akneelanjan / codefundopp_2018

Idea for 2018 codefundo++

codefundopp_2018

Idea for 2018 codefundo++

TITLE : Earthquake magnitude forecasting over the Indian subcontinent using Artificial Neural Network (ANN) methods from past seismic datasets

Source of Data: United States Geological Survey (USGS) earthquake catalog : Link: https://earthquake.usgs.gov/earthquakes/search/

Methodology and Steps involved:

Obtaining the catalog data (spatial and temporal) from USGS website.

Declustering the catalog to filter the mainshocks using method suggested by Gardner and Knopoff (1974).

Use of 7 seismic indicators as parameters-

> the frequency of foreshocks (n events) before a mainshock during the time under consideration
> Mean magnitude of the last n events
> Rate of seismic energy release
> b value obtained from Gutenberg-Richter relation 
> Difference between maximum occurred and maximum expected magnitude for an event
> Mean time difference between characteristic events
> Source depth

Division of study area in a requisite number of tiles.

Feature extraction and feature modelling out of the chosen 7 parameters

Seven seismic parameters shall be passed as an input to seven neurons and one set of input features corresponding to 1 month. In this case, two hidden layers will be, each with 12 neurons. We shall use Matlab for the training and testing. Total number of synaptic connections would be given by w=(I+1)a1 + (a1+1)a2 + (a2+1)O

Generating a hypothesis by using Artificial Neural Networks

Training this hypothesis with training dataset and optimizing the cost function.

Running the model on the same spatial location (same grid and same tilings) in a new time-stepping and with different seismic indicators to predict the probability of an earthquake to cross a certain magnitude threshold "m". This magnitude threshold is a user defined parameter.

Creating Heat maps on the basis of numerous iteration results, representing the probabilities of exceedance of a given magnitude.