Accurate automatic Phase picking for continuous, noisy microseismic recordings without prefiltering.
Abstract
Automated picking of low signal-to-noise ratio (SNR) seismic events remains a challenge among existing detection algorithms. While most algorithms can pick large events (M>4) with relatively high accuracy (within a few sample points), detection and picking for small events (M<2) remains problematic. Previous research (Kinali, 2019) involved testing a number of suggested approaches from the international literature. Results were excellent for microseismic events with clear impulsive arrivals and larger earthquakes. Yet, where microseismicity is involved, it is usually the events that are neither impulsive nor visible in the time domain that are of interest. For such events, existing methodologies do not perform well.
We present an automated algorithm which detects and picks the first arrivals of microseismic events without the requirement for pre-filtering of the data. Working in the frequency domain, events are detected by calculating the energy contained within small individual time segments of a recording and comparing it with the energy contained within a longer pre-existing time window. If the excess energy exceeds a given threshold criterion, which is determined dynamically based on the background noise of the data, then an event is detected. The picking accuracy lies within 10 samples. For example, the picking accuracy is 0.04 seconds for a sampling rate of 250Hz but becomes better than 0.01 sec for sampling rates of 1000Hz or more. The advantage of this algorithm is that its accuracy only depends on the sampling rate rather than the characteristics of a specific data set. Therefore, it is suitable for a number of applications that involve microseismic recordings. We have already successfully applied our methodology on available microseismic data sets and have evidence that our algorithm can perform satisfactorily even with data of low SNR. The implications on fully taking advantage of the potential of microseismic monitoring can be significant as this approach allows for the detection and picking of events that are not always visible or distinguishable in the time domain. Therefore, it minimizes the need of manual verification while at the same time provides information on microseismicity which under other circumstances could have remained difficult, if not impossible, to locate in time and space.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.S43D0683M
- Keywords:
-
- 0555 Neural networks;
- fuzzy logic;
- machine learning;
- COMPUTATIONAL GEOPHYSICS;
- 1910 Data assimilation;
- integration and fusion;
- INFORMATICS;
- 1914 Data mining;
- INFORMATICS;
- 1942 Machine learning;
- INFORMATICS