Turkey and adjacent area seismicity forecasts from earthquake inter-event time mean ratio statistics

Abstract

The alarm-based forecasting model for earthquakes called moment ratio (MR) is retrospectively tested on Turkey and adjacent area seismicity. This model uses the ratio of the mean inter-event time over the variance as a precursory alarm function to forecast future earthquakes in a given region. In a former study, the MR model was successfully tested in forecasting large earthquakes with magnitude M\u2009≥\u20097, occurred in Japan. In this study, it is tested on Turkey and adjacent area seismicity using lower magnitude thresholds, namely by learning from M\u2009≥\u20095 events, to forecast earthquakes with magnitude M\u2009≥\u20096. For this purpose, a composite earthquake data file is compiled using Kandilli Observatory and Earthquake Research Institute Regional Earthquake and the Tsunami Monitoring Center (KOERI-RETMC) provided catalogs, for the period 1900–2016, and the SHARE European Earthquake Catalog (SHEEC) for the historical period 1000–1899. In this catalog, earthquakes are listed using surface magnitude scale Ms. The time periods used in training and testing are selected by taking into consideration the completeness of the magnitude. Finally, Molchan error diagrams are used to evaluate the forecasting performance of the MR method in practice using a retrospective test. Obtained results are presented as standard MR forecasting maps showing the overall forecasts and optimal maps showing high alarm areas with minimal miss and alarm rates. In addition, the relative intensity (RI) forecasting method is applied to compare different results. Results show MR forecasts outscoring random guessing with good performance compared to RI forecasts. The forecasting maps point to a small high alarm area situated along the Hellenic arc subduction zone east of Crete Island.