Supervised Deep Learning for Detecting and Locating Passive Seismic Events Recorded with DAS: A Case Study

Abstract

Exploring shallow mineral resources requires acquiring denser seismic data, for which Distributed Acoustic Sensing is an effective enabler and relevant to mining operations monitoring. Passive seismic can be of interest in characterizing the subsurface; however, dealing with large amounts of data pushes against the limits of existing computational systems and algorithms, especially for continuous monitoring. Hence, more than ever, novel data analysis methods are needed. In this article, we investigate using synthetic seismic data, paired with real noise recordings, as part of a supervised deep-learning neural network methodology to detect and locate induced seismic sources and explore their potential use to reconstruct subsurface properties. Challenges of this methodology were identified and addressed in the context of induced seismicity applications. Data acquisition and modelling were discussed, preparation workflows were implemented, and the method was demonstrated on synthetic data and tested on relevant seismic monitoring field dataset from the Otway CO₂ injection site. Conducted tests confirmed the effects of time shifts, signal-to-noise ratios, and geometry mismatches on the performance of trained models. Those promising results showed the method's applicability and paved the way for potential application to more field data, such as seismic while drilling.

Keywords: DAS; deep learning; induced seismicity; passive seismic.

Figure 1

A high-level diagram demonstrating the application of supervised deep learning.

Figure 2

(a) A diagram of the four wells used to record induced events at the Otway Stage 3 CO₂ injection monitoring site, where green stars mark event locations. (b) Configuration schematics of the fiber-optic cable installations.

Figure 3

Upscaled 1D model of (a) P-wave and S-wave velocity and (b) density from vertical CRC-3 well. (c) An extended 2D model over 3000 m of offset with an overlay of source and receiver locations used for modelling, focusing on an area of interest around the induced event locations.

Figure 4

Three 1.5 s long strain-rate seismograms with matching geometries of (a) synthetic data for a source located at a depth of 1450 m and an offset of 720 m from CRC-3 well, (b) a strong induced event at an estimated depth of 1470 m and an offset of 720 m from CRC-3 well, and (c) a noise record from CRC-3 well.

Figure 5

A schematic demonstrating the application of the deep-learning approach for microseismic event detection, location, and subsurface property estimation using ResNet50 architecture.

Figure 6

Training and application data preparation workflow highlighting common steps in blue, steps unique to training data in purple, and steps unique to application data in green.

Figure 7

(a) Synthetic events location predictions compared to the ground truth locations used for training. (b) Reconstructed subsurface properties were overlayed on the true models.

Figure 8

Predictions of event locations for data from vertical CRC-3 well. Green stars represent the actual locations of induced events relative to the well positioned at zero offset. Stars denoting predictions of induced event locations are in red, and predicted noise locations are in blue. The magenta dots show the actual locations of synthetic data used for the training.

Figure 9

The effect of not including time shifts in the training data on (a) event location predictions with time-shift annotations, and (b) the resulting large prediction errors, as compared to the effect of including time shifts in the training data on (c) event location predictions, and (d) the resulting reasonably low prediction errors.

Figure 10

The effect of varying synthetics signal-to-noise ratios on detection.

Figure 11

Predictions of event and noise locations for data from slightly deviated (a) CRC-4 and (b) CRC-6 wells. Green stars represent the actual locations of induced events relative to the well positioned at zero offset.