Global Monitoring Division

GLOBALVIEW-CO2, 2008

1. Release Date: 27 August 2008
2. Data additions
   Discrete surface measurements:

   • HPB Hohenpeissenberg, Germany (ESRL)
   • SGP374 Southern Great Plains, Oklahoma, United States (ESRL)
   Continuous surface measurements:

   • HDPDTA Hidden Peak, Colorado, United States (NCAR) (Daytime Averages, 1200-1800 LST)
   • HDPNTA Hidden Peak, Colorado, United States (NCAR) (Nighttime Averages, 0000-0400 LST)
   • NWRDTA Niwot Ridge, Colorado, United States (NCAR) (Daytime Averages, 1200-1800 LST)
   • NWRNTA Niwot Ridge, Colorado, United States (NCAR) (Nighttime Averages, 0000-0400 LST)
   • SPLDTA Storm Peak, Colorado, United States (NCAR) (Daytime Averages using 1200-1800 LST)
   • SPLNTA Storm Peak, Colorado, United States (NCAR) (Nighttime Averages using 0000-0400 LST)
3. Site Code Change
   In January 2007, the ESRL aircraft site at Rowley, Iowa, United States (RIA) was moved to West Branch, Iowa (WBI). Data from RIA are now merged with data from WBI.
4. Lab Change
   • Collaborating laboratory 06, formerly called "Meteorological Service of Canada" (MSC), is now called "Environment Canada" (EC).
   • Collaborating laboratory 27, formerly called "Instituto Nacional de Meteorologia" (INM), is now called "Meteorological State Agency of Spain" (AEMET).

1. Release Date: 20 August 2007
2. Data additions
   Discrete surface measurements:

   • CYA Casey, Antarctica, Australia (CSIRO)
   • PTA Point Arena, California, United States (ESRL)
   Discrete measurements from aircraft:

   • BNE Beaver Crossing, Nebraska, United States (ESRL)
   • DND Dahlen, North Dakota, United States (ESRL)
   • HIL Homer, Illinois, United States (ESRL)
   • LEF Park Falls, Wisconsin, United States (ESRL)
   • RIA Rowley, Iowa, United States (ESRL)
   • THD Trinidad Head, California, United States (ESRL)
   Continuous measurements from a tall tower:

   • AMT Argyle, Maine, United States (ESRL)
3. Sampling at Izana Observatory (IZO)
   Please note that daily averages from Izana Observatory contributed by the Observatorio Atmosferico de Izana, Instituto Nacional de Meteorologia (INM), Spain are computed using only nighttime hours, i.e., 20-23 (previous day) and 00-07 (reported day) to assure free troposphere background conditions at this high altitude mountain site. Prior to 2002, NOAA discrete samples were typically collected between 20 and 23 hours. Since June 2002, the collection time of NOAA samples has changed to afternoon hours (15-16). NOAA samples collected in the afternoon are not directly comparable to the INM daily averages computed using only nighttime hours.
4. Site Code Change
   The Old Black Spruce, Saskatchewan, Canada site (OBS) operated by MSC was renamed Candle Lake, Saskatchewan, Canada (CDL).
5. Aircraft Sampling in the Western Pacific (WPO)
   The long-term aircraft sampling program over the Western Pacific is now managed by the Center for Global Environmental Research National Institute for Environmental Studies (NIES) in collaboration with the Meteorological Research Institute (MRI), Japan. Please note that the standard gas scale used to report 1993-2005 data is different from the scale used to report the 2006 data due to the change in the central analytical laboratory from MRI to NIES. The difference of the CO2 standard gas scale is estimated to be about -0.14 ppm (MRI minus NIES) around 380 ppm CO2 level and +0.10 ppm around 340 ppm. A consistent data set will be made available after a more careful evaluation of the standard gas scale between the two laboratories.
6. Data deletions from GLOBALVIEW-CO2
   The ESRL discrete measurements from Park Falls, Wisconsin (LEF) and Grifton, North Carolina (ITN) are no longer included in the GLOBALVIEW product. These sites are instead represented by the ESRL tall tower quasi-continuous measurements, which provide a more representative record of observed high-frequency variability and daily patterns.

1. Release Date: 31 August 2006
2. Data additions
   Discrete surface measurements:

   • BKT Bukit Kototagang, Indonesia (ESRL)
   • MKN Mt. Kenya, Kenya (ESRL)
   • PAL Pallas, Finland (ESRL)

   Discrete measurements from aircraft:

   • TGC Sinton, Texas, United States (ESRL)
   • NHA Worcester, Massachusetts, United States (ESRL)
3. Lab Name Change
   As of October 1, 2005, the Climate Monitoring and Diagnostics Laboratory (CMDL) has merged into the Earth System Research Laboratory (ESRL) as part of its Global Monitoring Division (GMD).

1. Release Date: 15 August 2006
2. Data additions
   Discrete measurements from aircraft:

   • ESP Estevan Point, British Columbia, Canada (ESRL)
   • HFM Harvard Forest, Massachusetts, United States (ESRL)
   • ZOT Zotino, Siberia, Russia (MPI-BGC)

   Continuous surface measurements:

   • SBL Sable Island, Canada (MSC)

   Continuous measurements from a short tower:

   • OBS023 Saskatchewan, Canada (MSC)
3. Lab Identification Number Change
   The Lab ID number for NOAA ESRL has been changed from 00 to 01.
4. Site Classification Change
   The Fraserdale, Ontario, Canada (FRD) site has been re-classified within GLOBALVIEW as a tower site. It is now referred to as FRD040 indicating the sample intake height is 40m above the surface.
5. Modifications to the Preparation and Use of Tower Data
   Preparation of semi-continuous measurements from sites designated as tower platforms has been modified in an effort to standardize the treatment of data from towers sampling at one or many levels.

   Tower data are now averaged with daily resolution using afternoon hours only. In earlier releases of this data product, daily averages were computed using all hours (e.g., 24-hour average).

   The residual distribution used to prepare the statistical summary of average diurnal cycles is now determined at each level by subtracting the afternoon-hour average mixing ratio for each day from every observation for that day. In earlier releases of this product, the residual distribution was determined at each level by subtracting the 24-hour average mixing ratio for each day from every observation for that day; for tall tower measurements, the 24-hour average was determined from measurements at the highest level.

1. Release Date: 15 August 2004
2. Data additions
   Discrete surface measurements:

   • PDM Pic Du Midi, France (LSCE)

1. Release Date: 15 August 2003
2. Data additions
   Discrete surface measurements:

   • BGU Begur, Spain (LSCE)
   • SUM Summit, Greenland (ESRL)

   Continuous surface measurements:

   • FRDRBC Fraserdale, Ontario, Canada (Restricted Baseline Condition, MSC)
   • PALCBC Pallas, Finland (Continental Baseline Condition, FMI)
   • PALMBC Pallas, Finland (Marine Baseline Condition, FMI)

   Continuous measurements from a tall tower:

   • WKT Moody, Texas, United States (ESRL)

1. Release Date: 15 August 2002
2. Data additions
   Discrete surface measurements:

   • BGU Begur, Spain (LSCE)
   • SUM Summit, Greenland (ESRL)
3. Change to the release policy
   Participants of the Cooperative Atmospheric Data Integration Project-CO2 agreed to change the current release policy of the GLOBALVIEW-CO2 data product. A single ?complete? version of the data product will now be freely available to everyone [Toru et al., In Preparation].
4. Site Code Change
   The Kosan, Republic of Korea (KSN) site was renamed Gosan, Republic of Korea (GSN).
5. Summary of Sample Collection Times
   This update includes a summary of sample collection times for discrete measurement records (See SUMMARY - SAMPLE COLLECTION TIMES for details).
6. Data additions
   Discrete surface measurements:

   • ALT Alert, Nunavut, Canada (SIO)
   • CBA Cold Bay, Alaska, United States (SIO)
   • CGO Cape Grim, Tasmania, Australia (SIO)
   • KUM Cape Kumukahi, Hawaii, United States (SIO)
   • LJO La Jolla, California, United States (SIO)
   • PSA Palmer Station, Antarctica, United States (SIO)
   • SBL Sable Island, Nova Scotia, Canada (MSC)
   • SMO American Samoa (SIO)
   • SPO South Pole, Antarctica, United States (SIO)
   • TRM Tromelin Island, France (LSCE)

   Discrete measurements from aircraft:

   • WPO Western Pacific Ocean (MRI)
   • RTA Rarotonga, Rarotonga (ESRL)

1. Release Date: 15 August 2001
2. Data additions
   Continuous surface measurements:

   • COI Cape Ochi-ishi, Japan (NIES)
   • HAT Hateruma Island, Japan (NIES)
   • CPT Cape Point, South Africa (SAWS)

   Discrete surface measurements:

   • KZD Sary Taukum, Kazakstan (ESRL)
   • KZM Plateau Assy, Kazakstan (ESRL)
   • TDF Tierra Del Fuego, Argentina (ESRL)

   Discrete measurements from aircraft:

   • HAA Molokai Island, Hawaii, United States (ESRL)
   • PFA Poker Flat, Alaska, United States (ESRL)

1. Release Date: 15 August 2000
2. Modifications to the Data Extension procedure

   The data extension approach used to prepare the GLOBALVIEW product extends measurement time series by filling periods of missing data for a specific site with values based on knowledge gained from measurements at the site itself and from measurements from marine boundary layer (MBL) sites at comparable latitude. This "latitude reference" method has been improved upon over that described in Masarie and Tans, [1995] (hereafter MT95).

   In GLOBALVIEW-CO2, 1999 we improved the technique used to construct reference MBL time series to reduce their sensitivity to changes in the distribution of sites and to minimize discontinuities in these reference curves resulting from periods of sporadic or interrupted sampling with existing MBL records. In GLOBALVIEW-CO2, 2000, we have made a minor change to the construction of the difference climatology to minimize discontinuities between smooth values and interpolated and extrapolated values.

   Summary of the difference climatology described by MT95

   Data were prepared by fitting a function, f(t) [Equation 1 in MT95 consisting of harmonics and a polynomial] to each measurement record. The residuals from this fit are smoothed to capture interannual variations in the seasonal cycle. These variations are added to f(t) to produce a smooth curve, SSTA(t) [Equation 2, MT95], which is our best fit representation of the data The reference MBL time series, MBLSTA(t), is constructed for the latitude of each sampling location using the methods described by MT95 and modified according to A.1999.2 (see below). The difference distribution, ΔSTA,REF(t)=SSTA(t)-MBLSTA(t), highlights features that distinguish the individual record from the reference. A difference climatology was then described by fitting a function, δSTA,REF(t) [Equation 9, MT95] to ΔSTA,REF(t). This difference climatology describes the average difference between the smooth curve, SSTA(t), and the reference MBLSTA(t). To account for interannual variability in the difference distribution, ΔSTA,REF(t), we digitally filter the residuals, ΔSTA,REF(t)-δSTA,REF(t) using a low-pass filter with FWHM of 40 days. The smoothed residuals are then combined with the difference climatology according to Equation 10, MT95 to produce a smoothed difference climatology, SSTA,REF(t).

   Data extension relies on the assumption that the difference climatology described by δSTA,REF(t) is valid for periods when there are no actual measurements. Limitations of the assumption are discussed in Sections 4 and 5 of MT95. Finally, the extended record is constructed using SSTA(t) where measurements exist and by combining MBLSTA(t) and the difference climatology where measurements do not exist. Specifically, interpolated values are constructed by combining the MBL reference, MBLSTA(t), with the smoothed difference climatology, SSTA,REF(t). Extrapolated values are constructed by combining the MBL reference, MBLSTA(t), with the difference climatology, δSTA,REF(t).

   Modifications to the use of the difference climatology

   The difference climatology, δSTA,REF(t), is computed from the difference distribution, ΔSTA,REF(t), as described by MT95 and summarized above. The method described by MT95 to construct extrapolated values, however, had a tendency to introduce discontinuities at the transition between smoothed values, S(t), and extrapolated values (Figure 1b). These discontinuities arise when extrapolated values based on average behavior join values derived from observations, which do not reflect average behavior. The largest discontinuities occur when the seasonal pattern of actual data at a transition deviates significantly from the long-term average seasonal cycle (Figure 1a). To minimize discontinuities at the boundary between extrapolated values and smooth values, we smooth the transitions between the difference climatology, δSTA,REF(t), and the difference distribution, ΔSTA,REF(t). This is accomplished by defining a relaxation period (RELAX=8 weeks) whereby we force the difference climatology to "relax" linearly from its value RELAX weeks away to the first value from the difference distribution following a gap or to the last value from a difference distribution before a gap in the actual data begins.

   Extrapolated values are required to "fill" external gaps in the observations that occur when a data record begins or ends within the data extension synchronization period. For example, since the ESRL [lab# 01] flask sampling effort on container ships in the Pacific Ocean (POC) began in 1987 and the synchronization period for GLOBALVIEW-CO2, 2000 is 1979 through 1999, there exists an external gap at the beginning of the POC extended record. The transition between δPOCN30,REF(ti) and ΔPOCN30,REF(ti) where ti is the weekly time step corresponding to the first actual observation in the POCN30 record is smoothed using the following strategy. Values from δPOCN30,REF(t) are used (as in MT95) for time steps before ti-RELAX. Between the time steps ti-RELAX and ti, we use values from linear interpolation between δPOCN30,REF(ti-RELAX) and ΔPOCN30,REF(ti). Figure 1c illustrates this technique.

   The method to construct interpolated values (described by MT95) did not introduce discontinuities at transitions. By using the smoothed difference climatology, SSTA,REF(t), continuity was imposed at the transition between SSTA,REF(t) and ΔSTA,REF(t) by the curve fitting methods as described by Thoning et al. [1989]. A more defensible approach for the extension of data records is to use only the difference climatology, δSTA,REF(t), which describes the average difference between all actual observations and the MBL reference. Thus, we now apply the smoothing strategy described above to the construction of interpolated values.

   
   

   Figure 1. (a) Portion of the POCN30 difference climatology (squares) derived using the method described by MT95. (b) POCN30 extended record derived from (a). (c) Portion of POCN30 difference climatology derived using the modified method described in the text. (d) POCN30 extended record derived from (b) showing minimal discontinuity at the transition between extrapolated values (squares) and smoothed values (circles).

   Interpolated values are required to "fill" internal gaps in a data record that occur when an interruption in the observations exceeds 8 weeks (as described in MT95). There are two cases to consider, which again, can be best illustrated using the ESRL POCN30 record. First, there are internal gaps in the POCN30 record that exceed 8 weeks but are less than 2 * RELAX weeks (e.g., 1987). In these cases, we linear interpolate between ΔPOCN30,REF(ti) and ΔPOCN30,REF(tii) where ti corresponds to the weekly time step before the gap in the record begins and tii is the weekly time step when the observations restart. Second, there are internal gaps exceeding 2 * RELAX weeks in length (e.g., 1988). In these cases, we linear interpolate between ΔPOCN30,REF(ti) and δPOCN30,REF(ti+RELAX) and δPOCN30,REF(tii-RELAX) and ΔPOCN30,REF(tii). Between δPOCN30,REF(ti+RELAX) and δPOCN30,REF(tii-RELAX), we use the values δPOCN30,REF(ti+RELAX : tii-RELAX). Figure 1c illustrates each of these cases.

   Discontinuities in extended records caused by jumps at transitions between the difference climatology and the difference distribution are artifacts of the data extension method and do not reflect instantaneous sources and sinks of carbon. It is reasonable then to minimize these discontinuities since models "inverting" GLOBALVIEW-CO2 will be required to interpret these jumps. To smooth these discontinuities, we assume that the transition to the actual difference distribution will be gradual and not instantaneous. Because we cannot justify using one model over another, we have chosen linear interpolation.

   By smoothing the transition between the difference climatology and the difference distribution at external and internal gaps, we have minimized discontinuities caused by the non-average behavior of actual observations (Figure 1d). This improvement is apparent in the extended records included in this data product. These modifications, however, still cannot overcome certain discontinuities in the extended records caused by limitations in the observational network itself (see Release Notes for GLOBALVIEW-CO2, 1999).

1. Release Date: 15 August 1999
2. Modifications to the Data Extension procedure The data extension approach used to prepare the GLOBALVIEW product extends measurement time series by filling periods of missing data for a specific site with values based on knowledge gained from measurements at the site itself and from measurements from marine boundary layer (MBL) sites at comparable latitude. This "latitude reference" method has been improved upon over that described in Masarie and Tans, [1995] (hereafter MT95). Specifically, the technique used to construct reference MBL time series has been modified to reduce their sensitivity to changes in the distribution of sites and to minimize discontinuities in these reference curves resulting from periods of sporadic or interrupted sampling within existing MBL records.

   Summary of latitude reference method described by MT95

   Data were prepared by fitting a function, f(t) [Equation 1 in MT95 consisting of harmonics and a polynomial] to each measurement record. The residuals from this fit are smoothed to capture interannual variations in the seasonal cycle. These variations are added to f(t) to produce a smooth curve, S(t) [Equation 2, MT95], which is our best fit representation of the data The residuals are also smoothed to capture variations in the long-term trend only and these are added to the polynomial terms of f(t) to give the deseasonalized long-term trend, T(t) [Equation 3, MT95]. A detrended seasonal cycle is computed as S(t)-T(t), and the average seasonal cycle, H(t), is represented by the harmonic components of f(t) [see Equation 1].

   A single measurement record extended using the latitude reference method (as described in Section 4.2, MT95) utilized the record itself as well as information gleaned from additional measurements available from the observational network. Fundamental to this approach is the difference climatology that characterizes the uniqueness of a site record relative to a MBL reference calculated at the site's latitude. Differences between the smooth curve, SSTA(t), and the MBL reference, MBLSTA(t) are calculated (Equation 8, MT95). This distribution, ΔSTA,REF(t), highlights features in the site record that are not represented by the MBL reference. A curve [Equation 9, MT95] is then fitted to this distribution to characterize the average offset and average seasonal cycle of ΔSTA,REF(t) and represents the difference climatology for the site. We then assume the difference climatology is valid for periods where there are no measurements; limitations of this assumption are discussed in Sections 4 and 5 of MT95. Finally, the extended record is constructed using SSTA(t) where measurements exist and by combining MBLSTA(t) and the difference climatology where measurements do not exist.

   Modifications to the derivation of the MBL reference

   Reference MBL time series continue to be constructed using observations from active MBL sampling sites during the synchronization period (fixed span of time into which measurement records will be extended, e.g., 1979-1998). The method described in MT95, however, had a tendency to introduce discontinuities into the derived reference time series that were due to changes in the distribution of MBL measurements. For example, during construction of reference MBL time series, each MBL measurement record contributed its smooth values, S(t), everywhere measurements existed; no values from the site were contributed if an interruption in the observations exceeded 8 weeks. Further, the smooth curve was not defined before sampling at a location begins or after it ends. Thus, during construction of reference MBL time series, values from the smooth curves from MBL sites would abruptly appear, disappear, and reappear depending on the continuity and distribution of actual MBL measurements. This was particularly a problem in the equatorial and southern tropical regions where sampling is already sparse. In these regions, site additions, deletions, or gaps in the few existing MBL records had considerable impact on the reference MBL time series and added noise to existing variability due to changes in carbon exchange and atmospheric circulation.

   Modifications to the latitude reference procedure minimize the affects of a changing observational network on the derived reference MBL time series. This is accomplished in two ways. First, instead of using the smooth curve, S(t), from MBL measurement records as described by MT95, we use the long-term trend, T(t), the detrended seasonal cycle, S(t)-T(t), and the average seasonal cycle, H(t) derived from each MBL measurement record. Because the trend curve is, by definition, less sensitive than the smooth curve to short-term interruptions, we utilize interpolated values from the trend curve during problematic sampling periods. The seasonal component of the measurement record is represented by the detrended seasonal cycle where there are measurements and by the average seasonal cycle where there are short-term interruptions in the record. By using average seasonal cycle patterns, interruptions or periods of infrequent sampling in a MBL record where the seasonal cycle may be poorly defined or entirely missing have minimal impact on the derived MBL reference. Second, instead of using weights (which depend on sampling density and measurement variability) with annual resolution as described, we now use a single weight at each site that is determined using the entire measurement record. This eliminates variability in the MBL reference that arises when assigned weights may change abruptly from one year to the next, again, due to changes in the observational network. Considered together, these modifications to the latitude reference procedure ensure that once measurements at a MBL location commence, they contribute uninterrupted to the construction of the reference MBL time series until sampling is discontinued. This point is clarified in the description that follows.

   First, weekly latitudinal distributions (mixing ratio versus latitude) of values extracted from the long-term trends, T(t), at MBL sites are compiled. A weighted curve as described by Tans et al. [1989] is then fitted to each weekly distribution to approximate the meridional distribution of trends. At each time step, values are extracted from the curve at intervals of 0.05 sine of latitude from 90°S to 90°N producing a matrix (T(t,l)) of trends as a function of time and latitude.

   Second, using the same MBL sites, weekly latitudinal distributions of values extracted from the detrended seasonal cycle where measurements exist and from the average seasonal cycle where there are interruptions in the data record are compiled. A weighted curve is then fitted to each weekly distribution to approximate the meridional distribution of seasonal cycle patterns. At each time step, values are extracted from the curve at intervals of 0.05 sine latitude from 90°S to 90°N producing a matrix [S(t,l)-T(t,l); H(t,l)] of detrended seasonal cycle patterns as a function of time and latitude.

   Third, we construct the MBL matrix REF(t,l) = T(t,l) + {S(t,l)-T(t,l); H(t,l)}. This matrix contains derived model fits to the latitude distribution of long-term trends and detrended average or actual seasonal cycles from all MBL sites at each time step and latitude interval.

   Finally, a reference MBL time series can be extracted from the MBL matrix at any latitude using linear interpolation. For example, as described in Section 4.2 of MT95, a reference MBL time series is constructed at Cape Grim (CGO), REFCGO(t), by extracting, at each time step, a mixing ratio from the MBL matrix at the sine (latitude) of CGO. The MBL reference at CGO is most influenced by CGO itself (because it is designated as a MBL site) during the period of measurements, by MBL sites nearby in latitude to CGO, and to a lesser extent by all other MBL values used in the curve fits. The MBL reference at CGO is included in the CGO extension file. Reference MBL time series are included in data extension files for all MBL and non-MBL sampling locations. The reference MBL matrix is also included in this GLOBALVIEW product (See REFERENCE MARINE BOUNDARY LAYER MATRIX for details).

   The reference MBL time series constructed using this technique are considerably smoother and more stable than those generated using the original technique. This new technique, however, still cannot overcome certain limitations in the observational network itself. For example, in late-1990, the NOAA sampling at AMS (38°S) was terminated. NOAA sample collection began at CRZ (46°S) in early-1991 as a replacement to the AMS location. The 4-month gap in MBL measurements in this latitude region, however, results in a discontinuous period of low CO2 values in the reference MBL time series at CGO (41°S) that is bracketed in latitude by CRZ to the south and AMS to the north. This discontinuity in the MBL reference at the latitude of CGO is substantially attenuated in GLOBALVIEW-CO2 where continuous measurements at AMS [1980-1997] contributed by the LSCE laboratory in France provide the continuity that was lacking in the NOAA sampling network.