Physica A 499 (2018) 28–39 Contents lists available at ScienceDirect Physica A journal homepage: www.elsevier.com/locate/physa Testing the structure of earthquake networks from multivariate time series of successive main shocks in Greece D. Chorozoglou a, *, D. Kugiumtzis b , E. Papadimitriou a a Department of Geophysics, School of Geology, Aristotle University of Thessaloniki, GR54124 Thessaloniki, Greece b Department of Electrical and Computer Engineering, Aristotle University of Thessaloniki, GR54124 Thessaloniki, Greece highlights • Randomizing the time series provides proper randomized networks. • Small-world index performs appropriately with time series based randomized networks. • Small-world structure tend to emerge about 100 days before strong earthquakes. • Small-world index may be useful in short term seismic hazard assessment. article info Article history: Received 6 December 2017 Received in revised form 22 January 2018 Available online 5 February 2018 Keywords: Small-world network Random network Time series Clustering coefficient Main shock abstract The seismic hazard assessment in the area of Greece is attempted by studying the earth- quake network structure, such as small-world and random. In this network, a node represents a seismic zone in the study area and a connection between two nodes is given by the correlation of the seismic activity of two zones. To investigate the network structure, and particularly the small-world property, the earthquake correlation network is compared with randomized ones. Simulations on multivariate time series of different length and number of variables show that for the construction of randomized networks the method randomizing the time series performs better than methods randomizing directly the original network connections. Based on the appropriate randomization method, the network approach is applied to time series of earthquakes that occurred between main shocks in the territory of Greece spanning the period 1999–2015. The characterization of networks on sliding time windows revealed that small-world structure emerges in the last time interval, shortly before the main shock. © 2018 Elsevier B.V. All rights reserved. 1. Introduction The seismic hazard assessment is one of the main targets of seismological research aiming to contribute in reducing the catastrophic consequences of strong earthquakes. It is usually considered as the probability for the occurrence of strong earthquakes (e.g. M ≥ 6.0) within a given space, time, and magnitude ranges. The most known model for seismic hazard assessment is the Poisson model, which assumes that the earthquake occurrences are independent in space and time. Another popular model is the Markov or hidden Markov model, introduced as a suitable means for earthquake probability estimation, assuming that consecutive events are dependent on one another in space and time. The seismic hazard assessment based on Markov model was studied by Tsapanos and Papadopoulou [1] for the Greek area and by Nava * Corresponding author. E-mail addresses: chorozod@geo.auth.gr (D. Chorozoglou), dkugiu@auth.gr (D. Kugiumtzis), ritsa@geo.auth.gr (E. Papadimitriou). https://doi.org/10.1016/j.physa.2018.01.033 0378-4371/© 2018 Elsevier B.V. All rights reserved.