We performed 3-D seismic tomography using teleseismic arrival time at Southwest Mexico. The Mexican subduction zone results from successive fragmentation events that affected the ancient Farallon plate as various segments of the East Pacific rise approached the paleo-trench off western North America. The complexity in this region is related to two subducting oceanic plates, the Rivera and Cocos plates, that have different ages, compositions, convergence velocities and subduction dip angles. In this study, we compared the 3-D raytracing tomography model with finite frequency tomography model. Final models show the differences in amplitude and pattern between the raytracing and finite frequency. 3D raytracing models produced sharper images of fast velocity structures in the mantle. The deeper slabs are more coherent and show less broadening with depth than using 1D finite frequency kernels. However, although the finite frequency and 3-D ray tracing models show some differences in amplitude and pattern, the overall agreement of the models supports the interpretation of Yang et al. (2009) that slab rollback is occurring in South Western Mexico. One possible different interpretation between the raytracing and finite frequency theory results concerns the deep structure of the Rivera slab. The finite frequency models show that the Rivera slab is clearly observable at a depth of about 300km but fades away at greater depths. However, the 3-D ray tracing model shows a clear fast velocity band down to a depth of 400 km and thus our model does not support a slab tear of the Rivera plate above 400 km depth
The Short Term Averaging/Long Term Averaging (STA/LTA) has been widely used to detect earthquake arrival time. The method simply calculates the ratio of moving average of the waveform amplitude at short and long-time windows. However, although STA/LTA signals can distinguish between real events and noise, we still recognize some lack of accuracies in first P wave arrival pickings. In this study, we attempt to implement one machine learning method popularly, Artificial Neural Network (ANN) that employ input, hidden and output layer similar as human brain works. Note that in this study, we also try to add input parameters with another derivative signal attributes such as Recursive STA/LTA and Carl STA/LTA. The processing step started by collecting event waveforms from the Agency of Meteorology, Climatology and Geophysics. We chose regional events with moment magnitude higher than 3 in the Maluku region Indonesia. Next, we apply all STA/LTA attributes to the input waveforms. We also tested our STA/LTA with synthetic data and additional noise. Further step, we manually picked the arrival of P wave events and used this as the output for ANN. In total, we used 100 events for arrival data training in P wave phases. In the validation process, an accuracy of more than 0.98 can be obtained after 200 iterations. Final outputs showed, that in average, the difference between manual picking and automatic picking from ANN is 0.45 s. We are able to increase the accuracy by band pass filter (0.1 – 3 Hz) all signal and improve the mean into 0.15s difference between manual picking and ANN picks.
The traditional method in determining first arrival time of earthquake dataset is time consuming process due to waveform manual inspection. Additional waveform attributes can help determine events detection. One of the widely used attribute is The Short Term Averaging/Long Term Averaging (STA/LTA) which is simply division moving average of waveform amplitude between short time and longer time. Alternatively, waveform attribute can also be built using kurtosis and skewness. The kurtosis attribute is defined as sharpness of data distribution. By definition, the maximum signal should be at or close to the P wave arrival. The skewness is typically used to show normal distribution of the data. The uniqueness of this method is that it has an ability to determine whether the first P wave arrival has positive of negative number. The skewness calculation is similar to kurtosis but it uses the power of 3 instead of 4. With the objective of generating efficient scheme to pick first time arrival, we explore use artificial neural network and a combination of kurtosis and skewness attributes. We use a collection of magnitude events with moment magnitude larger than 3 located close to Moluccas island, Indonesia. We collected all events information from the Indonesian Agency of Meteorology, Climatology and Geophysics. The process is started with manually pick all P wave arrivals using manual inspection. Next, we trained the artificial neural network with attributes numbers as inputs and arrival time we manually picked as the output. In total we used 100 regional events that has clear P wave phases. Then, we validated the results until reaching 0.99 accuracy. In the last step, we tested the once trained procedures on new waveforms contained events. Current result shows an average of 0.4s different between manual pick and trained picked from machine learning. The accuracy can be improved by applying a band pass 0.1-2 Hz filtering with an average of 0.2s.
Seismic events detection and phase picking play an essential role in earthquake studies. Typical event detection is done visually or manually on recorded seismogram by choosing a series of higher amplitude signals recorded on at least 4 stations. More sophisticated methods have been used in event detection and picking with additional attributes such as Short Time Average over Long Time Average (STA/LTA). This method is based on average number sampled at multiple predefined windows. However, STA/LTA is dependent on the window size which becomes its drawback. In this study, we explore one derivative attribute, popularly known as envelope or instantaneous amplitude. It has been extensively used in seismic reflection and refraction method. In principle, this method uses the Hilbert Transform to calculate complex seismic trace and take the magnitude of complex seismic trace as envelope amplitude that can be used to analyze P wave arrival time. We employed one of the machine learning methods, Artificial Neural Network (ANN). The ANN method works by analyzing various inputs and training them to recognize patterns in P wave arrival signals. We started our study by applying envelope attribute to synthetic data with noise addition. We found that with noisy data the envelope attribute still gives a clear signal for first-time arrival. Next, we trained 300 seismograms of teleseismic events recorded on IRIS-US networks and tested our trained program on 20 seismograms as a blind test. To compare performance between the two methods, we calculated the difference between the results of automatic picking and manual picking. The final calculation shows an average deviation of 0.355 seconds. Twenty-five percent of testing data (5 samples) has a deviation above 0.5 seconds, and 75% of the remainder (15 samples) already had a deviation under 0.5 seconds. The more significant deviations of the P wave picks are likely due to noisy signals in the data set and complex arrival signals. This study shows that the combination of envelope attribute and machine learning method is promising to distinguish teleseismic P wave arrival and automatically pick them.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.