Forecasting induced seismicity in Oklahoma using machine learning methods

Abstract

Oklahoma earthquakes in the past decade have been mostly associated with wastewater injection. Here we use a machine learning technique—the Random Forest to forecast induced seismicity rate in Oklahoma based on injection-related parameters. We split the data into training (2011.01–2015.05) and test (2015.06–2020.12) periods. The model forecasts seismicity rate during the test period based on input features, including operational parameters (injection rate and pressure), geological information (depth to basement), and modeled pore pressure and poroelastic stress. The results show overall good match with observed seismicity rate (adjusted $$R^2$$ R 2 of 0.75). The model shows that pore pressure rate and poroelastic stressing rates are the two most important features in forecasting. The absolute values of pore pressure and poroelastic stress, and the injection rate itself, are less important than the stressing rates. These findings further emphasize that temporal changes of stressing rates would lead to significant changes in seismicity rates.

Physics-informed deep learning to forecast $${\widehat{{\varvec{M}}}}_{{\varvec{m}}{\varvec{a}}{\varvec{x}}}$$ during hydraulic fracturing

“…In summary, our findings provide a proof-of-principle that a fixed-window PIDL approach could serve as a basis for an adaptive TLP system. While many studies have explored methods to forecast induced seismicity associated with industrial activity, including hydromechanical models that combine fluid pressurization and rate-and-state friction 40 or related mechanisms 41 , 42 , models to predict maximum seismic magnitude using injection data 12 , 15 , 24 , 25 and machine learning models to forecast induced seismicity rates using highly related features 43 , these models require access to injection data and/or geomechanical parameters, e.g. poroelastic stress, stress rate and rate-state friction parameters, which are typically not available at all or at least not in real-time.…”

Recent advances in earthquake seismology using machine learning

“…In addition to the important feature analysis, attempts have been made to construct ML-based models for estimating seismic potential or predicting event rates. For example, Qin et al (2022a)…”

Uncovering the factors that affect earthquake insurance uptake using supervised machine learning

“…The first is to analyze the factors that influence the decision to purchase earthquake damage insurance. The second objective is to predict which individuals are more likely to obtain earthquake insurance in Oklahoma, given the state's increased vulnerability to earthquakes from 2010 to 2020 26 , 27 . To realize these twin objectives, we use a variety of supervised machine learning techniques, encompassing logistic regression, ridge regression, and LASSO, to identify the influential factors of earthquake insurance uptake.…”

“…The state of Oklahoma is a compelling case study for this research due to the significant increase in earthquakes associated with fracking between 2010 and 2020 11 , 26 , 27 , 29 – 31 . While the state experienced less than two M3.0 earthquakes per year on average from 1978 to 2008, the Oklahoma Geological Survey 32 reported a sharp increase in earthquakes from 41 to 903 between 2010 and 2015, all which exceed M3.0 on the Richter scale.…”