Forecasting Induced Seismicity in Oklahoma Using Machine Learning Methods

Session: Mechanisms of Induced Seismicity: Pressure Diffusion, Elastic Stressing and Aseismic Slip [Poster]
Type: Poster
Date: 4/29/2020
Time: 08:00 AM
Room: Ballroom
Description: 

Many of the earthquakes in Oklahoma have been associated with wastewater injection. In our study, we use machine learning methods to forest the induced seismicity in Oklahoma from injection related parameters. First, we compute the injection-related features for each injection well, such as injection volume, injection depth, pore pressure and poroelastic stress. We then divide the study area into uniform grids and search for injection wells and earthquakes in each grid. The statistics of the injection-related features in each grid are used as inputs to predict earthquake number in different time windows using a Random Forest model. The data from 2011—2016 are used to train the model and the data from 2017—2018 are used for test. We show that our machine learning based model can forecast the seismicity rate well based on the injection history. The relative importance of injection parameters can also be derived from the model. We find that the pore pressure rate and poroelastic stress capture most of the seismicity change. The developed machine learning based model help us understand the important features related to induced seismicity and these injection-related features can be used to effectively forecast seismicity rate based on the injection history.

Presenting Author: Yan Qin

Authors