Support Vector Machine Classification of Seismic Events in the Tianshan Orogenic Belt

Abstract

Discriminating between various types of seismic events is of significant scientific and societal importance. We use a machine learning method employing support vector machine (SVM) to classify tectonic earthquakes (TEs), quarry blasts (QBs), and induced earthquakes (IEs) among 30,181 1.5 < M L <2.9 seismic events that occurred in the Tianshan orogenic belt in China from 2009 to 2017. SVM classifiers are derived based on discriminant features of a training data set consisting of 1,400 TEs selected from the aftershock sequences of 18 M L ≥ 5.0 earthquakes, 2,881 QBs from repeating events occurring in those areas with a percentage of event daytime occurrence greater than 0.9, and 987 IEs from events in the known oil/gas fields and water reservoirs. The discriminant features include spectral amplitudes of observed P and S wave signals in a frequency range of 1-15 Hz normalized by the P spectrum and averaged over the entire seismic network, and an optional feature of the percentage of event daytime occurrence. Statistics analyses indicate that the accuracies of the SVM classifiers are 99.81% for TEs, 99.93% for QBs, and 99.62% for IEs. Our classification indicates that 37.57% of the seismic events are QBs occurring in possible mine areas and appearing mostly as clusters with a percentage of event daytime occurrence greater than 0.9, 50.12% are TEs occurring in various thrust faults in the Tianshan orogenic belt, and 12.31% are IEs or shallow tectonic earthquakes occurring mostly as clusters near oil and gas fields and water reservoirs. We reevaluate b values in the region and obtain relatively uniform values for the classified TEs with most of them below 1.0, as opposed to a large range of values (0.5-2.7) when all the seismic events are used in the analysis. Key Points:• We establish support vector machine (SVM) classifiers to classify tectonic earthquakes, quarry blasts, and induced earthquakes in the Tianshan orogenic belt • Statistics analyses indicated that the accuracies of the SVM classifiers are more than 99% • We classify the events occurring from 2009 to 2017 in the Tianshan orogenic belt and reevaluate the b values of the tectonic earthquakes

Advancing Local Distance Discrimination of Explosions and Earthquakes With Joint P/S and ML‐MC Classification

“…The high‐transportability of our method may also mitigate potential challenges with network‐based discriminations, where location pattern recognition or site‐specific source effects could unintentionally dominate source type determination. For instance, a neural network could learn to associate explosive sources with a specific place if event location is included in end‐to‐end classification process along with other information such as waveforms or spectrograms (e.g., Reynen & Audet, 2017; Tang et al., 2020; Figure S11 in Supporting Information ). Our results suggest that attributes such as P/S ratios and M L ‐M C could be beneficial to include for training future classification models.…”

Nondestructive detection of Pleurotus geesteranus strain degradation based on micro-hyperspectral imaging and machine learning

“…This is achieved by maximizing the margins between different classes and reducing the distance between the hyperplane focuses. To identify a suitable SVM, it is necessary to find a i and b by minimization, and then optimize and express them as follows ( Tang et al., 2020 ):…”

Classifying and Forecasting Seismic Event Characteristics Using Artificial Intelligence

“…On the other hand, machine learning utilizes algorithmic models to analyze data and make predictions [3]. Various machine learning models have been proposed and employed in seismic studies, such as rule-based classifiers, K-means clustering, and Support Vector Machines (SVM) [4]. For instance, rule-based classifiers have been utilized to categorize earthquakes based on their distinct attributes.…”