11:00 AM - 1:00 PM
[SCG44-P20] Toward a comprehensive catalogue of tremors in shallow subduction zones that enable to identify tremor missing regions
Keywords:tremor, Nankai trough, envelope correlation method, slow earthquake
The envelope correlation method is a technique often used to automatically detect and locate tectonic tremors. The method has been originally applied to so-called deep tremors that occur at depths of 30-40 km, recorded by seismic networks on land. The validity of the method has been carefully investigated during such applications. More recently, shallow tremors that occur in shallow subduction zone at depths of < 10 km have been closely recorded by ocean-bottom seismometer networks (e.g. DONET). Then, the envelope correlation method has been again extensively applied for such tremors, although the validity of assumptions made in the process has not always been properly investigated. Rather, some of the temporally coherent signals (i.e., possible tremors) have been simply eliminated if they do not follow the assumptions made for deep tremors. Such eliminations of detections may lead to a bias in the resultant tremor catalogues. Also, such catalogues would not allow discussions on tremor missing regions, although the regions provide important information on the shallow subduction zone tectonics.
Here, we carefully investigated how the method performs on records of shallow tremors. The method was applied to eight days of DONET1 data in October 2015, and the results were visually inspected. In the method, first, the coherent signal arrivals are detected (referred to as "detections"), then their source locations are estimated by measuring differential travel times among stations. We first adjusted some parameters for the detection, such as the minimum number of station pairs whose correlation coefficient should be larger than a certain criterion, while visually checking the output. There were 1304 detections. By visual inspection and by comparing with some existing earthquake catalogues, we confirmed that 1238 detections were local seismic events, and 66 were far-field events located more than 50km away from DONET1.
After applying the locating process of the envelope correlation method, only 825 local events were successfully located as local events, and 13 far-field events were wrongly located as local events. Nevertheless, many of them were eliminated due to having a large location error after being located; thus, if we count them in, there were 1093 located local events instead of 825. The results suggest two problems. First, many (i.e. 145) detections were eliminated in the locating process. Even a simple grid-search method failed to locate these events. One of the reasons is the difficulty in measuring differential travel times among stations for shallow tremors. The method assumes envelopes look similar among stations; nevertheless, shallow tremors often exhibit very different waveforms. Another reason is that the structures in the shallow tremor zones are very heterogeneous, which cannot be handled properly. Careful inspections of tremor waveforms suggest that the sharp plated boundary structure strongly influences them. One way to overcome this difficulty is to use the maximum amplitude measurements to locate the shallow events. We show that the number of located local events increased from 1093 to 1144 using amplitude measurements. The second problem is that some far-field events get mixed in as local events. This problem can also be partly be solved by using the amplitude measurements; nevertheless, we haven't yet figured out a way to completely solve the problem.
Based on these observations, we finally discuss on possible semi-automated procedure toward creating a comprehensive tremor catalogue that even enables identifications of tremor missing regions.
Here, we carefully investigated how the method performs on records of shallow tremors. The method was applied to eight days of DONET1 data in October 2015, and the results were visually inspected. In the method, first, the coherent signal arrivals are detected (referred to as "detections"), then their source locations are estimated by measuring differential travel times among stations. We first adjusted some parameters for the detection, such as the minimum number of station pairs whose correlation coefficient should be larger than a certain criterion, while visually checking the output. There were 1304 detections. By visual inspection and by comparing with some existing earthquake catalogues, we confirmed that 1238 detections were local seismic events, and 66 were far-field events located more than 50km away from DONET1.
After applying the locating process of the envelope correlation method, only 825 local events were successfully located as local events, and 13 far-field events were wrongly located as local events. Nevertheless, many of them were eliminated due to having a large location error after being located; thus, if we count them in, there were 1093 located local events instead of 825. The results suggest two problems. First, many (i.e. 145) detections were eliminated in the locating process. Even a simple grid-search method failed to locate these events. One of the reasons is the difficulty in measuring differential travel times among stations for shallow tremors. The method assumes envelopes look similar among stations; nevertheless, shallow tremors often exhibit very different waveforms. Another reason is that the structures in the shallow tremor zones are very heterogeneous, which cannot be handled properly. Careful inspections of tremor waveforms suggest that the sharp plated boundary structure strongly influences them. One way to overcome this difficulty is to use the maximum amplitude measurements to locate the shallow events. We show that the number of located local events increased from 1093 to 1144 using amplitude measurements. The second problem is that some far-field events get mixed in as local events. This problem can also be partly be solved by using the amplitude measurements; nevertheless, we haven't yet figured out a way to completely solve the problem.
Based on these observations, we finally discuss on possible semi-automated procedure toward creating a comprehensive tremor catalogue that even enables identifications of tremor missing regions.