An Integrated Approach of GIS and Machine Learning to Assess the Spatio-Temporal Earthquake Vulnerability in South Africa

Abstract

South Africa is experiencing a high frequency of seismic events which have catastrophic effects on individuals and infrastructure. This study aims to utilize the integrated approach of Geographical Information Systems (GIS) and Machine Learning (ML) based algorithms to analyse the large amount of seismicity data to discover the meaningful patterns and assess the geo-vulnerability of earthquakes with good confidence level in South Africa. Several analytical and modelling techniques including Space-Time Pattern Mining, Artificial Neural Networks (ANN) based hot and cold spots were applied. The results of earthquake data from year 1973 to 2021 revealed that in total there were 1,680 earthquake events that occurred with magnitude ranges from mild to moderate (2 ≤ M ≤ 5). Earthquakes with higher magnitude were concentrated notably in the Gauteng (48%) followed by North-West (31%) provinces of the South Africa. Also, 63% of the magnitude and depth of earthquakes are oriented from North-East to South-West direction. A significant increasing trend of earthquake was observed in some areas of Free State (p ≤ 0.1), Limpopo (p ≤ 0.1), Western Cape (p ≤ 0.5) and Gauteng (p ≤ 0.5) provinces. Whereas decreasing trend was found in areas of North-West (p ≤ 0.1) and Mpumalanga (p ≤ 0.5). The ANN based hot spot analysis predicted the cluster of high magnitude earthquakes (hot spots) in North-West province and low magnitude earthquakes (cold spots) in Gauteng province. Although the earthquake vulnerability is low in Gauteng province but these cold spots could be related to the deep mining activities in the region and have the potential to trigger the rock burst phenomena at the mines. The results can help the disaster management authorities for smart decision making, and urban and regional planning of future activities in the region.