Subsequently, the time resolution has been increased in order to make use of
individual flask pair values or hourly data time series.
This required an iterative solution of the inverse problem.
Moreover, the a-priori constraints have been formulated as a linear ``flux model'',
to allow independent specification of different time scales (non-seasonal, seasonal, long-term).
This algorithm (v2.0) is described in
[Rödenbeck, Tech. Rep., 2005].
Version v4.1 marks the beginning of a phase of further development,
with a number of changes especially in versions v4.1 and v4.3
(addition of runs starting earlier, e.g., in 1957
[Rödenbeck et al., Phil.Trans.R.Soc.B (2018b)];
model-free priors;
n-sigma outlier selection, later replacing the "variability selection";
addition of runs incorporating a regression of NEE anomalies against climate anomalies).
From 2020, the year of release has been used as version counter,
reflecting that all CarboScope products of that year
(currently atmospheric CO2 inversion, pCO2-based ocean fluxes, and APO inversion)
have mutually consistent set-ups.
| v2023
|
- Period extented to end of 2022
- Slight change in sets of sites
(according to data availability)
- Name of "NEE-T" inversion changed into "NBE-T"
to reflect that disturbances have always been included into the variability
(and thus also in the temperature sensitivities).
At the same time, the Run-ID of the "NBE-T" inversion starts with nbet (rather than s)
to make it clearer that it does not only use a different station set but
also differs substantially from the standard inversions in its degrees of freedom.
- In the "NBE-T" inversion, re-addition of a relaxation term with a decay time constant of 2.5 years
(such a relaxation term was present in sEXTocNEET_v4.3 but had been dropped again afterwards).
|
| v2022
|
- Period extented to end of 2021
- Slight change in sets of sites
(according to data availability)
|
| v2021
|
- Period extented to end of 2020
- Slight change in sets of sites
(according to data availability)
- Land/ocean fractions now calculated from "Earth2014" (Hirt and Rexer, 2015)
surface cover classification (small differences at land below sea level)
- In runs with explicitly interannual degrees of freedom,
added constant degree of freedom (small change in long-term mean flux)
|
| v2020
|
- Period extented to end of 2019
- Slight change in sets of sites
(according to data availability)
- Since the pCO2-based ocean fluxes (oc_v2020) are now covering all the period
of the atmospheric CO2 inversion,
it is used as ocean prior without change,
(ie., there is no blending with an ocean process model any more).
|
| v4.3
|
- Period extented to end of 2018
- Slight change in set of sites
(according to data availability)
- Slight change in station classifications determining the weights of data in the cost function.
In particular, the long records at BRW, ALT, MBC
are now classified "remote", thereby having more weight than before (when they were "shore").
- The "variability selection" of in-situ records
(previously removing pairs of adjacent data points if the change in mole fraction
was more than 1 ppm/hr)
has now been removed.
It turned out that the "variability selection" removed unduely many summer values
at continental stations due to the large diurnal cycle there.
The "variability selection" had originally been introduced to remove spikes automatically;
however, this is now achieved anyway by the n-sigma selection.
- The fossil fuel emission prior has been changed from CDIAC
to the new gridded product by Jones et al. (in preparation)
compatible with the country and fuel-type totals used in GCP.
- The use of a fixed ocean (up to now available as a sensitivity run, such as s85oc_v4.2s)
has now been chosen for the standard runs.
From 1993, the ocean prior is taken from the pCO2-based flux estimate (oc_v1.7),
as in the sensitivity run before.
Before 1985, the mean sea-air flux and its seasonality are still taken from oc_v1.7,
while interannual variations are taken from simulated CO2 fluxes of the PlankTOM5 biogeochemical model by
[Buitenhuis et al. (2016)].
The period in between uses a linear transition.
This construction takes into account that the coverage of pCO2 data
is sparse before 1993 in many regions
(in particular the tropical Pacific having the largest interannual variability),
while the interannual variations of the PlankTOM5 model reasonably agree
with oc_v1.7 during the data-covered time.
- In the
Standard inversions,
long-term mean and seasonal cycle of NEE
are always taken from run sEXTocNEET_v4.3 based on the extended station set
(as sEXTocNEET_v4.3 is an
NEE-T inversion,
it is not sensitive to the changes in the number of stations in the extended set).
The Standard inversions then have only degrees of freedom to estimate interannual anomalies
around the fixed mean and seasonality.
|
| v4.2
|
- Period extented to end of 2017
- Slight change in set of sites
(according to data availability)
|
| v4.1
|
- Period extented to end of 2016
- Addition of a run starting in 1976
(to avoid a growing number of different runs, some of the shorter runs have been omitted instead)
- Slight change in set of sites
(according to data availability)
- The individual data points have been screened for outlieres by a 2-sigma criterion,
very similar to the outlier flagging done routinely by many data providers.
By doing it within the inversion,
the model insufficiencies to reproduce small-scale circulation are taken into account to some extent.
- New set-up
with changes in prior fluxes (e.g., completely model-free land prior) and prior uncertainty settings:
- The prior flux of all land NEE components is zero.
- The a-priori uncertainties of NEE are proportional to the fraction of vegetated land area in each pixel
taken as the sum of "crop", "dbf", "dnf", "ebf", "enf", "grass", and "shrub" fractions
from SYNMAP (Jung et al., 2006).
- The fossil fuel emission prior is taken from monthly values of CDIAC (Andres et al., 2016).
The years after 2013 have been extrapolated by global scaling factors
based on the ratios in the emission totals from the GCP carbon budget
(e.g., Le Quéré et al., 2017).
See
[Rödenbeck et al., Biogeosciences (2018a)]
for more details.
- TM3 atmospheric transport model driven again by NCEP meteorological fields
(rather than ERA-Interim), as only NCEP is currently available before 1980
|
| v3.8
|
- Period extented to end of 2015
- Slight change in set of sites
(according to data availability)
|
| v3.7
|
- Period extented to end of 2014
- Slight change in set of sites
(according to data availability)
- TM3 atmospheric transport model now driven by ERA-Interim
meteorological fields (rather than NCEP).
|
| v3.6
|
- Period extented to end of 2012
- Slight change in set of sites
(according to data availability)
- Fossil fuel emission prior updated to
EDGAR 4.2 (with FT2010 for 2009 and 2010,
extrapolation based on BP global totals for 2011 and 2012,
and 2%/yr increase afterwards).
Categories used:
1A1a,
1A1c_2G,
1A2,
1A3a,
1A3b,
1A3d,
1A3c_e,
1A4,
1B2a,
2A,
2B_3,
2C,
4C_4D,
6C,
7A
(i.e., all categories provided by EDGAR 4.2 except for 5A_C_D_F_4E = biomass burning).
|
| v3.5
|
- Period extented to end of 2011
- Slight change in set of sites
(according to data availability)
|
| v3.4
|
- Period extented to end of 2010
- Slight change in set of sites
(according to data availability)
- Constant ocean prior fluxes from ocean interior carbon inversion by Gloor et al. (2003)
replaced by:
- Long-term from ocean interior carbon inversion
provided by Sara Michaloff-Fletcher
("//quercus.igpp.ucla.edu/OceanInversion/"):
sum of the ocean uptake flux induced
by the anthropogenic perturbation as compiled
by Mikaloff-Fletcher et al. (2006),
the preindustrial air-sea fluxes from Mikaloff-Fletcher et al. (2007),
and the river fluxes of Jacobson et al. (2007).
- Seasonality from Takahashi et al, (2002).
Only interannual variations are adjustable for ocean fluxes.
(Ocean now identical to that of set-up Ana18 used for the
3D CO2 concentration fields)
|
| v3.3
|
- Period extented to end of 2009
- Slight change in set of sites
(according to data availability)
|
| v3.2
|
- Fossil fuel emissions replaced by EDGAR 4.0
(linearly extrapolated after 2005 using BP global totals).
- Minor change in the values of a-priori uncertainties (now fully independent from period length).
- Minor numerical differences in a few years from using re-downloaded NCEP meteorology files
to drive the TM3 atmospheric transport model.
|
| v3.2b
|
- Period extented to end of 2008
- Slight change in set of sites
(according to data availability)
- Threshold of flask pair difference increased to 0.34 ppm
- Correction of time-of-day at certain sites
|
| v3.1
|
- Period extented to end of 2007
- Slight change in set of sites
(according to data availability)
- Threshold of flask pair difference decreased to 0.2 ppm
- Night-time data from SCH and CMN
|
| v3.0
|
- Further atmospheric sites added
- A-priori sigma for the non-seasonal land and ocean flux components reduced
by a factor 2.83 (corresponding to a dampening μ = 8).
- Length scales of a-priori spatial corrrelations increased by a factor 3 in longitude direction
and by 1.5 in latitude direction.
- Minor: Initial condition flux pulse at first time step only.
|
| v2.0
|
- Use of individual flask pair or hourly CO2 data
- Flux-model formulation of the a-priori constraints,
slightly different choices for a-priori fluxes, sigma intervals, and correlation structure
- Solution by Conjugate Gradients minimization with re-orthogonalization after each step
Full documentation in
[Rödenbeck, Tech. Rep., 2005]
|
| v1.0
| As described in
[Rödenbeck et al., ACP, 2003]
|