Establishing site response-based micro-zonation by applying machine learning techniques on ambient noise data: a case study from Northern Potwar Region, Pakistan

Abstract

Several major earthquakes have jolted Pakistan during the last 30 years, destroyed infrastructure and severe damage to the economy. Despite advancement in data sciences and its linkages with other domains, no such study has been conducted in Pakistan, which incorporates the application of machine learning techniques on ambient noise data. This study presents the application of machine learning techniques on the ambient noise data to establish the micro-zones within the urban settlements of northern Potwar region of Pakistan, based on the site response parameters and seismic vulnerability. The ambient noise data from 148 sites in the study area, are acquired, processed and interpreted. Different clustering techniques namely, K-mean, fuzzy c-means and hierarchical clustering have been applied to the interpreted ambient noise data set. Arc GIS maps of the study area have been developed by making use of the interpretation of the ambient noise data and the resulted clustering solutions. The results showed that the fundamental frequency f0 ranges between 0.5 and 15 Hz, the H/V spectral amplification factor ranges between 0.8 and 5.9; the soft sediment thickness ranges from 1.6 to 316 m, whereas, the soil vulnerability index is observed between 0.1 and 63. These site response parameters indicated that the study area is moderate to highly vulnerable to site amplification, and any seismic event can lead to catastrophe within the study area. The clustering techniques also detected three groups from the interpretation of the ambient noise data set by separating the locations according to their vulnerability due to site amplification. The quality of cluster solutions was evaluated using cluster validity indexes and the results of these techniques were compared. These results present the similarities and dissimilarities among different sites and indicate the sites which are geographically distant but have very similar vulnerability characteristics or vice versa. The Arc GIS tool showed the spatial distribution of site response parameters and three zones were established as zone 1, zone 2 and zone 3 with low, intermediate and high values, respectively. The spatial distribution maps showed that the northeastern and northwestern parts of the study area are more vulnerable to site amplification.