# Development of a Global Spatio-Temporal Seismicity Model and Its Application to the Vrancea Seismic Zone, Romania

This study investigates the temporal behaviour of major earthquakes in the Vrancea Seismic Zone (VSZ)in Romania. I used the Romplus catalogue, which is a compilation of several sources and spans the time from 984 AD to the year 2005 and in which the data are of different quality. This catalogue contains only Vrancean earthquakes and consists of more than 8000 events. Qualities 'A', 'B' and 'C' were used to model the data. 'D' and '=' were found as too unreliable for modeling. Using the b-value, I concluded that 3.5 is the correct cut-off magnitude for earthquakes after 1980 and at depths of 60 km and greater. Thereby I detected an increase in the b-value after 1986 of about 0.2 units. The reason for this increase could not be found. Plotting the Gutenberg-Richter relation for several time and depth intervals, it was found that at larger depths than 60 km, there are too many M7 earthquakes as compared to small shocks. The shape of the Gutenberg-Richter relation is similar as to the one expected by the characteristic earthquake model (Schwarz and Coppersmith, 1984; Wesnousky, 1994). A strike of 53 degree was found and the earthquake coordinates were rotated correspondingly. The resulting view on the slab showed the confined volume in which the earthquakes happen and well as the 'aseismic part' of the slab between 40 km and 60 km of depth. The seismicity seems to reach a depth of 180 km. Only the earthquakes in the slab, below a depth of 60 km, show clustering behaviour. Furthermore, the M7 earthquakes all happened in the slab. Thus, a depth limit of 60 km was introduced for modeling. In order to find aftershocks in the catalogue, the temporal behaviour of the Vrancea earthquakes was examined. The mean magnitude increases after each major earthquake, indicating an aftershock process. This was confirmed by the rate of occurrence, which showed an increase in rate after the 1990 earthquakes. The rate of occurrence is too low for the first 580 days after 1980, possibly due to insufficient earthquake detection in this period of time. All the damaging M7 earthquakes all happened in the slab. Thus, shallow earthquakes had to be considered separately. A depth limit of 60 km was introduced and earthquake in shallower and deeper depths were considered separately. For the shallow earthquakes there was a sharp increase in the apparent b-value below the cut-off magnitude of 3.5. After reaching a value of 2.4, the b-value starts to fall steeply. This was attributed to biases in the magnitude calculation. I used the rounded value of 3.5 as a cut-off magnitude for the shallow earthquakes. Having found the magnitude cut-off, depth and time limit, modeling could be started. The model gives two important parameters: the proportion of aftershock and the time to the next earthquake. Using the Maximum Likelihood Method, a best fit was found for a data set starting at 1980 and consisting of earthquakes with a cut-off magnitude of 3.5 and a depth equal and greater than 60 km. According to the model, this data set consists of 13 plus or minus 5% aftershocks and has an inter-event time for new earthquakes of 13 plus or minus 1 days. Using several cut-off magnitudes, it was found that the calculated inter-event time for these earthquakes is consistent with the Gutenberg-Richter law. In contrast, the predicted value for the interevent time of M7 earthquakes does not match the one found in the catalogue. While the Maximum Likelihood Method leads to 814 years as recurrence time, the data shows a recurrence time of only 23 years. The model fits the data set of the 1990 aftershocks very well, too, leading to a aftershock proportion of 58 plus or minus 15%. The data set for the 1986 did not lead to good results, probably due to missing aftershocks shortly after the main shock. Comparing model and data with a pure Poisson model I could see that earthquakes tend to cluster in the first days after the major event. Several days later, their behaviour changes and then is similar to the one proposed by the seismic gap model. Looking at the ratio between the probabilities of the model of Smith and Christophersen and of the Poisson model, a clustering behaviour in the first 24 hours after the main shock was found, followed by a decreased seismicity, which reverts to be Poissonian after 100 days. Thus, I concluded that aftershock behaviour is only relevant after the first 24 hours following a major earthquake. After 24 hours, seismic hazard decreases to be less than as expected by the Poisson model in the following 100 days, until seismicity returns to be Poissonian again. Additionally, I suggest that the 1990 earthquake and its aftershocks should be considered as a 'model earthquake' for future earthquakes as it seems to be representative for earthquake behaviour in the VSZ.