Addressing Interdependency in a Multimodel Ensemble by Interpolation of Model Properties

The diverse set of Earth system models used to conduct the CMIP5 ensemble can partly sample the uncertainties in future climate projections. However, combining those projections is complicated by the fact that models developed by different groups share ideas and code and therefore biases. The author...

Full description

Saved in:
Bibliographic Details
Published in:Journal of climate 2015-07, Vol.28 (13), p.5150-5170
Main Authors: Sanderson, Benjamin M., Knutti, Reto, Caldwell, Peter
Format: Article
Language:English
Subjects:
Acceptability
Archives
Bias
Climate
Climate change
Climate models
Climate sensitivity
Climatology
Ensembles
Future climates
GEOSCIENCES
Global climate models
Global warming
Interpolation
Meteorology
Methods
Model evaluation/performance
Modeling
Modelling
Multivariate analysis
Sampling
Simulation
Simulations
Statistical techniques
Studies
Surface temperature
Two dimensional modeling
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The diverse set of Earth system models used to conduct the CMIP5 ensemble can partly sample the uncertainties in future climate projections. However, combining those projections is complicated by the fact that models developed by different groups share ideas and code and therefore biases. The authors propose a method for combining model results into single or multivariate distributions that are more robust to the inclusion of models with a large degree of interdependency. This study uses a multivariate metric of present-day climatology to assess both model performance and similarity in two recent model intercomparisons, CMIP3 and CMIP5. Model characteristics can be interpolated and then resampled in a space defined by independent climate properties. A form of weighting can be applied by sampling more densely in the region of the space close to the projected observations, thus taking into account both model performance and interdependence. The choice of the sampling distribution’s parameters is a subjective decision that should reflect the researcher’s prior assumptions as to the acceptability of different model errors.
ISSN:0894-8755
1520-0442
DOI:10.1175/jcli-d-14-00361.1