AI Case Study

Harvard university seismologists developed a deep learning algorithm to predict the location of aftershocks with an AUC of .849

A group of researchers have developed a deep learning solution to predict the location of aftershocks based on static stress. The model has improved AUC from .583 to .849 and precision from 3% to 6%.

Industry

Public And Social Sector

Education And Academia

Project Overview

"We find that the learned aftershock pattern is physically interpretable: the maximum change in shear stress, the von Mises yield criterion (a scaled version of the second invariant of the deviatoric stress-change tensor) and the sum of the absolute values of the independent components of the stress-change tensor each explain more than 98 per cent of the variance in the neural-network prediction. This machine-learning-driven insight provides improved forecasts of aftershock locations and identifies physical quantities that may control earthquake triggering during the most active part of the seismic cycle."

Reported Results

"Six percent of the areas identified as high-risk did in fact experience aftershocks, up from three percent using existing methods."

Technology

"The neural networks used here are fully connected and have six hidden layers with 50 neurons each and hyperbolic tangent activation functions (13,451 weights and biases in total). The first layer corresponds to the inputs to the neural network; in this case, these inputs are the magnitudes of the six independent components of the co-seismically generated static elastic stress-change tensor calculated at the centroid of a grid cell and their negative values. "

Function

Strategy

Data Science

Background

"Aftershocks can be more destructive than the quakes they follow, making it all the more important for experts to be able to predict them. But while seismologists have methods to forecast when aftershocks will hit and how strong they will be, there is more uncertainty about how to predict where they will strike."

Benefits

Data

"The deep-learning aftershock location forecasts that we have developed are trained and tested using co-seismic slip distributions from the SRCMOD online database of finite-fault rupture models.

Neural network trained on more than 131,000 mainshock–aftershock pairs can predict the locations of aftershocks in an independent test dataset of more than 30,000 mainshock–aftershock pairs more accurately (area under curve of 0.849) than can classic Coulomb failure stress change (area under curve of 0.583)."