The University of Southampton
University of Southampton Institutional Repository

Parameter optimisation techniques and the problem of underdetermination in marine biogeochemical models

Parameter optimisation techniques and the problem of underdetermination in marine biogeochemical models
Parameter optimisation techniques and the problem of underdetermination in marine biogeochemical models
Parameter values in marine biogeochemical models can strongly affect model performance, but can be hard to define accurately and precisely. When making quantitative comparisons between models it is helpful to objectively assign optimal parameter values, so it is the best model performance rather than the degree (or lack) of tuning which is assessed. The efficacy of two optimisation techniques, a variational adjoint (VA) and a micro genetic algorithm (?GA), was studied with respect to the calibration of two simple one-dimensional models for Arabian Sea data. Optimisations were randomly initialised a number of times, and given the level of uncertainty in the data, the two techniques performed equally well in terms of reducing model-data misfits.

When ten parameters were optimised for either model, the Arabian Sea data were insufficient to constrain unique solutions; several parameters could be set anywhere across a wide range of values while providing a similarly good fit to the data. The significance of this underdetermination was assessed by evaluating model solutions against unassimilated equatorial Pacific data. When no prior information was used to assist the optimisation, the underdetermined solutions led to highly variable and often poor performance at the equatorial Pacific. Prior information was used to gain a more reliable solution in two ways: (1) by fixing all poorly-constrained parameters to their default prior values, optimising only parameters that were well-constrained by the data; or (2) by placing broad limits on the search to exclude unrealistic parameter values. Using the first approach the optimisation routines could constrain unique solutions and model performance in the equatorial Pacific was very consistent. The precise results were however sensitive to the uncertain a priori values of the fixed parameters. The second approach was less prescriptive, and consequently led to a more variable performance in the equatorial Pacific. It is argued that the first approach is unrealistically precise as it ignores any uncertainty in the unconstrained parameters, while solutions from the second approach may be unnecessarily broad. In conclusion, unconstrained parameter optimisation procedures should be assisted by stating all that is known a priori about the parameters, but no more.
0924-7963
34-43
Ward, Ben A.
9063af30-e344-4626-9470-8db7c1543d05
Friedrichs, Marjorie A.M.
5560fb4c-699a-4251-996c-4fad43ad786d
Anderson, Thomas R.
dfed062f-e747-48d3-b59e-2f5e57a8571d
Oschlies, Andreas
75e18f55-3134-44a2-82ba-71334397727f
Ward, Ben A.
9063af30-e344-4626-9470-8db7c1543d05
Friedrichs, Marjorie A.M.
5560fb4c-699a-4251-996c-4fad43ad786d
Anderson, Thomas R.
dfed062f-e747-48d3-b59e-2f5e57a8571d
Oschlies, Andreas
75e18f55-3134-44a2-82ba-71334397727f

Ward, Ben A., Friedrichs, Marjorie A.M., Anderson, Thomas R. and Oschlies, Andreas (2010) Parameter optimisation techniques and the problem of underdetermination in marine biogeochemical models. Journal of Marine Systems, 81 (1-2), 34-43. (doi:10.1016/j.jmarsys.2009.12.005).

Record type: Article

Abstract

Parameter values in marine biogeochemical models can strongly affect model performance, but can be hard to define accurately and precisely. When making quantitative comparisons between models it is helpful to objectively assign optimal parameter values, so it is the best model performance rather than the degree (or lack) of tuning which is assessed. The efficacy of two optimisation techniques, a variational adjoint (VA) and a micro genetic algorithm (?GA), was studied with respect to the calibration of two simple one-dimensional models for Arabian Sea data. Optimisations were randomly initialised a number of times, and given the level of uncertainty in the data, the two techniques performed equally well in terms of reducing model-data misfits.

When ten parameters were optimised for either model, the Arabian Sea data were insufficient to constrain unique solutions; several parameters could be set anywhere across a wide range of values while providing a similarly good fit to the data. The significance of this underdetermination was assessed by evaluating model solutions against unassimilated equatorial Pacific data. When no prior information was used to assist the optimisation, the underdetermined solutions led to highly variable and often poor performance at the equatorial Pacific. Prior information was used to gain a more reliable solution in two ways: (1) by fixing all poorly-constrained parameters to their default prior values, optimising only parameters that were well-constrained by the data; or (2) by placing broad limits on the search to exclude unrealistic parameter values. Using the first approach the optimisation routines could constrain unique solutions and model performance in the equatorial Pacific was very consistent. The precise results were however sensitive to the uncertain a priori values of the fixed parameters. The second approach was less prescriptive, and consequently led to a more variable performance in the equatorial Pacific. It is argued that the first approach is unrealistically precise as it ignores any uncertainty in the unconstrained parameters, while solutions from the second approach may be unnecessarily broad. In conclusion, unconstrained parameter optimisation procedures should be assisted by stating all that is known a priori about the parameters, but no more.

Full text not available from this repository.

More information

Published date: April 2010
Organisations: Marine Systems Modelling, Ocean and Earth Science, National Oceanography Centre,Southampton

Identifiers

Local EPrints ID: 79807
URI: http://eprints.soton.ac.uk/id/eprint/79807
ISSN: 0924-7963
PURE UUID: 98016104-69ec-4eaa-9db5-1dfed9e14b81

Catalogue record

Date deposited: 19 Mar 2010
Last modified: 17 Jul 2019 00:11

Export record

Altmetrics

Contributors

Author: Ben A. Ward
Author: Marjorie A.M. Friedrichs
Author: Thomas R. Anderson
Author: Andreas Oschlies

University divisions

Download statistics

Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.

View more statistics

Atom RSS 1.0 RSS 2.0

Contact ePrints Soton: eprints@soton.ac.uk

ePrints Soton supports OAI 2.0 with a base URL of http://eprints.soton.ac.uk/cgi/oai2

This repository has been built using EPrints software, developed at the University of Southampton, but available to everyone to use.

We use cookies to ensure that we give you the best experience on our website. If you continue without changing your settings, we will assume that you are happy to receive cookies on the University of Southampton website.

×