← Back to Sessions

Automated Detection of Clipping in Broadband Earthquake Records

Session: Strong-Motion Data Processing and Dissemination: State-of-the-Art and Outlook

Type: Oral

Date: 4/23/2021

Presentation Time: 10:30 AM Pacific

Description: 

Waveform clipping, meaning that the amplitude exceeded the instrument’s dynamic range limit, renders records useless for many applications. While clipping is typically easy to identify visually, automated clip detection approaches have not yet received much attention. We propose an automated algorithm for determining if a ground motion record is clipped (i.e., if the recorded motions exceed the dynamic range of the instrument and/or digitizer). We consider multiple algorithms for classifying the records as clipped. These methods include: (1) an algorithm based on the percentage difference in adjacent data points, (2) the standard deviation of the data within a moving window, (3) the histogram of the data values, (4) the second derivative of the data, and (5) the amplitude of the data. To assess the accuracy of these algorithms and to optimize the parameters of each algorithm, we compile records from earthquakes across a range of geographic regions, tectonic environments, and instrument types. We then manually classify each record for the presence of clipping, use the classified records to optimize the different algorithm parameters, and develop a model that combines the parameters of all of the different models with artificial neural networks. We find the histogram approach to be the most accurate algorithm with an accuracy of 94.8%, and the artificial neural network method provides even further accuracy in clipping identification with a validation accuracy reaching 97%.

Presenting Author: James K. Kleckner

Student Presenter: Yes

Authors