Data integration for the development of a seismic loss prediction model for residential buildings in New Zealand

Samuel Roeslin, Quincy Ma, Jörg Wicker, Liam Wotherspoon: Data integration for the development of a seismic loss prediction model for residential buildings in New Zealand. In: Antonie, Luiza; Christen, Peter; Rahm, Erhard; Zaïane, Osmar (Ed.): DINA 2019 - Data Integration and Applications Workshop, Forthcoming.

Abstract

In 2010-2011, New Zealand experienced the most damaging earthquakes in its history. It led to extensive damage to Christchurch buildings, infrastructure and its surroundings; affecting commercial and residential buildings. The direct economic losses represented 20% of New Zealands GDP in 2011. Owing to New Zealands particular insurance structure, the insurance sector contributed to over 80% of losses for a total of more than NZ$31 billion. Amongst this, over NZ$11 billion of the losses arose from residential building claims and were covered either partially or entirely from the NZ government backed EQC cover insurance scheme. In the process of resolving the claims, the Earthquake Commission (EQC) collected detailed financial loss data, post-event observations and building characteristics for each of the approximately 434,000 claims lodged following the Canterbury Earthquake sequence (CES). Added to this, the active NZ earthquake engineering community treated the
event as a large scale outdoor experiment and collected extensive data on the ground shaking levels, soil conditions, and liquefaction occurrence throughout wider Christchurch. This paper discusses the necessary data preparation process preceding the development of a machine learning seismic loss model. The process draws heavily upon using Geographic Information System (GIS) techniques to aggregate relevant information from multiple databases interpolating data between categories and converting data between continuous and categorical forms. Subsequently, the database is processed, and a residential seismic loss prediction model is developed using machine learning. The aim is to develop a grey-box model enabling human interpretability of the decision steps.

    BibTeX (Download)

    @inproceedings{Roeslin2019,
    title = {Data integration for the development of a seismic loss prediction model for residential buildings in New Zealand},
    author = {Samuel Roeslin and Quincy Ma and Jörg Wicker and Liam Wotherspoon},
    editor = {Luiza Antonie and Peter Christen and Erhard Rahm and Osmar Zaïane},
    year  = {2019},
    date = {2019-09-20},
    booktitle = {DINA 2019 - Data Integration and Applications Workshop},
    abstract = {In 2010-2011, New Zealand experienced the most damaging earthquakes in its history. It led to extensive damage to Christchurch buildings, infrastructure and its surroundings; affecting commercial and residential buildings. The direct economic losses represented 20% of New Zealands GDP in 2011. Owing to New Zealands particular insurance structure, the insurance sector contributed to over 80% of losses for a total of more than NZ$31 billion. Amongst this, over NZ$11 billion of the losses arose from residential building claims and were covered either partially or entirely from the NZ government backed EQC cover insurance scheme. In the process of resolving the claims, the Earthquake Commission (EQC) collected detailed financial loss data, post-event observations and building characteristics for each of the approximately 434,000 claims lodged following the Canterbury Earthquake sequence (CES). Added to this, the active NZ earthquake engineering community treated the
    event as a large scale outdoor experiment and collected extensive data on the ground shaking levels, soil conditions, and liquefaction occurrence throughout wider Christchurch. This paper discusses the necessary data preparation process preceding the development of a machine learning seismic loss model. The process draws heavily upon using Geographic Information System (GIS) techniques to aggregate relevant information from multiple databases interpolating data between categories and converting data between continuous and categorical forms. Subsequently, the database is processed, and a residential seismic loss prediction model is developed using machine learning. The aim is to develop a grey-box model enabling human interpretability of the decision steps.},
    keywords = {computational sustainability, earthquakes},
    pubstate = {forthcoming},
    tppubtype = {inproceedings}
    }