Rapid earthquake loss assessment based on machine learning and representative sampling

Abstract

This article proposes a new framework for rapid earthquake loss assessment based on a machine learning damage classification model and a representative sampling algorithm. A random forest classification model predicts a damage probability distribution that, combined with an expert-defined repair cost matrix, enables the calculation of the expected repair costs for each building and, in aggregate, of direct losses in the earthquake-affected area. The proposed building representation does not include explicit information about the earthquake and the soil type. Instead, such information is implicitly contained in the spatial distribution of damage. To capture this distribution, a sampling algorithm, based on K-means clustering, is used to select a minimal number of buildings that represent the area of interest in terms of its seismic risk, independently of future earthquakes. To observe damage states in the representative set after an earthquake, the proposed framework utilizes a local network of trained damage assessors. The model is updated after each damage observation cycle, thus increasing the accuracy of the current loss assessment. The proposed framework is exemplified using the 2010 Kraljevo, Serbia earthquake dataset.

Testing machine learning models for seismic damage prediction at a regional scale using building-damage dataset compiled after the 2015 Gorkha Nepal earthquake

“…number of stories, age, height, and plinth area) associated with ground-motion intensity provided a relevant estimate of the damage grade with a significant level of accuracy. Similar observations have also been reported by Mangalathu et al (2020b) and Stojadinović et al (2022) in their studies using the 2014 South Napa and 2010 Kraljevo, Serbian earthquake building damage dataset, respectively. Moreover, the RFR model trained on a relatively small amount of dataset (5%–20% of the test dataset) resulted in a reasonable estimate of damage; similar observation has been reported by Stojadinović et al (2022).…”

“…The input ground motion used was the USGS ShakeMap intensity of the mainshock, whereas the overall quality of the NBDP database results were based on the cumulative effects of the mainshock and aftershock events, which might have affected the prediction efficacy. In addition, missing building-by-building information in the NBDP database relative to their localization and their associated site condition reduce the damage prediction efficacy of the machine learning model (Mangalathu et al, 2020b; Roeslin et al, 2020; Stojadinović et al, 2022). However, we also observed that by considering the geographic locations of buildings (ward-id in our case) slightly improved the damage prediction efficacy of the RFR model.…”

“…The total number of wards in 11 districts was 949. Considering building-by-building locations and the associated soil conditions, which may vary within a ward, can improve the damage prediction performance of a machine learning model (Mangalathu et al, 2020b; Roeslin et al, 2020; Stojadinović et al, 2022). However, having as objective the large-scale classification of damages (and not the classification and representation per building), and considering the macro-seismic intensities, the attribution of each building within a ward is consistent with our approach.…”

“…Xie et al (2020) summarized the ongoing research on the application of machine learning methods in earthquake engineering; they concluded that the implementation of machine learning in earthquake engineering is still in its early stage and needs further investigations. Limited numbers of studies are available concerning the post-earthquake damage classification using machine learning (Harirchian et al, 2021; Mangalathu et al, 2020b; Roeslin et al, 2020; Stojadinović et al, 2022). In particular, Mangalathu et al (2020b) tested the performance of machine learning techniques in a building damage survey performed after the 2014 South Napa earthquake and demonstrated the effectiveness of such tools in interpreting the patterns of seismic damage observed in buildings.…”

Extraction of Earthquake Damage Information and Mapping of Buildings from Single Post-earthquake Polarimetric Synthetic Aperture Radar Image Based on Polarimetric Decomposition and Texture Features

“…This information will provide a guideline for making emergency plans for response to an earthquake. (4) Traditionally, the main method of obtaining post-earthquake damage information has been through field survey. Although information obtained by this method is highly credible, this method often has a heavy workload and low efficiency over a large area.…”