A geostatistical analysis of seismicity in Oklahoma using regression trees and neural networks

Abstract

ABSTRACT Induced seismicity in the mid-continental United States remains an ongoing concern. In 2016, 4,672 small-magnitude earthquakes occurred in Oklahoma, which now ranks number one for earthquake frequency in the United States. It is thought that this rise in seismicity is related to wastewater disposal (WWD) into Oklahoma’s subsurface near previously inactive faults. Here we model these earthquake frequencies using Geographic Information Systems (GIS), spatial analysis, and machine learning statistical processes. Data representing WWD sites considered Area of Interest (AOI) by the Oklahoma Geological Survey (OGS) and Oklahoma Corporation Commission (OCC) were used, as well as the 2016 Oklahoma earthquake and fault data. Euclidean distance values from each earthquake to its nearest WWD site, nearest fault, and average fluid injection rate at each nearest AOI WWD site were used to develop two non-parametric regression models using Classification and Regression Trees (CART) and Neural Networks (NN). Results show (NN R2 = 0.745, RMSE = 0.47; CART R2 = 0.62, RMSE = 0.31) that proximity to AOI WWD sites, fluid injection rates, and adjacency to subsurface faults are sufficient to model seismicity in north-central Oklahoma. These results provide further support for the need for more restrictive guidelines related to WWD.