T. Ziehn

A global sensitivity study of cyclohexane oxidation under low temperature fuel-rich conditions using HDMR methods

This paper presents a global sensitivity analysis of simulations of low-temperature isothermal cyclohexane oxidation under fuel-rich conditions using the method of high-dimensional model representation (HDMR). The analysis is used to investigate the important features of the oxidation process, as well as possible factors underlying qualitative discrepancies between simulations and experiments. The particular feature of interest is the characteristic of quadratic autocatalysis, which is observed experimentally and leads to the maximum rate of reaction occurring at 50% consumption of the deficient reactant (oxygen), with the fuel consumption exerting only a weak dependence. The kinetic mechanisms tested do not exhibit this characteristic when simulating the experimental conditions. The models also exhibit shorter induction times than those observed in the experiment. The HDMR study demonstrates a sensitivity of these features to the A-factors of key reactions of the cyclohexylperoxy radical (C6H11OO). At the low temperatures studied here, these are peroxy–peroxy radical reactions rather than the isomerisation routes that have been the subject of other investigations at higher temperatures. The low temperature product channels for reactions of the cyclohexylperoxy radical are therefore an important area for future kinetic studies. The effects of wall reactions of peroxy and peroxide species were not found to outweigh the impact of the main A-factors, but including wall losses led to significant higher order interactions between input parameters. This constitutes an interesting and important area for further research.

Investigating spatial differentiation of model parameters in a carbon cycle data assimilation system

Better estimates of the net exchange of CO(2) between the atmosphere and the terrestrial biosphere are urgently needed to improve predictions of future CO(2) levels in the atmosphere. The carbon cycle data assimilation system (CCDAS) offers the capability of inversion, while it is at the same time based on a process model that can be used independent of observational data. CCDAS allows the assimilation of atmospheric CO(2) concentrations into the terrestrial biosphere model BETHY, constraining its process parameters via an adjoint approach. Here, we investigate the effect of spatial differentiation of a universal carbon balance parameter of BETHY on posterior net CO(2) fluxes and their uncertainties. The parameter, beta, determines the characteristics of the slowly decomposing soil carbon pool and represents processes that are difficult to model explicitly. Two cases are studied with an assimilation period of 1979 to 2003. In the base case, there is a separate beta for each plant functional type (PFT). In the regionalization case, beta is differentiated not only by PFT, but also according to each of 11 large continental regions as used by the TransCom project. We find that the choice of spatial differentiation has a profound impact not only on the posterior (optimized) fluxes and their uncertainties, but even more so on the spatial covariance of the uncertainties. Differences are most pronounced in tropical regions, where observations are sparse. While regionalization leads to an improved fit to the observations by about 20% compared to the base case, we notice large spatial variations in the posterior net CO(2) flux on a grid cell level. The results illustrate the need for universal process formulations in global-scale atmospheric CO(2) inversion studies, at least as long as the observational network is too sparse to resolve spatial fluctuations at the regional scale. (Less)The Carbon Cycle Data Assimilation System (CCDAS) allows the current fluxes of CO2 to the atmosphere to be mapped and the evolution of these fluxes into the future to be predicted. In this work we concentrate on the calibration mode of CCDAS where an optimal parameter set is derived from 10 years of atmospheric CO2 concentration observations using an adjoint approach. Global and regional process parameters are considered via a mapping routine. The parameters are then optimised by calculating the mismatch of the observations and prior knowledge of the parameters via a defined cost function. Further, parameter uncertainty estimates, which are obtained during the parameter optimisation step, can be propagated in order to estimate uncertainties of any given output such as of the predicted net CO2 fluxes.

The Use of Global Sensitivity Methods for the Analysis, Evaluation and Improvement of Complex Modelling Systems

Models which involve the coupling of complex chemical and physical processes are being increasingly used within engineering design and decision making. Improvements in available compute power have allowed us to represent such processes with increasing levels of model detail. However, our ability to accurately specify the required high dimensional input data often does not keep pace with the development of model structure. The analysis ofmodel uncertainty must therefore form a key part of the evaluation of such models. Furthermore, sensitivity analysis methods, which determine the parameters contributing most to output uncertainty, can inform the process of model improvement. In this paper we show by example that global sensitivity methods, and in particular methods based onquasi-random sampling high dimensional model representation (QRS-HDMR), are capable of contributing to the model evaluation and improvement process by highlighting key parameters and model subcomponents which drive the output uncertainty of complex models. The method of QRS-HDMR will be described and its application within the fields of combustion and reactive pollution dispersion will be demonstrated. The key points addressed in the work are (1) the potential for complexity reduction using QRS-HDMR methods, (2) global vs. local sensitivity indices for exploring the response to parameters in complex non-linear models, (3) the possibilities for parameter tuning or feasible set reduction via comparison of models with experiment whilst incorporating uncertainty/sensitivity analysis, (4) model improvement through parameter importance ranking coupled with further ab initio modelling studies, (5) robustness to model structure. The generation of a meta-model via QRS-HDMR is shown to be a reasonably efficient global sensitivity method for systems where effects are limited to second-order. Where higher order effects exist, simple transformations of model outputs are shown to improve the accuracy of the meta-modelling process.