Empirical Fragility and Vulnerability Assessment: Not Just a Regression

doi:10.1016/B978-0-12-804071-3.00004-5

Risk Modeling for Hazards and Disasters

2018, Pages 79-103

https://doi.org/10.1016/B978-0-12-804071-3.00004-5 Get rights and content

Abstract

Empirical approaches to asset fragility and vulnerability are extensively used in the insurance industry, where they are commonly reputed the “gold standard” and to have high credibility because they are developed from past event data. However, the reliability of published empirical vulnerability and fragility functions is seen to vary greatly depending on the quality and quantity of the empirical data used and how the data is treated. This chapter presents an overview of some significant observations made by the Authors while working on earthquake and tsunami empirical fragility and vulnerability functions over the last 10 years. Common biases in event damage and loss data sets are presented, the consequences of ignoring the biases are discussed, and possible ways of dealing with them are suggested. The impact of different statistical model fitting approaches is described and illustrated with examples drawn from empirical earthquake and tsunami fragility and vulnerability studies. Throughout, areas for further research are highlighted, and it is observed that some sources of uncertainty remain largely unexplored.

References (0)

Cited by (6)

Fragility curves derivation for masonry buildings damaged after 2009 L'Aquila earthquake accounting for the effect of construction age
2022, International Journal of Disaster Risk Reduction
Citation Excerpt :
The latter fosters a range of statistical models, namely parametric, semiparametric, and nonparametric models, thus requiring the use of goodness of fit tests toward the selection of that with the best match with post-event damage data. This approach was then succeeded by other studies, i.e., by Refs. [25,32] among the others. Therefore, the evaluation of the goodness-of-fit of the fragility functions, already applied in various studies ([33–38]), is performed using hypothesis testing via the Pearson Chi Squared test on ensuing damage probability matrices determined through the adopted statistical models.
The aim of this study is the analysis of seismic fragility of residential masonry buildings, with particular emphasis to the evolution of seismic behaviour over the years. To this purpose, a detailed and comprehensive taxonomy has been established, considering a restricted selection of building's features, avoiding the ineffective fragmentation of the database. The fragility assessment has been performed based on the data collected after the L'Aquila 2009 earthquake and made available by the Italian Department of Civil Protection through the online platform Da.D.O. (Database of Observed Damage). The database has undergone a refinement process to guarantee the completeness of the information, avoiding possible bias in the subsequent fitting procedure for fragility analysis. Thus, PGA (peak ground acceleration) from the ShakeMap has been used for ground motion characterization and 5 + 1 damage levels defined according to the European Macroseismic Scale have been considered for damage classification. Different regression models have been adopted to determine the parameters of lognormal fragility curves, measuring their goodness of fit with the observed damage probability matrices.
Starting from the unconstrained model, further regression constraints (i.e., a common value for logarithmic standard deviation and the respect of the hierarchy of median PGA with the construction age) have been introduced, thus leading to the definition of the constrained model. The benefits in the introduction of further regression constraints are counterposed to the effectiveness of constrained curves to model observational data through the comparison of the goodness of fit between the unconstrained and constrained models.
Empirical fragility curves: The effect of uncertainty in ground motion intensity
2020, Soil Dynamics and Earthquake Engineering
Empirical fragility curves derived from large post-disaster databases with data aggregated at municipality-level, commonly make the assumption that the ground motion intensity level is known and is determined at the centroid of each municipality from a ground motion prediction equation. A flexible Bayesian framework is applied here to the 1980 Irpinia database to explore whether more complex statistical models that account for sources of uncertainty in the intensity can significantly change the shape of the fragility curves. Through this framework the effect of explicitly modelling the uncertainty in the intensity, the spatial correlation of its intra-event component and the uncertainty due to the scatter of the buildings in the municipality are investigated. The analyses showed that the results did not change substantively with increased model complexity or the choice of prior. Nonetheless, informed decisions should be based on the defensible modelling of the significant variability in the data between municipalities.
Tephra cushioning of ballistic impacts: Quantifying building vulnerability through pneumatic cannon experiments and multiple fragility curve fitting approaches
2019, Journal of Volcanology and Geothermal Research
Citation Excerpt :
The advantages and disadvantages of each method are given in Table 4 along with references containing detailed reviews and discussion on each of these statistical models and how they are used in fragility curve fitting. All curves presented in this study share the typical log-normal cumulative density function (CDF) form used for the vast majority of published parametric fragility curves (Rossetto and Ioannou, 2018). One reason the log-normal cumulative distribution is widely used, besides its precedence in previous work, is because it has numerous convenient characteristics for modelling fragility.
Ballistic projectiles are the most frequently lethal volcanic hazard close to the vent. Recent eruptions of Ontake in 2014 and Kusatsu-Shirane in 2018 showed that un-reinforced, timber-framed buildings - those typically considered highly vulnerable to the dangerous penetration of ballistics - provided life-saving shelter from ballistic impact. Modelled kinetic energies of some non-penetrating impacts were an order of magnitude above expected penetration thresholds. It has been hypothesised that a pre-existing layer of tephra on the roofs cushioned impacts. To quantitatively test this, and improve our understanding of how buildings respond to projectile impacts, we used a pneumatic cannon to simulate block impacts to clay tiles and reinforced concrete roof slabs covered with tephra layers 0–20 cm thick. Substantially higher impact energies were resisted when tephra was present with 5 cm of tephra approximately tripling the penetration threshold of both building materials. Fragility curves, which relate ballistic hazard intensity with the probability of building damage, were developed from our experimental data following three curve fitting approaches: generalised link models, cumulative link models and data binning. A key benefit of these approaches is that confidence in these curves can be robustly quantified from the data – the first time that this has been attempted for volcanic fragility curves. This study shows how the extent of building damage can be strongly influenced by the sequence of volcanic hazards and provides an example of proactive risk management through testing of physical mitigation strategies in a laboratory environment.
Agriculture and forestry impact assessment for tephra fall hazard: fragility function development and New Zealand scenario application
2021, Volcanica
Remotely assessing tephra fall building damage and vulnerability: Kelud Volcano, Indonesia
2020, Journal of Applied Volcanology
Seismic fragility for Italian RC buildings based on damage data of the last 50 years
2020, Bulletin of Earthquake Engineering

View full text

Chapter 4 - Empirical Fragility and Vulnerability Assessment: Not Just a Regression

Abstract