Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Development of an ensemble-adjoint optimization approach to derive uncertainties in net carbon fluxes

Ziehn, T ; Scholze, Marko LU and Knorr, W LU (2011) In Geoscientific Model Development 4(4). p.1011-1018
Abstract
Accurate modelling of the carbon cycle strongly depends on the parametrization of its underlying processes. The Carbon Cycle Data Assimilation System (CCDAS) can be used as an estimator algorithm to derive posterior parameter values and uncertainties for the Biosphere Energy Transfer and Hydrology scheme (BETHY). However, the simultaneous optimization of all process parameters can be challenging, due to the complexity and non-linearity of the BETHY model. Therefore, we propose a new concept that uses ensemble runs and the adjoint optimization approach of CCDAS to derive the full probability density function (PDF) for posterior soil carbon parameters and the net carbon flux at the global scale. This method allows us to optimize only those... (More)
Accurate modelling of the carbon cycle strongly depends on the parametrization of its underlying processes. The Carbon Cycle Data Assimilation System (CCDAS) can be used as an estimator algorithm to derive posterior parameter values and uncertainties for the Biosphere Energy Transfer and Hydrology scheme (BETHY). However, the simultaneous optimization of all process parameters can be challenging, due to the complexity and non-linearity of the BETHY model. Therefore, we propose a new concept that uses ensemble runs and the adjoint optimization approach of CCDAS to derive the full probability density function (PDF) for posterior soil carbon parameters and the net carbon flux at the global scale. This method allows us to optimize only those parameters that can be constrained best by atmospheric carbon dioxide (CO2) data. The prior uncertainties of the remaining parameters are included in a consistent way through ensemble runs, but are not constrained by data. The final PDF for the optimized parameters and the net carbon flux are then derived by superimposing the individual PDFs for each ensemble member. We find that the optimization with CCDAS converges much faster, due to the smaller number of processes involved. Faster convergence also gives us much increased confidence that we find the global minimum in the reduced parameter space. (Less)
Please use this url to cite or link to this publication:
author
; and
organization
publishing date
type
Contribution to journal
publication status
published
subject
in
Geoscientific Model Development
volume
4
issue
4
pages
1011 - 1018
publisher
Copernicus GmbH
ISSN
1991-959X
DOI
10.5194/gmd-4-1011-2011
language
English
LU publication?
yes
id
42f3281e-1b87-46a6-9d8e-8906854d07e0 (old id 4448718)
date added to LUP
2016-04-01 09:59:40
date last changed
2024-02-14 13:33:25
@article{42f3281e-1b87-46a6-9d8e-8906854d07e0,
  abstract     = {{Accurate modelling of the carbon cycle strongly depends on the parametrization of its underlying processes. The Carbon Cycle Data Assimilation System (CCDAS) can be used as an estimator algorithm to derive posterior parameter values and uncertainties for the Biosphere Energy Transfer and Hydrology scheme (BETHY). However, the simultaneous optimization of all process parameters can be challenging, due to the complexity and non-linearity of the BETHY model. Therefore, we propose a new concept that uses ensemble runs and the adjoint optimization approach of CCDAS to derive the full probability density function (PDF) for posterior soil carbon parameters and the net carbon flux at the global scale. This method allows us to optimize only those parameters that can be constrained best by atmospheric carbon dioxide (CO2) data. The prior uncertainties of the remaining parameters are included in a consistent way through ensemble runs, but are not constrained by data. The final PDF for the optimized parameters and the net carbon flux are then derived by superimposing the individual PDFs for each ensemble member. We find that the optimization with CCDAS converges much faster, due to the smaller number of processes involved. Faster convergence also gives us much increased confidence that we find the global minimum in the reduced parameter space.}},
  author       = {{Ziehn, T and Scholze, Marko and Knorr, W}},
  issn         = {{1991-959X}},
  language     = {{eng}},
  number       = {{4}},
  pages        = {{1011--1018}},
  publisher    = {{Copernicus GmbH}},
  series       = {{Geoscientific Model Development}},
  title        = {{Development of an ensemble-adjoint optimization approach to derive uncertainties in net carbon fluxes}},
  url          = {{http://dx.doi.org/10.5194/gmd-4-1011-2011}},
  doi          = {{10.5194/gmd-4-1011-2011}},
  volume       = {{4}},
  year         = {{2011}},
}