Key Considerations in Modeling of Earthquake Risk in Turkey Fouad Bendimerad, PhD, PE Christian Mortgat, PhD Financing the Risks of Natural Disasters World Bank Washington DC June 2-3, 2003
Background Present key risk modeling consideration introduced in RMS RiskLink Turkey Earthquake Model This presentation is mainly focused on impact of hazard parameters on modeling risk around Marmara Sea Region. Illustration of variability of Loss Exceeding Probability (LEP) and Average Annual Loss (AAL) to residential exposure with respect to: Source parameters Rupture parameters Recurrence parameters The risk around the Marmara Sea is key to the insurance industry because a large proportion of the exposure is in this region, and because there is a large probability of a large earthquake hitting the region in the near future © 2003 Risk Management Solutions, Inc. Confidential
Background RMS RiskLink Turkey Earthquake model was released commercially in July 2001 Model and its applications were presented to the Turkish insurance industry during a seminar in Istanbul co-organized with the Association of Insurance and Reinsurance Companies of Turkey (TSRSB) on February 24, 2003 © 2003 Risk Management Solutions, Inc. Confidential
Analysis Resolution (Mahalle Resolution) B l a c k S e a The loss analysis is performed at the Mahalle (I.e., neighborhood) level. About 10,000 stochastic events are simulated and loss is calculated for each event. An LEP is develop from the loss to each event. AAL is calculated accordingly M a r m a r a S e a Number of Mahalles = 922 I s t a n b u l (Mahalle Resolution) B l a c k S e a I s t a n b u l © 2003 Risk Management Solutions, Inc. Confidential
Earthquake Exposure Commercial lines constitutes the largest exposure 58% of the total earthquake exposure is located around the Marmara Sea (Cresta Zones 1 to 4) ; including 40% in Cresta 1 (Istanbul). Hence, earthquake model warrants major focus on the risk around the Marmara Sea. According to our own estimate, commercial and industrial lines account for about 57% of the TSI in the country. The Government-run TCIP is about 30% of exposure; and the remaining 13% is in voluntary Personal Lines business. What also very interesting is that if one distribute the exposure by region, almost 60% of the TSI is around the Marmara Sea (a total of 4 crestas), which include Istanbul; Cresta 1, Istanbul by itself accounts for 40% of the total earthquake sum insured. Therefore capturing the risk around Marmara Sea and in Istanbul is clearly an important factor in the development of the model. This is a driving factor in the reason why we have higher analysis resolution in Istanbul. © 2003 Risk Management Solutions, Inc. Confidential
Seismic Activity In Turkey Earthquake risk in Turkey is characterized both by high severity and high frequency 1999 Izmit and Duzce earthquakes (M7.4 and 7.2, respectively) created the largest historical losses Historical Earthquakes 1850-1999 Turkey is one of the most active seismic regions in the World, probably recording more large earthquakes by unit area than any other country In the last 2000 years, Turkey experienced 42 earthquakes of M>=7 (i.e., one M>=7 every 47 years) The North Anatolia Fault Zone (NAFZ) accounts for a large proportion in the occurrence of large events 1900-1999 15 earthquakes of M>6.5 (i.e., one M>=6.5 quake every 7 years) [a rate comparable to the total number of M>=6.5 to hit California] Earthquakes in the last century caused more than 110,000 deaths and destroyed more than 600,000 housing units 1999 Izmit and Duzce earthquakes (M7.4 and 7.2, respectively) were the latest earthquakes in a cycle that started with the 1939 Erzincan Earthquake (M8.0) 1850 to 1999 M>=7.0 M>=6.5 North Anatolia Fault (Turkey) 1 Eq every 12 yrs 1 Eq every 7 yrs San Andres Fault (California) 1 Eq every 30 yrs 1 Eq every 13 yrs © 2003 Risk Management Solutions, Inc. Confidential
Earthquake Threat to Istanbul The cluster of earthquakes on the North Anatolia Fault in the last century identifies a seismic gap around Istanbul 1992 6.8 7.4 7.2 1999 1912 1967 1957 1951 7.3 1944 1943 7.3 Seismic Gap 1942 1939 7.8 Marmara Sea If one focuses on the NAFZ, we could see how earthquakes exhibit a clear pattern of successive ruptures. Starting with the M7.9 Erzincan earthquake (the largest earthquake of the century) and moving Westward. 1939 to 1942; 1943; and so on until the August 12, 1999 Izmit quake, closely followed by the Duzce earthquake in Nov of 1999. The only section of the NAFZ that has NOT ruptured in this century is the part that goes into the Marmara Sea offshore from Istanbul. There is therefore a clearly identified seismic gap in this region. This is a high significance because as mentioned in the beginning of this presentation, about 60% of the Sum Insured is located in the areas around the Marmara Sea. More than 10 million people and close to 60% of the economic value of Turkey would be impacted by an earthquake in the Marmara Sea region Max magnitude could be as high as M7.7, potentially causing major losses © 2003 Risk Management Solutions, Inc. Confidential
Historic Seismicity in the Marmara Sea Ambraseys (2002) identifies 54 earthquakes of M>6.8 taking place in the Marmara Sea Region in the last 20 centuries Earthquakes seemed to take place in sequences of clusters that repeat itself every 300 years approximately © 2003 Risk Management Solutions, Inc. Confidential
Source Modeling and Rupture Modeling Three source models are considered to take into consideration uncertainty in the NAFZ structure in the Marmara Sea “Cascade” rupturing (I.e., potential for rupture of more than one segment) is considered in the study. Probability of cascade is determined by looking at ruptures in past events © 2003 Risk Management Solutions, Inc. Confidential
Recurrence Modeling In Model (a) eq. recurrence is modeled as Poisson Model (b) and Model (c) consider characteristic events for M>6.5 Characteristic events are restricted on the fault segments M<=6.5 occur within the area source following Poisson In the Northwest Strand of the MSSZ, the rate of occurrence is estimated using four methods: Slip rate Slip rate + Time dependency (I.e., Renewal model) Same as 2. + permanent stress migration due to the 1999 earthquakes Same as 3. + transient stress migration due to the 1999 earthquakes In the Southern Strand, occurrence is based on slip rate © 2003 Risk Management Solutions, Inc. Confidential
Other Model Parameters Uniform slip rate on the NAFZ is estimated at 2.4cm/year. About 2.0 cm/year is assigned to the Northwest strand of the MSSZ Rate are calculated while preserving the energy balance in the fault The time lapsed since the last occurrence and the average recurrence time are based on historical data. Investigation of historical seismicity for the last 2000 years Maximum magnitude is adjusted to take into consideration the rupture length of “cascade” events A background source of Mmax=6.5 is added to all models to account for the possibility of an earthquake outside of the geometry of the defined sources Rate of earthquake occurrence in the background source is calculated using smoothed historical seismicity using and adaptive Gaussian Kernel technique © 2003 Risk Management Solutions, Inc. Confidential
Combining Model An event tree technique is used to combine different models and calculate an event rate © 2003 Risk Management Solutions, Inc. Confidential
Probability Results (1) and (2) are similar, but there only about 1/3rd of (6) (3) increases probabilities significantly (6) is very high for M>=7.0, but lowest for M>=7.5 Stress transfer (4) and (5) has a significant impact © 2003 Risk Management Solutions, Inc. Confidential
Loss Results Comparison of loss depends on the return period (6) does not produce the highest losses (1) produce the lowest losses “Cascade” has a significant impact on increase losses for the high return periods (5) produces the largest losses © 2003 Risk Management Solutions, Inc. Confidential
Historical Loss Reconstruction Calibration is achieved by successive approximation © 2003 Risk Management Solutions, Inc. Confidential
Model Calibration: Historical Loss Economic Loss -Residential 98 Adan Ceyhan (6.2) 99 Kocaeli (7.4) 99 Duzce (7.2) 92 Erzincan(6.8) Average © 2003 Risk Management Solutions, Inc. Confidential
Model Calibration: Scenario Analysis Istanbul Max Scenario (M7.5) D C B © 2003 Risk Management Solutions, Inc. Confidential
Model Calibration: Industry Losses Event Magnitude Calculated Observed 1999 KOCAELI 7.4 $506 $410 1999 DUZCE 7.2 $124 $125 2002 Sultandagi 6.3 TCIP $.80 Non-TCIP $1.5 Repeat 1999 Kocaeli TCIP 7.4 $400 Repeat 1939 Erzincan 7.9 TCIP $57 Non-TCIP $67 Worst Central Marmara 7.5 Case TCIP $1,100 Scenario Non-TCIP $2,000 © 2003 Risk Management Solutions, Inc. Confidential
Model Validation: Pure Premium Sub-Province Level High Medium Low Model Resolution allows to identifies significant differences in risk within the different geographical regions of Turkey © 2003 Risk Management Solutions, Inc. Confidential
Model Validation: Modeled OEP versus Scenario Events Return Period Loss $billion Equivalent Event 250 $10 M7.5 Central Marmara 100 $7.7 M7.2 Central Marmara 50 $5 M7.0 Central Marmara M6.8 Izmir 20 $2 M7.9 1939 Erzincan Model results are consistent with historical experience Model allows relationship between individual scenarios and probabilistic losses © 2003 Risk Management Solutions, Inc. Confidential
Conclusion Model takes into consideration all the plausible scientific assumptions In-depth treatment of the seismo-tectonic in the region Thorough calibration of hazard and vulnerability components Validation with respect to various benchmarks and historical experience © 2003 Risk Management Solutions, Inc. Confidential