Analysis of the correlation between fractal attributes of complex seismotectonic variables may offer insights into seismic hazard assessment. The Gutenberg-Richter, moment-magnitude, and moment-source area relations yield a direct fractal relationship among the Gutenberg-Richter b-value, occurrence rate, and the characteristic linear dimension of the fault plane (square root of fault surface area). In contrast, temporal variation in the correlation dimension of epicenters (D C) is found, in several studies, to correlate negatively with the b-value in different regions of the world. Spatial variations between the b-value and D C also tend to oppose each other. In Japan, negative correlations are also observed in the regional scale comparisons of the capacity dimension (D o) of active fault systems and the b-value. However, at local scales, the relationship yields both positive and negative correlation. The occurrence of positive or negative correlation appears to be controlled by different modes of failure within the active fault complex. Spatial variations between the b-value and D C along the Northern Anatolian Fault Zone (NAFZ) suggest that, on average from 1900 to 1992, earthquake magnitudes were higher and epicenters more scattered within the central NAFZ than in its eastern and western segments. Temporal analysis reveals that the relationship between the b-value and D C are nonstationary. Temporal correlations are generally negative. A period of positive correlation is observed between 1976 and 1988. During the last 3 yr of this period (1985-1988), both the b-value and D C rose significantly, suggesting that stress release occurs through increased levels of low-magnitude and increasingly scattered seismicity. This dispersed pattern of seismicity, in combination with higher slip rates in the central NAFZ, may be one that did not adequately relieve stress along the main fault zone. This change in behavior and the tendency during the last century for the seismicity to migrate westward along the NAFZ may point to an increased risk of larger magnitude events such as the İzmit earthquake.

Summary We investigated precursory seismicity in and around the epicentral zone of the Mw= 7.4 1999 August 17 Izmit (Turkey) earthquake, by applying a statistical method—the RTL (Region–Time–Length) algorithm—to earthquake catalogues derived from that for the period 1981–1999 of Kandilli Observatory and Earthquake Research Institute (KOERI). The derived catalogues are complete for events MD≥ 3 in most of western Turkey. After declustering aftershocks, we investigated the seismicity patterns preceding the Izmit event at local (Izmit tectonic zone) and national (Turkey) scales. The RTL parameter indicates that a period of seismic quiescence started at the end of 1995 and reached a minimum in December 1996. An activation phase lasting about three months followed. The main shock in Izmit and vicinity did not occur when the seismicity returned to its background level, but occurred with a delay of nearly 2.5 yr. We present a new parameter to quantify the spatial distribution of seismic quiescence. The results from both catalogues indicate that a significant quiescence anomaly appeared in 1996 around the epicenter of the Izmit earthquake. The primary characteristics of the seismicity patterns prior to the Izmit earthquake are similar to those obtained for large events in Russia and Japan. The variations of seismicity patterns revealed by the RTL algorithm may offer better understanding of the physical nature of seismo‐tectonics and provide useful information for seismic hazard estimation. The varying characteristics of the Izmit and other events may reflect the difference between seismo‐tectonics in Turkey and in other regions such as Russia and Japan.

Summary We compare the rupture location of the Mw 7.4 Izmit earthquake to the local seismic hazard estimated by the technique of mapping local recurrence time, TL, based on the microseismicity. After correcting for a magnitude shift in 1990, the declustered earthquake catalogue, produced by the University of Istanbul for the Marmara Sea region, is homogeneous for Md ≥ 2.9 during 1983–1999. We mapped TL in the area bounded by 40°–41° latitude and 27.6°–30.5° longitude. TL is the probabilistic estimate of recurrence time, calculated from the a‐ and b‐values of the frequency–magnitude relation of the seismicity within a radius of 20 km from every point on a grid with 5 km spacing. TL varies strongly as a function of space, since a‐ and b‐values also vary strongly. In our interpretation, the 5–20 per cent of locations with the shortest recurrence times map major asperities. In the Marmara region, we mapped four anomalies of short TL, together covering about 12 per cent of the total area. They are centred near 40.25°/29.4°, 40.8°/28.3°, 40.75°/28.8° and 40.7°/29.8°. The last two of these coincide with the western end of the rupture and the epicentre location of the Izmit earthquake, respectively. Thus, we suggest that the major asperity of this rupture and a point past which it could not propagate were mapped out by the background seismicity during the years before the event as locations that produced more large microearthquakes than average, and hence showed anomalously short TL. The TL method does not contain information about when earthquakes are expected, and the absolute values of the recurrence time could be inaccurate. The method only specifies the most likely locations of main shocks. Since the method is new, it will have to be tested for many cases and in many areas before its reliability can be assessed.

No abstract