Detection of stick-slip events within the Whillans Ice Stream using an artificial neural network

Abstract:

Temporal changes in the periodic stick-slip events on the Whillans Ice Stream (WIS) help to understand the hydrosphere-cryosphere coupling in West Antarctica. Previous studies have shown that the periodic behavior has been ongoing for a number of years but the record of slip events is incomplete. Rayleigh waves from WIS grounding line events exhibit different patterns than events from the interior of the glacier. An algorithm using a backpropagation neural network is proposed to efficiently extract surface waves that are a result of stick slip events. A neural network approach has its advantages of machine learning, simplified mathematics, and eliminates the need for an analyst to correctly pick first arrivals. Training data has been assembled using 107 events occuring during the 2010 austral summer that were previously identified to correspond to stick slip events at the grounding line as well as the interior of the WIS. A 0.1 s moving window of 3 s of each of the preprocessed attributes is input into the neural network for automated surface wave detection. Following surface wave detection a much longer 30 minute sliding window is used to classify surface wave detections as grounding line, interior, or non-stick slip events. Similar to the automatic detection algorithms for body waves, preprocessing using STA/LTA ratio, degree of polarization, variance, and skewness exhibit obvious patterns during the onset of surface waves. The The automated event detection could lead to more cost effective means of data collection in future seismic experiments especially with an increase in array density in cold weather regions.