Meeting held at Hadley Centre, Met Office, Bracknell, UK on 29th January
Office: Peter Cox, Chris Jones, Venkata Jogireedy
LSCE: Pierre Friedlingstein, Philippe Peylin, Andrew Friend
MPI-BGC: Martin Heimann, Wolfgang Knorr, Jens Kattge
FastOpt: Thomas Kaminski
ALTERRA: Ronald Hutjes, Gert-Jan Nabuurs, Mart-Jan Schelhaas
UNITUS: Dario Papale
EFI: Ari Pussinen, Sergey Zudin
CEH: Richard Harding, Chris Huntingford
JRC: Michel Verstraete,Nadine Gobron
EC: Claus Bruning
Valentini (UNITUS), Philippe Ciais (LSCE)
Date of next meeting:
Lisbon, 19th March.
CAMELS is a project, part of the CARBOEUROPE Cluster, motivated by
Article 3.4 of Kyoto Protocol concerning contribution of managed
ecosystems to a countrys carbon balance. In particular, CAMELS will try
to deal with the problem of how to distinguish anthropogenic influence
on terrestrial carbon budgets (i.e. examine mechanisms behind
sources/sinks due to e.g. CO2 fertilisation, Nitrogen deposition,
climate change etc.)
CAMELS will make use of data (both forest inventory and flux data) and
modelling. It will combine the advantages of inversion modelling (large
scale, data constrained) and mechanistic forward modelling
1. best estimates and uncertainties of contemporary
and historical carbon sinks in Europe and elsewhere
2. prototype C-cycle data assimilation system
(CCDAS) exploiting data and models to produce estimates of Kyoto
Work Package 1. Ronald
Hutjes (Alterra). Data Consolidation.
i) validation data, e.g. FLUXNET tower sites some sites have 8
years of data, many just 1 or 2. European tower coverage mainly at
forest sites. Carbodata project has already processed some of this
data.Simple model such as:
NEE = (aR + becT)(1-e-dP) calibrated against data.
ii) N deposition EMEP data over Europe, 1985 onwards. Some data
prior to this, but before 1980 mainly use reconstruction. Landuse
database Hyde database has maps for 1750, 1800, 1850, 1900, 1950,
1970, 1990. Natural vegetation combination of Ramankutty data and
BIOME model output.
iii) forest inventory data used with soil data, emissions and land use
data to give above/below ground carbon stocks. Changes in stocks give
annual fluxes. Some of the maps, especially soil maps, have marked
national boundaries likely due to measurement/recording practices.
Role of this in CAMELS is a validation. Key first deliverable for WP1 is
flux data to validate terrestrial models for WP2.
Work Package 2. Wolfgang
Knorr (MPI-BGC). Model Validation & Uncertainty.
Aims to improve TEMs and come up with optimal parameters and estimates
of their uncertainty. The plan is
i) to identify key uncertainties from literature
ii) to determine optimal params from flux obs (at individual site
iii) to improve process description
iv) to repeat (ii)
v) to validate against inventory data (at regional scale)
vi) to final determination of parameters and uncertainties for use in
Measurement sites (for flux data) mainly in temperate locations, and
mainly forest. Some in Boreal and tropical areas, again mainly forest.
issues of data availability were discussed not all sites have long
timescale data series. Suggested that short timeseries could still be
useful - even just diurnal data may be enough to constrain the models
for WP2. WP3 then does the long timescale processes.
- from WP1 met data to drive model
- soil/veg/water data
- satellite (FAPAR)
- fluxes & inventory data
- to WP 3&4 improved processes
- optimal model params (by PFT)
Work Package 3. Pierre
Friedlingstein (LSCE). Modelling C-20 land carbon balance.
- Especially over Europe, but global too.
- Spin up TEMs to 1900 (as in C4MIP)
- Simulate 1900-2000 land C balance with high/low
param values from WP2. (may use CCMLP protocol, but Martin H question
appropriateness of this).
- Use results to refine parameter values from WP2
- Try to separate impact of mechanisms e.g.
land use change, nitrogen deposition, CO2 fertilisation, climate change.
- WP1 to provide forcing data, land-use, N, CO2
- WP1 to provide calibration & validation data
(fluxes & inventories)
- WP2 to provide high/low estimates of parameters
- WP3 provides long-term constraints on parameters to
compliment WP2s short term constraint
- WP3 to provide initial conditions for WP4 (i.e.
state of contemporary biosphere).
Work Package 4. Peter Cox
(Hadley). Development of DA system.
Combine data & models to estimate sources/sinks @ resolution
required by Kyoto.
- need inverse of TEMs and atmos transport models.
- Use within CCDAS to adjust parameters in models
(use cycle of 25 years of flux data). offline approach.
- online approach more like existing NWP data
assimilation schemes nudge model Carbon values in real time to
The inputs for this work package are
- data from WP1
- TEMs from WP2
- Initial conditions from WP3
Discussion on whether or not site data-derived parameters scale up to
grid scale (i..e. do we still get good agreement with original flux data
at site level if we put final parameters back into WP2?)
Work Package 5. Dario
Papale (UNITUS). Information dissemination.
- construct website (for use by both scientists and
- produce policy-maker summaries of annual
reports suitable for use at, e.g. CoP etc.
Contributions to other WPs:
- gap filling of data for WP1
- neural net to process site data to regional scale.
- Carbodata: http://carbodat.ei.jrc.it
- Tacos: http://gaia.agraria.unitus.it/cpz/index.asp
1. European Forest
Institute (EFI). Ari Pussinen.
Forest inventory based carbon budgeting. Combine inventory data with
Get from inventory (volume) data to carbon balance by model using wood
density, % C content, age class info
Discussion do we need age class information in the models? It was
agreed that this could become a significant factor in models (especially
if they are to cope with forest management), and hence this data would
become an even more valuable validation source.
2. CEH. Richard Harding.
Mostly involved in WP2, will be using MOSES.
See this work as different from previous studies that try to find
optimal parameters at a single site. Here we want parameters to be
consistent across many sites. Will present a challenge as this has not
been done before.
Possible problem with energy budget data from tower sites? They
generally cant close their budget, and it is suspected (by RH) that the
problem is in the latent heat component. Therefore we should exercise
caution when deciding what data to be used.
There was then some discussion at this stage about whether our models
tend to be over-parameterised, and if so whether we need to confine the
procedure to a subset of parameters (i.e. choose some to be held fixed).
3. JRC. Nadine Gobron.
Measure FAPAR from remote sensing.
Aim to produce useful, quantitative, reliable and accurate data from
remote sensing for use in WP4.
2 km data from 1997 to 2001.
Method gives a lower rms error and higher signal to noise ratio (by
factor 3) than antiquated method.
Daily data has cloud gaps, and only partial satellite coverage, so plan
to produce 10-day composite. Monthly could also be done.
Aims to be able to use different sensors (e.g. SeaWiFS, MISR, MERIS)
with results being transparent. This is not the case for NDVI data,
which is sensor-dependent.
Discussion on data format required decided that 10 day data would
be better. Still to be agreed whether JRC to produce high
resolution data for modellers to process to their own grid, or JRC to
produce it already at coarse resolution. Some preference for the former
because different models have different grids.
Nadine finished with some questions.
1. Which associated products, i.e. Rectified channels
and BRF TOA, may be required by WP4?
2. What temporal resolution (10 days or monthly) is
useful for European maps?
3. What spatial resolution is required for specific
sites (both in and outside Europe)?
4. How should the products be provided, e.g. ftp
site, support media and formats, contact point, etc.?
4. FastOpt. Thomas
Calibration step of CCDAS via cost function and variational
approach to find optimal parameters. Iterative minimisation of cost
function by gradient descent method.
While dJ/dx gives gradient information for descent algorithm, can also
get a measure of uncertainty from d2J/dx2 (Hessian matrix) which
measures curvature at xopt.
The technique requires derivative code i.e. code to get the derivative
of the TEM. This can be constructed automatically. FastOpts role is to
support this process.
End of afternoon - Discussion.
i) Wolfgang suggested that some of the role of
inventory data be moved from WP2 to WP3. Pierre and Gert-Jan agreed in
principle (for large scale inventory data WP2 just requires site
data). Claus warned against structural changes to the WPs. ACTION: WK,
PF, G-JN to agree on distribution of inventory work between WP2 and WP3
ii) Wolfgang noted need for error estimates on flux
data. Richard said that this is included under Consolidation so WP1 will
be producing this. Jens Kattge has produced a questionnaire about
requirements such as this.
iii) There is a need to select some sites. It was
suggested that one for each PFT be chosen.
iv) CAMELS website can be set up in Jena, and
be run from Bracknell. Suggested that it should be part free access and
part password protected.
Meeting closed at 4.30pm.