Predicting ground shaking intensities using DYFI data and estimating event terms to identify induced earthquakes

Home / Artificial Intelligence & ML / Predicting ground shaking intensities using DYFI data and estimating event terms to identify induced earthquakes

Predicting ground shaking intensities using DYFI data and estimating event terms to identify induced earthquakes

There has been a dramatic increase in seismicity in CEUS in recent years (Ellsworth 2013). There is a possibility that this increased seismicity in CEUS is caused by anthropogenic processes and is referred to as induced or triggered seismicity. The earthquakes are a nuisance for people and some larger magnitude earthquakes have also caused structural damage. Hence, it is important to quantify seismic hazard and risk from this increased seismicity. One of the major components in determining seismic hazard and risk is the expected level of ground shaking at a site. Level of ground shaking from a given earthquake is typically estimated using previously collected ground motion data in a region. However, in CEUS due to historically low seismicity and sparse seismic network, there is not enough ground motion data available to constrain the prediction models. In this study, we use DYFI data which is more widely available to develop an intensity prediction model and to assess if intensities from induced events differ from natural events. Since DYFI data is recorded on a continuous scale, regression models are best suited for prediction. Assessing the difference in predictions from different earthquakes can be achieved by using a random effects model. Additionally, another random effect for regions can also be calculated to assess how intensities vary across regions. Thus, the intensity level can be predicted from a combination of fixed effects which are a function of earthquake magnitude, depth and earthquake-to-site distance, and random effects which are computed for each earthquake and each region. Hence, a mixed effects model is utilized as the primary model to achieve the objectives of this study. The mixed-effects regression model is also compared to a support vector regression (SVR) model. This is primarily done to assess the effectiveness of mixed-effects regression compared to other popular regression models. However, it should be noted that SVR does not have the functionality to assess random effects. Hence the second component of this study to evaluate whether intensities differ for induced events could not be achieved using the SVR model.

 

Research Paper Link: Download Paper

Related Post

Leave a Reply