Microseismic monitoring is a method for analyzing the geometry of cracks in the earth’s crust. Finding the precise location of the crack source is one of the crucial jobs

What kind of data is used?

The data can be described as follows: sensors positioned on the ground detect P and S waves coming from the crack’s source. The image below illustrates one potential location for sensors on the ground.

An example of a potential location for sensors on the ground

A microseismic event (the crack) underground can be described, for instance, by the speed at which waves travel to the surface. The source of the crack in the following illustration is at point (1,1,1)

Plot describing the travel time of the S-wave from the crack source to the surface

As a result, we may get the following information for each crack source.

Features

• (X, Y) coordinates of the sensors
• Arrival time for P and S waves (the authors use the difference between the time at each sensor and the time of the first registration)
• The variation in P and S waves registration time for each sensor

Target variables

• (X,Y,Z) coordinates of the source of the crack

The data is supplied to train the random forest model in this format. A random forest is essentially a collection of random trees that are trained on a random, repeating subsample of the original data and then used together to perform predictions.

Experiments

Experiments are carried out on artificial and real-world data.

Artificial data has already been mentioned (see two previous pictures). The experiments’ major focus is on determining if combining P and S waves is advantageous or if using just one kind is adequate. Below is a visualization of the experiment’s findings using artificial data.