Dissertation Abstract
Quaternary climate changes : A global sensitivity analysis-based emulation approach of climate model response to forcing
Bounceur, Nabila 2015
Earth and Life institute - Georges Lemaître Centre for Earth and Climate Research, Université catholique de Louvain (Belgium), 360 pp.
Since the 19th century, many studies are being based both on modeling and on natural archives to explain causes of Quaternary climate changes. From a modeling perspective, sensitivity analysis is necessary to gain understanding on the effect of external and internal factors relevant to the changes both at the global and regional scales. In order to systematically and fully explore the space of these factors and identify their separate and combined effects, a global sensitivity analysis is considered for the first time in this domain.
We have provided the analysis of the LOVECLIM input-output climate model responses to astronomical forcing, pCO2 and ice sheet volume. We gave a special care to the presentation and spatial visualization of the sensitivity measures of the different factors with an estimation of their uncertainty, as well as the determination of the dominant factors. Moreover, we defined for the first time the notion of fingerprints which are characteristic modes extracted from the variation of the sensitivity measures. The phasing between the factors and the climate changes is suggested by the analysis of these fingerprints.
Challenges of the global sensitivity analysis using the expensive simulators lead us to combine several techniques: (i) space filling design to sample computer experiments for which we proposed an approach and a new algorithm for sampling in a non hypercube volume defined by the factor, (ii) multivariate analysis to reduce the dimension of the climate model outputs as they are redundant and highly correlated and, (iii) emulation, to approximate the expensive climate simulator. Emulation is necessary to overcome the computational burden caused by the multi-integrals needed for the calculations of the sensitivity measures.
The combination of the three methods led to an efficient way to analyze the sensitivity of the complex climate model to the systematic variation in the factors. Compared to standard approaches based on a small number of simulations for well-defined past epochs, we showed that the methodology is robust regarding modeling assumptions, allows us to identify more systematically regions susceptible of experiencing nonlinear climate changes in response to the smooth astronomical forcing and internal factors, and examine the response phase of climate changes to forcing, without having to rely on transient experiment.
We have provided the analysis of the LOVECLIM input-output climate model responses to astronomical forcing, pCO2 and ice sheet volume. We gave a special care to the presentation and spatial visualization of the sensitivity measures of the different factors with an estimation of their uncertainty, as well as the determination of the dominant factors. Moreover, we defined for the first time the notion of fingerprints which are characteristic modes extracted from the variation of the sensitivity measures. The phasing between the factors and the climate changes is suggested by the analysis of these fingerprints.
Challenges of the global sensitivity analysis using the expensive simulators lead us to combine several techniques: (i) space filling design to sample computer experiments for which we proposed an approach and a new algorithm for sampling in a non hypercube volume defined by the factor, (ii) multivariate analysis to reduce the dimension of the climate model outputs as they are redundant and highly correlated and, (iii) emulation, to approximate the expensive climate simulator. Emulation is necessary to overcome the computational burden caused by the multi-integrals needed for the calculations of the sensitivity measures.
The combination of the three methods led to an efficient way to analyze the sensitivity of the complex climate model to the systematic variation in the factors. Compared to standard approaches based on a small number of simulations for well-defined past epochs, we showed that the methodology is robust regarding modeling assumptions, allows us to identify more systematically regions susceptible of experiencing nonlinear climate changes in response to the smooth astronomical forcing and internal factors, and examine the response phase of climate changes to forcing, without having to rely on transient experiment.