Pattern recognition technique for prediction of strong earthquakes in the North Anatolian fault zone, Turkey

Abstract

Bu <;al~mada Kuzey Anadolu Fay Zonu (KAFZ) boyunca u<; bolgede Ortintu Tamma(Pattern Recognition) teknigine dayah bir algoritma (CN) kullamlarak buyuk depremlerin ortavadede (1 ay ila u<; yIl) onceden belirlenilmesi konusunda <;all~Ilml~ ve uygulamalarneticesinde elde edilen sonu<;lar sunulm~tur.KAFZ boyunca bir<;ok ar~tlflCI tarafmdan sismik bo~luk olarak nitelenen ve buy uk birdepremin olma olaslhgmm diger bolgelere nazaran daha yiiksek oldugu d~uniilen bu bolgelerse<;ilerek yontem uygul~trr. Birinci bOlge olarak Marmara bolgesi ve <;evresi se<;ilmi~tir.Bilindigi uzere bu bolge ekonomik ve yerle~im a<;lsmdan Ulkemizin en onemli yoresidir.ikinci bolge olarak KAFZ'nun orta kesimi se<;ilmi~tir. U<;uncu bOlge ise Erzincan ve Karllovamerkez olmak uzere geni~ <;evresi almml~trr.Bu <;al~mada birinci ve u<;Uncu bolgelerde yakm ge<;mi~te olm~ aletsel buyu.klugu 6.8 vedaha buyuk depremler; ikinci bOlgede ise aletsel buyuklugu 5.2 ve daha buyuk depremlerinorta vadede once den belirlenilmesine <;ah~Ilml~trr. Uygulanan algoritma sonucunda birincibolgede 6 depremden 5'i; ikinci bolgede 5 depremden 1 'i; ve u<;uncu bOlgede ise 6 depremden4' u onceden belirlenebilmi~tir. Artan Olaslhk Zamam (TIP), <;al~llan ikinci donemin (1964-1994) birinci bolge i<;in yuzde 40, ikinci bOlge i<;in yuzde 16 ve u<;Uncu bolge i<;in ise yuzde36'Slill kapsaml~trr. CN algoritmasl birinci ve u<;Uncu bolgelerde 1995 ylll i<;in alarmgostermektedir. Bu sonu<; bu bolgelerde bir deprem olma olaslhgmm arttlglill gostermektedir.Bu uygulamadan elde edilen sonu<;larm duyarhhgl kullamlan deprem kataloglarma ve<;ah~llan sureye bagh olarak degi~im gostermektedir. Bu yontem ile birinci ve u<;Uncubolgelerde geriye doniik olarak depremlerin once den belirlenebilmesi olanakh goriinm~tur.Elde edilen sonu<;lar, diger sonu<;larla uyum gostermektedir. Diger yandan bu algoritmaKAFZ'nun orta kesiminde ve Marmara bolgesinde e~ik magnitfidu 5 ve 5 ten buyukdepremlerin onceden belirlenmesinde b~anslz sonu<;lar vermi~tir. Bu sonu<;lar bolgedekaydedilen depremlerin saylca az olmasma ve deprem olu~ duzenlerinin geryek olarakbelirlenememesine baglanmaktadlr. Bu bOlge ve magnitud smlfl i<;in algoritmanm standartparametrelerle uyguianmasl imkanslzdrr. Bu amaylar iyin algoritmanm gozden geyirilip,yeniden diizenlenerek ve veri tabamm guncelle~tirerek yontemin bolgelere uygulanmaslbolgenin deprem olU§ duzenleri ve deprem riski hakkmda yararh bilgiler saglayacaktrr.Artan OlasIhk Zamam'nm duyarhhgl ileriye donlik olarak test edilmelidir. Buradageryekle~tirilen yah~ma sadece algortimanm geriye donlik olarak test edilmesiniamaylamaktadrr. ate yandan kullamlan algoritmanm bolgesellige baghhgl, onemliparametrelerin kUylik degi~imlerle kontrol edilmesi ile elde edilen veriler, yal~llan bOlge iyinduyarhhgm degi~medigini gostermektedir. C;al~manm yeni verilerle guncelle~tirilerek tumbOlgelerde (ozellikle alarm gosteren yerlerde) sfuekli uygulanmasl, KAFZ' nun orta kesimindeise mlimklinse yeni algoritmanm denenmesi onerilmektedir. Birinci bOlge iyin elde edileceksonuylarm iznik-Mekece yoresinde yurlitlilen depremlerin onceden belirleme yah~malanverileri ile kar~11~tlfllmaSl sonucu orta ve klsa vadede depremlerin onceden belirlenmesinebuyuk katkl saglayabilecektir.

A pattern recognition algorithm called the CN (California-Nevada) for intermediate- termearthquake prediction in three areas situated in the westernmost, the central, and easternmostportions of the NAFZ. One of them is the Marmara region and its vicinity (Region 1), whichcontain the western Anatolian graben complex and part of the Anatolian Trough. The NAFZsplays into three strands and extends to mainland Greece. The Marmara region and itsvicinity are the most important area of Turkey from economical and sociological points ofview. Existence ofa seismic gap in a densely populated region of Turkey, attempts havebeen made to collect data for predicting future earthquakes in this region. The second one(Region 2) is the central section of the NAFZ. Also, there is a probable seismic gap in thisarea (specifically between the longitudes of 33 °E and 35 °E). The third area is the Karliovatriple junction and surrounding area (Region 3), where three active fault zones intersect andhave produced many destructive earthquakes in the past. There are two unbroken segmentsof the NAFZ, (proposed as seismic gaps) located in the regions 1 and 3. Recently, adestructive earthquake occurred on March 13, 1992, causing about 650 deaths and at least2000 injures. The epicenter of this earthquake and its aftershocks are located around thewestern boundary of the proposed seismic gap in Region 3. The 13.3.1992 earthquake is notconsidered as the potentially greatest earthquake that can take place in Erzincan accordingto data from strain accumulation of the segment of the NAFZ to the east of Erzincan, andthe OvacIk Fault has not ruptured for more than 1000 years. These regions were selectedbecause many destructive earthquakes have occurred in these areas, which have highseismotectonic activity relative to their surroundings according to information fromtectonical and seismological studies.In the present work we investigate the possibility of the application of CN algorithm forintermediate-term prediction of earthquakes with magnitude Mo=6.8 in region 1 and 3;Mo=5.2 in Region 2 for the period of 1934-64 and 1964-1994. The first period used only fortesting and learning the seismicity patterns of the NAFZ.The CN algorithm has diagnosed the Time of Increased Probability of a strongearthquake (TIP) which preceded 5 out of 6 earthquakes with Mo=6.8 in Region 1, lout of5 strong earthquakes with 110=5.2 in Region 2 and 4 out of 6 strong earthquakes with 110=6.8in Region 3. The TIPs occupied 40% of the total time for Region 1, 27% for Region 2 and36% for Region 3 for the second period. The eN algorithm shows current alarms for regions1 and 3.The stability of the results depends on using different catalogues and time periods ofthe algorithm considered. I believe that all experiments demonstrate that the results of theTIP's diagnosis are stable for Region 1 and Region 3. The retrospective analysis of seismicdata in the Marmara region and the Karliova triple junction show that the eN algorithm canbe applied without any readaptation for these regions of Turkey. The results of TIPdiagnosis by the eN algorithm for both regions coincide well with the result of a worldwidetest of the eN algorithm. The total diagnosis time of TIPs is slightly higher than the actualones. This algorithm has given unsuccessful results for the prediction of earthquakes withmagnitude 5 or higher in the middle part ofthe NAFZ and Marmara region. These results canbe explained by incompleteness of the earthquakes recorded in the region and existence of thedifferent seismicity patterns in the central part ofthe NAFZ. Application of the algorithm withstandard parameters is not possible for this region and given magnitude range. Therefore, theeN algorithm must be reconsidered for adaptation and adjustment and database must be alsoupdated.The reliability of TIP diagnostics needs further monitoring of earthquake flow, i.e. by forwardprediction. The analysis carried out in this work should be considered only as the basis for suchtesting. The results of eN were stable when I changed the threshold value for the definition ofstrong earthquakes and made a slight change in the regionalization. I recommend that thealgorithm should be applied continuously with monthly catalogues for the purpose of monitoringexisting current alarms in the regions and for short-term prediction of strong shocks in Turkeycompared with the observation results obtained from multidisciplinary earthquake predictionstudies in iznik -Mekece area.