Siamese Neural Network to Detecting Spatial Similarities in Earthquake Patterns: A Case Study of Maluku and Sulawesi

Earthquake occurrences in a given region can be regarded as spatial point pattern data, with prior studies indicating a correlation between seismic events and geological features such as volcanoes, faults, and subduction zones, employing point process methodologies. In addition to these geological factors, earthquakes display a periodicity influenced by annual environmental forces, including hydrological, atmospheric, thermal, and tidal changes. This allows for year-to-year pattern analysis. Recent advances in spatial point pattern similarity analysis, particularly the use of Siamese neural networks, have demonstrated superior performance compared to traditional methods such as intensity and K-function analysis, as evidenced by studies in ecology. This research employs a comparable neural network architecture to examine the spatial point pattern similarities of earthquakes in Maluku and Sulawesi from 1993 to 2022. The regions in question are situated at the junction of three tectonic plates, which results in a high frequency of seismic activity. The data pertaining to earthquakes was employed to train a one-shot learning model, which proved effective in differentiating point pattern images. However, it did not clearly reveal any periodic groupings. Nevertheless, some pattern similarities were identified in years with one-, three-, six-, or nine-year gaps.