This event is over.
Artemii Novoselov: Deep Learning with Uncertainties for Onset Time Prediction
The Deep Learning has become an important tool for seismology and earthquake monitoring. We present a deep learning method that detects seismic phases and estimates the corresponding picking uncertainties.
Seismic phase picking is the process of identifying the onset time of a seismic phase in a seismogram. The arrival time of a seismic phase is one of the most important pieces of information in seismology and is used in a wide range of applications. These include locating earthquakes, determining earth structure, and identifying nuclear explosions. The detection and picking of seismic phases is a time-consuming and labor-intensive task when performed manually. The need for automation has driven the development of several algorithms and neural networks that attempt to solve this problem.
Deep Learning methods have been shown to be able to outperform conventional methods of phase picking. However, those methods usually lack an estimation of model’s uncertainty. The knowledge of the uncertainty of the phase pick helps to improve the accuracy of the location. If the arrival time is overestimated, the distance between the seismic station and the earthquake is overestimated and the estimated location of the earthquake will be further away from the seismic station than it is. If the arrival time is underestimated, the distance between the seismic station and the earthquake is underestimated and the estimated location of the earthquake will be closer to the seismic station than it is.
In this work, we are demonstrating Deep Learning phase picker that can estimate the uncertainty of the pick.
So Ozawa: Slow Slip Events Driven by Fault Valving
Pore fluids are thought to have a significant role on the sliding of faults. We study the coupled dynamics of pore fluid flow and permeability evolution on a rate-and-state fault. Linearized stability analysis shows that steady sliding is unstable even for velocity-strengthening faults due to fault valving. By using numerical simulations, we find the emergence of aseismic slip pulses for sufficiently large background fluid flow.
Jeong-Ung Woo: Microseismic Monitoring of Induced Earthquakes in Texas and Causation of Fluids in Faulting
Fluid-injection and extraction activities during the operation of hydrocarbon development sites have been associated with many induced earthquakes in the central and eastern United States, substantially increasing earthquake rates beyond the background level of seismic activity since 2011. While elevated pore pressure on faults due to fluids directly diffused through permeable reservoirs and remote triggering by poroelastic or thermoelastic effect are generally applicable mechanisms for induced earthquakes, each case study requires a meticulous examination of earthquakes, fluid injections, and their relevance in order to provide the most appropriate requirements for reducing seismic risks for both operators and regulators.
The Midland basin in Western Texas, where thousands of oil and gas wells have been operated until now, has recently experienced a rapid increase of seismic activity during 2020, presumably caused by fluid injections at nearby hydrocarbon sites. From local seismic networks operated throughout the Midland basin, earthquakes are detected and located by using a machine-learning phase picker, a source-scanning phase associator, and a locally calibrated velocity model. The majority of earthquakes in this region occur on optimally oriented faults in Precambrian basement bedrock, below injection intervals for deep wastewater injections. Focal mechanism solutions and centroid source depths of earthquakes with the best-fitting waveforms are matched with the distribution of microseismicity. In some clusters, the distribution of earthquakes has expanded along strike and dip with time, increasing seismogenic zones. Given that seismic activity is roughly correlated with fluid volume increasing with time and earthquakes are located below injection interval locations, remote triggering would be a plausible mechanism for inducing earthquakes in the Midland basin. In the geophysics seminar, I will present up-to-date observations of seismic activities in the Midland basin, especially more focused on the Stanton sequence, and describe possible causality with injection fluids from nearby wells.