\n– XCO2 ,ref is a reference value for the column-average mole fraction of CO2 . We use 4 × 10−4 (400 ppm) to approximate\n\nthe value seen in TCCON data during 2015 (the same year as the fO2 ,ref value), though as discussed below, it is not\n\n610\n\ncrucial that the O2 and CO2 reference values be for exactly the same time.\n– XCO2 is the “raw” measured TCCON XCO2 with airmass correction and assuming fO2 = fO2 ,ref = 0.209341.\nTo validate this approach, we also compute the change in fO2 (including the effect of CO2 dilution) using δ(O2 /N2 ) data\nmeasured by the Scripps Institution of Oceanography at Alert, NWT, Canada (station code ALT); La Jolla Pier, California,\nUSA (LJO); and Cape Grim, Australia (CGO) and NOAA CO2 annual trend data (Lan et al., 2023). To approximate a global\n\n615\n\nmean δ(O2 /N2 ) value, we follow §5.15.4.2 of Keeling and Manning (2014) and combine the data from these stations as (ALT\n+ LJO)/4 + CGO/2.\nThe results of this comparison are shown in Fig. 13. The black line shows the change in fO2 computed using the Scripps\nδ(O2 /N2 ) data (see Appendix E2 for the methodology), while the other three lines represent fO2 calculated with Eq. (10)\n\n620\n\nand various values of α. We can see that Eq. (10) with α = −1/0.4575 gives quite good agreement with the change in fO2\ncomputed using the Scripps δ(O2 /N2 ) and NOAA global CO2 data.\n\nThe final step in adopting the variable O2 mole fraction was to recompute the in situ correction factor once more, using the\nvariable O2 mole fraction in the TCCON Xgas values for the comparison. Doing so ensures that any constant multiplicative\nbias introduced by incorrect or inconsistent values for the fO2,ref or XCO2 ,ref values is scaled out. This is why, in the discussion\nabove about the choice of those reference values, we note that it is not critical to have the O2 and CO2 reference value exactly\n625\n\nconsistent.\nThe orange lines in Fig. 12 show the effects of the change from a fixed O2 mole fraction to the variable one. For XCO2\nvalues around 400 ppm, the change is generally small in all three CO2 products. If CO2 mixing ratios continue to increase in\n30\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nthe future, the difference between using the incorrectly fixed and correctly varying O2 mole fraction would increase to 0.75 to\n1 ppm in magnitude.\n630\n\nThe green lines in Fig. 12 show the combined effect of the CO2 calibration scale change and the switch to a variable O2\nmole fraction. For low “raw” XCO2 values (i.e. values without the in situ bias correction and using a fixed O2 mole fraction)\nthe two effects reinforce each other, but as the raw XCO2 increases, the O2 mole fraction change starts to counteract part of the\nCO2 scale change.\nXCO2 , XwCO2 , and XlCO2 on the X2019 scale and accounting for the variable O2 mole fraction are now available in the pub-\n\n635\n\nlic data set as variables xco2_x2019, xwco2_experimental_x2019, and xlco2_experimental_x2019. Users\ncomparing to other data or model simulations/assimilations on the X2019 scale should use these variables. Anyone needing\nto compare against data still on the X2007 scale can use xco2, xwco2_experimental, and xlco2_experimental\ninstead.\n7.3.3\n\n640\n\nN2 O in situ comparisons\n\nTo derive an in situ correction for N2 O, we adopted a different approach than the other gases due to the small number of\nN2 O profiles over TCCON sites which our matching algorithm found in the NOAA CCGG Aircraft Program v1.0 ObsPack\n(Sweeney et al., 2018). Figure 14a shows the 10 profiles identified from the ObsPack, and Fig. 14b shows the TCCON/in situ\nratio vs. Xluft relationship for these profiles. We note that this scarcity of profiles was partly due to the criteria used to filter\nfor good quality profiles (Appendix C1.1). However, given how well-mixed N2 O is in the troposphere, the criteria intended to\n\n645\n\nensure a profile had enough vertical resolution to capture plumes of CO2 or CH4 could be relaxed for N2 O in future TCCON/in\nsitu comparisons to increase the number of available N2 O profiles for comparison.\nThe available profiles were further restricted by our criteria for coincidence with good quality TCCON observations. 2 of\nthese 10 profiles do not meet the coincidence criteria for inclusion in Fig. 14b, and 5 of the remaining 8 fall outside the allowed\nXluft range of 0.996 to 1.002. With the available data, it is difficult to distinguish whether there is significant correlation\n\n650\n\nbetween Xluft and TCCON XN2 O bias, and therefore whether those 5 comparisons below Xluft = 0.996 should be excluded.\nAs their exclusion would significantly alter the in situ correction for XN2 O , we tested a second approach to derive the N2 O\ncorrection.\nThis alternate approach uses NOAA surface N2 O data from the NOAA Halocarbons and other Atmospheric Trace Species\n(HATS) program (Dutton et al., 2023) combined with the GGG2020 priors to generate pseudo-in situ profiles. This takes\n\n655\n\nadvantage of the limited vertical variation in N2 O up to the tropopause seen in Fig. 14a and the good accuracy of the GGG2020\npriors in the stratosphere (Laughner et al., 2023). These pseudo-in situ profiles use the HATS N2 O data for the tropospheric\nVMRs, the GGG2020 priors for VMRs above 380 K potential temperature, and linearly interpolates in between. These pseudoin situ profiles are then integrated following Eq. (8) to produce a pseudo-in situ XN2 O and compared to TCCON in the same\nmanner as the other gases. As we are not limited by when an aircraft provided an N2 O profile over a TCCON site, we can\n\n660\n\ncompare to TCCON observations from any time. We use spectra from the same sites and days as the other gases, filtered for\nthe following criteria:\n31\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n(a)\n\n(b)\nTCCON XN2O / in situ XN2O\n\nPressure (hPa)\n\n200\n400\n600\n\nEast Trout Lake\nLauder 02\nLamont\nTroposphere\nAbove tropopause\n1000\n270 280 290 300 310 320 330\nN2O DMF (ppb)\n800\n\n(c)\nTCCON XN2O / in situ XN2O\n\n1.02\n1.01\n\nEast Trout Lake\n\nLauder 02\n\n0.990\n0.985\n0.980\n0.975\nr = 0.98250\n0.970 2 r = 0.00750\n0.986\n\n0.988\n\n0.990\n\n0.992 0.994\nXluft (mol/mol)\n\n0.996\n\n0.998\n\n1.000\n\ny = (0.782 +/- 0.0023) + (0.000646 +/- 7.53e-06)x\nFit = 0.98216 at x = 310.000\n\n1.00\n0.99\n0.98\n0.97\n0.96\n0.95\n\nLamont\n\n0.995\n\nr = 0.97905\n2 r = 0.00024\n285\n\n290\n\n295\n\n300\n\n700\n\n(K)\n\n305\n\n310\n\n315\n\nAscension Island\nAnmyeondo\nBremen\nPasadena\nDarwin\nArmstrong\nEast Trout Lake\nGarmisch\nLauder 01\n\nLauder 02\nManaus\nNicosia\nLamont\nOrleans\nPark Falls\nRikubetsu\nSodankyla\nWollongong\n\n320\n\nFigure 14. (a) The available N2 O profiles over TCCON sites from the NOAA CCGG Aircraft Program v1.0 ObsPack (Sweeney et al., 2018).\n(b) TCCON/in situ ratio vs. Xluft similar to Fig. 11, but for N2 O. (c) The TCCON/in situ XN2 O ratio derived using surface NOAA N2 O\ndata versus mid-tropospheric potential temperature. The dashed gray line is a robust fit to the data. The text in the lower right hand corner\ngives the mean TCCON/in situ ratio (denoted also by the horizontal dashed black line) and its 2σ standard deviation. The points are colored\nby TCCON site.\n\n– FVSI ≤ 0.05, as for the other gases\n– Xluft between 0.996 and 1.002, as for the other gases\n– The difference between prior HF column density and retrieved HF column density is < 2 × 1014 molec. cm−2 .\n665\n\nThe filtering on HF column helps to remove cases where the stratosphere prior N2 O used in the pseudo-in situ profiles is\nincorrect. HF is a gas found almost exclusively in the stratosphere, and in GGG2020, the HF and N2 O stratospheric priors\nare coupled. Thus, when the retrieved HF column is substantially different from the prior, that indicates that the HF prior was\nincorrect, which implies the same for the N2 O profile. HF columns tend to be between 1 and 2×1015 molec. cm−2 , so 2×1014\nmolec. cm−2 represents a 10% to 20% error in the HF prior. Given that the stratosphere component of N2 O is < 20% of the\n\n670\n\ncolumn, and assuming that the percent error in the N2 O prior is similar, this keeps the random error in the pseudo-in situ XN2 O\nto less that 2% to 4%. All together, these filtering criteria retain approximately 8600 TCCON observations from the initial set\nof ∼ 20, 000 observations used in the in situ correction analysis.\n\nThis larger sample set for N2 O allowed us to identify a correlation in XN2 O bias with atmospheric temperature. Figure 14c\n\nshows how the TCCON/in situ XN2 O ratio varies with potential temperature at 700 hPa. As in the ADCF analysis (§7.1), these\n675\n\npotential temperature values come from the GEOS FP-IT meteorology used as input to the GGG retrievals. The presence of\nthis bias suggests that there is an error in the temperature dependence of the N2 O cross sections (similar to that we identified\n32\n\n\fTCCON / in situ XN2O\n\nhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n1.04 (a)\n1.02\n\nGGG2020 fit: y = 0.00066x + 0.79616\nGGG2020.1 fit: y = 0.00000x + 0.99985\n\n1.00\n0.98\n290\n\n300\n700\n\n(K)\n\n310\n\n320\n\n1.02 (b)\nGGG2020\nNotional GGG2020.1\nEast Trout Lake\nLauder 02\nLamont\nRatio = 1\n\n1.01\n1.00\n\n330\n\n285 290 295 300 305 310 315\n700 (K)\n\n(c)\n\n102\n# observations\n\nNotional GGG2020.1 XN2O (ppb)\n\n0.99\n\n320\n\n101\n\n310\n300\n300\n\n310\n320\nGGG2020 XN2O (ppb)\n\n330\n\n100\n\nFigure 15. Future correction for XN2 O . (a) Similar to Fig. 14c, except showing the ratio between TCCON and the surface-derived XN2 O\nfrom GGG2020 with the in situ correction factor of 0.9821 applied in blue, and the expected temperature-corrected XN2 O in orange, with\ntheir respective fits. (b) Similar to Fig. 14b, but like panel (a) of this figure, comparing the ratios of GGG2020 and temperature-corrected\nXN2 O to in situ. (c) A 2D histogram comparing the current and notional corrected XN2 O\n\nand removed for O2 , §3.3). In the near term, we plan to develop a post-processing correction for this temperature bias in\nN2 O for inclusion in a minor update to the TCCON GGG2020 data within 2–3 years. Long term, the underlying error in the\nspectroscopic model will be corrected so that the next major TCCON data release will have improved XN2 O data.\n680\n\nFor GGG2020, we elected to choose the XN2 O in situ correction as the value of the fit in Fig. 14c at 310 K potential\ntemperature. This is consistent with the choice of ADCF values at the same temperature (§7.1). The value of 0.9822 is very\nclose to the mean TCCON/in situ ratio using the 8 true in situ profiles in Fig. 14b. That both methods agree gives us confidence\nthat this is a reasonable value to use for the in situ correction. We are also investigating applying the slope from Fig. 14c to\nTCCON XN2 O as a temperature-based bias correction. Figure 15 demonstrates the difference this correction would make, both\n\n685\n\nin comparison to the in situ data (Fig. 15a,b) and to the column-average dry mole fractions themsleves (Fig. 15c).\n\n33\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nXgas product\n\nCorrection factor\n\nCF error\n\nCalibration scale\n\nN\n\nfO2\n\nXCO2\n\n1.0101\n\n0.0005\n\nWMO X2007\n\n67\n\nFixed\n\nXCO2 _x2019\n\n1.0090\n\n0.0005\n\nWMO X2019\n\n70\n\nVar.\n\nXwCO2\n\n1.0008\n\n0.0005\n\nWMO X2007\n\n67\n\nFixed\n\nXwCO2 _x2019\n\n0.9996\n\n0.0005\n\nWMO X2019\n\n69\n\nVar.\n\nXlCO2\n\n1.0014\n\n0.0007\n\nWMO X2007\n\n67\n\nFixed\n\nXlCO2 _x2019\n\n1.0006\n\n0.0007\n\nWMO X2019\n\n69\n\nVar.\n\nXCH4\n\n1.0031\n\n0.0014\n\nWMO X2004\n\n40\n\nFixed\n\nXN2 O\n\n0.9821\n\n0.0098\n\nNOAA 2006A\n\nN/A\n\nFixed\n\nXCO\n\n1.000\n\n0.0526\n\nN/A\n\n31\n\nFixed\n\nXH2 O\n\n0.9883\n\n0.0157\n\nARM Radiosondes\n\n94\n\nFixed\n\nTable 2. In situ correction factors and their errors for each Xgas product evaluated against in situ data. The “Calibration scale” column\nindicates which scale or source these data are tied to by the AICFs. The N column indicates how many profiles are used to calculate the\nAICF for that gas. The fO2 column indicates what O2 mole fraction was used in the column density to column-average mole fraction\nconversion: “Fixed” means fO2 = 0.2095 in Eq. (1) and “Var.” means that the variable mole fraction described in §7.3.2 was used.\n\n7.3.4\n\nIn situ bias correction summary\n\nA summary of the in situ correction factors, their errors, and what in situ calibration scales each product is tied to are given\nin Table 2. Because these correction factors are the mean TCCON/in situ ratio, dividing the airmass-corrected and windowaveraged values by these correction factors removes the mean TCCON-in situ bias.\n690\n\nIn the TCCON data, users will find two sets of XCO2 variables. Those with the _x2019 suffix (xwco2_experimental_x2019,\nxlco2_experimental_x2019, and xco2_x2019) are those tied to the WMO X2019 CO2 scale and which use the variable O2 mole fraction. Those CO2 variables without the _x2019 suffix remain tied to the WMO X2007 CO2 scale and still\nuse the fixed O2 mole fraction. All other gases (xch4, xco, etc.) also still use the fixed O2 mole fraction.\nReleasing the rescaled XCO2 as new variables, rather than creating a new TCCON data version with the existing variables\n\n695\n\nrescaled, was chosen for several reasons. First, it is logistically simpler, allowing us to provide this update more quickly.\nSecond, during this transitional period when existing CO2 data is available on both the X2007 and X2019 scales, having\nboth X2007 and X2019 XCO2 allows users to switch back and forth easily if they need to match up with other datasets on\na mix of both scales. Third, this approach provides for release of more recent TCCON data without disrupting existing users\nworkflows—users do not have to worry about existing variables changing, but can switch their analyses to use the updated\n\n700\n\nXCO2 variables if and when they wish. Incorporating the variable O2 mole fraction for all gases is planned for an upcoming\nminor revision of the TCCON data (tentatively called “GGG2020.1”). Likewise, a temperature-corrected XN2 O product will\nbe included in GGG2020.1 or the follow on GGG2020.2, depending on development time.\n\n34\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n6000\n\n90\n\n40\n\n80\n\n30\n\n5000\n\n-20\n\n20\n\n60\n10\n50\n\n3000\n40\n2000\n\nSZA ( °)\n\nX H O (ppm)\n2\n\n-10\n\n°C)\n\n4000\n\nSurface temperature (\n\n70\n\n0\n\n30\n20\n\n1000\n\n0\nJan Feb Mar Apr May Jun Jul Aug\n\n10\n\n-30\n\n0\n\n-40\n\n2019\nFigure 16. The three dates chosen for the error budget calculations are from East Trout Lake on February 18 (blue), June 11 (red), and July\n23 (yellow), 2019. These dates were chosen to span a range of water vapor, solar zenith angle, and surface temperature. In the left panel, the\nblack data points show the full East Trout Lake record between Jan and Aug 2019 for reference.\n\n8 Uncertainty budget\nTo calculate an uncertainty in the GGG2020 dataset, we selected three days from the East Trout Lake dataset spanning a\n705\n\nrange of atmospheric water vapor, surface temperature and solar zenith angle (Figure 16). Each known source of uncertainty is\nmodeled or perturbed by a realistic amount in the GFIT forward model (the quantitative amounts are described in the following\nparagraphs), and we compute the percent fractional difference in Xgas between the perturbed and unperturbed value. The total\nuncertainty is computed as the sum in quadrature of the individual uncertainties. For each gas, we have plotted the contributions\nof each source as a function of solar zenith angle for the June 11, 2019 date in Figures 18–20. The same figures for cold, dry\n\n710\n\nFebruary 18 are in the Appendix in Figures B1–B3, and for warm, wet July 23 are in Figures B4–B6. The sum in quadrature\nof all the sources of error for each gas are plotted for the three days together in Figures 21–23. Each source of uncertainty\nincluded in our error budget is described below.\nField of view\nThe field of view (FOV) is the maximum solid angle viewed by the detector element, and its value is set by the field stop\n\n715\n\ndiameter inside the instrument. It is an important parameter in the GFIT forward model because it defines the extent of off-axis\nrays that pass through the interferometer, ultimately limiting the spectral resolution of a spectrum. The field stop diameter is\n\n35\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nset by a physical pinhole ranging from 0.5-1.3 mm drilled into a thin plate within the instrument, and its size can be in error by\na few percent. Here, we increase FOV by 7% to reflect any uncertainty in the field stop diameter.\nContinuum basis functions\n720\n\nIn GGG2020, the number of continuum basis functions has been optimized to improve the spectral fits without over fitting the\ndata (see §6). Here, we increase the number of continuum basis functions fitted by 1 in all windows that have widths >5 cm−1\nto assess the sensitivity of our choice of the number of basis functions to the retrieved Xgas value. The gases excluded from\nthis test because of their fitting window widths are HF, HCl, and some H2 O and HDO windows.\nSolar pointing\n\n725\n\nThe observer-sun Doppler stretch (OSDS) is a calculation made by GFIT based on the Earth-Sun radial velocity and the\nEarth’s rotational velocity component, under the assumption that the solar tracker is imaging the centre of the Sun. It defines\nthe Doppler stretch of the solar absorption lines relative to the telluric (atmospheric) absorption lines. If the solar tracker is not\nimaging the exact centre of the Sun, the solar lines may be Doppler-shifted relative to the telluric lines, creating systematic\nresiduals in the spectral fits. Here we increase the OSDS by 2 ppm to assess the sensitivity of the retrievals to a small pointing\n\n730\n\nerror from the Doppler stretch component alone. This error affects carbon monoxide more than the other gases because for\nevery telluric CO line in the spectrum, there is also a solar CO absorption line beneath, making it difficult to distinguish solar\nfrom telluric CO absorption. In GGG2014 and previous versions, this was a particular problem, because the pointing was\nassumed to be in the centre of the solar disk. In GGG2020, however, the solar-gas stretches are now fitted, reducing the impact\nof an OSDS error on the CO retrievals (see Wunch et al., 2015, Fig. 13).\n\n735\n\nSolar tracker pointing offsets also affect the ray tracing in GFIT, causing errors in the airmasses calculated for a given\nspectrum. This error impacts all gas retrievals, but should mostly cancel in the ratio between the gas of interest and oxygen.\nHere, we add a 0.05 degree pointing offset (poff), which represents a pointing error of about 20% of the solar radius.\nPrior\nWe modify the priors in several ways to estimate the uncertainties caused by various errors in the a priori profiles.\n\n740\n\n– A priori pressure profile (prior pressure). We multiplied the pressure at each atmospheric level in the prior by 1.002 to\nscale up the pressure by 0.2% at all altitudes. For the HCl cell pressure error, we added 0.14 hPa (0.138 atm) to the cell\npressure, following the “pessimistic” uncertainty budget in Hase et al. (2013, P3565). (The purpose of the HCl cells will\nbe described in §8.10.)\n– A priori temperature profile (prior temperature). We added 1 K to each atmospheric level in the prior.\n\n745\n\n– A priori profile shape (prior shift). We shifted the a priori profiles down by one atmospheric level. In GGG2014, we\nshifted the priors down by 1 km, so this is a slightly different approach, but the level spacing is about 1 km in altitude\nnear the tropopause, where this shift is most important for well-mixed tropospheric gases like N2 O and CH4 , and HF, a\nstratospheric gas. H2 O and HDO are not shifted as part of this process, but are modified in an independent test.\n\n36\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n– A priori boundary layer CO (prior CO enhanced). The GEOS FP-IT CO profiles are created using an old emission\ninventory, and tend to significantly overestimate emissions in urban regions that have reduced their emissions over time\n\n750\n\n(e.g., Los Angeles). However, because of the coarse spatial resolution of GEOS FP-IT, sites that are located near to an\nurban centre can be affected by the urban enhancements in the model. We therefore add an additional test that affects\nonly the CO error budget, in which we add 25 ppb to the altitudes below 2 km to estimate the uncertainty caused by the\nincorrect lower atmosphere shape in the GGG2020 CO prior profiles.\n755\n\n– A priori H2 O and HDO (prior h2o/hdo). We modified the water and HDO profiles by reducing the values in the first\n1 km by 50%.\nSurface pressure\nThe surface pressure measurements we collect as part of our on-site meteorological data are important for calculating the total\ncolumn of air overhead. The largest surface pressure uncertainty permitted by the TCCON data protocol is 0.3 hPa, but we\n\n760\n\nhave seen these instruments drift by up to 1 hPa. Here, we add 1 hPa to the surface pressure (pout) to calculate the sensitivity\nof the retrievals to this error.\nNonlinearity\nDetector nonlinearities, described in §5.1, cause a discrepancy between the low-resolution spectral envelope and the high\nresolution spectral lines, resulting in an offset at zero in the spectrum. These zero level offsets are most readily observed in\n\n765\n\nregions of the spectrum where strong absorption lines absorb all the incident light (Abrams et al., 1994). Here, we add 0.001\n(0.1%) to the zero level offset (ZLO) parameter in GFIT, a large ZLO observed in the network.\nInstrument line shape\nThe instrument line shape (ILS) of a Fourier transform spectrometer quantifies the optical alignment of the instrument, and is\nindependent of the alignment of the solar image. The ILS is characterized by two parameters: the modulation efficiency and\n\n770\n\nphase error. The modulation efficiency is the broadening or narrowing of the ideal spectral line width in the instrument, and the\nphase error is the asymmetrical component of the spectral line that is caused by the misalignment. It is not currently possible\nto model phase error within GFIT, but we can model imperfect modulation efficiency. The TCCON data protocol requires that\nthe instrument modulation efficiencies must be within 5% of a perfect alignment. The modulation efficiency of a perfectly\naligned interferometer is defined as a value of 1.0 at all optical path differences, taking self-apodization into account, and\n\n775\n\ntherefore the maximum and minimum modulation efficiency acceptable in the network is 1.05 and 0.95, respectively. Here we\nmodel two cases: a “shear” misalignment, where the modulation efficiency of the spectrometer increases linearly to 1.05 as a\nfunction of optical path difference, and an “angular” misalignment, where the modulation efficiency drops linearly to 0.95 as a\nfunction of optical path difference. We confirmed the misalignment by passing synthetic spectra generated by GFIT with these\nmisalignments through LINEFIT (v14.8 Hase et al., 1999), a program designed to assess instrument line shapes (see Figure\n\n780\n\n17).\nBecause GGG2020 cannot model phase errors, these sensitivity studies are likely to underestimate the full effect of ILS\nerrors, and therefore we include both the “shear” and “angular” misalignment in the sum.\n37\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nFigure 17. Synthetic spectra were generated using GFIT to simulate shear and angular misalignment with 5% change from the ideal line\nshape at a maximum optical path difference of 45 cm. These spectra were then passed through LINEFIT 14.8 to confirm that the modulation\nefficiency and phase errors were as expected.\n\n8.1\n\nGeneral comments on the results\n\nThe results of this error budget analysis only account for changes using a single instrument, and therefore cannot assess\n785\n\nimprovements to GGG2020 that affect inter-instrument precision, such as consistent continuum fitting across the network,\nchannel fringe fitting when needed, a priori shape improvements, and so on. Site-to-site variability have been assessed in\nsections §7.1–§7.3. The results in this section quantify the errors incurred by uncertainties in a single instrument setup.\nThe method of simulating modulation efficiency errors in GGG2014 (Wunch et al., 2015) was incorrect, resulting in an\ninferred uncertainty from ILS errors that is too large, likely by about a factor of 2 (see Appendix B1 for details). The change\n\n790\n\nfrom the errant ILS modeling to our current model, on its own, will produce an apparent overall uncertainty reduction for\nGGG2020 when compared with GGG2014, but there have been no improvements in GGG2020 with respect to fitting imperfect\nILS. However, there are several other improvements in GGG2020 that have resulted in systematic reductions in the uncertainty,\nincluding higher order continuum fitting (§6), solar-gas stretch fitting (§8), and gas-specific spectroscopy (§3.1) and line shape\nfitting improvements (§3.2).\n\n795\n\nIn GGG2014, our retrievals were performed on a 1 km grid, and we shifted the profiles down by 1 level (or 1 km at all\naltitudes). In GGG2020, our retrievals are on a grid that increases in spacing with altitude, and a shift down by 1 level is\n\n38\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nxco2 - 20190611\n\nxluft - 20190611\n0.25\n\n1\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.6\n\n%\n\n0.4\n\n0.2\n\n0\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.2\n0.15\n0.1\n\n%\n\n0.8\n\n0.05\n0\n-0.05\n\n-0.2\n\n-0.1\n-0.15\n\n-0.4\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n0\n\n90\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n°)\n\n60\n\n70\n\n80\n\n90\n\n°)\n\nxco - 20190611\n\nxch4 - 20190611\n2.5\n\n0.5\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.3\n0.2\n\n%\n\n0.1\n0\n-0.1\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n2\n\n1.5\n\n1\n\n%\n\n0.4\n\n0.5\n\n0\n\n-0.2\n\n-0.5\n\n-0.3\n\n-1\n\n-0.4\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n0\n\n90\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n°)\n\n60\n\n70\n\n80\n\n90\n\n°)\n\nFigure 18. June 11, 2019 error budget from East Trout Lake. The figures show the percent difference between the perturbed test and the\nstandard retrieval plotted as a function of solar zenith angle. “Sum” in the legend means the quadrature sum of the other terms. The retrievals\nplotted here are Xluft , XCO2 , XCH4 , and XCO .\n\n39\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nxh2o - 20190611\n\nxhdo - 20190611\n4\n\n2\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n1\n\n%\n\n0.5\n0\n-0.5\n\n3\n\n2\n\n1\n\n%\n\n1.5\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0\n\n-1\n\n-1\n\n-2\n\n-1.5\n-2\n\n-3\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n90\n\n0\n\n10\n\n20\n\n30\n\nxn2o - 20190611\n\n50\n\n60\n\n70\n\n80\n\n90\n\n70\n\n80\n\n90\n\n°)\n\nxhf - 20190611\n\n1.2\n\n5\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.8\n0.6\n0.4\n0.2\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n4\n3\n2\n\n%\n\n1\n\n%\n\n40\n\nSolar Zenith Angle (\n\n°)\n\n1\n0\n\n0\n\n-1\n\n-0.2\n\n-2\n\n-0.4\n\n-3\n\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n90\n\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n°)\n\n60\n\n°)\n\nFigure 19. As in Figure 18, but for XH2 O , XHDO , XN2 O , and XHF .\n\nroughly 1 km at the tropopause, but smaller below and larger above. This change is most likely to affect the retrievals of gases\nfor which there is a rapid change in abundance near the tropopause and above: N2 O, CH4 , and HF. Therefore, our shift for the\nGGG2020 error budget represents a larger perturbation to the a prior shape for these gases, which will cause larger errors in\n800\n\nretrievals. However, because HF is a species found primarily in the stratosphere, and N2 O and CH4 are species found primarily\nin the troposphere, retrievals of HF can be used to diagnose and reduce the impact of the profile shift errors on XN2 O and XCH4\n(e.g., Washenfelder et al., 2003; Saad et al., 2014, 2016; Wang et al., 2014).\nIn each section below, we will discuss the results for each gas, keeping in mind the reductions in error from the ILS model,\nand the inflation of error from the prior shifts.\n\n40\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nxwco2 - 20190611\n\nxlco2 - 20190611\n0.6\n\n0.4\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.2\n\n%\n\n0.1\n\n0\n\n-0.1\n\n0.5\n0.4\n0.3\n0.2\n\n%\n\n0.3\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.1\n0\n-0.1\n-0.2\n\n-0.2\n\n-0.3\n-0.4\n\n-0.3\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n0\n\n90\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n°)\n\n60\n\n70\n\n80\n\n90\n\n°)\n\nvsf_hcl - 20190611\n4\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n3\n\n%\n\n2\n\n1\n\n0\n\n-1\n\n-2\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n90\n\n°)\n\nFigure 20. As in Figure 18, but for XlCO2 , XwCO2 , and HCl scale factors (vsf_hcl).\n805\n\n8.2\n\nXluft\n\nXluft is the column-averaged amount of dry air, and is equivalent to the parameter Xair in GGG2014. The error budget for\nXluft (Figures 18 and 21) is very similar to that of Xair in GGG2014, with uncertainties smaller than 0.7% for all solar zenith\nangles less than 82◦ . The error is dominated by pointing offsets at large solar zenith angles, and zero level offsets contribute\nsignificantly to the error at all solar zenith angles.\n810\n\n8.3\n\nXCO2\n\nThe XCO2 error budget is smaller than for GGG2014 (Wunch et al., 2015), mostly from the reduced continuum fitting errors.\nThe GGG2020 errors are below 0.16% (∼0.6 ppm) for solar zenith angles less than 82◦ , though if extrapolated linearly to\n41\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nsmaller solar zenith angles, the error could become larger than 0.15% at 0 degrees (Figures 18 and 21). The largest sources\nof error at lower solar zenith angles are from prior pressure offsets and misalignment. At larger solar zenith angles, the error\n815\n\nbecomes dominated by prior temperature errors and zero level offsets.\n8.4\n\nXCH4\n\nThe XCH4 error budget is smaller than for GGG2014 (Wunch et al., 2015). There is a significant reduction in the errors\nassociated with observer-sun Doppler stretch (OSDS) offsets and continuum fitting errors. The GGG2020 errors are below\n0.4% (∼7 ppb) for solar zenith angles less than 82◦ (Figures 18 and 21). The largest sources of error at lower solar zenith\n820\n\nangles are from prior profile shifts and prior pressure errors. At larger solar zenith angles, the error is dominated by prior\nprofile shifts. Errors caused by profile shifts can be mitigated by extracting the tropospheric partial column of XCH4 using the\nSaad et al. (2014) or Wang et al. (2014) methods.\n8.5\n\nXCO\n\nThe XCO spectral fitting has been substantially improved in GGG2020, largely because of our reduced sensitivity to errors in\n825\n\nthe observer-sun Doppler stretch (OSDS), and also because we removed one of the fitted windows from our standard analysis\nin GGG2020 that had relatively poorer spectral fits. The GGG2020 errors are below 2% (∼ 2 ppb assuming a 100 ppb column)\nfor all SZA < 82◦ . The largest sources of error are the prior CO enhancement, the prior shift, prior temperature, and shear\nmisalignment (Figures 18 and 21).\n8.6\n\n830\n\nXH2 O and XHDO\n\nThe error budget for water and HDO is roughly the same as for GGG2014 and earlier, with total errors under 2% in XH2 O and\n3% in XHDO over all solar zenith angles less than 82◦ . The largest component of the error budget for water vapor and HDO is\nthe shape of the a priori profile, which dominates the error budget for all solar zenith angles below 75◦ for water, and over all\nsolar zenith angles below 82◦ for HDO (Figures 19 and 22).\n8.7\n\n835\n\nXN2 O\n\nThe XN2 O error budget is roughly the same as in GGG2014, with total errors less than 1.25% (∼4 ppb) over all solar zenith\nangles. The largest source of error is the prior shift, which is not surprising, given the rapid chemical destruction of N2 O above\nthe tropopause, though the magnitude of the error is about twice as large as it was for GGG2014. As discussed above, this\nis likely caused by differences in the way we shift the profile, and could be mitigated by extracting the tropospheric partial\ncolumn by adapting the Saad et al. (2014) approach. Other contributors to the total error include the prior pressure, and shear\n\n840\n\nand angular misalignments (Figures 19 and 22).\n\n42\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n8.8\n\nXHF\n\nHF has only a single absorption line (4038.96 cm−1 ) that is located on the wing of a strong water absorption feature, so the\nretrievals tend to be noisy, especially at high solar zenith angles and under wet conditions. The XHF error budget has reduced\nin GGG2020 compared with GGG2014, with total errors now less than 5% over all solar zenith angles. In GGG2014, the errors\n845\n\nwere typically below 8%, but that error was dominated by the much larger shear misalignment. The largest source of error in\nGGG2020 is the prior shift, followed closely by shear misalignment (Figures 19 and 22).\n8.9\n\nXlCO2 and XwCO2\n\nIn GGG2014 and previous versions, we did not retrieve strong (“lCO2 ”) and weak (“wCO2 ”) CO2 bands. The strong CO2\nretrieval errors are dominated by prior temperature errors, and the weak CO2 errors are dominated by both shear and angular\n850\n\nmisalignments, errors in the prior pressure, adjustments to the continuum curvature, and zero level offsets (Figures 20 and 23).\nThe strong lCO2 retrieval errors are less than 0.3% over all solar zenith angles, and the weak wCO2 retrievals have around\n0.5% errors at all solar zenith angles, declining slightly at higher angles.\n8.10\n\nVSF HCl\n\nIn this error budget, we have included the scale factors retrieved for HCl (vsf_hcl in Figs. 20 and 23). In the East Trout Lake\n855\n\ninstrument and most others in the network, a sealed HCl cell filled with a known quantity of gas (Hase et al., 2013) is placed\npermanently in the solar beam inside the evacuated spectrometer to monitor long-term changes in ILS. Because the quantity\nof gas in the cell is significantly larger than the atmospheric abundance, the atmospheric component is negligible and largely\nindependent of surface pressure or other atmospheric adjustments. To assess the HCl retrieval sensitivity to changes in ILS and\nother parameters, we include the HCl scale factors in our error budget.\n\n860\n\nThe retrieval errors in the scaling factors retrieved for HCl in a sealed cell are dominated by errors in the instrument line\nshape with no significant solar zenith angle dependence. This is a comforting result, showing that our HCl retrievals are a good\ndiagnostic for instrument line shape drift. The HCl retrievals are not included in the standard public data files as they are used\nprimarily for diagnostic purposes.\n8.11\n\n865\n\nUncertainty estimate comparison\n\nFor six products (XCO2 , XwCO2 , XlCO2 , XCH4 , XCO , and XH2 O ) we can compare the uncertainty estimates derived from the\nerror budgets with those computed from in situ comparisons similar to those in §7.3 but with one difference: the comparisons\nin §7.3 use the in situ vertical profiles as the prior trace gas profiles in the TCCON retrievals; the in situ comparisons in this\nsection use standard TCCON GGG2020 prior profiles. For the in situ uncertainty, we use the median absolute deviation of the\nTCCON Xgas values from the in situ Xgas values after removing the mean bias for each Xgas (i.e. the correction factor in\n\n870\n\nTable 2). To convert the percent error from the error budget into a column-average mole fraction, we use the mean total percent\nerror across all three days used in the error budget (18 Feb, 11 June, and 23 July 2019) binned by SZA in 5° increments. We\n43\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nxco2\n\nxluft\n0.16\n\n0.7\n20190218\n20190611\n20190723\n\n0.65\n\n20190218\n20190611\n20190723\n\n0.15\n\n0.6\n\n0.14\n\n0.55\n\n%\n\n%\n\n0.5\n\n0.13\n\n0.45\n\n0.12\n\n0.4\n0.35\n\n0.11\n0.3\n\n0.1\n\n0.25\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n0\n\n90\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n90\n\n70\n\n80\n\n90\n\n°)\n\nxco\n\nxch4\n\n2\n\n0.34\n20190218\n20190611\n20190723\n\n0.32\n\n10\n\n°)\n\n20190218\n20190611\n20190723\n\n1.9\n\n0.3\n\n1.8\n\n0.28\n\n1.7\n\n0.26\n\n%\n\n%\n\n1.6\n0.24\n\n1.5\n0.22\n\n1.4\n0.2\n\n1.3\n\n0.18\n\n1.2\n\n0.16\n\n1.1\n\n0.14\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n0\n\n90\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n°)\n\n60\n\n°)\n\nFigure 21. These figures show the sum in quadrature of all the errors plotted in Figure 18 for all three dates. The errors plotted here are for\nXluft , XCO2 , XCH4 , and XCO .\n\n44\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nxh2o\n\nxhdo\n\n1.8\n\n3\n20190218\n20190611\n20190723\n\n1.6\n\n20190218\n20190611\n20190723\n\n2.8\n2.6\n\n1.4\n2.4\n2.2\n\n%\n\n%\n\n1.2\n1\n\n2\n1.8\n\n0.8\n\n1.6\n0.6\n1.4\n0.4\n\n1.2\n\n0.2\n\n1\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n90\n\n0\n\n10\n\n20\n\n30\n\n40\n\nxn2o\n1.2\n20190218\n20190611\n20190723\n\n60\n\n70\n\n80\n\n90\n\n70\n\n80\n\n90\n\n°)\n\nxhf\n\n4.4\n\n1.15\n\n50\n\nSolar Zenith Angle (\n\n°)\n\n20190218\n20190611\n20190723\n\n4.2\n4\n\n1.1\n3.8\n3.6\n\n1\n\n%\n\n%\n\n1.05\n\n3.4\n3.2\n\n0.95\n\n3\n\n0.9\n2.8\n\n0.85\n\n2.6\n\n0.8\n\n2.4\n\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n90\n\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n°)\n\n60\n\n°)\n\nFigure 22. As in Figure 21, but for XH2 O , XHDO , XN2 O , and XHF . XHF values above 68° SZA are not available on 2019-07-23 because\nthe HF lines were blacked out by H2 O absorbance.\n\n45\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nxwco2\n\nxlco2\n0.48\n\n0.3\n20190218\n20190611\n20190723\n\n0.28\n\n20190218\n20190611\n20190723\n\n0.47\n0.46\n\n0.26\n\n0.45\n0.44\n\n%\n\n%\n\n0.24\n0.22\n\n0.43\n0.42\n\n0.2\n\n0.41\n0.18\n\n0.4\n0.16\n\n0.39\n0.38\n\n0.14\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n0\n\n90\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n°)\n\n60\n\n70\n\n80\n\n90\n\n°)\n\nvsf_hcl\n2.77\n20190218\n20190611\n20190723\n\n2.76\n2.75\n2.74\n\n%\n\n2.73\n2.72\n2.71\n2.7\n2.69\n2.68\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n90\n\n°)\n\nFigure 23. As in Figure 21, but for XlCO2 , XwCO2 , and HCl scale factors (vsf_hcl).\n\ninterpolate this to the mean SZA of all spectra used in the in situ comparison for that gas and multiply this interpolated mean\npercentage by the mean TCCON Xgas value across all the in situ comparisons. The results are presented in Table 3.\nIt is important to acknowledge that the error amounts calculated from the in situ comparison are (for most gases) conserva875\n\ntive, for several reasons. First, in situ profiles are usually taken when the target TCCON station is near optimal performance, so\nthose comparisons are unlikely to capture the full range of error sources. Second, the in situ profiles are heavily concentrated\nover certain TCCON sites, also limiting how representative they are. Finally, the TCCON Xgas values compared against the\nin situ values are averaged over a minimum of 2 hours. This will reduce sources of random error. However, we believe this\nis still a worthwhile evaluation of measurement accuracy because (a) there is real physical variation in the atmosphere during\n\n880\n\nthe in situ profile, and the time averaging is necessary to account for that and (b) many of the factors considered in the error\n\n46\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nGas\n\nSZA\n\nMean abs. dev.\n\nError budget\n\n\u000fin situ\n\n\u000fmeas\n\n\u000fFT\n\n\u000fstrat\n\nXCO2\n\n46°\n\n0.42 ppm\n\n0.47 ppm\n\n0.053 (0.30) ppm\n\n0.033 (0.16) ppm\n\n0.032 (0.12) ppm\n\n0.061 (0.072) ppm\n\nXwCO2\n\n46°\n\n0.43 ppm\n\n1.8 ppm\n\n0.062 (0.36) ppm\n\n0.037 (0.16) ppm\n\n0.038 (0.15) ppm\n\n0.075 (0.10) ppm\n\nXlCO2\n\n46°\n\n0.75 ppm\n\n0.66 ppm\n\n0.038 (0.24) ppm\n\n0.025 (0.14) ppm\n\n0.020 (0.067) ppm\n\n0.057 (0.060) ppm\n\nXCH4\n\n46°\n\n4.9 ppb\n\n3.9 ppb\n\n2.0 (9.6) ppb\n\n0.65 (3.1) ppb\n\n0.19 (0.49) ppb\n\n3.4 (6.3) ppb\n\nXCO\n\n43°\n\n8.1 ppb\n\n1.7 ppb\n\n2.8 (14.0) ppb\n\n1.9 (9.3) ppb\n\n0.13 (0.39) ppb\n\n0.24 (4.8) ppb\n\nXH2 O\n\n52°\n\n140 ppm\n\n33 ppm\n\n100 (950) ppm\n\n100 (950) ppm\n\n0 (0) ppm\n\n0 (0) ppm\n\nTable 3. A comparison of typical errors calculated from the differences between TCCON and in situ Xgas values (“Mean abs. dev.” in the\ntable) and errors calculated from the error budget (“Error budget” in the table). The text gives details on how each error metric was computed.\n“SZA” gives the solar zenith angle for which the error budget percent was taken to calculate the “Error budget” column. The remaining four\ncolumns give the total 2σ uncertainty on the in situ data (\u000fin situ ), followed by the 2σ uncertainty components coming from the in situ measurement itself (\u000fmeas ), the unmeasured free troposphere (\u000fFT ), and the unmeasured stratosphere (\u000fstrat ). The last two components are 0 for\nXH2 O because the radiosonde measurements used always cover the free troposphere, and we assume that error in the meteorological model\nused to obtain the stratospheric H2 O profile is negligible. The parenthetical numbers give the mean value per TCCON/in situ comparison\nfor each uncertainty component; the non-parenthetical errors are smaller because they are calculated by formally propagating the error from\n√\nindividual comparisons to the mean absolute deviation, thus reducing by n. Appendix C6 describes how the uncertainty components from\nthe last 3 columns are calculated and combined to give \u000fin situ in the fifth column.\n\nbudget will not average out over the coincidence window. For example, angular or shear misalignment of the instrument would\nbe essentially constant over an entire day.\nFor three Xgas products (XCO2 , XlCO2 , and XCH4 ) the in situ and error budget estimates are similar, which gives us\nconfidence in the error budget estimates. For XwCO2 , the error budget estimate is much larger than the in situ error estimate.\n885\n\nIt may be that the error budget tested larger errors in the stratosphere temperature or VMR prior profile than were observed\nduring the in situ comparisons, as the XwCO2 product is more sensitive to the upper atmosphere than the other CO2 products\nin GGG2020. (Pressure errors could be another source of the overestimate, but the pressure perturbation test was designed to\navoid introducing an overly large perturbation to the stratosphere.) As we treat the in situ-derived errors as conservative, this\nsituation is acceptable, but will be investigated in the future.\n\n890\n\nBoth XCO and XH2 O had larger errors in comparison with in situ data than through the error budget. For XCO , the difference\nin error estimates is 6.4 ppb. Almost half of that is attributed to uncertainty in the in situ measurements. The uncertainty in\nindividual comparisons (the parenthetical numbers in Table 3) is quite a bit larger; if part of this error is systematic (such as\nfrom drift in calibration tanks, e.g., Andrews, 2019), that could explain the remaining difference. For XH2 O , this is because of\nuncertainty in the radiosondes used to compare against. The radiosondes used at ARM have a 4 or 5% uncertainty in relative\n\n895\n\nhumidity (https://www.arm.gov/publications/tech_reports/handbooks/sonde_handbook.pdf, last accessed 10 Apr 2023). When\nwe propagate this uncertainty to the mean absolute deviation, it works out to 103 ppm—very nearly the missing 110 ppm\nbetween the two error estimates.\n47\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n9\n\nMiscellaneous changes\n\n9.1\n900\n\nAK binning\n\nThe public GGG2020 TCCON files now include one averaging kernel (AK) per observation. This is a change from GGG2014,\nwhere the public files included a table of canonical AKs for a limited set of SZAs, and users were required to interpolate the\nAKs to the SZA of each spectrum. This was done in response to user requests to simplify the use of the averaging kernels. This\ndoes not mean that averaging kernels are computed by GGG for every TCCON observation (they are not). Internally, we still\nuse a table of precomputed AKs, which are interpolated as needed to provide per-spectrum AKs in the public files. This affords\n\n905\n\nsignificant saving in data storage, as the files GGG requires for AK calculation are very large.\nThough users of public TCCON data no longer need to know how the AK tables work, there are two changes from GGG2014\nthat we wish to document here.\nFirst, in GGG2020, the bin coordinate has changed from solar zenith angle (SZA) to “slant Xgas ,” which is defined as:\n\nSlant Xgas = airmass · Xgas\n910\n\n(11)\n\nwhere “airmass” is the airmass calculated by GGG in the O2 window and “Xgas ” is the column average mole fraction of\nthe gas of interest. Using slant Xgas as the bin coordinate correctly accounts for cases where the dynamic range of a gas’s\nconcentrations is large enough to change the AK at a single SZA. This can be seen in Fig. 24. For CO2 (Fig. 24a,b), the AKs\nvary smoothly and monotonically with either SZA or slant XCO2 . However, for H2 O, the AKs do not vary monotonically with\nSZA (Fig. 24c) but do with slant XH2 O (Fig. 24d). Therefore, slant Xgas was adopted as the binning coordinate for all AKs for\n\n915\n\nconsistency.\nSecond, in order to provide per-spectrum AKs in the public TCCON data files without significantly increasing the file size,\nit was necessary to ensure that observations with similar slant Xgas values had identical AKs so that the netCDF compression\nalgorithm could operate effectively. We achieved this by “quantizing” the slant Xgas values that we interpolated the AKs to;\nthat is, we select 500 slant Xgas values that cover the expected range of slant Xgas , plus 50 additional points to cover extreme\n\n920\n\nvalues. Each observation then uses the AK corresponding to the one of those 550 slant Xgas values closest to its true slant Xgas\nvalue. This scheme keeps the difference between the quantized and full resolution AKs to < 1% in 90% of observations while\nonly increasing file size by ∼ 20%.\n9.2\n\nA priori profiles and AK corrections\n\nAs described in §4.3, the a priori profiles reported in the published GGG2020 netCDF files are in wet mole fraction. When\n925\n\napplying an averaging kernel correction to calculate the Xgas value that would be retrieved by TCCON for an arbitrary gas\nprofile, that gas profile must be converted into wet mole fraction. This can be done using either the TCCON H2 O a priori\nprofile provided or an H2 O profile measured or modeled coincidentally with the gas profile for which an Xgas value is desired.\n\n48\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n90\n\n0.0\n\nlog10(XCO2 airmass)\n\n50\n\n2.8\n\nSZA\n\nPressure (atm)\n\n2.6\n\n60\n0.4\n\n40\n\n0.6\n\n3.0\n\n30\n\n0.8\n\n20\n\n3.2\n\n10\n\n1.0\n0.2\n\n0.4\n\n0.6 0.8 1.0\nCO2 Column AK\n\n1.2\n\n1.4\n\n0\n\n0.2\n\n90\n\n0.0\n\n0.4\n\n0.6 0.8 1.0\nCO2 Column AK\n\n1.2\n\n1.4\n\n(d)\n\n1.2\n\n80\n\n1.4\n\n70\n\n0.2\n\n1.6\n\n60\n0.4\n\n50\n\n1.8\n\n40\n\n2.0\n\n30\n\n2.2\n\nSZA\n\nPressure (atm)\n\n2.4\n\n70\n\n0.2\n\n(c)\n\n(b)\n\n80\n\nlog10(XH2O airmass)\n\n(a)\n\n0.6\n0.8\n\n20\n\n2.4\n\n10\n\n1.0\n0.2\n\n0.4\n\n0.6\n0.8\nH2O Column AK\n\n1.0\n\n1.2\n\n2.6\n\n0\n\n0.2\n\n0.4\n\n0.6\n0.8\nH2O Column AK\n\n1.0\n\n1.2\n\nFigure 24. CO2 and H2 O AKs from four days’ measurements at the TCCON site in Lamont, OK, USA. (a) CO2 AKs binned by SZA. (b)\nCO2 AKs binned by slant XCO2 . (c) H2 O AKs binned by SZA. (d) H2 O AKs binned by slant XH2 O .\n\nUsers who are unsure which is appropriate for their application are encouraged to reach out to the TCCON network chairs\n(listed at https://tccon-wiki.caltech.edu/Main/SteeringCommitteeMembership) for assistance.\n930\n\n9.3\n\nChanges to quality flags\n\nAs in GGG2014, a spectrum is flagged as being poor quality if any of the retrieved Xgas or Xgas error values, or ancillary\nvariables pertaining to instrument operation or local observation conditions are outside of expected ranges. Such spectra are\nnot included in the public data files. In GGG2020, spectra may also be flagged as poor quality and withheld if:\n– the staff at the TCCON site identify a hardware issue affecting that spectrum\n935\n\n– during pre-release data review, a time period containing that spectrum is identified as out-of-family for TCCON data.\nThe latter case focuses on a smoothed timeseries of Xluft and DIP. As shown in §7.3 and §8, deviation of Xluft from the\nnetwork median correlate with bias in the other Xgas products. (See §3.3 for a definition of Xluft .) Therefore, when a 500spectrum rolling median of Xluft falls consistently outside the nominal range of 0.995 to 1.003, that time period is rejected,\nas the Xgas products will likely have biases exceeding the expected TCCON accuracy. Likewise, DIP is a measure of detector\n\n940\n\nnonlinearity (§5.1), and testing has shown that increasing magnitude of DIP increases bias in XCO2 (Fig. 26). Thus, data where\nDIP consistently exceeds ±5 × 10−4 are removed in order to keep the XCO2 bias less than 0.25 ppm.\n49\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n104\n\n0.2\n\n0.4\n\n0.6 0.8 1.0 1.2\nXCO2 column AK (unitless)\n\n1.4\n\n104\n\n400\n\n103\n\n600\n800\n\n102\n\n0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75\nXlCO2 column AK (unitless)\n\n104\n103\n\n400\n600\n\n102\n\n800\n0.0\n\n0.2\n\n0.4 0.6 0.8 1.0\nXHF column AK (unitless)\n\n1.2\n\n1.4\n\n104\n\n200\nSlant XN2O\n\nPressure (hPa)\n\n(g) 0\n400\n\n103\n\n600\n800\n1000\n\n0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25 2.50\nXN2O column AK (unitless)\n\nSlant XH2O\n\nPressure (hPa)\n\n400\n\n104\n\n600\n800\n1000\n\n0.2\n\n0.4 0.6 0.8 1.0 1.2\nXH2O column AK (unitless)\n\n1.4\n\nSlant XwCO2\n\n800\n0.5\n\n0.6\n\n0.7 0.8 0.9 1.0 1.1\nXCH4 column AK (unitless)\n\n1.2\n\n107\n\n400\n\n106\n\n600\n800\n0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75\nXO2 column AK (unitless)\n\n105\n104\n\n200\n\n103\n\n400\n600\n\n102\n\n800\n0.6\n\n0.8\n\n1.0 1.2 1.4 1.6 1.8\nXCO column AK (unitless)\n\n101\n\n2.0\n\n105\n\n(j) 0\n200\n400\n\n104\n\n600\n800\n1000\n\n103\n\n103\n\n200\n\n1000\n\n102\n\n200\n\n104\n\n600\n\n(h) 0\n\n105\n\n(i) 0\n\n400\n\n1000\n\n101\n\nPressure (hPa)\n\n1000\n\n105\n\n(f) 0\nPressure (hPa)\n\n200\n\nSlant XHF\n\nPressure (hPa)\n\n(e) 0\n\n102\n\n200\n\n1000\n\nPressure (hPa)\n\n1000\n\n0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25\nXwCO2 column AK (unitless)\n\n(d) 0\nPressure (hPa)\n\n200\nSlant XlCO2\n\nPressure (hPa)\n\n(c) 0\n\n800\n1000\n\n102\n\n103\n\n600\n\nSlant XCH4\n\n800\n1000\n\n400\n\nSlant XO2\n\n600\n\n200\n\nSlant XCO\n\n103\n\nSlant XHDO\n\n400\n\n104\n\n(b) 0\nPressure (hPa)\n\n200\nSlant XCO2\n\nPressure (hPa)\n\n(a) 0\n\n1.0\n\n1.2\n1.4\n1.6\n1.8\nXHDO column AK (unitless)\n\n2.0\n\n103\n\nFigure 25. Precomputed column AKs for TCCON Xgas products: (a) XCO2 , (b) XwCO2 , (c) XlCO2 , (d) XCH4 , (e) XHF , (f) XO2 , (g)\nXN2 O , (h) XCO , (i) XH2 O , and (j) XHDO .\n\n50\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nFigure 26. Detector nonlinearity can cause a bias in XCO2 . This figure shows an example of the difference between the XCO2 retrieved after\ncorrecting the nonlinearity and prior to the nonlinearity correction as a function of the DIP parameter, that is a proxy for nonlinearity. Prior\nto correction, the Indianapolis data had DIP values that were almost exclusively negative. To limit the XCO2 bias caused by nonlinearity to\nless than 0.25 ppm, the absolute value of the DIP must be smaller than 0.5×10−3 .\n\n10\n\nConclusions\n\nThe GGG2020 TCCON data product incorporate numerous improvements to the GGG retrieval, based both on first-principle\nunderstanding and empirical evaluation. To review:\n945\n\n– The interferogram-to-spectrum conversion has added checks and diagnostics for detector nonlinearity or saturation, as\nwell as a modification to the phase correction that reduces bias between forward and reverse scans of the interferometer.\n– The solar and telluric spectroscopic linelists used in the GGG forward model have been updated to reflect new laboratory\nand atmospheric/solar observing studies, to include non-Voigt lineshapes, and to reduce an observed temperature and\nwater dependence in the O2 column amounts.\n\n950\n\n– The a priori inputs of atmospheric state (temperature, pressure, and composition) have increased temporal resolution and\nthe trace gas profiles have been updated to better reflect both atmospheric growth rates of key species and gradients in\ntheir mixing ratios across the tropopause.\n– Improvements to fitting the continuum and channel fringes in the spectra.\n– A more flexible airmass correction applied to Xgas value from individual spectral windows, rather than multi-window\n\n955\n\naverages of said values.\n– A change to how retrieved Xgas values from multiple spectral windows measuring the same gas are averaged together\nthat eliminates a dependence on how many observations were averaged at once.\n51\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n– An updated in situ correction factor that increases the number of profiles used to tie TCCON to the calibration scales\nused by in situ GHG measurements.\n– Improvements to user-friendliness in how AKs and prior profiles are reported in public files.\n\n960\n\nThere remains work to be done to further improve the TCCON data product. Implementing the capability in GGG to account\nfor errors in ILS remains a high priority. This was planned for inclusion in GGG2020, but could not be completed in time. It\nis expected that this capability will be an important tool to eliminate the XCO2 bias seen in comparison with in situ profiles\nas Xluft deviates from its nominal 0.999 value. A second high priority objective is to investigate the temperature dependence\n965\n\nseen in the N2 O and (to a much lesser extent) CH4 data and correct the underlying spectroscopic terms.\nWe currently plan to develop a minor release, GGG2020.1, within the next several years that will include additional postprocessing bias corrections to address the bias of XCO2 versus Xluft and XN2 O and XCH4 versus temperature. We expect these\nwill allow us to release data from the early years of several sites, which is currently flagged as poor quality due to out-of-bounds\nXluft as well as improve the XN2 O data substantially. As this would be a post-processing-only update, the reprocessing could\n\n970\n\nbe completed very rapidly.\nAt time of writing, 26 TCCON sites have reprocessed their existing data with GGG2020. Several sites are still in the process\nof carrying out this reprocessing, in many cases to improve the data quality based on new diagnostics available in GGG2020.\nWork is ongoing towards completing these sites’ reprocessing. Extensions to the existing data records will be released monthly\ngoing forward.\n\n975\n\n11\n\nCode and data availability\n\nAll TCCON GGG2020 data is linked through tccondata.org and stored as DOI-tagged datasets on CaltechDATA (data.caltech.\nedu). Each TCCON site has a separate repository and DOI on CaltechDATA; these are listed in Table 1. If a future correction\nrequires a revision of previously published data, that revision will receive a new DOI. Users are encouraged to check tccondata.\norg for the latest revisions of data rather than relying on Table 1. A repository containing the full set of TCCON GGG2020\n980\n\ndata is also available on CaltechDATA with the DOI 10.14291/TCCON.GGG2020 (Total Carbon Column Observing Network\n(TCCON) Team, 2022). Users are asked to cite the individual sites’ data records rather than the combined record as this\nhelps track usage of site data and thus support the ongoing operation of these sites. We provide a citation generator at https:\n//tccondata.org/metadata/siteinfo/genbib/. All data is provided in netCDF format, and additional documentation for the data is\navailable at https://tccon-wiki.caltech.edu/. The GGG2020 retrieval software is archived on CaltechDATA (Toon, 2023) as well\n\n985\n\nas publicly available through GitHub at https://github.com/TCCON/GGG.\nAppendix A: Abbreviations\nAbbreviations used in this paper are listed in Table A1.\n\n52\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nAbbreviation\n\nMeaning\n\nNotes\n\nADCF\n\nAirmass dependent correction factor\n\nSee SZA note\n\nAICF\n\nAirmass independent correction factor\n\nAlso call the “in situ correction factor”\n\nAK\n\nAveraging Kernel\n\nRefers to column averaging kernels unless otherwise indicated\n\nCBF\n\nContinuum basis function\n\nFT\n\nFree troposphere\n\nFFT\n\nFast Fourier transform\n\nFOV\n\nField of view\n\nFTIR\n\nFourier transform infrared\n\nFTS\n\nFourier transform spectrometer\n\nFVSI\n\nFraction variation in solar intensity\n\nILS\n\nInstrument line shape\n\nIR\n\nInfrared\n\nGGG\n\n-\n\nGHG\n\nGreenhouse gas\n\nLM\n\nLine mixing\n\nMIR\n\nMid infrared\n\nMOPD\n\nMaximum optical path difference\n\nMOPITT\n\nMeasurements of Pollution in the Troposphere\n\nNDIR\n\nNondispersive infrared\n\nNIR\n\nNear infrared\n\nOSDS\n\nObserver-sun Doppler stretch\n\nRH\n\nRelative humidity\n\nRMS\n\nRoot mean square/squared\n\nqSDV\n\nQuadratice speed-dependent Voigt\n\nSZA\n\nSolar zenith angle\n\nTCCON\n\nTotal Carbon Column Observing Network\n\nUTC\n\nCoordinated Universal Time\n\nVMR\n\nVolume mixing ratio\n\nVSF\n\nVMR scale factor\n\nXgas\n\nColumn-average mole fraction\n\nZLO\n\nZero level offset\n\nZPD\n\nZero path difference\n\nThe name of the retrieval, not an abbreviation\n\nAn instrument on the Terra satellite\n\n“SZA-” and “airmass-dependence” are used equivalently\n\n“Xgas ” is generic; “XCO2 ”, “XCH4 ”, etc. are specific\n\nTable A1. Abbreviations used in this paper.\n\n53\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nxco2 - 20190218\n\nxluft - 20190218\n0.25\n\n1\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.6\n\n%\n\n0.4\n\n0.2\n\n0\n\n0.2\n0.15\n0.1\n\n%\n\n0.8\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.05\n0\n-0.05\n\n-0.2\n\n-0.1\n-0.15\n\n-0.4\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n0\n\n90\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n°)\n\n60\n\n70\n\n80\n\n90\n\n°)\n\nxco - 20190218\n\nxch4 - 20190218\n2.5\n\n0.5\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.3\n0.2\n\n%\n\n0.1\n0\n-0.1\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n2\n\n1.5\n\n1\n\n%\n\n0.4\n\n0.5\n\n0\n\n-0.2\n\n-0.5\n\n-0.3\n\n-1\n\n-0.4\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n0\n\n90\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n°)\n\n60\n\n70\n\n80\n\n90\n\n°)\n\nFigure B1. February 18, 2019 error budget from East Trout Lake. The figures show the percent difference between the perturbed test and the\nstandard retrieval plotted as a function of solar zenith angle. The retrievals plotted here are Xluft , XCO2 , XCH4 , and XCO .\n\nAppendix B: Error budget\nFor completeness, we include the error budget figures equivalent to Figures 18–20 for February and July at East Trout Lake in\n990\n\nFigs. B1 to B6. February is extremely cold (-30 to -15◦ C) and dry (<500 ppm XH2 O ), with short days and large solar zenith\nangles. July is warm (20 to 30◦ C) and humid (3000 to 4500 ppm XH2 O ), causing the HF absorption feature to be blacked out\nby adjacent H2 O lines at higher solar zenith angles, causing unreliable retrievals of HF.\n\n54\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nxh2o - 20190218\n\nxhdo - 20190218\n4\n\n2\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n1\n\n%\n\n0.5\n0\n-0.5\n\n3\n\n2\n\n1\n\n%\n\n1.5\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0\n\n-1\n\n-1\n\n-2\n\n-1.5\n-2\n\n-3\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n90\n\n0\n\n10\n\n20\n\n30\n\nxn2o - 20190218\n\n50\n\n60\n\n70\n\n80\n\n90\n\n70\n\n80\n\n90\n\n°)\n\nxhf - 20190218\n\n1.2\n\n5\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.8\n0.6\n0.4\n0.2\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n4\n3\n2\n\n%\n\n1\n\n%\n\n40\n\nSolar Zenith Angle (\n\n°)\n\n1\n0\n\n0\n\n-1\n\n-0.2\n\n-2\n\n-0.4\n\n-3\n\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n90\n\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n°)\n\n60\n\n°)\n\nFigure B2. As in Figure B1, but for XH2 O , XHDO , XN2 O , and XHF .\n\nB1\n\nILS\n\nWe created synthetic spectra in GGG2020 with different ILS errors, following the formulation for the “shear” and “angular”\n995\n\nmisalignments tested for the GGG2014 error budget, and for the new formulation in GGG2020. We then passed these synthetic\nspectra through an ILS quantification program called LINEFIT (v14.8) (Hase et al., 1999), which calculates the modulation\nefficiency and phase error of the spectra. Here, we plot the LINEFIT-derived modulation efficiencies for these four cases\nin Figure B7. The GGG2020 shear and angular misalignments represent a ramp-up and ramp-down from 1.0 at zero path\ndifference to 5% offsets at 45 cm optical path difference, as expected. Unfortunately, the GGG2014 “shear” and “angular”\n\n1000\n\nmisalignments both model shear misalignments of different magnitudes. The GGG2014 “shear” case is, in fact, more like a\n\n55\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nxwco2 - 20190218\n\nxlco2 - 20190218\n0.6\n\n0.4\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.2\n\n%\n\n0.1\n\n0\n\n-0.1\n\n0.5\n0.4\n0.3\n0.2\n\n%\n\n0.3\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.1\n0\n-0.1\n-0.2\n\n-0.2\n\n-0.3\n-0.4\n\n-0.3\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n0\n\n90\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n°)\n\n60\n\n70\n\n80\n\n90\n\n°)\n\nvsf_hcl - 20190218\n4\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n3\n\n%\n\n2\n\n1\n\n0\n\n-1\n\n-2\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n90\n\n°)\n\nFigure B3. As in Figure B1, but for XlCO2 , XwCO2 , and HCl scale factors (vsf_hcl).\n\n15% ramp up as a function of optical path difference, and the GGG2014 “angular” case is more like a 3% ramp up. This will\nessentially double the inferred error from the ILS in GGG2014, when compared with GGG2020.\nAppendix C: AICF profile selection\nC1\n1005\n\nCO2 , CH4 , CO\n\nIn situ profiles for CO2 , CH4 , and CO were drawn primarily from the NOAA CO2 ObsPack (Cooperative Global Atmospheric\nData Integration Project, 2019), NOAA CH4 ObsPack (Cooperative Global Atmospheric Data Integration Project, 2020),\nNOAA AirCore dataset (Baier et al., 2021), additional AirCore launches at the Sodanklyä and Nicosia TCCON sites, the\n\n56\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nxco2 - 20190723\n\nxluft - 20190723\n0.25\n\n1\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.6\n\n%\n\n0.4\n\n0.2\n\n0\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.2\n0.15\n0.1\n\n%\n\n0.8\n\n0.05\n0\n-0.05\n\n-0.2\n\n-0.1\n-0.15\n\n-0.4\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n0\n\n90\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n°)\n\n60\n\n70\n\n80\n\n90\n\n°)\n\nxco - 20190723\n\nxch4 - 20190723\n2.5\n\n0.5\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.3\n0.2\n\n%\n\n0.1\n0\n-0.1\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n2\n\n1.5\n\n1\n\n%\n\n0.4\n\n0.5\n\n0\n\n-0.2\n\n-0.5\n\n-0.3\n\n-1\n\n-0.4\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n0\n\n90\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n°)\n\n60\n\n70\n\n80\n\n90\n\n°)\n\nFigure B4. July 23, 2019 error budget from East Trout Lake. The figures show the percent difference between the perturbed test and the\nstandard retrieval plotted as a function of solar zenith angle. The retrievals plotted here are Xluft , XCO2 , XCH4 , and XCO .\n\n57\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nxh2o - 20190723\n\nxhdo - 20190723\n4\n\n2\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n1\n\n%\n\n0.5\n0\n-0.5\n\n3\n\n2\n\n1\n\n%\n\n1.5\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0\n\n-1\n\n-1\n\n-2\n\n-1.5\n-2\n\n-3\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n90\n\n0\n\n10\n\n20\n\n30\n\nxn2o - 20190723\n\n50\n\n60\n\n70\n\n80\n\n90\n\n70\n\n80\n\n90\n\n°)\n\nxhf - 20190723\n\n1.2\n\n5\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.8\n0.6\n0.4\n0.2\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n4\n3\n2\n\n%\n\n1\n\n%\n\n40\n\nSolar Zenith Angle (\n\n°)\n\n1\n0\n\n0\n\n-1\n\n-0.2\n\n-2\n\n-0.4\n\n-3\n\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n90\n\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n°)\n\n60\n\n°)\n\nFigure B5. As in Figure 18, but for XH2 O , XHDO , XN2 O , and XHF .\n\nInfrastructure for Measurement of the European Carbon Cycle (IMECC) campaign, and the GO-AMAZON campaign. The\nObsPack contains data from numerous providers across different institutions; Tables C1 and C2 provide a detailed breakdown.\n1010\n\nFor the NOAA ObsPack Aircraft and AirCore profiles, the procedure used to match these data to TCCON sites will be detailed\nin the following subsections. For the remaining sources, the profiles were already associated with specific TCCON sites, so no\ncolocation was required.\nAll airborne data sources used for these profiles are listed in Tables C1 and C2. Ground data used to extend some of the\nprofiles to the surface are listed in Table C3.\n\n58\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nxwco2 - 20190723\n\nxlco2 - 20190723\n0.6\n\n0.4\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.2\n\n%\n\n0.1\n\n0\n\n-0.1\n\n0.5\n0.4\n0.3\n0.2\n\n%\n\n0.3\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n0.1\n0\n-0.1\n-0.2\n\n-0.2\n\n-0.3\n-0.4\n\n-0.3\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n0\n\n90\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n°)\n\n60\n\n70\n\n80\n\n90\n\n°)\n\nvsf_hcl - 20190723\n4\n\nfov\ncontinuum\nosds\npointing offset\nprior co enhanced\nprior h2o/hdo\nprior pressure\nprior shift\nprior temperature\npsurf\nangular\nshear\nzlo\nsum\n\n3\n\n%\n\n2\n\n1\n\n0\n\n-1\n\n-2\n0\n\n10\n\n20\n\n30\n\n40\n\n50\n\nSolar Zenith Angle (\n\n60\n\n70\n\n80\n\n90\n\n°)\n\nFigure B6. As in Figure B4, but for XlCO2 , XwCO2 , and HCl scale factors (vsf_hcl).\n1015\n\nC1.1\n\nObsPack\n\nThe ObsPack data is provided as a single time series per measurement campaign or similar source. To extract individual profiles\nfrom these files, we:\n1. Scan all files for data points within 2° (total distance) of an active TCCON site. When one is found, we store the list of\ndata points surrounding it in time that fall a box 10° longitude width and 5° latitude tall centered on the TCCON site as\n1020\n\na “chunk.” A chunk extends forward and backward in time from the point closest to the TCCON site and stops at the\nfirst data point in each direction that is outside the 10◦ × 5◦ box. Any profiles derived from this chunk are assigned to\n\nthe TCCON station it passes closest to.\n\n59\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nFigure B7. The modulation efficiencies tested in GGG2014 and GGG2020.\n\n2. Further filter the chunks based on the lowest altitude, highest altitude, number of data points, and minimum distance to\na TCCON site. This step was done interactively to find the filtering criteria that gave the best balance between number\n1025\n\nof chunks retained and the usefulness of the profile(s) within the chunk. The final criteria used were:\n– Minimum altitude below 2000 m\n– Maximum altitude above 7500 m\n– At least 20 data points\n– Approached within 0.1° of a TCCON station\n\n1030\n\n3. These filtered chunks were then individually evaluated and specific data points within them chosen by hand to use as\nprofiles. In this process, we considered the latitude/longitude position of the aircraft, the profile of altitude versus time,\nand the profile of CO2 or CH4 versus altitude. We generally selected as profiles times when the aircraft was consistently\nascending or descending, and excluded times of level flight. However, this had to be handled on a case-by-case basis to\nallow for profiles with a period of level flight in between two legs of an ascent or descent. If a chunk contained multiple\n\n1035\n\nascending/descending legs, we would split them if:\n– there was a clear separation in time, or\n– the legs measured different airmasses (evidenced by different CO2 or CH4 mole fractions)\n4. For each profile, we check for ground data in the ObsPack that can be used to extend the profile to the surface. We\nidentified which ground files in the ObsPack are near which TCCON sites by hand. We interpolate any data within 4\n60\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nTable C1. Airborne profile data used in the AICF calculation. “CO2 Obspack” is the CO2 GLOBALVIEWplus v5.0 ObsPack (Cooperative\nGlobal Atmospheric Data Integration Project, 2019) and “CH4 ObsPack” the CH4 GLOBALVIEWplus v2.0 ObsPack (Cooperative Global\nAtmospheric Data Integration Project, 2020). The “TCCON sites” column indicates at which sites profiles were used; the IDs are mapped\nto locations in Table 1 and numbers of profiles per site are given in Tables C4 and C5. In the “Providers” column, affiliations are given\nin parentheses. If only one affiliation is listed, it applies to all individuals named. Abbrevations: NASA = National Aeronautics and Space\nAdministration; LaRC = Langley Research Center; Harvard U. = Harvard University; CSUSB = California State University San Bernadino;\nGSFC = Goddard Space Flight Center; NCAR = National Center for Atmospheric Research; NOAA = National Oceanic and Atmospheric\nAdministration; GML = Global Monitoring Laboratory; FMI = Finnish Meteorological Institure; CARE-C = Climate and Atmosphere Research Center; LSCE/IPSL = Laboratoire des Sciences du Climat et de l’Environnement.\nSource\n\nCampaign or ID\n\nProviders\n\nTCCON sites\n\nCO2 ObsPack\n\nCO2 Budget and Regional Airborne Study -\n\nSteve Wofsy (Harvard U.)\n\npa\n\nDeep Convective Clouds & Chemistry\n\nAndreas Beyersdorf (CSUSB) & Yonghoon\n\noc\n\n(DC3), DC8 aircraft\n\nChoi (SSAI)\n\nCO2 ObsPack\n\nGoddard Space Flight Center (GSFC)\n\nStephan Randolph Kawa, James Brice Ab-\n\nCO2 ObsPack\n\nHIAPER\n\nMaine (COB2004)\nCO2 ObsPack\n\ndf, pa\n\nshire, & Haris Riris (NASA GSFC)\n\nCO2 ObsPack\nCO2 ObsPack\nCO2 ObsPack\n\nPole-to-Pole\n\nObservations\n\nSteve Wofsy (Harvard U.), & Britton\n\n(HIPPO)\n\nStephens (NCAR)\n\nIntercontinental Chemical Transport Exper-\n\nStephanie A. Vay (NASA LaRC) &\n\niment - North America (INTEX-NA)\n\nYonghoon Choi (SSAI)\n\nKorea-United States Air Quality Study\n\nJoshua P. DiGangi, & Yonghoon Choi\n\n(KORUS-AQ)\n\n(SSAI)\n\nO2 /N2 Ratio and CO2 Airborne Southern\n\nBritton Stephens (NCAR), Colm Sweeney\n\nOcean Study (ORCAS)\n\n(NOAA GML), Kathryn McKain (NOAA\n\nll, wg\npa\nan, df, rj\noc\n\nGML), Eric Kort (U. Michigan)\nCO2 ObsPack\n\nStudies of Emissions and Atmospheric\n\nSteve Wofsy (Harvard U.)\n\ndf\n\nStudies of Emissions and Atmospheric\n\nAndreas Beyersdorf (CSUSB) & Yonghoon\n\noc\n\nComposition, Clouds and Climate Coupling\n\nChoi (SSAI)\n\nComposition, Clouds and Climate Coupling\nby Regional Surveys (SEAC4RS), ER-2 aircraft\nCO2 ObsPack\n\nby Regional Surveys (SEAC4RS), DC8 aircraft\n\n1040\n\nhours of the lowest altitude measurement in a profile to the time of the lowest altitude profile measurement. In cases\n\n61\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nTable C2. Table C1, continued. ARM = Atmospheric Radiation Monitoring facility.\nSource\n\nCampaign or ID\n\nProviders\n\nTCCON sites\n\nCO2 ObsPack\n\nStratosphere-Troposphere Analyses of\n\nSteve Wofsy (Harvard U.)\n\npa\n\nKathryn McKain (NOAA GML), Colm\n\nae, df, eu, ll, oc,\n\nSweeney (NOAA GML), Steve Wofsy\n\npa\n\nRegional Transport (START-08)\nCO2 ObsPack\n\nAtmospheric\n\nTomography\n\nMission\n\n(ATom)\n\n(Harvard U.), Bruce Daube (Harvard\nU.), Roisin Commane (Harvard U.)\nOther CO2\n\nNOAA Manaus\n\nJohn Miller (NOAA GML)\n\nma\n\nCH4 ObsPack\n\nHIAPER Pole-to-Pole Observations\n\nSteve Wofsy, Greg Santoni, & Jasna\n\nll,oc,pa,wg\n\n(HIPPO)\n\nPittman (Harvard U.)\n\nStratosphere-Troposphere Analyses of\n\nSteve Wofsy (Harvard U.)\n\npa\n\nKathryn McKain & Colm Sweeney\n\nae,ci,df,eu,ll,oc,pa\n\nCH4 ObsPack\n\nRegional Transport (START08)\nCH4 ObsPack\nIMECC\n\nAtmospheric\n\nRepository\n\nTomography\n\nMission\n\n(ATom)\n\n(NOAA GML)\n\nInfrastructure for Measurement of the\n\nVarious\n\nbi,br,gm,je,ka,or\n\nBianca Baier & Colm Sweeney (NOAA\n\ndf,oc,pa,so\n\n(CO2 , CH4 , CO)\n\nEuropean Carbon Cycle (IMECC)\n\nNOAA AirCores (CO2 ,\n\nN/A\n\nCH4 , CO)\nSodankylä\n\nGML)\nAirCores\n\nN/A\n\nHuilin Chen (RUG) & Rigel Kivi (FMI)\n\nso\n\nN/A\n\nPierre-Yves Quéhé (CARE-C, Cyl) &\n\nni\n\n(CO2 , CH4 , CO)\nNicosia AirCores (CO2 ,\nCH4 , CO)\nRadiosondes (H2 O)\n\nThomas Laemmel (LSCE/IPSL)\nSouthern Great Plains (SGP) Lamont\nCentral Facility and Tropical Western\nPacific (TWP) Darwin Facility\n\n62\n\nARM\n\ndb, oc\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nwhere ground data is only available before or after the lowest profile measurement, we use the closest ground data in\ntime.\nC1.2\n\nAirCore\n\nAs AirCore data intrinsically provides discrete profile, matching these data to TCCON sites was much simpler. For NOAA\n1045\n\nAirCores, we search all files for those where the mean latitude and longitude of the profile were within 1° (total distance) of\na TCCON site. We use a looser distance compared to the aircraft as it is unlikely that an AirCore would be within 1° of a\nTCCON site by happenstance if it was not intended to match with that TCCON. However, since it is possible that the balloon\ntrajectory drifted significantly depending on the winds, we use the looser distance criterion to allow for that.\nC2\n\n1050\n\nH2 O\n\nProfiles for the H2 O AICF come from radiosonde data provided by the Department of Energy Atmospheric Radiation Measurement (ARM) facility (Keeler and Burk). The data were downloaded from https://adc.arm.gov/discovery/#/results/instrument_\nclass_code::sonde%2Fprimary_meas_type_code::atmtemp in March 2021. Two ARM sites are close enough to TCCON locations to be useful: the Southern Great Plains (SGP) site’s Central Facility (facility code C1) is near the Lamont, OK, USA\nTCCON site, and the Tropical Western Pacific (TWP) site’s Darwin facility (code C3) is near the Darwin, Australia TCCON\n\n1055\n\nsite.\nThese facilities produce more radiosonde observations than we can feasibly use in the AICF calculation, so we must choose\na subset. We use the following steps for each site:\n1. Identify radiosonde profiles that are coincident with another trace gas profile (CO2 , CO, CH4 , or N2 O).\n\n1060\n\n2. Identify radiosonde profiles not in the set identified in Step 1 that have at least 30 TCCON spectra within ±3 hours of\nthe time of the profile’s lowest altitude measurement and\n\n3. Combine the profiles from step 1 with randomly selected profiles from step 2 to collect 50 total profiles. (We use a seed\nof 42—chosen in reference to “The Hitchhiker’s Guide to the Galaxy”—to ensure repeatability across runs.)\n4. Finally, remove any profiles from this set of 50 that have a maximum altitude < 15 km.\nOnce we have assembled a pool of radiosonde profiles, we convert the relative humidity (RH) values stored in the files to\n1065\n\nwater mole fractions. Based on the convention described in Miloshevich et al. (2006), we assume that the definition of RH\nis the ratio of water vapor pressure to the saturation water vapor pressure over liquid water and calculate the H2 O dry mole\nfraction as\nRH · SVP\np\nfH2 O,wet\nfH2 O,dry =\n1 − fH2 O,wet\n\nfH2 O,wet =\n\n(C1)\n(C2)\n\n63\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n1070\n\nwhere RH is the relative humidity as a fraction (i.e. 0 to 1), SVP is the saturation vapor pressure of water over liquid water\ncalculated using Eq. 6 of Miloshevich et al. (2004) (see also Eq. 15 of Wexler, 1976), and p is the atmospheric pressure (in the\nsame units as SVP).\nC3 Constructing full profiles\nIn order to ensure a proper comparison between the in situ and TCCON column amounts, the in situ profiles must extend to the\n\n1075\n\ntop of the TCCON retrieval altitude grid, 70 km. No aircraft or balloon-borne profile reaches this altitude, therefore, similarly\nto Wunch et al. (2010), we extend the in situ profiles using the GGG2020 prior profiles (Laughner et al., 2023).\nThe differences between Wunch et al. (2010) and our approach stem from (1) the GGG2020 priors do a better job of\nrepresenting trace gas profiles in the stratosphere and (2) we have enough additional profiles over TCCON sites to be selective\nabout which ones we use. This is why we filtered the ObsPack data to “chunks” that have data up to at least 7500 m altitude\n\n1080\n\n(§C1.1), to limit the altitude that needs to be filled in above the top of the profile.\nThere are three ways that profiles are extended up to 70 km altitude, depending on their top altitude:\n1. If the profile’s top is above 380 K potential temperature (i.e. reaches the stratospheric overworld), then we append the\nGGG2020 priors for levels above the profile top.\n2. If the profile’s top is below 380 K potential temperature but at or above 7.5 km, then the in situ profile’s values are\n\n1085\n\nbinned to the same altitude grid (see below) and then we do a constant value extrapolation of the top binned value up\nto the tropopause altitude. We use the GGG2020 prior above 380 K potential temperature again, and connect the two\nparts of the profile by linearly interpolating the trace gas mole fractions with respect to potential temperature between\nthe tropopause and 380 K. This case covers profiles where the top of the measured profile is expected to be a better\nrepresentation of the unmeasured free troposphere than the GGG2020 priors.\n\n1090\n\n3. If the profile’s top is below 7.5 km, then we use the GGG2020 priors for all levels above the profile top. The case assumes\nthat profiles that do not reach above 7.5 km do not constrain the free troposphere well enough to supplant the GGG2020\npriors. While we filtered the ObsPack data for “chunks” that have data above 7.5 km, we still have a few profiles with\nceilings below 7.5 km from chunks that needed to be split into multiple profiles.\nFor #2, we calculate the binned in situ profile values for the highest altitude of the GGG retrieval grid below the in situ\n\n1095\n\nprofile’s ceiling (zGGG,k ) as:\nPnobs\nwf\nPnobsi obs,i\nf obs = i=1\ni=1 wi\n\n\n\n(zobs,i − zGGG,k−1 )/(zGGG,k − zGGG,k−1 ) if zGGG,k−1 ≤ zobs,i < zGGG,k\nwi = (zGGG,k+1 − zobs,i )/(zGGG,k+1 − zGGG,k ) if zGGG,k ≤ zobs,i < zGGG,k+1\n\n\n\n0\notherwise\n64\n\n(C3)\n\n(C4)\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n5000\n\nGGG altitude grid level\nObservation weighting\n\nAltitude (km)\n\n4000\n3000\n2000\n1000\n0.0\n\n0.1\n\n0.2\n0.3\nWeight\n\n0.4\n\n0.5\n\n0.6\n\nFigure C1. An example of the weighting functions from Eq. (C4). Lines indicate the weights applied to the observed mole fractions and\ncircles indicate the GGG altitude grid levels that correspond to those weights—like colors match.\n\nFigure C1 shows an example of the weights for one short profile at the Armstrong TCCON site.\nThere is a special case for CH4 applied when integrating the in situ profile to calculate the in situ-derived XCH4 . Previous\n1100\n\nwork (e.g. Washenfelder et al., 2003; Saad et al., 2014, 2016) established that there is a strong correlation between CH4 and\nHF in the stratosphere. Since this correlation is encoded into the GGG2020 priors (Laughner et al., 2023), we can use the\ndifference between the prior and posterior HF column (which is almost entirely found in the stratosphere) from the TCCON\nretrievals to adjust the levels in the in situ CH4 profiles that use the GGG2020 profiles.\nSpecifically, when calculating the in situ XCH4 , we get the slope of CH4 vs. HF mixing ratios used by the GGG2020 priors\n\n1105\n\nfor the year and region (tropics, midlatitudes, or polar vortex) of the profile (see §3.5 and Fig. 11 of Laughner et al., 2023). We\nthen multiply this slope by the difference between the prior and median posterior HF profile of all the TCCON observations\nmatched with the in situ profile in question in order to get the expected change in the CH4 priors to better match the true\nstratospheric profile. Finally, we multiply this profile difference by the TCCON AK and integrate only the levels in the total in\nsitu profile obtained from the GGG2020 priors. The integration uses Eq. (8) and add the integrated change to the in situ XCH4\n\n1110\n\nas a posterior adjustment.\nAgain, note that this correction is only applied when integrating the in situ profiles to obtain the true XCH4 value to compare\nthe TCCON retrievals against. When using the in situ profiles as priors in the TCCON retrievals, the levels taken from the\nGGG2020 priors are not adjusted in this fashion.\nC4 Grouping temporally proximate profiles\n\n1115\n\nThere are several cases where multiple profiles are available within a short time of each other (such as different legs of a\nmissed approach or duplicate AirCore launches). Because we use the observed profiles as the prior in the TCCON retrievals\nfrom which the AICF is derived, this presents a technical challenge. Ideally, we want to use the same prior for all retrievals\nmatched up with a given profile for comparison. Our temporal coincidence criterion can be up to ±3 hours, therefore, in cases\n65\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nwith two or more profiles within a few hours, if for each TCCON retrieval we used the observed priors closest in time to it, this\n1120\n\nwould result in a change of prior partway through our coincidence window.\nOur solution was to merge profiles close enough in time for this to occur, but only for use as priors. Each individual observed\nprofile still contributes one point on Fig. 11. This does mean that the prior will not exactly match any of the observed profiles\nthose retrievals are compared against, but we consider that an acceptable error, given that we do apply an AK correction to the\nintegrated in situ profile.\n\n1125\n\nTo find profiles that need to be merged, we first identify which TCCON observations would match with that profile. We ignore\nthe quality filtering criteria from §7.3.1 during this step, and only try to find the time window (± 1, 2, or 3 hours) necessary\nto match at least 30 TCCON observations to each profile. If any two profiles from the same TCCON site are matched to any\nof the same TCCON observations, they are grouped together in the list of profiles, to be averaged together when creating the\ncustom priors in §C5. This initial list is written out to a text file so that it can be modified by hand later, as needed.\n\n1130\n\nC5 Running custom TCCON retrievals\nAs mentioned in §7.3, when we run the TCCON retrievals for the AICF calculation, we use as custom priors the in situ profiles\nthat a given TCCON observation will be compared against. This reduces error in the TCCON Xgas value that arises from an\nincorrect prior profile and thus improves the accuracy of the AICF. There are several technical considerations in how we handle\nthis matching. In order to make those considerations clear, let us first describe how the GGG retrieval accepts inputs describing\n\n1135\n\nboth the prior profiles and the TCCON observations to retrieve on.\nGGG takes a list of TCCON spectra to retrieve as input in the “runlog” file. This lists each spectrum on which to run the\nretrieval in order. For the AICF retrievals, we combined all the spectra from all the relevant TCCON sites into a single runlog.\nThe priors (including temperature and pressure as well as trace gas mixing ratios) are written to a “.mav” file. This file is\norganized into blocks. Each block indicates the first spectrum from the runlog which the priors contained in the .mav block\n\n1140\n\napply to. During the retrieval, GGG iterates through the spectra contained in the runlog. When it reaches the spectrum defined\nas the first spectrum of the next block in the .mav file, it loads the priors from that block before continuing.\nIn inserting the in situ profiles into the .mav file as priors, we had three objectives:\n1. Retain the standard priors for gases and times that we did not have in situ profiles available.\n2. Ensure that the in situ profiles were used as priors for any spectra that they might be compared against.\n\n1145\n\n3. Ensure that any in situ profiles were only applied to the TCCON site where and day when they were measured.\nTo meet these objectives, our approach to inserting the in situ profiles as priors was:\n– Divide the runlog into chunks by site and day, so that each chunk only has spectra from one site on one day.\n– For each unique site/day chunk, collect all the in situ profiles from that day.\n\n66\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n– Average together any in situ profiles grouped together in the list created in §C4. For this, we used an approach that\nconsiders whether each in situ profile contributed observations to a given level in the regridded profile. For a level on the\n\n1150\n\nretrieval grid where none of the in situ profiles provided any data points (i.e. the observed profiles were extrapolated or\nhad the GGG2020 prior appended to it), both profiles are weighted equally. For a level where at least one of the in situ\nprofiles had observed data, each profile is weighted by the fraction of data for that level that came from observations.\n– For gases that only have one profile (after averaging) for that site/day, assign that profile to all the .mav blocks for that\nsite/day.\n\n1155\n\n– For gases that have multiple profiles that are not merged together (§C4), use the first profile in the day for all .mav blocks\nup until the first spectrum that could be compared with the second profile in the day (for our coincidence criteria, this\nwill be the spectra 3 hours before the floor time of the second profile). Introduce a new .mav block on that profile that\nswitches to the second profile. Repeat for third, fourth, etc. profiles if present. Assign the last profile to cover all .mav\nblocks through the end of the day.\n\n1160\n\nOnce the profiles are assigned to their .mav blocks, they must be averaged from their native vertical resolution to the GGG\nretrieval altitude grid and, if multiple profiles for the same gas were present for the same block, they must be averaged together.\nFor the vertical regridding, we use the same approach as described in §C3 where we do a weighted average of the observed\nmixing ratios, where the weights are maximized when the observed altitude equals the altitude of the GGG retrieval level they\n1165\n\nare being averaged to, and which decrease linearly to the adjacent GGG retrieval levels (Fig. C1, Eq. C4).\nWe found that it is crucial that we use geopotential height as the altitude for the regridding, as that did a better job ensuring\nthat the observed profiles followed hydrostatic balance. To compute geopotential height for the in situ profiles, we take pressure\nand geopotential height from the two GEOS FP-IT files (Lucchesi, 2015) that bound the profile’s lowest altitude in time and\naverage the GEOS FP-IT data, and weight each by the time difference between the GEOS FP-IT profile and the time of the\n\n1170\n\nlowest altitude measurement in the in situ profile, giving greater weight to profiles nearer in time to the in situ profile. We then\ninterpolate the GEOS FP-IT geopotential altitude on the logarithm of pressure to the pressures in the in situ profile.\nThe final consideration in preparing the custom priors is that we always retain the pressure and temperature profiles from\nthe standard GEOS FP-IT priors used in GGG2020. This is because our testing found it very difficult to maintain hydrostatic\nbalance if we used the observed pressure and temperature. This, in turn, caused greater error in the retrieved Xgas values, as\n\n1175\n\nthe air column would be incorrect.\nOnce the custom priors were generated, the TCCON retrievals could be run as normal. The standard post processing corrections for airmass dependence (§7.1) and window-to-window averaging (§7.2) were applied as well. AKs were calculated for\neach spectrum retrieved as used to smooth the in situ profiles and account for the TCCON vertical sensitivity (§7.3).\nC6\n\n1180\n\nUncertainty in TCCON/in situ comparisons\n\nFor the TCCON/in situ ratios in §7.3, we considered five sources of uncertainty for the comparisons:\n\n67\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n1. In situ measurement error (\u000fmeas ): This accounts for the error in individual in situ measurements that make up the\nprofiles. To be conservative, we assume the worst-case scenario with 100% correlated error at all levels. The uncertainty in\nXgas is then calculated as:\nZ\nZ\n\u000fmeas = c(p) + 2σ(p) dp − c(p) dp\n1185\n\n(C5)\n\nwhere c(p) is the measured mixing ratio and σ(p) the uncertainty at each level. The integrals represent the pressure-weighted\nintegration, Eq. (8). The uncertainty values are those reported in the original data files where available or a typical value chosen\nin consultation with the data providers.\n2. Unmeasured free troposphere (\u000fFT ): This accounts for uncertainty due to the portion of the free troposphere not measured by a given profile. For each profile, we first calculate σobs,FT , the standard deviation of measurements above 750 hPa\n\n1190\n\nand below the tropopause (as determined by GEOS FP-IT meteorology). We then create a perturbed profile,\n\nc(p) + 2σ\nobs,FT if interp/extrap at p\nc0 (p) =\n\nc(p)\notherwise\n\n(C6)\n\nwhich adds this standard deviation to interpolated or extrapolated levels above the top of the measured profile. The uncertainty\nin Xgas is calculated as:\nZ\nZ\n\u000fFT = c0 (p) dp − c(p) dp\n\n1195\n\n(C7)\n\nThis error will be zero for profiles that do not require extrapolation or interpolation to reach the stratospheric overworld (i.e.\naltitudes with potential temperature ≥ 380 K).\n\n3. Bias in stratospheric prior (\u000fstrat ): This represents expected bias in the column from the use of GGG2020 priors for\n\nlevels in the stratosphere. This uses the retrieved vs. prior HF column as a proxy for error in the stratospheric prior. As discussed\nin §7.3.3, HF is predominately found in the stratosphere, so the difference between the retrieved and prior HF columns gives\n1200\n\ninformation about whether the stratospheric profile was biased high or low. We calculate the bias as:\n\u000fstrat = 2 · (XHF,post − XHF,prior ) ·\n\n∂Xgas\n∂XHF\n\n(C8)\n\nThe derivative ∂Xgas /∂XHF has to be calculated for each gas. For CO2 we use 8.09×103 , which was derived from East Trout\nLake TCCON data by comparing prior and posterior wCO2 and HF columns. (East Trout Lake is positioned to see significant\nstratospheric variability due to the polar vortex, and wCO2 is the GGG2020 CO2 product with enhanced sensitivity to the\n1205\n\nstratosphere.) For CH4 , this is drawn from the CH4 :HF slopes used in the GGG2020 priors (Laughner et al., 2023).\nAirCore profiles are treated specially, as they always reach into the stratosphere. For these profiles, we create a perturbed\nprofile, c0 (p), where the levels in the stratosphere filled by the GGG2020 priors have the difference between the top of the\nAirCore profile and the corresponding level in the prior added to them. The difference between the integral of these profiles\n\n68\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nbecome the stratospheric error. Mathematically, that is\n\nc(p) + 2[c\nprior (pobs. top ) − cAirCore (pobs. top )]\n1210\nc0 (p) =\n\nc(p)\nZ\nZ\n\u000fstrat,AirCore = c0 (p) dp − c(p) dp\n\nif using prior at p\n\n(C9)\n\notherwise\n\n(C10)\n\n4. Random error in TCCON Xgas value (\u000fstd. xgas ): This represents random error in the TCCON observations. Because we\nrequire at least 30 TCCON observations coincident with a profile for a valid comparison, we use twice the standard deviation\namong those coincident observations as the metric of random error. The coincidence windows vary between 2 and 6 hours\n1215\n\nwide, so the standard deviation likely includes some true change in the data, and can therefore be considered conservative.\n5. Bias in TCCON derived from Xluft (\u000fXluft ): This represents bias in retrieved Xgas values resulting from instrument\nhardware issues diagnosed from deviations in Xluft from the nominal network value (0.999, see §7.3). The bias is calculated\nas:\n\u000fXluft =\n\n1220\n\n∂r\n· (Xluft,median − 0.999) · Xgas,median\n∂Xluft\n\n(C11)\n\nHere, Xluft,median and Xgas,median are the median values of TCCON Xluft and the target Xgas across the 30+ coincident\nobservations for the comparison. 0.999 is the nominal value of Xluft that represents a well-operating instrument. The ∂r/∂Xluft\nvalue is how the TCCON/in situ ratio changes with Xluft , and was derived for XCO2 , XwCO2 , XlCO2 , and XCH4 by an\nunweighted robust fit through similar plots of TCCON/in situ ratio vs. Xluft as Fig. 11, but with TCCON retrievals that used\nthe standard trace gas priors instead of custom ones built from the in situ profiles. The values used are given in Table C6.\n\n1225\n\nFull error calculation: As the error terms include a mix of random (\u000fmeas , \u000fFT , \u000fstd. xgas ) and systematic (\u000fstrat , \u000fXluft )\nerrors, the in situ and TCCON total errors are calculated as:\nq\n\u000fin situ = \u000f2meas + \u000f2FT + |\u000fstrat |\nq\n\u000fTCCON = \u000f2std. xgas + |\u000fXluft |\n\n(C12)\n(C13)\n\nThe first term in the second equation is written as a root of a square to indicate that if additional random TCCON error terms\n1230\n\nwere to be added in the future, they should add in quadrature. The uncertainty in the TCCON/in situ ratio (Xgas,TCCON /Xgas,in situ )\nP\nfollows standard error propagation (\u000ftotal = i (σx · ∂f (x)/∂x)2 ):\n\u000ftotal =\n\n2\n\u000f2TCCON \u000f2in situ Xgas,TCCON\n+\n\u000f2in situ\n\u000f4in situ\n\n(C14)\n\nNote that Eq. (C12) is applied to each individual TCCON/in situ comparison, while the statistics in Table 3 are averaged\nover all the comparisons for a given gas. Therefore, the values of \u000fin situ , \u000fmeas , \u000fFT , and \u000fstrat in Table 3 do not directly relate\n1235\n\nto each other through Eq. (C12). As noted in the caption for Table 3, the non-parenthetical values in the last four columns\n\n69\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nformally propagate the error from the individual comparisons, such that the values shown in the table (which we will denote\ngenerally as \u000fformal ) are calculated from the individual comparisons’ values with\n\n\u000f2formal =\n\nn \u0012\nX\n1\ni=1\n\nn\n\n\u000findiv,i\n\n\u00132\n\n(C15)\n\nwhere \u000findiv,i denotes individual comparisons’ error values and n is the number of individual observations. Conversely, the\n1240\n\nparenthetical numbers in Table 3 give the simple mean, i.e.:\nn\n\n1X\n\u000fmean =\n\u000findiv,i\nn i=1\n\n(C16)\n\nAppendix D: Comparison between TCCON and NOAA surface N2 O\nFor Fig. 14, we constructed N2 O profiles to compare TCCON XN2 O against using NOAA surface data. This approach takes\nadvantage of how well-mixed N2 O is in the troposphere to build a large set of comparison. The approach, in detail, is as\n1245\n\nfollows.\nThe TCCON vs. in situ comparison shown in Fig. 14 calculates an in situ XN2 O from N2 O profiles using Eq. (7) as with the\nother Xgas quantities in §7.3. These N2 O profiles are constructed using the NOAA surface N2 O VMR from the surface to the\ntropopause, the GGG2020 N2 O prior for levels with potential temperature greater than 380 K, and linearly interpolating the\nN2 O VMR with respect to potential temperature between the tropopause and 380 K level.\n\n1250\n\nFor the tropospheric N2 O VMRs, we obtained monthly average NOAA global N2 O data from https://gml.noaa.gov/hats/\ncombined/N2O.html (last access 10 May 2021). For sites at latitudes north of 23◦ N or south of 23◦ S, we use the northern and southern hemispheric averages, respectively (GML_NH_N2O and GML_SH_N2O in the combined NOAA N2 O file).\nFor equatorial latitudes between 23◦ S and 23◦ N, we used the average of the Mauna Loa and American Samoa N2 O data\n(GML_mlo_N2O and GML_smo_N2O in the combined file). For each comparison point in Fig. 14, we used the N2 O VMR\n\n1255\n\nfrom that month as the tropospheric VMR of the profile.\nThe comparisons selected for Fig. 14 meet the following criteria:\n– The difference between the prior and posterior HF column must be < 2 × 1014 molec. cm−2 in magnitude. Since HF is\n\nalmost entirely in the stratosphere, this limits the comparisons to cases where the GGG2020 prior stratospheric profiles\nare reasonably accurate, thus limiting error in the in situ XN2 O from an incorrect assumed stratosphere\n\n1260\n\n– Xluft must be in the range [0.996, 1.002). This ensures we are considering data when the TCCON instrument was well\naligned, as discussed in §7.3.1\n– FVSI must be ≤ 0.05. This limits the comparison to mostly cloud-free observations.\n\n70\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nAppendix E: Variable O2 mole fraction derivations\nE1\n1265\n\nTrends in O2 mole fraction from trends in XCO2\n\nThe derivation of Eq. (10) begins from the definition of fO2 :\n\nfO2 =\n\nNO2\nN + NO2 + NCO2\n\n(E1)\n\nwhere:\n– NO2 and NCO2 are the number of moles of O2 and CO2 , respectively,\n– N is the number of moles of gases other than O2 or CO2 in H2 O-free air, and\n– Ntot (used below) is N + NO2 + NCO2\n\n1270\n\nDefining α = ∂NO2 /∂NCO2 , taking the derivative of fO2 with respect to NCO2 , and simplifying gives:\n∂fO2\n=\n∂NCO2\n\n\u0012\n\nα(N + NCO2 ) NO2\n−\nNtot\nNtot\n\n\u0013\n\n·\n\n1\nNtot\n\n(E2)\n\nRecognizing that NO2 /Ntot = fO2 and (N + NCO2 )/Ntot = 1 − fO2 as well as converting the derivative to a ratio of small\n\nbut finite differences (represented by δ in place of ∂) gives:\n\n1275\n\n1\nδfO2\n= (α − α · fO2 − fO2 ) ·\nδNCO2\nNtot\nδNCO2\n⇒ δfO2 = (α − α · fO2 − fO2 ) ·\nNtot\n\n(E3)\n(E4)\n\nFinally, to convert δNCO2 /Ntot into terms of XCO2 and XCO2 ,ref , we start by defining:\n\nXCO2 ,ref =\n\nNCO2\nNtot\n\n(E5)\n\nand\n1280\n\nXCO2 =\n\nNCO2 + δNCO2\nNtot + δNCO2 + δNO2\n\n(E6)\n\nas well as δNO2 = α · δNCO2 . Substituting this and NCO2 = XCO2 ,ref · Ntot from Eq. (E5) in Eq. (E6) and rearranging gives:\nXCO2 − XCO2,ref\nδNCO2\n=\nNtot\n1 − XCO2 − α · XCO2\n\n(E7)\n\nSubstituting Eq. (E7) in Eq. (E4) yields the final version of Eq. (10).\n71\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nE2\n1285\n\nO2 mole fraction from O2 /N2 data\n\nMeasurements of atmospheric O2 concentration are commonly reported as 10−6 relative deviations in the O2 /N2 ratio (denoted\nδ(O2 /N2 ) and given in units of per meg) to avoid the complexities of diluation effects from changes in CO2 and other trace\nspecies on the O2 mole fraction. To convert from available measurements of trends in δ(O2 /N2 ), we must convert to units of\nppm and account for the diluting effect of trends in CO2 . The equation for the black line in Fig. 13, based on Scripps δ(O2 /N2 )\nand NOAA global mean CO2 data, is slightly different from Eq. (10). As above, the derivation starts with Eq. (E1), but now\n\n1290\n\nsince we have measured values for the change in NO2 and NCO2 , our change in fO2 will instead be:\n\nδfO2 =\n\n∂fO2\n∂fO2\n· δNO2 +\n· δNCO2\n∂NO2\n∂NCO2\n\n(E8)\n\nIn this case, both ∂NO2 /∂NCO2 and ∂NCO2 /∂NO2 are 0, since we have measurements of both O2 and CO2 and therefore\ncan treat their changes as orthogonal. That leads to the following expressions for the derivatives in Eq. (E8):\n\n1295\n\n1 − fO2,ref\n∂fO2\n=\n∂NO2\nNtot\nf\n∂fO2\nO\n= − 2,ref\n∂NCO2\nNtot\n\n(E9)\n(E10)\n\nInserting these into Eq. (E8) gives:\n\nδfO2 = (1 − fO2,ref ) ·\n\nδNCO2\nδNO2\n− fO2,ref ·\nNtot\nNtot\n\n(E11)\n\nδNO2 /Ntot can be expressed in terms of δ(O2 /N2 ) values by using the definition of δ(O2 /N2 ) (Keeling et al., 1998):\nδ(O2 /N2 ) =\n1300\n\n(O2 /N2 )sample\n−1\n(O2 /N2 )reference\n\n(E12)\n\nand assuming that the amount of N2 in the atmosphere does not change. Multiplying this definition by fO2,ref gives:\n\nδ(O2 /N2 ) · fO2,ref\n\n\u0014\n\n\u0015\n(NO2 + δNO2 )/NN2\nNO2\n=\n−1 ·\nNO2 /NN2\nNtot\nδNO2\n=\nNtot\n\n(E13)\n(E14)\n\nδNCO2 /Ntot can be expressed as in Eq. (E7) except with α = 0 (again, this is because we have measurements of mole\nfractions of CO2 and O2 ). The final equation used for the “best estimate” line in Fig. 13 is therefore:\n\n1305\n\nfO2 ,ref + δfO2 = fO2 ,ref + (1 − fO2 ,ref ) · δ(O2 /N2 ) · fO2 ,ref −\n\nXCO2 − XCO2 ,ref\n· fO2 ,ref\n1 − XCO2\n\n72\n\n(E15)\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nwhere fO2 ,ref is the 0.209341 value obtained in §7.3.2 by adjusting Aoki et al. (2019). As noted in §7.3.2, the δ(O2 /N2 ) data\nused is a weighted average of the ALT, LJO, and CGO sites with weights of 41 , 14 , and 12 , respectively. Note that the NOAA\nglobal mean CO2 (rather than TCCON XCO2 ) is used for XCO2 and XCO2 ,ref in this equation.\n\nAuthor contributions. J.L. Laughner led the development of the new airmass correction (§7.1), window-to-window averaging (§7.2), in situ\n1310\n\nscaling (§7.3), and miscellaneous changes in §9. G.C. Toon is the main developer of GGG. J. Mendonca developed the non-Voigt treatment\nof the spectral line shape (§3.2). C. Petri contributed to the development of the phase correction update (§5.2). S. Roche developed the new\nretrieval grid (§4.1), meteorological resampler (§4.2), and netCDF writer. D. Wunch carried out the sensitivity tests (§8). D. Wunch, C.M.\nRoehl, G.C. Toon, P.O. Wennberg, and J.L. Laughner conducted the O2 study in §3.3. J.-F. Blavier is a key developer of I2S. D.W.T. Griffith\ncontributed to various aspects of GGG2020 development. P. Heikkinen, R. Kivi, and M.K. Sha first diagnosed the nonlinearity issue from\n\n1315\n\n§5.1 and developed a correction methodology. R.F. Keeling and B.B. Stephens consulted on the approaches to parameterize the change in O2\nmole fraction (§7.3.2). M. Kiel performed tests of the phase correction threshold (§5.2) and choices of NCBFs (§6) C.M. Roehl, N. Deutscher,\nP. Jeseck, D. Pollard, M. Rettinger, S. Roche, M.K. Sha, Y. Té and D. Wunch all participated in a beta test of GGG2020. N.M. Deutscher,\nJ. Gross, B. Herkommer, P. Jeseck, I. Morino, H. Ohyama, C. Petri, J. Notholt, D. Pollard, M. Rettinger, S. Roche, E. McGee, K. Strong,\nC.M. Roehl, M.K. Sha, K. Shiomi, R. Sussmann, Y. Té, V. Velazco, D. Wunch, and M. Zhou provided data used to derive the corrections in\n\n1320\n\n§7.1 and §7.2. B.C. Baier, B.B. Stephens, H. Chen, Y. Choi, X. Lan, T. Laemmel, K. McKain, J. Miller, H. Riris, C. Rousogenous, and S.C.\nWofsy provided in situ data used in §7.3. P.O. Wennberg provided input to all elements of this work. All authors reviewed the manuscript.\n\nCompeting interests. The authors declare no competing interests.\n\nAcknowledgements. The authors gratefully acknowledge the use of GNU Parallel (Tange, 2011) in the GGG processing. The authors also\nthank James Abshire for providing CO2 data used in deriving the in situ correction (§7.3). A portion of this research was carried out\n1325\n\nat the Jet Propulsion Laboratory (JPL), California Institute of Technology, under a contract with NASA (80NM0018D0004). Government\nsponsorship is acknowledged. Support for Caltech TCCON sites and partial support for JLL, MK, CMR, and POW provided by NASA grants\nNNX17AE15G and 80NSSC22K1066. Material from BBS and RFK is based upon work supported by the National Center for Atmospheric\nResearch, which is a major facility sponsored by the National Science Foundation under Cooperative Agreement No. 1852977.\nMR and RS acknowledge funding by the German Helmholtz Research Program “Changing Earth – Sustaining our Future” within the\n\n1330\n\nResearch Field “Earth and Environment”. The Paris TCCON site has received funding from Sorbonne Université, the French research center\nCNRS and the French space agency CNES. The Cyprus TCCON site and AirCore flights have received funding from the European Union’s\nHorizon 2020 research and innovation programme under grant agreement No. 856612 and the Cyprus Government. The TCCON site at\nRéunion Island has been operated by the Royal Belgian Institute for Space Aeronomy with financial support since 2014 by the EU project\nICOS-INIWRE (Grant agreement ID: 313169), the ministerial decree for ICOS (FR/35/IC1 to FR/35/C6), ESFRI-FED ICOS-BE project\n\n1335\n\n(EF/211/ICOS-BE) and local activities supported by LACy/UMR8105 and by OSU-R/UMS3365 – Université de La Réunion. The Eureka\nTCCON measurements were made at the Polar Environment Atmospheric Research Laboratory (PEARL) by the Canadian Network for the\n\n73\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nDetection of Atmospheric Change (CANDAC), primarily supported by the Natural Sciences and Engineering Research Council of Canada,\nEnvironment and Climate Change Canada, and the Canadian Space Agency. TCCON sites at Tsukuba, Rikubetsu and Burgos are supported\nin part by the GOSAT series project. Burgos is supported in part by the Energy Development Corporation Philippines.\n\n74\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n1340\n\nReferences\nAbrams, M. C., Toon, G. C., and Schindler, R. A.: Practical example of the correction of Fourier-transform spectra for detector nonlinearity,\nAppl. Opt., 33, 6307–6314, https://doi.org/10.1364/AO.33.006307, 1994.\nAllen, D. R. and Nakamura, N.: Tracer Equivalent Latitude: A Diagnostic Tool for Isentropic Transport Studies, Journal of the Atmospheric\nSciences, 60, 287 – 304, https://doi.org/10.1175/1520-0469(2003)060<0287:TELADT>2.0.CO;2, 2003.\n\n1345\n\nAndrews, A.: Technical Procedure: Analysis of Carbon Monoxide in Air, https://www.gml.noaa.gov/ccl/docs/TP_analysis_CO_v2.0.pdf, last\naccess: 13 Sept 2022, 2019.\nAoki, N., Ishidoya, S., Matsumoto, N., Watanabe, T., Shimosaka, T., and Murayama, S.: Preparation of primary standard mixtures for\natmospheric oxygen measurements with less than 1 µmol mol−1 uncertainty for oxygen molar fractions, Atmos. Meas. Tech., 12, 2631–\n2646, https://doi.org/10.5194/amt-12-2631-2019, 2019.\n\n1350\n\nBaier, B., Sweeney, C., Tans, P., Newberger, T., Higgs, J., Wolter, S., and NOAA Global Monitoring Laboratory: NOAA AirCore atmospheric\nsampling system profiles (Version 20201223) [Data set]. NOAA GML, https://doi.org/10.15138/6AV0-MY81, 2021.\nBenner, D. C., Devi, V. M., Sung, K., Brown, L. R., Miller, C. E., Payne, V. H., Drouin, B. J., Yu, S., Crawford, T. J., Mantz, A. W., et al.: Line\nparameters including temperature dependences of air-and self-broadened line shapes of 12 C16 O2 : 2.06-µm region, Journal of Molecular\nSpectroscopy, 326, 21–47, 2016.\n\n1355\n\nBlumenstock, T., Hase, F., Schneider, M., Garc{\\’\\i}a, O., and Sep\\’ulveda, E.: TCCON data from Izana, Tenerife, Spain, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A.,\nhttps://doi.org/10.14291/tccon.ggg2020.izana01.R1, 2017.\nBuschmann, M., Petri, C., Palm, M., Warneke, T., Notholt, J., and Engineers, A. S.: TCCON data from Ny-Alesund, Svalbard, Norway, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A.,\n\n1360\n\nhttps://doi.org/10.14291/tccon.ggg2020.nyalesund01.R0, 2022.\nByrne, B., Baker, D. F., Basu, S., Bertolacci, M., Bowman, K. W., Carroll, D., Chatterjee, A., Chevallier, F., Ciais, P., Cressie, N., Crisp, D.,\nCrowell, S., Deng, F., Deng, Z., Deutscher, N. M., Dubey, M. K., Feng, S., García, O. E., Griffith, D. W. T., Herkommer, B., Hu, L., Jacobson, A. R., Janardanan, R., Jeong, S., Johnson, M. S., Jones, D. B. A., Kivi, R., Liu, J., Liu, Z., Maksyutov, S., Miller, J. B., Miller, S. M.,\nMorino, I., Notholt, J., Oda, T., O'Dell, C. W., Oh, Y.-S., Ohyama, H., Patra, P. K., Peiro, H., Petri, C., Philip, S., Pollard, D. F., Poulter,\n\n1365\n\nB., Remaud, M., Schuh, A., Sha, M. K., Shiomi, K., Strong, K., Sweeney, C., Té, Y., Tian, H., Velazco, V. A., Vrekoussis, M., Warneke,\nT., Worden, J. R., Wunch, D., Yao, Y., Yun, J., Zammit-Mangion, A., and Zeng, N.: National CO2 budgets (2015–2020) inferred from\natmospheric CO2 observations in support of the global stocktake, Earth System Science Data, 15, 963–1004, https://doi.org/10.5194/essd15-963-2023, 2023.\nChen, Y., Cheng, J., Song, X., Liu, S., Sun, Y., Yu, D., and Fang, S.: Global-Scale Evaluation of XCO2 Products from GOSAT, OCO-2 and\n\n1370\n\nCarbonTracker Using Direct Comparison and Triple Collocation Method, Remote Sensing, 14, 5635, https://doi.org/10.3390/rs14225635,\n2022.\nCooperative Global Atmospheric Data Integration Project: Multi-laboratory compilation of atmospheric carbon dioxide data for the period 1957-2018; obspack_co2_1_GLOBALVIEWplus_v5.0_2019_08_12; NOAA Earth System Research Laboratory, Global Monitoring\nDivision, https://doi.org/10.25925/20190812, 2019.\n\n75\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n1375\n\nCooperative Global Atmospheric Data Integration Project: Multi-laboratory compilation of atmospheric methane data for the period 19572018; obspack_ch4_1_GLOBALVIEWplus_v2.0_2020-04-24; NOAA Earth System Research Laboratory, Global Monitoring Division,\nhttps://doi.org/10.25925/20200424, 2020.\nCorredera, P., Hernanz, M. L., lez Herr ez, M. G., and Campos, J.: Anomalous non-linear behaviour of InGaAs photodiodes with overfilled\nillumination, Metrologia, 40, S150–S153, https://doi.org/10.1088/0026-1394/40/1/334, 2003.\n\n1380\n\nDeutscher, N. M., Griffith, D. W. T., Bryant, G. W., Wennberg, P. O., Toon, G. C., Washenfelder, R. A., Keppel-Aleks, G., Wunch, D., Yavin,\nY., Allen, N. T., Blavier, J.-F., Jiménez, R., Daube, B. C., Bright, A. V., Matross, D. M., Wofsy, S. C., and Park, S.: Total column CO2\nmeasurements at Darwin, Australia ndash; site description and calibration against in situ aircraft profiles, Atmos. Meas. Tech., 3, 947–958,\nhttps://doi.org/10.5194/amt-3-947-2010, 2010.\nDeutscher, N. M., Griffith, D. W., Paton-Walsh, C., Velazco, V. A., Wennberg, P. O., Blavier, J.-F., Washenfelder,\n\n1385\n\nR. A., Yavin, Y., Keppel-Aleks, G., Toon, G. C., and et al.: TCCON data from Darwin (AU), Release GGG2020.R0,\nhttps://doi.org/10.14291/tccon.ggg2020.darwin01.R0, 2023a.\nDeutscher, N. M., Griffith, D. W. T., Paton-Walsh, C., Jones, N. B., Velazco, V. A., Wilson, S. R., Macatangay, R. C., Kettlewell, G. C.,\nBuchholz, R. R., Riggenbach, M. O., Bukosa, B., John, S. S., Walker, B. T., and Nawaz, H.: TCCON data from Wollongong, Australia, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A.,\n\n1390\n\nhttps://doi.org/10.14291/tccon.ggg2020.wollongong01.R0, 2023b.\nDevi, V. M., Benner, D. C., Brown, L., Miller, C., and Toth, R.: Line mixing and speed dependence in CO2 at 6348 cm−1 : Positions,\nintensities, and air-and self-broadening derived with constrained multispectrum analysis, Journal of Molecular Spectroscopy, 242, 90–\n117, 2007a.\nDevi, V. M., Benner, D. C., Brown, L., Miller, C., and Toth, R.: Line mixing and speed dependence in CO2 at 6227.9 cm−1 : Constrained\n\n1395\n\nmultispectrum analysis of intensities and line shapes in the 30013←− 00001 band, Journal of Molecular Spectroscopy, 245, 52–80, 2007b.\nDevi, V. M., Benner, D. C., Sung, K., Crawford, T. J., Yu, S., Brown, L. R., Smith, M. A. H., Mantz, A. W., Boudon, V., and Ismail, S.:\nSelf-and air-broadened line shapes in the 2ν3 P and R branches of 12 CH4 , Journal of Molecular Spectroscopy, 315, 114–136, 2015.\nDevi, V. M., Benner, D. C., Sung, K., Brown, L. R., Crawford, T. J., Yu, S., Smith, M. A. H., Mantz, A. W., Boudon, V., and Ismail, S.:\nSpectral line parameters including line shapes in the 2ν3 Q branch of 12 CH4 , Journal of Quantitative Spectroscopy and Radiative Transfer,\n\n1400\n\n177, 152–169, 2016.\nDubey, M., Henderson, B., Green, D., Butterfield, Z., Keppel-Aleks, G., Allen, N., Blavier, J.-F., Roehl, C., Wunch, D., and Lindenmaier,\nR.: TCCON data from Manaus, Brazil, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of\nTechnology, Pasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.manaus01.R0, 2022a.\nDubey, M., Lindenmaier, R., Henderson, B., Green, D., Allen, N., Roehl, C., Blavier, J.-F., Butterfield, Z., Love, S., Hamelmann, J., and\n\n1405\n\nWunch, D.: TCCON data from Four Corners, NM, USA, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California\nInstitute of Technology, Pasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.fourcorners01.R0, 2022b.\nDubey, M. K., Parker, H. A., Wennberg, P. O., Wunch, D., Jacobson, A. R., Kawa, S. R., Keppel-Aleks, G., Basu, S., O’Dell, C., Frankenberg, C., Michalak, A. M., Baker, D. F., Christofferson, B., Restrepo-Coupe, N., Saleska, S. R., De Araujo, A. C., and Miller, J. B.:\nSeasonal Daily Amazon Column CO2 & CO Observations from Ground Space Used to Evaluate Tropical Ecosystem Models, American\n\n1410\n\nGeophysical Union, Fall Meeting 2016, abstract B33C-0623, last access 13 Sep 2022, 2016.\n\n76\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nDutton, G., Hall, B., Dlugokencky, E., Lan, X., Nance, J., and Madronich, M.: Combined Atmospheric Nitrous Oxide Dry Air Mole Fractions\nfrom the NOAA GML Halocarbons Sampling Network, 1977-2023, Version: 2023-04-13, https://doi.org/10.15138/GMZ7-2Q16, accessed\n10 May 2021, 2023.\nFeist, D. G., Arnold, S. G., John, N., and Geibel, M. C.: TCCON data from Ascension Island, Saint Helena, Ascension and Tristan da\n1415\n\nCunha, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A. ,\nhttps://doi.org/10.14291/tccon.ggg2020.ascension01.R0, 2017.\nFleurbaey, H., Reed, Z. D., Adkins, E. M., Long, D. A., and Hodges, J. T.: High accuracy spectroscopic parameters of the 1.27 µm band\nof O2 measured with comb-referenced, cavity ring-down spectroscopy, Journal of Quantitative Spectroscopy and Radiative Transfer, 270,\n107 684, 2021.\n\n1420\n\nForman, M. L., Steel, W. H., and Vanasse, G. A.: Correction of Asymmetric Interferograms Obtained in Fourier Spectroscopy∗, J. Opt. Soc.\nAm., 56, 59–63, https://doi.org/10.1364/JOSA.56.000059, 1966.\nFox, N. P.: Improved Near-Infrared Detectors, Metrologia, 30, 321, https://doi.org/10.1088/0026-1394/30/4/019, 1993.\nGoo, T.-Y., Oh, Y.-S., and Velazco, V. A.: TCCON data from Anmyeondo, South Korea, Release GGG2020R0. TCCON data archive, hosted\nby CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.anmeyondo01.R0,\n\n1425\n\n2017.\nHall, B. D., Crotwell, A. M., Kitzis, D. R., Mefford, T., Miller, B. R., Schibig, M. F., and Tans, P. P.: Revision of the World Meteorological\nOrganization Global Atmosphere Watch (WMO/GAW) CO2 calibration scale, Atmospheric Measurement Techniques, 14, 3015–3032,\nhttps://doi.org/10.5194/amt-14-3015-2021, 2021.\nHase, F.: Inversion von Spurengasprofilen aus hochaufgelöstenbodengebundenen FTIR-Messungen in Absorption, Ph.D. the-\n\n1430\n\nsis,\n\nInstitut\n\nfür\n\nMeteorologie\n\nund\n\nKlimaforschung,\n\nForschungszentrum\n\nKarlsruhe\n\nGmbH,\n\nKarlsruhe,\n\nhttp://www.imk-\n\nasf.kit.edu/downloads/bod/disshase.pdf, 2000.\nHase, F., Blumenstock, T., and Paton-Walsh, C.: Analysis of the instrumental line shape of high-resolution Fourier transform IR spectrometers\nwith gas cell measurements and new retrieval software, Appl. Opt., 38, 3417–3422, https://doi.org/10.1364/AO.38.003417, 1999.\nHase, F., Drouin, B. J., Roehl, C. M., Toon, G. C., Wennberg, P. O., Wunch, D., Blumenstock, T., Desmet, F., Feist, D. G., Heikkinen,\n1435\n\nP., De Mazière, M., Rettinger, M., Robinson, J., Schneider, M., Sherlock, V., Sussmann, R., Té, Y., Warneke, T., and Weinzierl, C.:\nCalibration of sealed HCl cells used for TCCON instrumental line shape monitoring, Atmospheric Measurement Techniques, 6, 3527–\n3537, https://doi.org/10.5194/amt-6-3527-2013, 2013.\nHase, F., Blumenstock, T., Dohe, S., Gro{\\ss}, J., and Kiel, M.: TCCON data from Karlsruhe, Germany, Release\nGGG2020R1. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A.,\n\n1440\n\nhttps://doi.org/10.14291/tccon.ggg2020.karlsruhe01.R1, 2022.\nHashemi, R., Gordon, I. E., Tran, H., Kochanov, R. V., Karlovets, E. V., Tan, Y., Lamouroux, J., Ngo, N. H., and Rothman, L. S.: Revising\nthe line-shape parameters for air-and self-broadened CO2 lines toward a sub-percent accuracy level, Journal of Quantitative Spectroscopy\nand Radiative Transfer, 256, 107 283, 2020.\nHedelius, J. K., He, T.-L., Jones, D. B. A., Baier, B. C., Buchholz, R. R., Mazière, M. D., Deutscher, N. M., Dubey, M. K., Feist, D. G.,\n\n1445\n\nGriffith, D. W. T., Hase, F., Iraci, L. T., Jeseck, P., Kiel, M., Kivi, R., Liu, C., Morino, I., Notholt, J., Oh, Y.-S., Ohyama, H., Pollard, D. F.,\nRettinger, M., Roche, S., Roehl, C. M., Schneider, M., Shiomi, K., Strong, K., Sussmann, R., Sweeney, C., Té, Y., Uchino, O., Velazco,\nV. A., Wang, W., Warneke, T., Wennberg, P. O., Worden, H. M., and Wunch, D.: Evaluation of MOPITT Version 7 joint TIR–NIR\n\n77\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nX&ampltsub&ampgtCO&amplt/sub&ampgt retrievals with TCCON, Atmos. Meas. Tech., 12, 5547–5572, https://doi.org/10.5194/amt12-5547-2019, 2019.\n1450\n\nIraci, L., Podolske, J., Hillyard, P., Roehl, C., Wennberg, P. O., Blavier, J.-F., Landeros, J., Allen, N., Wunch, D., Zavaleta, J., Quigley, E., Osterman, G., Barrow, E., and Barney, J.: TCCON data from Indianapolis, Indiana, USA, Release GGG2020R0. TCCON data archive, hosted\nby CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.indianapolis01.R1,\n2022a.\nIraci, L., Podolske, J., Roehl, C., Wennberg, P. O., Blavier, J.-F., Allen, N., Wunch, D., and Osterman, G.: TCCON data from Armstrong\n\n1455\n\nFlight Research Center, Edwards, CA, USA, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of\nTechnology, Pasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.edwards01.R0, 2022b.\nKarion, A., Sweeney, C., Tans, P., and Newberger, T.: AirCore: An Innovative Atmospheric Sampling System, J. Atmos. Oceanic Technol.,\n27, 1839–1853, https://doi.org/10.1175/2010jtecha1448.1, 2010.\nKeeler, E. and Burk, K.: Balloon-Borne Sounding System (SONDEWNPN), https://doi.org/10.5439/1595321.\n\n1460\n\nKeeling, R. and Manning, A.: Studies of Recent Changes in Atmospheric O2 Content, in: Treatise on Geochemistry, pp. 385–404, Elsevier,\nhttps://doi.org/10.1016/b978-0-08-095975-7.00420-4, 2014.\nKeeling, R. F., Manning, A. C., McEvoy, E. M., and Shertz, S. R.: Methods for measuring changes in atmospheric O2 concentration and their\napplication in southern hemisphere air, J. Geophys. Res. Atmos., 103, 3381–3397, https://doi.org/10.1029/97jd02537, 1998.\nKeppel-Aleks, G., Toon, G. C., Wennberg, P. O., and Deutscher, N. M.: Reducing the impact of source brightness fluctuations on spectra\n\n1465\n\nobtained by Fourier-transform spectrometry, Appl. Opt., 46, 4774–4779, https://doi.org/10.1364/AO.46.004774, 2007.\nKeppel-Aleks, G., Wennberg, P. O., and Schneider, T.: Sources of variations in total column carbon dioxide, Atmos. Chem. Phys., 11,\n3581–3593, https://doi.org/10.5194/acp-11-3581-2011, 2011.\nKeppel-Aleks, G., Wennberg, P. O., Washenfelder, R. A., Wunch, D., Schneider, T., Toon, G. C., Andres, R. J., Blavier, J.-F., Connor, B.,\nDavis, K. J., Desai, A. R., Messerschmidt, J., Notholt, J., Roehl, C. M., Sherlock, V., Stephens, B. B., Vay, S. A., and Wofsy, S. C.: The im-\n\n1470\n\nprint of surface fluxes and transport on variations in total column carbon dioxide, Biogeosciences, 9, 875–891, https://doi.org/10.5194/bg9-875-2012, 2012.\nKiel, M., Hase, F., Blumenstock, T., and Kirner, O.: Comparison of XCO abundances from the Total Carbon Column Observing Network\nand the Network for the Detection of Atmospheric Composition Change measured in Karlsruhe, Atmos. Meas. Tech., 9, 2223–2239,\nhttps://doi.org/10.5194/amt-9-2223-2016, 2016a.\n\n1475\n\nKiel, M., Wunch, D., Wennberg, P. O., Toon, G. C., Hase, F., and Blumenstock, T.: Improved retrieval of gas abundances from\nnear-infrared solar FTIR spectra measured at the Karlsruhe TCCON station, Atmospheric Measurement Techniques, 9, 669–682,\nhttps://doi.org/10.5194/amt-9-669-2016, 2016b.\nKivi, R. and Heikkinen, P.: Fourier transform spectrometer measurements of column CO2 at Sodankylä, Finland, Geoscientific Instrumentation, Methods and Data Systems, 5, 271–279, https://doi.org/10.5194/gi-5-271-2016, 2016.\n\n1480\n\nKivi, R., Heikkinen, P., and Kyro, E.: TCCON data from Sodankyla, Finland, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.sodankyla01.R0, 2022.\nLan, X., Dlugokencky, E., Mund, J., Crotwell, A., Crotwell, M., Moglia, E., Madronich, M., Neff, D., and Thoning, K.: Atmospheric Methane\nDry Air Mole Fractions from the NOAA GML Carbon Cycle Cooperative Global Air Sampling Network, 1983-2021, Version: 2022-1121, https://doi.org/10.15138/VNCZ-M766, 2022a.\n\n78\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n1485\n\nLan, X., Dlugokencky, E., Mund, J., Crotwell, A., Crotwell, M., Moglia, E., Madronich, M., Neff, D., and Thoning, K.: Atmospheric Carbon\nDioxide Dry Air Mole Fractions from the NOAA GML Carbon Cycle Cooperative Global Air Sampling Network, 1968-2021, Version:\n2022-11-21, https://doi.org/10.15138/wkgj-f215, 2022b.\nLan, X., Dlugokencky, E., Mund, J., Crotwell, A., Crotwell, M., Moglia, E., Madronich, M., Neff, D., and Thoning, K.: Atmospheric Nitrous\nOxide Dry Air Mole Fractions from the NOAA GML Carbon Cycle Cooperative Global Air Sampling Network,1997-2021, Version:\n\n1490\n\n2022-11-21, https://doi.org/10.15138/9N0H-ZH07, 2022c.\nLan, X., Tans, P., and Thoning, K.: Trends in globally-averaged CO2 determined from NOAA Global Monitoring Laboratory measurements.\nVersion 2023-04, https://doi.org/10.15138/9N0H-ZH07, 2023.\nLaughner, J. L., Roche, S., Kiel, M., Toon, G. C., Wunch, D., Baier, B. C., Biraud, S., Chen, H., Kivi, R., Laemmel, T., McKain, K., Quéhé,\nP.-Y., Rousogenous, C., Stephens, B. B., Walker, K., and Wennberg, P. O.: A new algorithm to generate a priori trace gas profiles for the\n\n1495\n\nGGG2020 retrieval algorithm, Atmos. Meas. Tech., 16, 1121–1146, https://doi.org/10.5194/amt-16-1121-2023, 2023.\nLiu, C., Wang, W., and Sun, Y.: TCCON data from Hefei, China, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA,\nCalifornia Institute of Technology, Pasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.hefei01.R0, 2022.\nLorente, A., Borsdorff, T., Martinez-Velarte, M. C., Butz, A., Hasekamp, O. P., Wu, L., and Landgraf, J.: Evaluation of the methane\nfull-physics retrieval applied to TROPOMI ocean sun glint measurements, Atmospheric Measurement Techniques, 15, 6585–6603,\n\n1500\n\nhttps://doi.org/10.5194/amt-15-6585-2022, 2022.\nLucchesi, R.: File Specification for GEOS-5 FP-IT (forward processing for instrument teams), Tech. rep., NASA Goddard Space Flight\nCenter, Greenbelt, MD, USA„ https://gmao.gsfc.nasa.gov/pubs/docs/Lucchesi865.pdf, last access: 13 Oct 2020, 2015.\nMartínez-Alonso, S., Deeter, M. N., Baier, B. C., McKain, K., Worden, H., Borsdorff, T., Sweeney, C., and Aben, I.: Evaluation of MOPITT\nand TROPOMI carbon monoxide retrievals using AirCore in situ vertical profiles, Atmospheric Measurement Techniques, 15, 4751–4765,\n\n1505\n\nhttps://doi.org/10.5194/amt-15-4751-2022, 2022.\nMaziere, D., M., M. K. S., Desmet, F., Hermans, C., Scolas, F., Kumps, N., Zhou, M., Metzger, J.-M., Duflot, V., and Cammas, J.-P.: TCCON\ndata from Reunion Island (La Reunion), France, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute\nof Technology, Pasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.reunion01.R0, 2022.\nMendonca, J., Strong, K., Toon, G., Wunch, D., Sung, K., Deutscher, N., Griffith, D., and Franklin, J.: Improving atmospheric CO2 retrievals\n\n1510\n\nusing line mixing and speed-dependence when fitting high-resolution ground-based solar spectra, Journal of Molecular Spectroscopy, 323,\n15–27, https://doi.org/https://doi.org/10.1016/j.jms.2016.01.007, atmospheric Spectroscopy, 2016.\nMendonca, J., Strong, K., Sung, K., Devi, V., Toon, G., Wunch, D., and Franklin, J.: Using high-resolution laboratory and ground-based solar\nspectra to assess CH4 absorption coefficient calculations, Journal of Quantitative Spectroscopy and Radiative Transfer, 190, 48–59, 2017.\nMendonca, J., Strong, K., Wunch, D., Toon, G. C., Long, D. A., Hodges, J. T., Sironneau, V. T., and Franklin, J. E.: Using a speed-dependent\n\n1515\n\nVoigt line shape to retrieve O2 from Total Carbon Column Observing Network solar spectra to improve measurements of XCO2 , Atmos.\nMeas. Tech., 12, 35–50, 2019.\nMertz, L.: Transformations in optics, John Wiley and Sons, Inc., 1965.\nMertz, L.: Auxiliary computation for Fourier spectrometry, Infrared Physics, 7, 17–23, https://doi.org/https://doi.org/10.1016/00200891(67)90026-7, 1967.\n\n1520\n\nMiloshevich, L. M., Paukkunen, A., Vömel, H., and Oltmans, S. J.: Development and Validation of a Time-Lag Correction for Vaisala Radiosonde Humidity Measurements, J. Atmos. Oceanic Technol., 21, 1305–1327, https://doi.org/10.1175/15200426(2004)021<1305:davoat>2.0.co;2, 2004.\n\n79\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nMiloshevich, L. M., Vömel, H., Whiteman, D. N., Lesht, B. M., Schmidlin, F. J., and Russo, F.: Absolute accuracy of water vapor measurements from six operational radiosonde types launched during AWEX-G and implications for AIRS validation, J. Geophys. Res., 111,\n1525\n\nhttps://doi.org/10.1029/2005jd006083, 2006.\nMorino,\n\nI.,\n\nOhyama,\n\nH.,\n\nHori,\n\nA.,\n\nand\n\nIkegami,\n\nH.:\n\nTCCON\n\ndata\n\nfrom\n\nRikubetsu,\n\nHokkaido,\n\nJapan,\n\nRelease\n\nGGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A.,\nhttps://doi.org/10.14291/tccon.ggg2020.rikubetsu01.R0, 2022a.\nMorino, I., Ohyama, H., Hori, A., and Ikegami, H.: TCCON data from Tsukuba, Ibaraki, Japan, 125HR, Release\n1530\n\nGGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A.,\nhttps://doi.org/10.14291/tccon.ggg2020.tsukuba02.R0, 2022b.\nMorino, I., Velazco, V. A., Hori, A., Uchino, O., and Griffith, D. W. T.: TCCON data from Burgos, Philippines, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A.,\nhttps://doi.org/10.14291/tccon.ggg2020.burgos01.R0, 2022c.\n\n1535\n\nNotholt, J., Petri, C., Warneke, T., Deutscher, N., Buschmann, M., Weinzierl, C., Macatangay, R., and Grupe, P.: TCCON data from Bremen,\nGermany, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A.,\nhttps://doi.org/10.14291/tccon.ggg2020.bremen01.R0, 2022.\nParker, H. A., Laughner, J. L., Toon, G. C., Wunch, D., Roehl, C. M., Iraci, L. T., Podolske, J. R., McKain, K., Baier, B. C., and Wennberg,\nP. O.: Inferring the vertical distribution of CO and COsub2/sub from TCCON total column values using the TARDISS algorithm, Atmo-\n\n1540\n\nspheric Measurement Techniques, 16, 2601–2625, https://doi.org/10.5194/amt-16-2601-2023, 2023.\nPeiro, H., Crowell, S., and III, B. M.: Optimizing 4 years of COsub2/sub biospheric fluxes from OCO-2 and in situ data in TM5: fire emissions\nfrom GFED and inferred from MOPITT CO data, Atmospheric Chemistry and Physics, 22, 15 817–15 849, https://doi.org/10.5194/acp22-15817-2022, 2022.\nPetri, C., Deutscher, N., Notholt, J., Messerschmidt, J., Weinzierl, C., Warneke, T., , Grupe, P., and Katrynski, K.: TCCON data from\n\n1545\n\nBialystok, Poland, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena,\nCA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.bialystok01.R0, 2017.\nPetri, C., Vrekoussis, M., Rousogenous, C., Warneke, T., Sciare, J., and Notholt, J.: TCCON data from Nicosia, Cyprus, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A.,\nhttps://doi.org/10.14291/tccon.ggg2020.nicosia01.R0, 2023.\n\n1550\n\nPollard, D., Robinson, J., and Shiona, H.: TCCON data from Lauder, New Zealand, 125HR, Release GGG2020R0. TCCON data archive,\nhosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.lauder03.R0,\n2022.\nRousogenous, C.: Automated ground-based remote sensing measurements of greenhouse gases at the Nicosia site in comparison with collocated in situ measurements and model data, Atmos. Chem. Phys., in prep.\n\n1555\n\nSaad, K. M., Wunch, D., Toon, G. C., Bernath, P., Boone, C., Connor, B., Deutscher, N. M., Griffith, D. W. T., Kivi, R., Notholt, J., Roehl,\nC., Schneider, M., Sherlock, V., and Wennberg, P. O.: Derivation of tropospheric methane from TCCON CH4 and HF total column\nobservations, Atmos. Meas. Tech., 7, 2907–2918, https://doi.org/10.5194/amt-7-2907-2014, 2014.\nSaad, K. M., Wunch, D., Deutscher, N. M., Griffith, D. W. T., Hase, F., De Mazière, M., Notholt, J., Pollard, D. F., Roehl, C. M., Schneider, M.,\nSussmann, R., Warneke, T., and Wennberg, P. O.: Seasonal variability of stratospheric methane: implications for constraining tropospheric\n\n80\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n1560\n\nmethane budgets using total column observations, Atmos. Chem. Phys., 16, 14 003–14 024, https://doi.org/10.5194/acp-16-14003-2016,\n2016.\nScripps O2 Program: Flask O2 /N2 data from the Alert, NWT, Canada; La Jolla Pier, California; and Cape Grim, Australia stations, https:\n//scrippso2.ucsd.edu/data.html, 2022.\nSha, M. K., De Mazière, M., Notholt, J., Blumenstock, T., Chen, H., Dehn, A., Griffith, D. W. T., Hase, F., Heikkinen, P., Hermans, C.,\n\n1565\n\nHoffmann, A., Huebner, M., Jones, N., Kivi, R., Langerock, B., Petri, C., Scolas, F., Tu, Q., and Weidmann, D.: Intercomparison of lowand high-resolution infrared spectrometers for ground-based solar remote sensing measurements of total column concentrations of CO2 ,\nCH4 , and CO, Atmospheric Measurement Techniques, 13, 4791–4839, https://doi.org/10.5194/amt-13-4791-2020, 2020.\nSherlock, V., Connor, B., Robinson, J., Shiona, H., Smale, D., and Pollard, D.: TCCON data from Lauder, New Zealand, 120HR,\nRelease GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A.,\n\n1570\n\nhttps://doi.org/10.14291/tccon.ggg2020.lauder01.R0, 2022a.\nSherlock, V., Connor, B., Robinson, J., Shiona, H., Smale, D., and Pollard, D.: TCCON data from Lauder, New Zealand, 125HR,\nRelease GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A.,\nhttps://doi.org/10.14291/tccon.ggg2020.lauder02.R0, 2022b.\nShiomi, K., Kawakami, S., Ohyama, H., Arai, K., Okumura, H., Ikegami, H., and Usami, M.: TCCON data from Saga, Japan,\n\n1575\n\nRelease GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A.,\nhttps://doi.org/10.14291/tccon.ggg2020.saga01.R0, 2022.\nStrong, K., Roche, S., Franklin, J., Mendonca, J., Lutsch, E., Weaver, D., Fogal, P., Drummond, J., Batchelor, R., and Lindenmaier, R.: TCCON data from Eureka, Canada, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology,\nPasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.eureka01.R0, 2022.\n\n1580\n\nSussmann, R. and Rettinger, M.: TCCON data from Garmisch, Germany, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.garmisch01.R0, 2017a.\nSussmann, R. and Rettinger, M.: TCCON data from Zugspitze, Germany, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.zugspitze01.R0, 2017b.\nSweeney, C., McKain, K., Higgs, J., Wolter, S., Crotwell, A., Neff, D., Dlugokencky, E., Lang, P., Novelli, P., Mund, J., Moglia,\n\n1585\n\nE., and Crotwell, M.: NOAA Carbon Cycle and Greenhouse Gases Group aircraft-based measurements of CO2 , CH4 , CO,\nN2 O, H2\n\nSF6 in flask-air samples taken since 1992. NOAA Earth System Research Laboratory, Global Monitoring Division.,\n\nhttps://doi.org/10.7289/V5N58JMF, 2018.\nTange, O.: GNU Parallel - The Command-Line Power Tool, ;login: The USENIX Magazine, February, 42–47, 2011.\nTans, P.: System and method for providing vertical profile measurements of atmospheric gases., U.S. Patent 7,597,014, filed 15 Aug 2006,\n1590\n\nissued 6 Oct 2009, 2009.\nTe, Y., Jeseck, P., and Janssen, C.: TCCON data from Paris, France, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA,\nCalifornia Institute of Technology, Pasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.paris01.R0, 2022.\nToon, G.: Atmospheric Non-Voigt Line List for the TCCON 2020 Data Release (GGG2020.R0) [Data set]. CaltechDATA,\nhttps://doi.org/10.14291/TCCON.GGG2020.ATMNV.R0, 2022a.\n\n1595\n\nToon,\n\nG.:\n\nSolar\n\nLine\n\nList\n\nfor\n\nthe\n\nTCCON\n\n2020\n\nhttps://doi.org/TCCON.GGG2020.SOLAR.R0, 2022b.\n\n81\n\nData\n\nRelease\n\n(GGG2020.R0)\n\n[Data\n\nset].\n\nCaltechDATA,\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nToon, G.: Atmospheric Voigt Line List for the TCCON 2020 Data Release (GGG2020.R0) [Data set]. CaltechDATA,\nhttps://doi.org/10.14291/TCCON.GGG2020.ATM.R0, 2022c.\nToon, G.: TCCON/GGG – GGG2020, https://doi.org/10.14291/tccon.ggg2020.stable.R0, 2023.\n1600\n\nTotal\n\nCarbon\n\nColumn\n\nObserving\n\nNetwork\n\n(TCCON)\n\nTeam:\n\n2020\n\nTCCON\n\nData\n\nRelease\n\n(GGG2020),\n\nhttps://doi.org/10.14291/TCCON.GGG2020, 2022.\nTran, D., Delahaye, T., Armante, R., Hartmann, J.-M., Mondelain, D., Campargue, A., Fleurbaey, H., Hodges, J., and Tran, H.: Validation of\nspectroscopic data in the 1.27 µm spectral region by comparisons with ground-based atmospheric measurements, Journal of Quantitative\nSpectroscopy and Radiative Transfer, 261, 107 495, 2021.\n1605\n\nTran, D. D., Tran, H., Vasilchenko, S., Kassi, S., Campargue, A., and Mondelain, D.: High sensitivity spectroscopy of the O2 band at 1.27\nµm:(II) air-broadened line profile parameters, Journal of Quantitative Spectroscopy and Radiative Transfer, 240, 106 673, 2020.\nTran, H., Ngo, N. H., and Hartmann, J.-M.: Efficient computation of some speed-dependent isolated line profiles, Journal of Quantitative\nSpectroscopy and Radiative Transfer, 129, 199–203, 2013.\nWang, Z., Deutscher, N., Warneke, T., Notholt, J., Dils, B., Griffith, D. W., Schmidt, M., Ramonet, M., and Gerbig, C.: Retrieval of tropo-\n\n1610\n\nspheric column-averaged CH 4 mole fraction by solar absorption FTIR-spectrometry using N 2 O as a proxy, Atmospheric Measurement\nTechniques, 7, 3295–3305, 2014.\nWarneke, T., Messerschmidt, J., Notholt, J., Weinzierl, C., Deutscher, N., Petri, C., Grupe, P., Vuillemin, C., Truong, F., Schmidt, M., Ramonet, M., and Parmentier, E.: TCCON data from Orleans, France, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA,\nCalifornia Institute of Technology, Pasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.orleans01.R0, 2022.\n\n1615\n\nWashenfelder, R. A., Wennberg, P. O., and Toon, G. C.: Tropospheric methane retrieved from ground-based near-IR solar absorption spectra,\nGeophys. Res. Lett., 30, https://doi.org/10.1029/2003GL017969, 2003.\nWeidmann, D., Brownsword, R., and Doniki, S.: TCCON data from Harwell, Oxfordshire (UK), Release GGG2020.R0,\nhttps://doi.org/10.14291/tccon.ggg2020.harwell01.R0, 2023.\nWennberg, P. O., Roehl, C., Blavier, J.-F., Wunch, D., Landeros, J., and Allen, N.: TCCON data from Jet Propulsion Laboratory, Pasadena,\n\n1620\n\nCalifornia, USA, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology, Pasadena, CA,\nU.S.A., https://doi.org/10.14291/tccon.ggg2020.jpl02.R0, 2022a.\nWennberg, P. O., Roehl, C., Wunch, D., Toon, G. C., Blavier, J.-F., Washenfelder, R., Keppel-Aleks, G., Allen, N., and Ayers, J.: TCCON\ndata from Park Falls, Wisconsin, USA, Release GGG2020R1. TCCON data archive, hosted by CaltechDATA, California Institute of\nTechnology, Pasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.parkfalls01.R1, 2022b.\n\n1625\n\nWennberg, P. O., Wunch, D., Roehl, C., Blavier, J.-F., Toon, G. C., and Allen, N.: TCCON data from California Institute of Technology,\nPasadena, California, USA, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology,\nPasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.pasadena01.R0, 2022c.\nWennberg, P. O., Wunch, D., Roehl, C., Blavier, J.-F., Toon, G. C., Allen, N., Dowell, P., Teske, K., Martin, C., and Martin, J.: TCCON data\nfrom Lamont, Oklahoma, USA, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of Technology,\n\n1630\n\nPasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.lamont01.R0, 2022d.\nWennberg, P. O., Wunch, D., Yavin, Y., Toon, G. C., Blavier, J.-F., Allen, N., and Keppel-Aleks, G.: TCCON data from Jet Propulsion\nLaboratory, Pasadena, California, USA, Release GGG2020R0. TCCON data archive, hosted by CaltechDATA, California Institute of\nTechnology, Pasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.jpl01.R0, 2022e.\n\n82\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nWexler, A.: Vapor Pressure Formulation for Water in Range 0 to 100◦ C. A Revision, J. Res. Nat. Bur. Stand. A Phys. Chem., 80A,\n1635\n\nhttps://doi.org/10.6028/jres.080A.071, 1976.\nWunch, D., Toon, G. C., Wennberg, P. O., Wofsy, S. C., Stephens, B. B., Fischer, M. L., Uchino, O., Abshire, J. B., Bernath, P., Biraud,\nS. C., Blavier, J.-F. L., Boone, C., Bowman, K. P., Browell, E. V., Campos, T., Connor, B. J., Daube, B. C., Deutscher, N. M., Diao,\nM., Elkins, J. W., Gerbig, C., Gottlieb, E., Griffith, D. W. T., Hurst, D. F., Jiménez, R., Keppel-Aleks, G., Kort, E. A., Macatangay, R.,\nMachida, T., Matsueda, H., Moore, F., Morino, I., Park, S., Robinson, J., Roehl, C. M., Sawa, Y., Sherlock, V., Sweeney, C., Tanaka,\n\n1640\n\nT., and Zondlo, M. A.: Calibration of the Total Carbon Column Observing Network using aircraft profile data, Atmos. Meas. Tech., 3,\n1351–1362, https://doi.org/10.5194/amt-3-1351-2010, 2010.\nWunch, D., Toon, G. C., Blavier, J.-F. L., Washenfelder, R. A., Notholt, J., Connor, B. J., Griffith, D. W., Sherlock, V., and Wennberg,\nP. O.: The Total Carbon Column Observing Network, Philosophical Transactions of the Royal Society A: Mathematical, Physical and\nEngineering Sciences, 369, 2087–2112, https://doi.org/10.1098/rsta.2010.0240, 2011.\n\n1645\n\nWunch, D., Toon, G. C., Sherlock, V., Deutscher, N. M., Liu, C., Feist, D. G., and Wennberg, P. O.: Documentation for the 2014 TCCON\nData Release, https://doi.org/10.14291/TCCON.GGG2014.DOCUMENTATION.R0/1221662, 2015.\nWunch, D., Wennberg, P. O., Osterman, G., Fisher, B., Naylor, B., Roehl, C. M., O'Dell, C., Mandrake, L., Viatte, C., Kiel, M., Griffith, D.\nW. T., Deutscher, N. M., Velazco, V. A., Notholt, J., Warneke, T., Petri, C., Maziere, M. D., Sha, M. K., Sussmann, R., Rettinger, M.,\nPollard, D., Robinson, J., Morino, I., Uchino, O., Hase, F., Blumenstock, T., Feist, D. G., Arnold, S. G., Strong, K., Mendonca, J., Kivi,\n\n1650\n\nR., Heikkinen, P., Iraci, L., Podolske, J., Hillyard, P. W., Kawakami, S., Dubey, M. K., Parker, H. A., Sepulveda, E., García, O. E., Te, Y.,\nJeseck, P., Gunson, M. R., Crisp, D., and Eldering, A.: Comparisons of the Orbiting Carbon Observatory-2 (OCO-2) XCO2 measurements\nwith TCCON, Atmospheric Measurement Techniques, 10, 2209–2238, https://doi.org/10.5194/amt-10-2209-2017, 2017.\nWunch, D., Mendonca, J., Colebatch, O., Allen, N., Blavier, J.-F. L., Kunz, K., Roche, S., Hedelius, J., Neufeld, G., Springett, S., Worthy, D., Kessler, R., and Strong, K.: TCCON data from East Trout Lake, Canada, Release GGG2020R0. TCCON data archive, hosted\n\n1655\n\nby CaltechDATA, California Institute of Technology, Pasadena, CA, U.S.A., https://doi.org/10.14291/tccon.ggg2020.easttroutlake01.R0,\n2022.\nZhou, M., Wang, P., Nan, W., Yang, Y., Kumps, N., Hermans, C., and De Mazière, M.: TCCON data from Xianghe,\nhttps://doi.org/10.14291/tccon.ggg2020.xianghe01.R0, 2022.\n\n83\n\n\f84\n\nCRDS on tower\n\nFlask\n\nCRDS on tower\n\nLicor 7000 NDIR (CO2 ), in situ GHG FTS\n\nCH4 ObsPack\n\nCH4 ObsPack\n\nCH4 ObsPack\n\nNIWA (direct)\n\n(CH4 )\n\nProgrammable flask packages\n\nCH4 ObsPack\n\nLi-cor NDIR on tower\n\nCO2 ObsPack\n\nCRDS on tower\n\nProgrammable flask packages\n\nCO2 ObsPack\n\nCO2 ObsPack\n\nMeasurement type\n\nSource\n\nNIWA = National Institute of Water & Atmospheric Research Ltd.\n\nGreat\n\nDan Smale (NIWA)\n\nGreat\n\nPlains\n\nPlains\n\nPlains\n\nLauder, New Zealand\n\nARM site, OK, USA\n\nSouthern\n\nARM site, OK, USA\n\nBiraud (LBNL), & Margaret Torn (LBNL)\nSebastien Biraud & Margaret Torn (LBNL)\n\nSouthern\n\nGreat\n\nPark Falls, WI, USA\n\nPark Falls, WI, USA\n\nARM site, OK, USA\n\nSouthern\n\nPark Falls, WI, USA\n\nPark Falls, WI, USA\n\nLocation\n\nEd Dlugokencky (NOAA GML), Sebastien\n\nsai (U. of WI), & Dan Baumann (USGS)\n\nArlyn Andrews (NOAA GML), Ankur De-\n\nWI), & Dan Baumann (USGS)\n\nkencky (NOAA GML), Ankur Desai (U. of\n\nArlyn Andrews (NOAA GML), Ed Dlugo-\n\nSebastien Biraud & Margaret Torn (LBNL)\n\n(USGS)\n\nAnkur Desai (U. of WI), & Dan Baumann\n\nkencky (NOAA GML), Ken Davis (PSU),\n\nArlyn Andrews (NOAA GML), Ed Dlugo-\n\nWI), & Dan Baumann (USGS)\n\nwin, Ken Davis (PSU), Ankur Desai (U. of\n\nArlyn Andrews (NOAA GML), Peter Bak-\n\nProviders/partners\n\nll\n\noc\n\noc\n\npa\n\npa\n\noc\n\npa\n\npa\n\nTCCON site\n\nGeological Survey; LBNL = Lawrence Berkeley National Laboratory; ARM = Atmospheric Radiation Measurement; CRDS = cavity ring-down spectroscopy;\n\nand Atmospheric Administration Global Monitoring Laboratory; PSU = Pennsylvania State University; U. of WI = University of Wisconsin; USGS = United States\n\nparentheses. If only one affiliation is listed, it applies to all individuals named. Abbrevations: NDIR = Nondispersive infrared; NOAA GML = National Oceanic\n\n“TCCON sites” column indicates which sites profile were used at, the IDs are mapped to locations in Table 1. In the “Providers” column, affiliations are given in\n\nIntegration Project, 2019) and “CH4 ObsPack” the CH4 GLOBALVIEWplus v2.0 ObsPack (Cooperative Global Atmospheric Data Integration Project, 2020). The\n\nTable C3. Ground in situ data used in validating the priors. “CO2 Obspack” is the CO2 GLOBALVIEWplus v5.0 ObsPack (Cooperative Global Atmospheric Data\n\nhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nTable C4. The number of profiles in the CO2 in situ correction from each campaign or other data source identified and used for each TCCON\nsite. The “Found” column gives the number of profiles identified for that campaign & site, the “Used” column gives the number of those\nprofiles which could be used in the in situ comparison after matching with TCCON data. The definitions of the site IDs can be found in Table\n1; “we” refers to an instrument in Jena, Germany for which GGG2020 data is not available at time of writing.\nCampaign\n\nSite\n\nFound\n\nUsed\n\nCampaign\n\nSite\n\nATom\n\nae\n\n4\n\ndf\neu\n\nFound\n\nUsed\n\n0\n\nINTEX-NA\n\npa\n\n3\n\n3\n\n1\n\n1\n\nKORUS-AQ\n\n2\n\n0\n\nan\n\n1\n\n1\n\ndf\n\n1\n\n1\n\nll\n\n4\n\n4\n\nrj\n\n2\n\n2\n\noc\n\n1\n\n0\n\nORCAS\n\noc\n\n1\n\n1\n\npa\n\n1\n\n1\n\nSEAC4RS\n\ndf\n\n1\n\n1\n\nCOB2004\n\npa\n\n5\n\n4\n\noc\n\n2\n\n0\n\nDC3\n\noc\n\n3\n\n2\n\nSTART-08\n\npa\n\n2\n\n0\n\nGO-Amazon\n\nma\n\n2\n\n1\n\nAirCore\n\ndf\n\n3\n\n3\n\nGSFC\n\ndf\n\n8\n\n7\n\nni\n\n3\n\n2\n\npa\n\n2\n\n2\n\noc\n\n19\n\n13\n\nll\n\n7\n\n5\n\npa\n\n2\n\n2\n\nwg\n\n1\n\n0\n\nso\n\n16\n\n9\n\nbi\n\n2\n\n2\n\nbr\n\n2\n\n0\n\ngm\n\n1\n\n1\n\nje\n\n1\n\n0\n\nka\n\n1\n\n0\n\nor\n\n2\n\n0\n\nHIPPO\nIMECC\n\n85\n\n\fhttps://doi.org/10.5194/essd-2023-331\nPreprint. Discussion started: 24 August 2023\nc Author(s) 2023. CC BY 4.0 License.\n\nTable C5. Same as Table C4 but for the CH4 in situ correction.\nCampaign\n\nSite\n\nATom\n\nae\n\nHIPPO\n\nFound\n\nUsed\n\n4\n\n0\n\nCampaign\n\nSite\n\nFound\n\nUsed\n\nIMECC\n\nbi\n\n2\n\n2\n\nci\n\n2\n\n1\n\nbr\n\n2\n\n0\n\ndf\n\n1\n\n1\n\ngm\n\n1\n\n0\n\neu\n\n1\n\n0\n\nje\n\n1\n\n0\n\nll\n\n1\n\n1\n\nka\n\n1\n\n0\n\noc\n\n1\n\n0\n\nor\n\n2\n\n0\n\npa\n\n1\n\n1\n\nSTART-08\n\npa\n\n2\n\n1\n\nll\n\n5\n\n3\n\nAirCore\n\ndf\n\n3\n\n3\n\noc\n\n4\n\n1\n\nni\n\n3\n\n2\n\npa\n\n1\n\n0\n\noc\n\n19\n\n13\n\nwg\n\n1\n\n0\n\npa\n\n2\n\n2\n\nso\n\n16\n\n9\n\nTable C6. Values of ∂r/∂Xluft in Eq. (C11). Gases not listed here use 0 for ∂r/∂Xluft .\nGas\n\n∂r/∂Xluft\n\nCO2\n\n0.363\n\nwCO2\n\n0.206\n\nlCO2\n\n0.928\n\nCH4\n\n0.0609\n\n86\n\n\f"