Project

Back to overview

Quantifying climate change uncertainty from the CMIP3 ensemble of global coupled climate models

Applicant Knutti Reto
Number 119952
Funding scheme Project funding (Div. I-III)
Research institution Institut für Atmosphäre und Klima ETH Zürich
Institution of higher education ETH Zurich - ETHZ
Main discipline Climatology. Atmospherical Chemistry, Aeronomy
Start/End 01.04.2008 - 31.03.2011
Approved amount 306'450.00
Show all

Keywords (6)

climate change; uncertainty quantification; climate projection; probabilistic methods; climate modeling; climate model uncertainty

Lay Summary (English)

Lead
Lay summary
The Earth’s climate is changing as a response to anthropogenic emissions of fossil fuels, and is very likely to do so over the next decades to centuries. The projected changes from multiple numerical climate models differ significantly, because some feedbacks and processes are poorly understood or cannot be resolved in the models. Yet a quantitative picture of the uncertainty associated with the expected changes on regional and global scales is crucial, in order to quantify impacts, and decide on adaptation and mitigation measures.The objective of this project is to assess the uncertainty in large-scale regional to global climate change projections based on the CMIP3 dataset, a collection of simulations of about twenty global climate models coordinated for the recent IPCC report. While the amount of data produced by climate models and the desire for ever more detailed projections and probabilistic information is growing rapidly, there is a lack of understanding on how to define the skill of a climate model for a forecast that in practice cannot be verified. So far, models are often averaged regardless of their ability to satisfy even basic requirements, and the community is reluctant to define a set of metrics that would seem important for a model to be credible. Also, a consensus of uncertainty in climate change projections on regional to global scales is missing.We propose to develop a framework to evaluate climate models for their suitability in predicting changes by a set of metrics that are important to be simulated correctly from a physical point of view, rather than comparing present day simulations to observations. In addition, by treating one model M as observations, statistical methods can be used to find optimal predictors and to define model weight, such that a weighted mean or a probabilistic method using the rest of the models can best predict M. The issue of common model error and model dependence will be addressed. We also propose novel methods to aggregate results in regional averages using clustering techniques, and a more flexible approach to communicating results using climate change indices.As milestones of the project, we will attempt to answer the following questions:How large are the uncertainties of model projections on different spatial and temporal scales?On what spatial scale can the current models provide robust information?How can we quantify model performance for projections, and use that to define model weight?Is a small set of ‘good’ models more useful than a large number of ‘good’ and ‘bad’ models?How do model dependence, common biases, and overconfidence affect projection uncertainties?How can we attach confidence to climate change indices? The proposed work combines efforts in climate research and statistics. The results will be used in improving a Bayesian hierarchical model to generate probabilistic climate change information for the spatial scales on which the models can actually provide robust information. The potential of such models has been demonstrated, and the availability of the data and the expertise of the people involved make the project feasible.The proposed work will result in an improved understanding of the uncertainty associated with climate change and will thus be of immediate benefit to those users who rely on climate change projections. As the number and diversity of climate models are increasing, the project will contribute to build a framework on how multiple model results can be best aggregated to provide policy relevant information. The results will also be made avail-able on the web for educational purposes. Finally, the interdisciplinary nature of the project and collaboration between climate scientists and statisticians, and the involvement of international collaborations will enhance the networks across disciplines and make the project an intellectual experience for the students involved.
Direct link to Lay Summary Last update: 21.02.2013

Responsible applicant and co-applicants

Employees

Associated projects

Number Title Start Funding scheme
135067 Quantification of climate uncertainty for deep mitigation scenarios 01.04.2011 Project funding (Div. I-III)
144332 Towards impact-relevant climate projections based on the new CMIP5 global climate model simulations 01.03.2013 Project funding (Div. I-III)
129921 Quantifying the sensitivity of the hydrological cycle to increasing temperature 01.04.2010 Project funding (Div. I-III)

Abstract

The Earth’s climate is changing as a response to anthropogenic emissions of fossil fuels, and is very likely to do so over the next decades to centuries. The projected changes from multiple numerical climate models differ significantly, because some feedbacks and processes are poorly understood or cannot be resolved in the models. Yet a quantitative picture of the uncertainty associated with the expected changes on regional and global scales is crucial, in order to quantify impacts, and decide on adaptation and mitigation measures.The objective of this project is to assess the uncertainty in large-scale regional to global climate change projections based on the CMIP3 dataset, a collection of simulations of about twenty global climate models coordinated for the recent IPCC report. While the amount of data produced by climate models and the desire for ever more detailed projections and probabilistic information is growing rapidly, there is a lack of understanding on how to define the skill of a climate model for a forecast that in practice cannot be verified. So far, models are often averaged regardless of their ability to satisfy even basic requirements, and the community is reluctant to define a set of metrics that would seem important for a model to be credible. Also, a consensus of uncertainty in climate change projections on regional to global scales is missing.We propose to develop a framework to evaluate climate models for their suitability in predicting changes by a set of metrics that are important to be simulated correctly from a physical point of view, rather than comparing present day simulations to observations. In addition, by treating one model M as observations, statistical methods can be used to find optimal predictors and to define model weight, such that a weighted mean or a probabilistic method using the rest of the models can best predict M. The issue of common model error and model dependence will be addressed. We also propose novel methods to aggregate results in regional averages using clustering techniques, and a more flexible approach to communicating results using climate change indices.As milestones of the project, we will attempt to answer the following questions:How large are the uncertainties of model projections on different spatial and temporal scales?On what spatial scale can the current models provide robust information?How can we quantify model performance for projections, and use that to define model weight?Is a small set of ‘good’ models more useful than a large number of ‘good’ and ‘bad’ models?How do model dependence, common biases, and overconfidence affect projection uncertainties?How can we attach confidence to climate change indices? The proposed work combines efforts in climate research and statistics. The results will be used in improving a Bayesian hierarchical model to generate probabilistic climate change information for the spatial scales on which the models can actually provide robust information. The potential of such models has been demonstrated, and the availability of the data and the expertise of the people involved make the project feasible.The proposed work will result in an improved understanding of the uncertainty associated with climate change and will thus be of immediate benefit to those users who rely on climate change projections. As the number and diversity of climate models are increasing, the project will contribute to build a framework on how multiple model results can be best aggregated to provide policy relevant information. The results will also be made avail-able on the web for educational purposes. Finally, the interdisciplinary nature of the project and collaboration between climate scientists and statisticians, and the involvement of international collaborations will enhance the networks across disciplines and make the project an intellectual experience for the students involved.
-