Earthquake Probability Assessment for the Indian Subcontinent Using Deep Learning
Center for Advanced Modeling and Geospatial Information Systems (CAMGIS), Faculty of Engineering and Information Technology, University of Technology Sydney, Sydney 2007, Australia
Department of Energy and Mineral Resources Engineering, Sejong University, Choongmu-gwan, 209, Neungdong-ro Gwangin-gu, Seoul 05006, Korea
Department of Geology and Geophysics, College of Science, King Saud University, P.O. Box 2455, Riyadh 11451, Saudi Arabia
Division of Science Education, Kangwon National University, Kangwondaehak-gil, Chuncheon-si, Gangwon-do 24341, Korea
Authors to whom correspondence should be addressed.
Sensors 2020, 20(16), 4369; https://doi.org/10.3390/s20164369
Received: 22 May 2020 / Revised: 24 July 2020 / Accepted: 1 August 2020 / Published: 5 August 2020
(This article belongs to the Special Issue Advances in Sensors and Intelligent Techniques for Natural Hazard Modeling and Management)
Earthquake prediction is a popular topic among earth scientists; however, this task is challenging and exhibits uncertainty therefore, probability assessment is indispensable in the current period. During the last decades, the volume of seismic data has increased exponentially, adding scalability issues to probability assessment models. Several machine learning methods, such as deep learning, have been applied to large-scale images, video, and text processing; however, they have been rarely utilized in earthquake probability assessment. Therefore, the present research leveraged advances in deep learning techniques to generate scalable earthquake probability mapping. To achieve this objective, this research used a convolutional neural network (CNN). Nine indicators, namely, proximity to faults, fault density, lithology with an amplification factor value, slope angle, elevation, magnitude density, epicenter density, distance from the epicenter, and peak ground acceleration (PGA) density, served as inputs. Meanwhile, 0 and 1 were used as outputs corresponding to non-earthquake and earthquake parameters, respectively. The proposed classification model was tested at the country level on datasets gathered to update the probability map for the Indian subcontinent using statistical measures, such as overall accuracy (OA), F1 score, recall, and precision. The OA values of the model based on the training and testing datasets were 96% and 92%, respectively. The proposed model also achieved precision, recall, and F1 score values of 0.88, 0.99, and 0.93, respectively, for the positive (earthquake) class based on the testing dataset. The model predicted two classes and observed very-high (712,375 km2) and high probability (591,240.5 km2) areas consisting of 19.8% and 16.43% of the abovementioned zones, respectively. Results indicated that the proposed model is superior to the traditional methods for earthquake probability assessment in terms of accuracy. Aside from facilitating the prediction of the pixel values for probability assessment, the proposed model can also help urban-planners and disaster managers make appropriate decisions regarding future plans and earthquake management.