Prof. Dr. Burak Berk Üstündağ's research project titled "Development of a Seismic Risk Monitoring Method with Terrestrial Stress-Strain Measurements and Data Fusion Based on Deep Learning (Yersel Gerilme-Şekil Değiştirme Ölçümleri ve Derin Öğrenmeye Dayalı Veri Füzyonu ile Sismik Risk İzleme Yöntemi Geliştirilmesi)" was entitled to be supported within the scope of TUBITAK ARDEB-1001 Earthquake Research Special Call.

We congratulate Prof. Dr. Üstündağ and wish him success in the project.



Project Abstract:
Since the North Anatolian Fault and its neighboring fault systems cause relatively shallow earthquakes which increase the potential damage rate. Depending on the occurrence frequency of this kind of earthquake and redundancy problems of existing buildings, earthquakes have become the main disaster risk in Turkey. Approximate seismic risk estimations can be done at decades level according to fault sizes, displacement speeds, and the past earthquake periods. However, any method that reliably determines seismic risk variation in shorter times has not been proven yet. Electrostatic Stress Monitoring (ESM) system has first been applied for the patent in 1999 and granted for funding by ITU Scientific Projects Unit starting from 2000. Its 20 years long acquired data has just begun to enable machine learning techniques for analysis of seismic risk relations due to the requirement for a satisfactory amount of independent earthquakes having magnitudes greater than 3. Preliminary studies have shown that ESM field patterns provide at least 4 times higher probability concerning the expected value of earthquake occurrence probability characterized by the Poisson distribution. For example, when 3000 independents (short-term) M>3 earthquakes occurred within 100km range to ESM stations have been investigated for predictability in two hours time interval, it has been determined that approximate normal distribution value of 4% can be increased to 17% level in real-time and prediction rate of 50%-50% balanced data set can be increased to 68%. 

A global stress-strain relation will be achieved by the proposed regional seismic risk monitoring system. Stresses will be measured by ESM stations whereas strains achieved by InSAR images. In the method, the CORS network, ground GPSs and InSAR data will be utilized to be input data. Hereafter, the spatio-temporal geographical data layer will be created by deep learning algorithms. The existing stations will be utilized for the development of the procedure. Microseismic activities will be measured from the stations in Nilüfer while data of microseismic activities will be adopted from National Earthquake Network. The data achieved by ground GPSs will be shared with the coordinated research project. 

The total project duration is suggested as 24 months for the four work packages. In the first six months, existing measuring stations will be calibrated and integrated. Deep learning algorithms for data fusion and real-time time-series analyses will be developed between the 5th and 13th months. At the end of the first year, the coordinated project will share the processed InSAR and strain data for the evaluated pilot regions. Hereafter, model training will be performed. Using the collected data, a global-stress relation will be generated for the pilot regions. Three journal papers will be prepared from this project which focus on “data fusion using real-time ground measurements and deep learning algorithms”, “effect of distinct seismic risk indexes on time series”, and all comprehensive work to reach a regional seismic risk monitoring procedure. After the project, the project team will apply a patent for the developed procedure. In line with the 11th development plan, it will be possible to reduce loss of life and property by the proposed procedure since an earthquake will be sensed a few hours before its rupture. New jobs will be created by spreading the procedure throughout the country.