Study Of Detecting Weak Seismic Signals With Array Processing And Matched Filter Techniques

Differing from the large earthquakes with great energy releasing in a short time, small earthquakes release less energy, and may not draw sufficient attention. However, the frequent occurrence of small earthquakes may suggest a critical stress state and an increased seismic hazard locally. The detection of small earthquakes could benefit to seismic tomography and earthquakes forecasting. On the other hand, it helps to detect the micro-fractures during shale gas exploitation and identify the relationship between fluid-injection and induced seismicity. Due to the low amplitude of small seismic signals and the strong background level, we need to apply some techniques to increase the signal to noise ratio (SNR) and detect the low-amplitude seismic signals accordingly.Here we introduce two techniques for low-amplitude seismic signals detection. The first is the array processing technique, which stacks the waveforms across a dense seismic array to enhance the coherent weak signals. The second is the matched filter technique (MFT), which uses known catalog events as templates to scan through continuous waveforms and identify events with high similarities.In order to test the feasibility of weak signals detection with array processing technique, we analyze seismic data from ChinArrray (Phase I), which is a dense portable broadband seismic array deployed in Yunnan Province and its vicinity from 2011-2013. The weak signal from the 15 February Chelyabinsk (Russian) meteor was contaminated on the ChinArray recordings by a Mw5.8 Tonga earthquake occurred-20 minutes earlier, which provides us an interesting challenge to evaluate the detection capability of weak signals on ChinArray. We apply the array processing techniques of vespagram and F-K analysis, and identify the surface wave from the Russian meteor event, with back azimuth of 329.7° and slowness of 34.73 sec/deg. We also conduct the similar F-K analysis for the dense broadband array F-net in Japan, and compute the back azimuth of surface wave from the Russian meteor event to be 316.61°. Then we locate the meteor event at 58.80°N,61.41°E, where the two ray paths defined by the back azimuths of ChinArray and F-net overlap. It has a large mislocation (-438 km) compared with the meteor location by USGS (55.15°N,61.41°E), mainly due to the bending propagation path of surface waves. Nevertheless, our results suggest that with appropriate array processing techniques, ChinArray could be used for weak signals detection.In the second study, we apply the MFT to conduct a detection of micro-earthquakes in Southern Tibet around the 2015 Mw7.8 Gorkha (Nepal) earthquake. With the continuous waveforms 1.6 days before and 4.4 days after the Nepal mainshock, we identify five times more earthquakes than listed on the China Earthquake Network Center (CENC) catalog after applying the MFT. The seismicity in Southern Tibet increases significantly immediately following the Nepal mainshock, including the Mw5.8 Tingri earthquake and the Mw5.3 Nyalam earthquake within a few hours after the mainshock. The two Southern Tibet earthquakes are normal faulting earthquakes, consistent with extensional deformation within the Tibet plateau. The static Coulomb stress changes have a slightly better correlation with seismicity rate changes than the dynamic stress changes, while the absolute value of dynamic stress changes at the site of the Mw5.8 Tingri earthquake (-2.2 MPa) is two orders larger than the static Coulomb stress changes (-10 kPa). With the MFT, more earthquakes could be detected, which contributes to better understand the triggering mechanism of large earthquakes.Besides near-field triggering, the MFT could be used to detect the remotely triggered seismicity. With the seismicity analysis on CENC catalog and spectral analysis following the 2012 Mw8.6 Indian Ocean earthquake, we find a lack of widespread triggering in Yunnan during the Indian Ocean mainshock. With the MFT, we conduct a detection of microseismic activity in Tengchong volcanic region 1.7 days before and 3.3 days after the mainshock. With 17 earthquakes recorded on the CENC catalog as templates, we detect 12 more earthquakes. The seismicity also shows no significant change after the detection, suggesting that there is a lack of instantaneous triggering in Tengchong within 3.3 days following the Indian Ocean mainshock. Comparing with the clear dynamic triggering in Yunnan following the 2004 Mw9.1 Sumatra earthquake, we suggest that the different dynamic stress changes or different background rates may contribute to the different behaviors of dynamic triggering.In summary, based on the similarity of waveforms, the array processing technique and the matched filter technique enhance the seismic signals and suppress noises by stacking or scanning the waveforms with correlation signals. With the two techniques, low-amplitude seismic signals could be identified effectively, and they could be used to better understand earthquake interactions at nearby and long-range distances.