Exploration of diverse solutions for the calibration of imperfect climate models

The calibration of Earth system model parameters is subject to data, time, and computational constraints. The high dimensionality of this calibration problem, combined with errors arising from model structural assumptions, makes it impossible to find model versions fully consistent with historical o...

Full description

Saved in:
Bibliographic Details
Published in:Earth system dynamics 2024-08, Vol.15 (4), p.987-1014
Main Authors: Peatier, Saloua, Sanderson, Benjamin M, Terray, Laurent
Format: Article
Language:English
Subjects:
Calibration
Climate
Climate change
Climate models
Climate prediction
Climatic evolution
Climatology
Configuration management
Earth Sciences
Eigenvectors
Error analysis
Future climates
General circulation models
Parameter identification
Physics
Principal components analysis
Sciences of the Universe
Simulation
Tradeoffs
Variables
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The calibration of Earth system model parameters is subject to data, time, and computational constraints. The high dimensionality of this calibration problem, combined with errors arising from model structural assumptions, makes it impossible to find model versions fully consistent with historical observations. Therefore, the potential for multiple plausible configurations presenting different trade-offs between skills in various variables and spatial regions remains usually untested. In this study, we lay out a formalism for making different assumptions about how ensemble variability in a perturbed physics ensemble relates to model error, proposing an empirical but practical solution for finding diverse near-optimal solutions. A meta-model is used to predict the outputs of a climate model reduced through principal component analysis. Then, a subset of input parameter values yielding results similar to a reference simulation is identified. We argue that the effective degrees of freedom in the model performance response to parameter input (the “parametric component”) are, in fact, relatively small, illustrating why manual calibration is often able to find near-optimal solutions. The results explore the potential for comparably performing parameter configurations that have different trade-offs in model errors. These model candidates can inform model development and could potentially lead to significantly different future climate evolution.
ISSN:2190-4987
2190-4979
2190-4987
DOI:10.5194/esd-15-987-2024