Artificial neural networks (ANN) and stochastic techniques to estimate earthquake occurrences in Northeast region of India

Abstract

The paper presents the probability of earthquake occurrences and forecasting of earthquake magnitudes size in northeast India, using four stochastic models (Gamma, Lognormal, Weibull and Log-logistic) and artificial neural networks, respectively considering updated earthquake catalogue of magnitude Mw ≥ 6.0 that occurred from year 1737 to 2015 in the study area. On the basis of past seismicity of the region, the conditional probabilities for the identified seismic source zones (12 sources) have been estimated using their best fit model and respective model parameters for various combinations of elapsed time (T) and time interval (t). It is observed that for elapsed time T=0 years, EBT and Kabaw zone shows highest conditional probability and it reaches 0.7 to 0.91 after about small time interval of 3-6 years (2014-2017; since last earthquake of Mw ≥ 6.0 occurred in the year 2011) for an earthquake magnitude Mw ≥ 6.0.Whereas, Sylhet zone shows lowest value of conditional probability among all twelve seismic source zones and it reaches 0.7 after about large time interval of 48 years (year 2045, since last event of Mw ≥ 6.0 occurred in the year 1999). While for elapsed time up to 2016 from the occurrence of the last earthquake of magnitude Mw ≥ 6.0, the MBT and MCT region shows highest conditional probability among all twelve seismic source zones and it reaches 0.88 to 0.91 after about 6-7 (2022-2023) years and in the same year (2022-2023) Sylhet zone shows lowest conditional probability and it reaches 0.14-0.17. However, we have proposed Artificial Neural Network (ANN) technique, which is based on feedforward backpropagation neural network model with single hidden layer to estimate the possible magnitude of future earthquake in the identified seismic source zones. For conditional probability of earthquake occurrence above 0.8, the neural network gives the probable magnitude of future earthquake as Mw 6.6 in Churachandpur-Mao fault (CMF) region in the years 2014 to 2017, Mw 6.8 in Myanmar Central Basin (MCB) region in the years 2013 and Kabaw region in the years 2015-2018, respectively. The epicentre of recently occurred 4 January 2016 Manipur earthquake (M 6.7), 13 April 2016 Myanmar earthquake (M 6.9) and the 24 August 2016 Myanmar earthquake (M 6.8) are located in Churachandpur-Mao fault (CMF) region Myanmar Central Basin (MCB) region and EBT and Kabaw region, respectively and that are the identified seismic source zones in the study area which show that the ANN model yields good forecasting accuracy.

Forecasting the Future Tsunami Potential of the Indo-Pacific Region Using Artificial Neural Network

“…The training is considered satisfactory when the Mean Square Error (MSE) between the Observed and Predicted Target date (M W ) becomes minimum (≤ 0.0003). Tsunamigenic zones with less than ten event data (Zones 2 and 3) are not included for training; hence, the M W of future tsunamigenic earthquakes in these zones cannot be predicted (Zarola & Sil 2017).…”

Spatiotemporal Evolution of Earthquakes in Longmenshan Fault and Adjacent Area, before and after the 2008 Wenchuan Earthquake

“…However, due to its rarity, the understanding of foreshock activity of this type of earthquake is still very shallow. Seismicity following a main earthquake carries useful information regarding the triggering mechanism of earthquakes [23][24][25][26][27][28]. In recent years, many studies have focused on the historical earthquake sequence, including the Landers earthquake [29,30], the Chi-Chi earthquake [31], the mid-Niigata Prefecture earthquake [32], the Tohoku earthquake [33], and the Rigan earthquake [34].…”