Miroslav Sýkora - Academia.edu (original) (raw)

Uploads

Papers by Miroslav Sýkora

Environmental and climate change is a global issue that will and has already impacted the frequen... more Environmental and climate change is a global issue that will and has already impacted the frequency and intensity of natural hazards in many regions throughout the world. Consequently, the actions on structures will be changing and present design practices will need to be adapted to provide for reliable structures with service lifetimes spanning over decades and centuries. A submitted review of present environmental and climate change information is focused on the distinct, but complementary climatic conditions of Central Europe and South Africa. An outline of the basis of structural design accounting for extreme wind and snow loads is presented and recommendations for future risk-based design procedures are discussed. Issues to consider include: a) use of extreme value models; b) implementation of changes in distribution parameters to obtain extremes with long return periods; c) consideration of the rate of the change. The example of a representative structure illustrates the effects of climatic actions on structural reliability. It appears that uncertainties related to the lack of observations hinder drawing strong conclusions concerning appropriate modifications of design procedures due to environmental and climate change. Uncertainties in the prediction of environmental and climate change have a direct bearing on optimal levels of reliability and the subsequent derivation of design values.

Statistical uncertainties, arising from the uncertainty of parameter estimation and model selecti... more Statistical uncertainties, arising from the uncertainty of parameter estimation and model selection, are often neglected in probabilistic assessment of engineering structures. However, few previous studies indicate that this might cause severe underestimation of extreme loads and lead to insufficient structural reliability. This contribution aims to qualitatively and quantitatively investigate the effect of this simplification on extreme values of wind speed that are commonly associated with design values. The probabilistic modelling of basic wind speeds is thoroughly investigated. Moderately high temporal resolution data – daily 10 min averaged maxima from three distinct one hour long measurement sessions – are obtained from the Carpatclim database, covering a 50-year observation period. Data for Budapest are taken into account as a representative example. Block maxima and peak over threshold approaches are applied to extract maxima and to fit associated distributions. Frequentist and Bayesian statistics are used to assess the effect of statistical uncertainties. The parameter estimation uncertainty is quantified by uncertainty intervals. Statistical model uncertainty is explored using different distribution types and taken into account by Bayesian model averaging. The conducted analyses imply that neglecting statistical uncertainties might yield to considerable underestimation of extreme values. Using the currently widespread annual maxima approach, the parameter estimation uncertainty can lead to underestimation of 1000-year return period values by about 20%. The commonly adopted Gumbel model yields 20% larger values with a return period of 1000 years than those based on the generalized extreme value distribution. The latter fits better to data though unambiguous, fully data-driven recommendation on model selection cannot be made. Bayesian posterior predictive distribution is recommended for accounting parameter estimation uncertainty. Moreover, if viable, smaller than one year block size, multiple maxima in a block, or peak over threshold methods are recommended to increase sample size and reduce statistical uncertainties. This leads to 70% reduction in the range of a 90% confidence interval for 1000-year extremes for the selected location.

Observations are inevitably contaminated by measurement uncertainty, which is a predominant sourc... more Observations are inevitably contaminated by measurement uncertainty, which is a predominant source of uncertainty in some cases. In reliability analysis, probabilistic models are typically fitted to measurements without considering this uncertainty. Hence, this paper intends to explore the effect of this simplification on structural reliability and to provide recommendations on its treatment. Statistical and interval-based approaches are used to quantify and propagate measurement uncertainty. They are critically compared by analyzing ground snow measurements that are often affected by large measurement uncertainty. It is propagated through the mechanical model of a generic structure to investigate its effect on reliability. Parametric studies facilitate to analyze the effect of key parameters, such as measurement uncertainty, coefficient of variation of ground snow load, and distribution type. The interval analysis is performed as a hybrid interval-probabilistic analysis. Measurements are represented as intervals and probabilistic model is then fitted to them. Thus, snow parameters and the reliability index are also interval variables; other random variables are described by standard probabilistic distributions. Implementation of the statistical approach is based on the frequentist paradigm where the contamination mechanism is expressed in terms of random variables. This approach allows decoupling measurement uncertainty from a variable of interest. The results indicate that measurement uncertainty may lead to significant (order of magnitude) underestimation of failure probability and should be taken into account in reliability analysis. If more information than interval endpoints is available, a statistical approach is recommended; otherwise the interval representation should be used. Ranges of the key parameters are identified where measurement uncertainty should be considered. For practical applications, the lower interval bound and predictive reliability index are recommended as point estimates using interval and statistical analysis, respectively. The point estimates should be accompanied by uncertainty intervals, which convey valuable information about the credibility of results. Although general recommendations are given, treatment of measurement uncertainty should be handled on a case-specific basis.

This paper studies the effect of statistical uncertainties on fully probabilistic models and exam... more This paper studies the effect of statistical uncertainties on fully probabilistic models and examines the current practice in civil engineering that typically neglects these uncertainties. Frequentist and Bayesian approaches are applied to quantify parameter estimation and model selection uncertainties. Focusing on annual maxima of ground snow loads, representative fractiles for different models and methods are compared. Four distribution types are considered and model uncertainty is accounted by model averaging. The numerical results indicate that for small to medium sample sizes the effect of statistical uncertainties can be substantial and should be accounted for in reliability studies. For three-parameter distributions the models without statistical uncertainty may underestimate the 1000-year return period fractiles by 20% compared to the models which incorporate this uncertainty. This corresponds to about 0.5-times smaller return period for the same fractile. The study indicates that unlike the frequentist statistics, the Bayesian paradigm offers a coherent and rational way to incorporate statistical uncertainties. It is argued that the posterior predictive distribution should be applied in reliability studies, and broader agreement on the distribution type for annual maxima is needed.

This paper studies the effect of commonly neglected statistical uncertainties on struct... more This paper studies the effect of commonly neglected statistical uncertainties on structural reliability. The failure probability of a generic structural member, subjected to snow load is analysed using frequentist and Bayesian techniques to quantify parameter estimation and model selection uncertainties in ground snow load. Various variable to dead load ratios are considered to cover a wide range of real structures. The analysis reveals that statistical uncertainties may have a substantial effect on reliability. By accounting for parameter estimation uncertainty, the failure probability can increase by more than an order of magnitude. Bayesian posterior predictive distribution is recommended to incorporate parameter estimation uncertainty in reliability studies.

Climate change affects not only the natural but also the built environment. The latter comprises ... more Climate change affects not only the natural but also the built environment. The latter comprises large part of societal wealth, and it is a crucial component of developed economies. The focus of this paper is the quantitative assessment of the reliability of load bearing structures in changing climate. Despite its significance, relatively few quantitative studies are available on this topic, and particularly the Carpathian Region has been analyzed insufficiently. Therefore, the aim of this paper is (i) to present two quantitative studies on structures and climate change for the Carpathian Region, and (ii) to give an overview about approaches in civil engineering in relation to climate sciences, thus to trigger and facilitate future cooperation. The first part of the study is about the carbonation-induced corrosion of reinforced concrete structures analyzed considering six climate change scenarios. The results show that the depassivation probability can double from the beginning to the end of the 21st century. For structures executed in 2000, the effects will be subtle within the first half of the century, whilst the considerable changes are expected in the other 50 years. The second part of the study is about ground snow load and its effect on structural failure probability. It focuses on probabilistic models and statistical uncertainties, and draws attention to the significance of uncertainties arising from the insufficient number of observations. These uncertainties are typically neglected in current civil engineering practice, and they are especially important for climate change, for which the historical observations are not representative of the future environment. Bayesian statistical approach is used to handle these uncertainties. The analyses show that statistical uncertainties can have several order of magnitude effect on failure probability, thus their neglect is not justified. Additionally, long-term trends in historical snow observations are analyzed using stationary and non-stationary generalized extreme value distributions. Statistically significant decreasing trends (p < 10%) are found for numerous locations, but they are practically significant only for a few in respect of structural reliability. The results of both studies indicate that climate change can have significant practical consequences on structures and should be
considered by civil engineering profession. Revision of design standards and further research in cooperation with meteorologist seem to be needed to explore and reduce the
impacts of climate change on load bearing structures in the Carpathian Region.

Parameter estimation uncertainty is often neglected in reliability studies, i.e. point estimates ... more Parameter estimation uncertainty is often neglected in reliability studies, i.e. point estimates of distribution parameters are used for representative fractiles, and in probabilistic models. A numerical example examines the effect of this uncertainty on structural reliability using Bayesian statistics. The study reveals that the neglect of parameter estimation uncertainty might lead to an order of magnitude
underestimation of failure probability.

Journal of Performance of Constructed Facilities, Oct 1, 2010

The current structural design provisions are prevalently based on experience and on the assumptio... more The current structural design provisions are prevalently based on experience and on the assumption of stationary meteorological conditions. However, the observations of past decades and advanced climate models show that this assumption is debatable. Therefore, this paper examines the historical long-term trends in ground snow load maxima, and their effect on structural reliability. For this purpose, the Carpathian region is selected, and data from a joint research effort of nine countries of the region are used. Annual maxima snow water equivalents are taken, and univariate generalized extreme value distribution is adopted as a probabilistic model. Stationary and five non-stationary distributions are fitted to the observations utilizing the maximum likelihood method. Statistical and information theory based approaches are used to compare the models and to identify trends. Additionally, reliability analyses are performed on a simple structure to explore the practical significance of the trends. The calculations show decreasing trends in annual maxima for most of the region. Although statistically significant changes are detected at many locations, the practical significance - with respect to structural reliability - is considerable only for a few, and the effect is favourable. The results indicate that contrary to the widespread practice in extreme event modelling, the exclusive use of statistical techniques on the analysed extremes is insufficient to identify practically significant trends. This should be demonstrated using practically relevant examples, e.g. reliability of structures.

Journal of Performance of Constructed Facilities, Oct 1, 2010